[ { "id": "1dbd3e38-1c81-4cf1-94eb-d1d24f9a2d49", "case_id": 10181, "language": "global", "system_prompt": "", "question": "A high-speed stamping production line at an automotive parts factory uses a PROFINET industrial communication system. The main equipment includes: one Siemens S7-1500 controller; eight KUKA robots (controlling die open/close and part handling); and four Keyence vision inspection cameras (real-time measurement of stamped part dimensions). These devices are connected via three Hirschmann industrial switches into a distributed network, using standard unshielded Category 5 Ethernet cables, with a maximum cable run of about 80 meters. Three 200-kilowatt stamping presses are situated around the line, generating strong electromagnetic interference (EMI) during operation.\nRecently, the line has encountered issues: when the presses run at high speed (60 strokes per minute), robot response slows from the original 5 milliseconds to over 25 milliseconds, and the vision cameras lose inspection data 3 to 5 times per hour (causing defects to go undetected). However, when the presses run at low speed (30 strokes per minute), communications are normal. The factory mandates no large-scale retrofit and no shutdown; the problem must be solved at minimal cost on the existing hardware, with communication response stabilized within 10 milliseconds and a data loss rate not exceeding 0.01%.\nPlease analyze root causes and provide a step-by-step diagnostic workflow and concrete optimization plan from four core perspectives: PROFINET protocol characteristics, industrial network topology/connection methods, EMI mitigation design, and hardware configuration parameters. Include detailed operational points for segmenting switches by area via VLAN-based network partitioning, rectifying cabling, and adjusting communication protocol parameters.", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "Accurately identify that the protocol issue is operation in Real Time (RT) rather than the more interference-resilient Isochronous Real Time (IRT), resulting in data errors and retransmissions.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "Accurately point out that the network connection problem is shared network bandwidth contention among devices plus unshielded Cat5 cabling and long runs up to 80 meters.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "The diagnostic workflow includes checking switch error counters under high-speed (60 strokes/min) and low-speed (30 strokes/min) conditions to distinguish fault types.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 4, "rubric_detail": "The diagnostic workflow includes checking the distance between Ethernet cables and press power lines and whether cables run close to the machine body.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 5, "rubric_detail": "The diagnostic workflow includes verifying the robots’ and cameras’ update cycle/interval and priority/QoS settings.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 6, "rubric_detail": "In zero-cost optimization, device parameter adjustments include setting the robots’ update cycle to 2 milliseconds, the cameras to 8 milliseconds, and transmitting only inspection results.", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 7, "rubric_detail": "Low-cost remediation includes replacing the 2–3 segments (20–50 meters) near the presses with shielded Ethernet cables, using metal-shell connectors, and bonding the shield to ground with ground resistance ≤ 4 ohms.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 8, "rubric_detail": "The optimization plan clearly prioritizes zero-cost configuration tuning first, followed by low-cost physical remediation.", "rubric_weight": 4, "rubric_tag": "Instructions Following" }, { "rubric_number": 9, "rubric_detail": "Lack of bulleting or subheadings, causing diagnostic steps and parameter settings to be mixed together with poor readability.", "rubric_weight": -4, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 10, "rubric_detail": "Proposes large-scale shutdown retrofit solutions (such as rewiring the entire network or replacing switches/controllers).", "rubric_weight": -10, "rubric_tag": "Instructions Following" }, { "rubric_number": 11, "rubric_detail": "The diagnostic workflow is missing the step to distinguish fault types by comparing high-speed versus low-speed operating conditions.", "rubric_weight": -4, "rubric_tag": "Factual Information" }, { "rubric_number": 12, "rubric_detail": "The answer contains extensive popular-science background or generic definitions of the PROFINET protocol, leading to severe redundancy.", "rubric_weight": -4, "rubric_tag": "Structure and Formatting" } ] }, { "id": "6c34d48e-cc4e-4193-be63-a1094f541fe8", "case_id": 10298, "language": "global", "system_prompt": "", "question": "In order to tap into new business growth areas, a provincial operator plans to carry out a trial deployment of Rel-18 5G-Advanced NR-DC technology in the live network. This trial will use the FR2 high-frequency band as the primary coverage band, with the core objective of fully supporting MR-DC scenarios. In combination with the planned introduction of high-definition VR/AR and high-throughput data services, several key technical requirements are proposed as follows:\nFirst, in FR2-band UE mobility scenarios, the interruption latency must be controlled within 10 ms, and the signaling overhead must be reduced by more than 40% compared with the Rel-17 version. This is the fundamental prerequisite for assuring the service experience of low-latency services such as high-definition VR/AR. Second, the propagation characteristics of the FR2 band result in relatively frequent SCG changes. Preliminary statistics on the live network show that, on average, one change occurs every 30 seconds, and technical optimization is required to avoid additional latency incurred by frequent network reconfiguration. Third, in conditional handover (CHO) scenarios, it is not sufficient to focus only on the channel quality of a single cell; it is necessary to ensure the stability of both the PCell and the PSCell, with the ultimate objective of achieving a UE throughput gain of more than 25%. Finally, the UE side must support the storage of at least 8 CHO-related configurations and must be fully compatible with the Rel-17 CPAC/CPC procedures, so as to ensure smooth access for existing UEs and avoid additional modification costs on the UE side.\nBased on the standardization achievements of 3GPP Rel-18 in mobility enhancement, and in view of the actual requirements of the current live network, the following key technical issues need to be addressed:\n1. For the core requirement of low-latency mobility in the FR2 band, which technical solution should be selected as the main approach? A detailed description is required of the key content of the UE pre-configuration and the methods for acquiring Early TA—here both reliability and low latency must be considered. In addition, what are the key fields in the handover command? If the UE has already obtained a valid TA value through autonomous measurement, which handover type should be used in this case? How should the grant for the first uplink transmission be selected? Finally, it is necessary to analyze the fundamental reasons why this solution can reduce latency compared with the traditional Rel-17 handover.\n2. For scenarios with frequent SCG changes, how should the optimization solution be designed? First, the technical option must be clearly defined, followed by the core execution procedure and configuration update mechanism. A specific scenario needs to be considered here: if the UE first accesses candidate PSCell1 via the CPC procedure, and within 10 seconds candidate PSCell2 meets the access conditions, how should the UE perform the access? Is it necessary for the network side to re-deliver configuration? At the same time, the update procedure of the SN Counter in the security mechanism needs to be elaborated in detail.\n3. To achieve the throughput improvement target in CHO scenarios, what configuration structure should be adopted for CHO? Which network element determines the execution conditions for the PCell and for the PSCell, respectively? Which triggering events are supported for each? Suppose there is a candidate PCell that meets the execution conditions, but among its three associated candidate PSCells only PSCell3 meets the condEventA4 threshold requirement, while the other two do not. How should the UE proceed in this case? If, at the same time, the conditions of the traditional Rel-17 CHO are also met, which handover type should ultimately be triggered? What key information needs to be reported to the network?\n4. The live network requires UEs to support storage of 8 CHO-related configurations. Suppose a given UE currently stores configurations consisting of 3 CHO configurations with candidate SCG, 2 Rel-17 CHO configurations, and 3 CPA configurations. What is the storage compatibility mechanism for these different types of configurations in the UE? If, in one CHO configuration with candidate SCG, only the PCell meets the execution conditions while the PSCell does not, how should the UE handle this? In addition, if the network side needs to add one new CPC configuration, how should this be done without exceeding the UE’s storage limit?\n5. The trial network is planned to evolve subsequently to support inter-CU LTM handover functionality. Please design a smooth evolution方案 from Rel-18 to Rel-19, focusing on clearly identifying the key enhancement points at three levels: interfaces, measurements, and functions. At the same time, a differentiated scheduling mechanism on the network side for Rel-18 and Rel-19 UEs needs to be formulated, with detailed explanation from three dimensions: LTM trigger mechanisms, measurement configuration, and handover procedures.", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Weakly time-sensitive", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The model enumerates the Early TA acquisition methods as primarily UE autonomous measurement, with PDCCH order–triggered RACH procedure as a fallback.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 2, "rubric_detail": "The model analyzes the core reasons why the LTM solution reduces latency, namely that it bypasses L3 measurements and RRC signaling interaction, and does not require execution of the RACH procedure.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "The model explains the core execution procedure of SCPAC: after accessing PSCell1, other candidate configurations are not released; when conditions are met, the UE directly accesses PSCell2 without requiring network-side reconfiguration.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "The model describes the SN Counter update procedure, including selecting the SK-counter from a list and reporting it via the MN RRC Reconfiguration Complete message.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 5, "rubric_detail": "The model states that each configuration consists of a combination of “candidate PCell + 1 candidate PSCell”.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 6, "rubric_detail": "For the low-latency mobility requirement in the FR2 band, the model explicitly identifies LTM (L1/L2 Triggered Mobility) as the core technical option.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 7, "rubric_detail": "For scenarios with frequent SCG changes, the model indicates that SCPAC (Subsequent CPAC) technology should be adopted.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "The model clearly states that the PCell execution conditions are determined by the source MN, while the PSCell execution conditions are determined by the target MN.", "rubric_weight": 4, "rubric_tag": "Factual Information" }, { "rubric_number": 9, "rubric_detail": "The model explicitly states that the UE should execute other stored Rel-17 configurations.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "The model analyzes the priority when multiple conditions are satisfied and indicates that when a CHO with candidate SCG and a traditional Rel-17 CHO are both satisfied, the CHO with candidate SCG should be triggered with higher priority.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "The model differentiates the LTM trigger mechanisms of Rel-18 and Rel-19 UEs, stating that Rel-19 supports inter-CU triggers, while Rel-18 supports only intra-CU triggers.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "The model lists three enhancement points for the evolution from Rel-18 to Rel-19: interface enhancement (inter-CU interaction), measurement enhancement (event trigger/CSI-RS), and functional enhancement (conditional LTM).", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 13, "rubric_detail": "The answer contains a large amount of redundant information unrelated to the specific technical solutions, such as 5G background and Rel-18 standard history, resulting in excessive length.", "rubric_weight": -5, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 14, "rubric_detail": "The answer is not organized in a logical order, for example mixing the Rel-19 evolution方案 with the FR2 low-latency方案, making it difficult to read.", "rubric_weight": -5, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 15, "rubric_detail": "The model does not list the key fields in the handover command (such as TCI State ID).", "rubric_weight": -5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 16, "rubric_detail": "When describing technical parameters, the model does not use accurate professional terminology (such as DG, CG, TCI state, etc.) and instead uses non-technical, colloquial expressions.", "rubric_weight": -5, "rubric_tag": "Structure and Formatting" } ] }, { "id": "2b4583f1-f332-43c6-999e-9166e6d2c6e5", "case_id": 10308, "language": "global", "system_prompt": "", "question": "Title: Review the Fatal Bug in This LLaMA W8A8 Quantized Operator\n\nProblem Description:\nI am an architect responsible for inference engine optimization. Recently, in order to run LLaMA-7B on an edge device (NVIDIA Orin), I asked an intern to hand-write a custom W8A8 (INT8 weights, INT8 activations) matrix multiplication (GEMM) CUDA kernel.\n\nThe goal was to replace cuBLAS and reduce binary size with an extremely minimal implementation. However, the submitted code produces completely garbled outputs (perplexity explodes), and it is even slower than FP16.\n\nBelow is the intern’s quantization description and the core CUDA code snippet (simplified):\n\n1. Quantization Scheme:\nBoth weights and activations use Per-Tensor Symmetric Quantization.\nFormula: Q = clip(round(X / scale), -127, 127), where scale = max(abs(X)) / 127.\n\n2. CUDA Kernel (C++) Snippet:\n```cpp\n__global__ void w8a8_matmul_kernel(const int8_t* A, const int8_t* B, float* C,\n float scale_a, float scale_b, int N, int K) {\n // A: M x K (Row Major)\n // B: K x N (Column Major, transposed to optimize reads)\n // C: M x N\n\n int row = blockIdx.y * blockDim.y + threadIdx.y;\n int col = blockIdx.x * blockDim.x + threadIdx.x;\n\n if (row < N && col < N) {\n // accumulator\n int8_t sum = 0;\n\n for (int k = 0; k < K; ++k) {\n // naive dot product\n int8_t a_val = A[row * K + k];\n int8_t b_val = B[col * K + k]; // B is column-major, so we do this\n\n // multiply-accumulate\n sum += a_val * b_val;\n }\n\n // dequantize and write to global memory\n C[row * N + col] = (float)sum * scale_a * scale_b;\n }\n}\n```\n\nAs the Tech Lead, identify at least three fatal errors in the above approach that cause either catastrophic accuracy degradation or poor performance. Explain, from first principles, why these design choices are fundamentally flawed, and then provide the correct remediation direction tailored to the architectural characteristics of LLaMA.\n\nRequirements:\n- Do not provide generic code optimization advice (e.g., \"add comments\"); focus only on hard-core mathematical computation and CUDA hardware mechanisms.\n- You must clearly explain why the intern’s quantization strategy does not work for a model like LLaMA.\n- You must point out the severe mathematical fallacy in the code regarding data type handling.", "tags": { "topics": [ "Industry", "Machine Learning", "Machine Learning" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The answer should compute an arithmetic intensity of approximately 1 Op/Byte (or 2 Ops / 2 Bytes), or explicitly state that the kernel falls into the memory-bound region of the Roofline model and is limited by global memory bandwidth.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "In the optimization suggestions, explicitly proposes using the Ampere-specific asynchronous copy instruction `cp.async` to hide global memory load latency.", "rubric_weight": 10, "rubric_tag": "Factual Information" }, { "rubric_number": 3, "rubric_detail": "In the optimization suggestions, explicitly proposes using padding or swizzling techniques to prevent shared-memory bank conflicts after introducing shared-memory tiling.", "rubric_weight": 10, "rubric_tag": "Factual Information" }, { "rubric_number": 4, "rubric_detail": "Recommends double buffering or multi-stage pipelining strategies to maximize instruction-level parallelism.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "Recommends using vectorized load instructions (LDS.128 / float4 / int4) or `ld.global.nc` (non-coherent loads) to improve bandwidth utilization.", "rubric_weight": 10, "rubric_tag": "Factual Information" }, { "rubric_number": 6, "rubric_detail": "Explicitly points out that using an `int8_t` accumulator will cause integer overflow, and provides a concrete fix recommendation to change the accumulator to `int32_t` (or int32).", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 7, "rubric_detail": "Points out that LLaMA activations have outlier characteristics, causing per-tensor symmetric quantization to fail (catastrophic accuracy loss).", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "Explicitly recommends using the `ldmatrix` instruction to efficiently load data from shared memory into registers to match Tensor Core computation.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 9, "rubric_detail": "Points out that the access pattern for matrix B causes non-coalesced memory access, resulting in low bandwidth utilization.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 10, "rubric_detail": "Explicitly recommends leveraging NVIDIA Orin (Ampere) 2:4 structured sparsity to further boost W8A8 inference throughput.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 11, "rubric_detail": "Recommends using shared-memory tiling for blocked optimization.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 12, "rubric_detail": "The proposed remediation includes well-structured code blocks (typically wrapped in Markdown) containing complete loop structures or kernel signatures, rather than only scattered one-line patch suggestions.", "rubric_weight": 5, "rubric_tag": "Instructions Following" }, { "rubric_number": 13, "rubric_detail": "The answer should indicate that the formula must reflect element-wise multiplication, and clearly state that the activation scale corresponds to the row/token index, while the weight scale corresponds to the column/channel index.", "rubric_weight": 10, "rubric_tag": "Instructions Following" }, { "rubric_number": 14, "rubric_detail": "The answer cites non-existent CUDA functions (e.g., `__int8_mul`, `cudaQuantize`) or fabricates API libraries.", "rubric_weight": -8, "rubric_tag": "Factual Information" }, { "rubric_number": 15, "rubric_detail": "Uses fabricated academic citations (e.g., [[1]], [[5]]) or provides invalid reference links.", "rubric_weight": -5, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 16, "rubric_detail": "The answer incorrectly recommends using float16 (half) as the accumulator type.", "rubric_weight": -10, "rubric_tag": "Factual Information" }, { "rubric_number": 17, "rubric_detail": "The remediation incorrectly recommends asymmetric quantization or introducing a zero point, ignoring that this adds extra instruction overhead in Tensor Core computation and severely reduces inference throughput.", "rubric_weight": -5, "rubric_tag": "Factual Information" } ] }, { "id": "364cfadc-3764-4ffd-a2c2-2688ae800950", "case_id": 10447, "language": "global", "system_prompt": "", "question": "I am a data development engineer and have recently been responsible for maintaining a user data pipeline. In this pipeline, user registration data is written to MongoDB (the `users` collection). The `created_at` field is provided by the frontend in the format \"2026-01-06 10:30:00\" (without any timezone information), and the backend converts it into a Date object using `new Date(str)` before storing it.\n\nThe backend server runs in the UTC-5 timezone. Every day at 06:00 UTC, an ETL job automatically synchronizes data from MongoDB into the Impala `dim_users` table. The partition field is defined as `dt = DATE(created_at)`, and users with `status = 'deleted'` are excluded from synchronization. User behavior events are written into the `user_events` index in Elasticsearch, where the `timestamp` field stores a Unix timestamp in milliseconds, and the `user_id` field stores `MongoDB _id.toString()`.\n\nNow, the product manager reports that the data for January 6 does not match across systems: BI (Impala) shows 1,247 users, MongoDB shows 1,302 users, and Elasticsearch shows 1,180 users.\n\nI have conducted some preliminary analysis and found the following facts: about 90% of users are in the UTC-5 timezone and are active between 09:00 and 22:00 local time; registrations between 00:00 and 05:00 account for approximately 3% of the total; about 8% of newly registered users have no behavior events on the same day; MongoDB `_id` is of type ObjectId; and on January 6, 23 users were soft-deleted.\n\nCan you help me write a discrepancy analysis report for the product manager? I would like you to explain the reasons for the data discrepancies across different databases separately, and also point out any other potential issues that may still exist. The report should also include remediation plans, with the requirement that after the fixes, queries for the same day return consistent results across all three data sources. Do not ask follow-up questions; answer directly.", "tags": { "topics": [ "Industry", "Databases and Data Engineering", "Databases and Data Engineering" ], "time_sensitivity": { "time_sensitivity": "Weakly time-sensitive", "year_month": "2026-01", "day": "6" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The report explicitly states that the number of soft-deleted users is 23.", "rubric_weight": 7, "rubric_tag": "Factual Information" }, { "rubric_number": 2, "rubric_detail": "Based on MongoDB’s total user count of 1,302, the answer correctly calculates that approximately 8% of users (about 104 users) have no behavior events.", "rubric_weight": 7, "rubric_tag": "Factual Information" }, { "rubric_number": 3, "rubric_detail": "The model explains the timezone misalignment mechanism: users who register after 19:00 in the UTC-5 timezone will have their UTC time fall into the next day, resulting in incorrect partitioning.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "The model identifies that one of the causes of the discrepancy between MongoDB and Impala is that the ETL logic filters out users with `status = 'deleted'`.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "The model provides a potential technical cause for the remaining discrepancy in Elasticsearch, using the mismatch between MongoDB ObjectId and string representations in Elasticsearch as an example.", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "The model provides a fix at the Impala/SQL layer, using timezone conversion functions (such as `from_utc_timestamp`).", "rubric_weight": 5, "rubric_tag": "Instructions Following" }, { "rubric_number": 7, "rubric_detail": "The model provides a corrected MongoDB query strategy by adjusting the query time range to the corresponding UTC window (e.g., from 05:00 to 05:00 the next day).", "rubric_weight": 7, "rubric_tag": "Instructions Following" }, { "rubric_number": 8, "rubric_detail": "The model provides a corrected Elasticsearch query strategy and explicitly points out the need to specify the `time_zone` parameter in the query.", "rubric_weight": 7, "rubric_tag": "Instructions Following" }, { "rubric_number": 9, "rubric_detail": "The model suggests modifying the ETL logic to no longer directly filter out deleted users, but instead retain deletion flags to align statistical definitions across systems.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "The answer fails to calculate that the total discrepancy between MongoDB and Elasticsearch is 122 users.", "rubric_weight": -5, "rubric_tag": "Factual Information" }, { "rubric_number": 11, "rubric_detail": "The report does not explicitly calculate that the number of users affected by timezone misalignment is 32.", "rubric_weight": -5, "rubric_tag": "Factual Information" }, { "rubric_number": 12, "rubric_detail": "The answer does not calculate that the remaining discrepancy in Elasticsearch is approximately 18 users.", "rubric_weight": -5, "rubric_tag": "Factual Information" }, { "rubric_number": 13, "rubric_detail": "The model points out that converting ObjectId to string may introduce format inconsistency issues.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 14, "rubric_detail": "The answer contains chain-of-thought content, a conversational opening, or other redundant descriptions unrelated to the final report format.", "rubric_weight": -3, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 15, "rubric_detail": "The answer explains the same concept more than three times.", "rubric_weight": -3, "rubric_tag": "Structure and Formatting" } ] }, { "id": "deabcd1e-64ae-4659-82ac-646b517cdfc2", "case_id": 10631, "language": "global", "system_prompt": "", "question": "Given a non-empty integer array `nums`, where each element can be any integer, and the array may contain one or multiple elements, fully decompose and regroup every digit of every number in the array, and concatenate them to form the smallest possible integer, subject to all the following rules:\n\n[1] Every digit of every number in the array must be used exactly once—no more and no less. For example, -123 is decomposed into [-1, 2, 3], 0 is decomposed into [0], and 44 is decomposed into [4, 4].\n\n[2] For the decomposition of each number, the negative sign of a negative number can only be placed immediately before the most significant digit. For example, -78 can only be decomposed into [-7, 8].\n\n[3] The concatenated integer must not have a leading zero. For example, the result cannot be 012. If all digits are 0, the final result is 0.\n\n[4] The concatenated integer must be valid. If it cannot be formed according to the requirements, return the reason directly. What constitutes an impossible case is left for you to determine.\n\nIf it can be formed, return the smallest integer found as a string.\n\nNote: Optimize for the best possible complexity.\n\nRequirements:\nFirst, describe your approach in natural language, covering all cases, and provide the time and space complexity.\nThen, provide a Python code example that follows coding conventions and uses paragraph comments only.\nFinally, provide test cases that fully cover all types of scenarios.", "tags": { "topics": [ "Industry", "Backend Development", "Backend Development" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The code section uses line-level comments (# comments), violating the requirement that only paragraph comments are allowed.", "rubric_weight": -10, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 2, "rubric_detail": "The answer is organized in the order of: approach description, Python code, and test cases.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 3, "rubric_detail": "In the test cases, the input and the corresponding output of each case are completely correct: any number of zeros should output 0; two or more negative numbers should output an \"impossible\" message; for exactly one negative number, the negative sign and its most significant digit must not be separated, and the output should be a negative number with the maximum absolute value; for all-positive inputs, the output should be the smallest possible positive number. Also ensure that the number of digits in the output matches the number of digits in the input (except for the all-zero array).", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "The answer does not provide test cases that fully cover all scenario types. At minimum, it should include: an impossible case, a case with no negatives, a case with exactly one negative, and a large-number stress test.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "The approach discussion mentions ultra-large-number scenarios, including how to handle potential overflow if converting to a string is extremely long (in Python, integers do not overflow, so no special handling is required).", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "The code contains reasoning errors, such as producing a number that is not the minimum under the problem requirements (example: for [100, 20], outputting 10020 is not minimal; 10002 is the minimal result).", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "The stated time complexity is O(n), where n is the number of elements or the total number of characters/digits.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 8, "rubric_detail": "The stated space complexity is O(1) (excluding storage for the output).", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 9, "rubric_detail": "Explicitly states that the output for input 0 is 0.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 10, "rubric_detail": "For cases where the result is positive, proposes a concatenation strategy that sorts all digits in ascending order and places the smallest non-zero digit first.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "For cases where the result is negative, proposes a strategy that places the minus sign first and sorts the remaining digits in descending order to maximize the absolute value.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "The code does not comply with Python indentation rules.", "rubric_weight": -5, "rubric_tag": "Structure and Formatting" } ] }, { "id": "c68fc03a-3fdc-4cca-824f-f14db3730027", "case_id": 10704, "language": "global", "system_prompt": "", "question": "I am a graduate student in the Department of Road and Bridge Engineering, currently participating in professional practice to design a concrete mix proportion for a project. While learning concrete mix design methods, my supervisor explained the specific surface area method. However, preliminary tests indicated issues with the mix proportion designed using this method; the state of the prepared concrete was relatively poor (flowability, water retention, and cohesiveness did not meet construction requirements). Preliminary analysis suggests that this method is too outdated to adapt to the rapid development of concrete technology. Please help me analyze the specific reasons for the failure of this method and the resulting significant errors. What errors will occur? What are the reasons?\n\nMy supervisor believes the reasons for the method's failure may lie in changes across the following three aspects:\n1. Differences in the components and content of concrete in different periods;\n2. Continuous expansion of concrete application fields and advancements in construction technology;\n3. Improvements in concrete workability, mechanical properties, and durability.\n\nPlease answer two questions based on the hints provided by the supervisor:\n1. What errors will the old method produce (specifically referring to the mix proportion and the state of the initially mixed concrete)?\n2. The underlying reasons (in conjunction with the supervisor's hints).\n\n regarding the logic of the answer, please first explain the specific surface area method and its principles, then explain the errors, followed by the reasons, and finally recommend other effective methods for designing concrete mix proportions.", "tags": { "topics": [ "Industry", "Civil Engineering", "Civil Engineering" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The answer explicitly identifies the specific surface area values of modern cement and explains that cement fineness has changed, increasing from less than 300 m²/kg in the past to the current 330~350 m²/kg (for 42.5 cement) or exceeding 380 m²/kg (for 52.5 cement).", "rubric_weight": 4, "rubric_tag": "Factual Information" }, { "rubric_number": 2, "rubric_detail": "Mentions the specific upgrade direction of coarse aggregate crushing technology, noting the significant transformation from jaw crushers to impact crushers or hammer crushers.", "rubric_weight": 3, "rubric_tag": "Factual Information" }, { "rubric_number": 3, "rubric_detail": "Lists mainstream changes in concrete strength grades, noting the widespread use of high-grade cement and the shift from C30 and below in the past to C40~C80 currently.", "rubric_weight": 3, "rubric_tag": "Factual Information" }, { "rubric_number": 4, "rubric_detail": "Points out that the application of polycarboxylate superplasticizers allows the water-binder ratio to be reduced to below 0.3, emphasizing the significant role of water-reducing agents.", "rubric_weight": 3, "rubric_tag": "Factual Information" }, { "rubric_number": 5, "rubric_detail": "When explaining the reasons, mentions changes in the mineral composition proportions of cement clinker, specifically the increase in tricalcium aluminate and tricalcium silicate content, and the relative decrease in tetracalcium aluminoferrite and dicalcium silicate content.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "Explains the basic principle of the specific surface area method (calculating the amount of cement paste required to coat the aggregates based on their total surface area) according to the logical requirements set in the prompt.", "rubric_weight": 5, "rubric_tag": "Instructions Following" }, { "rubric_number": 7, "rubric_detail": "Points out that one of the direct errors in mix calculation is the inaccurate estimation of the sand ratio (sand percentage), which affects the state and strength of the initially mixed concrete. It also notes that the traditional inverse relationship between sand ratio and strength often fails in high-performance concrete (i.e., high sand ratio does not necessarily imply low strength).", "rubric_weight": 7, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "Points out that in some modern engineering practices, within the low water-binder ratio range (e.g., 0.38~0.45), the traditional water-cement ratio law weakens, and there is a lack of significant linear correlation between strength and the water-cement ratio.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 9, "rubric_detail": "When describing errors in the state of initially mixed concrete, points out specific manifestations of poor workability, such as bleeding, segregation, poor flowability, or poor cohesiveness.", "rubric_weight": 7, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "When analyzing the reasons for failure, points out that the introduction of mineral admixtures (such as fly ash and slag) has expanded the concept of \"cement\" to \"total cementitious materials.\" The complex influencing factors of admixtures cause calculations based on pure cement to fail.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "Points out that changes in aggregate morphology (especially fine aggregate) are a significant cause of error; specifically, modern engineering uses smaller particle size stones with better quality features, but manufactured sand tends to be more angular and porous, resulting in higher water absorption compared to natural river sand.", "rubric_weight": 9, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "Analyzes that the fineness of modern cement is too high (ground increasingly fine to pursue high early strength), leading to excessively fast early hydration, which weakens the potential for long-term strength growth; this is one of the reasons for the failure of predictions using the specific surface area method.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 13, "rubric_detail": "Points out that the application fields of modern concrete engineering are continuously expanding with a pursuit of multi-objective high quality (taller, larger, and more durable structures). Consequently, performance requirements have increased, shifting the mix design philosophy from a singular \"strength-oriented\" approach to a \"durability-oriented\" approach focusing on indicators like crack resistance and impermeability.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 14, "rubric_detail": "Provides more scientifically accurate mix design methods at the end of the answer, such as the paste-to-aggregate ratio method, absolute volume or weight method, or performance-based design methods (rheology, durability, etc.).", "rubric_weight": 6, "rubric_tag": "Instructions Following" }, { "rubric_number": 15, "rubric_detail": "The answer strictly follows the content sequence required by the prompt: \"Explain the principle of the specific surface area method -> Explain errors (direct calculation errors and initial state errors) -> Analyze reasons (combining hints) -> Provide new methods (effective methods).\"", "rubric_weight": 5, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 16, "rubric_detail": "The answer contains excessive lengthy preparatory material regarding the history of concrete development or general textbook definitions with low relevance to the core questions (errors and reasons), or there is significant overlap between different parts of the answer, causing serious redundancy.", "rubric_weight": -3, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 17, "rubric_detail": "The answer is not bulleted or lacks clear subheadings, causing the core examination points of error manifestations and root causes to be mixed together, resulting in a chaotic logical structure, and failing to meet the requirement to explain errors and reasons separately.", "rubric_weight": -3, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 18, "rubric_detail": "The specific surface area method determines cement usage based on aggregate surface area, whereas the water-cement ratio law controls strength based on the water-cement ratio. If the model presents 'the lower the water-cement ratio, the higher the strength' as the core principle of the specific surface area method, it counts as a confusion of concepts.", "rubric_weight": -6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 19, "rubric_detail": "If the model answers issues related to construction operations such as \"inaccurate metering,\" \"failure to adjust water content,\" or \"uneven mixing,\" it counts as confusing \"construction errors\" with \"theoretical errors.\"", "rubric_weight": -6, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "0d785b81-fc88-438c-964b-e4c8256f9f42", "case_id": 10724, "language": "global", "system_prompt": "", "question": "I am a graduate student at Beijing Institute of Technology. My current research focuses on solid electrolytes for sodium-ion batteries, mainly the sodium superionic solid electrolyte NZSP. The core issues to address are low ionic conductivity and the wettability between the electrolyte and the metal electrode. I intend to design an experimental scheme. My initial idea is to replace NHSP with NZSP for surface modification experiments based on the following reference; however, due to laboratory limitations, I cannot use screen printing. Our lab only has ball milling and heating equipment.\nMy questions are: 1) Please analyze the feasibility of this substitution and the relevant principles; 2) Design a reasonable scheme based on the available equipment, using solid-state sintering to prepare NZSP, and include the necessary parameters.\nReference: NiO powder and NHSP powder were ball-milled with ethanol as the dispersant in a mass ratio of 3:7 for 24 h. The mixture was blown dry and ground into powder, then the mixed powder was combined with the organic carrier in a mass ratio of 3:7; the organic carrier consisted of terpineol and ethyl cellulose at a mass ratio of 9:1. The mixed slurry was stirred on a magnetic stirrer for 48 h. The mixed slurry was coated onto the NHSP electrolyte surface by screen printing, and then the electrolyte was sintered at high temperature at 1200 °C for 3 h with a heating rate of 3 °C/min.", "tags": { "topics": [ "Industry", "Chemical Engineering and Materials", "Chemical Engineering and Materials" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "Both NZSP and NHSP electrolytes belong to the NASICON-type (sodium superionic conductor) crystal structure.", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 2, "rubric_detail": "The microstructure of NZSP consists of PO4/SiO4 tetrahedra and ZrO6 octahedra sharing oxygen vertices, while NHSP consists of PO4/SiO4 tetrahedra and HfO6 octahedra sharing oxygen vertices.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 3, "rubric_detail": "Using ZrO2 instead of HfO2 as the raw material can significantly reduce experimental costs.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "In raw material stoichiometry, the sodium source (e.g., Na2CO3) should be provided with a 0.1 (10%) molar excess.", "rubric_weight": 10, "rubric_tag": "Factual Information" }, { "rubric_number": 5, "rubric_detail": "A binder such as PVB must be added prior to pressing to ensure the powder compacts without becoming friable or dispersing; after pressing, binder removal (debinding/binder burnout) is required, typically at 650 °C for 3 h.", "rubric_weight": 7, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "During high-temperature sintering, employ a packing powder (sacrificial powder bed; NZSP precursor powder) to prevent sodium volatilization that can deform the electrolyte or alter its composition.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "Interface modification primarily relies on NiO: NiO acts as a hydrophilic phase and, after the reduction reaction (NiO + Na → Ni + Na2O), converts to an electronic conductor, thereby improving wetting between sodium and the electrolyte and reducing interfacial resistance. The organic carrier mainly serves as an intermediate layer to achieve good contact.", "rubric_weight": 4, "rubric_tag": "Factual Information" }, { "rubric_number": 8, "rubric_detail": "The response contains content unrelated to the requested experimental scheme and principle analysis, such as unrequested testing/measurement steps.", "rubric_weight": -3, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 9, "rubric_detail": "The experimental scheme lacks clear stepwise separation, mixing the NZSP bulk preparation process with the surface modification process, resulting in unclear logic.", "rubric_weight": -2, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 10, "rubric_detail": "The response fails to mention adding ethanol or similar solvents to the organic carrier to reduce system viscosity.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "Specify the temperatures and hold times for removing terpineol and ethyl cellulose: Terpineol, 217 °C (no hold); Ethyl cellulose, 350–500 °C for 1 h.", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 12, "rubric_detail": "The model’s NZSP preparation steps are incorrect, failing to properly include key solid-state steps such as mixing, pre-calcination, pressing, debinding (binder burnout), and high-temperature sintering.", "rubric_weight": -5, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "8276aec4-036f-459b-8770-2f15f693edd2", "case_id": 10767, "language": "global", "system_prompt": "", "question": "Graph Neural Networks (GNNs) are a class of machine learning methods that take graph structure into account. One of the most classic algorithms is the Graph Convolutional Network (GCN). This GitHub repository (https://github.com/iDC-NEU/iDC-MlSys_interview) contains an unoptimized GCN implementation written in C++. I am currently training it in a CPU–GPU heterogeneous environment, and I need you to perform performance optimizations by reviewing the code in that repository. Please note that you only need to propose optimization ideas; you do NOT need to write concrete code.\n\nYour tasks:\n1. Step 1: Provide general optimizations that are useful for both CPU and GPU.\n2. Step 2: Provide CPU-side optimizations tailored to Intel CPU characteristics.\n3. Step 3: Provide performance optimizations tailored to NVIDIA A-series GPUs.\n4. Step 4: Provide optimizations for distributed settings.\n\nPlease follow these instructions:\n1. Reiterating: you only need to propose optimization ideas; you do NOT need to write concrete code.\n2. You must optimize under all four scenarios above.\n3. Do not attempt to use mature external machine learning libraries; you may only use solutions achievable with basic built-in libraries.\n4. After the optimizations, provide a theoretical performance analysis.\n\nThe core code is as follows; you may access the webpage to obtain the full code.\n\n#include \n#include \n#include \n#include \n#include \n#include \n#include \n#include \n#include \n#include \n\nusing namespace std;\n\ntypedef std::chrono::time_point TimePoint;\n\nint v_num = 0;\nint e_num = 0;\nint F0 = 0, F1 = 0, F2 = 0;\n\nvector> edge_index;\nvector> edge_val;\nvector degree;\nvector raw_graph;\n\nfloat *X0, *W1, *W2, *X1, *X1_inter, *X2, *X2_inter;\n\nvoid readGraph(char *fname)\n{\n ifstream infile(fname);\n\n int source;\n int end;\n\n infile >> v_num >> e_num;\n\n // raw_graph.resize(e_num * 2);\n\n while (!infile.eof())\n {\n infile >> source >> end;\n if (infile.peek() == EOF)\n break;\n raw_graph.push_back(source);\n raw_graph.push_back(end);\n }\n}\n\nvoid raw_graph_to_AdjacencyList()\n{\n\n int src;\n int dst;\n\n edge_index.resize(v_num);\n edge_val.resize(v_num);\n degree.resize(v_num, 0);\n\n for (int i = 0; i < raw_graph.size() / 2; i++)\n {\n src = raw_graph[2*i];\n dst = raw_graph[2*i + 1];\n edge_index[dst].push_back(src);\n degree[src]++;\n }\n}\n\nvoid edgeNormalization()\n{\n for (int i = 0; i < v_num; i++)\n {\n for (int j = 0; j < edge_index[i].size(); j++)\n {\n float val = 1 / sqrt(degree[i]) / sqrt(degree[edge_index[i][j]]);\n edge_val[i].push_back(val);\n }\n }\n}\n\nvoid readFloat(char *fname, float *&dst, int num)\n{\n dst = (float *)malloc(num * sizeof(float));\n FILE *fp = fopen(fname, \"rb\");\n fread(dst, num * sizeof(float), 1, fp);\n fclose(fp);\n}\n\nvoid initFloat(float *&dst, int num)\n{\n dst = (float *)malloc(num * sizeof(float));\n memset(dst, 0, num * sizeof(float));\n}\n\nvoid XW(int in_dim, int out_dim, float *in_X, float *out_X, float *W)\n{\n float(*tmp_in_X)[in_dim] = (float(*)[in_dim])in_X;\n float(*tmp_out_X)[out_dim] = (float(*)[out_dim])out_X;\n float(*tmp_W)[out_dim] = (float(*)[out_dim])W;\n\n for (int i = 0; i < v_num; i++)\n {\n for (int j = 0; j < out_dim; j++)\n {\n for (int k = 0; k < in_dim; k++)\n {\n tmp_out_X[i][j] += tmp_in_X[i][k] * tmp_W[k][j];\n }\n }\n }\n}\n\nvoid AX(int dim, float *in_X, float *out_X)\n{\n float(*tmp_in_X)[dim] = (float(*)[dim])in_X;\n float(*tmp_out_X)[dim] = (float(*)[dim])out_X;\n\n for (int i = 0; i < v_num; i++)\n {\n vector &nlist = edge_index[i];\n for (int j = 0; j < nlist.size(); j++)\n {\n int nbr = nlist[j];\n for (int k = 0; k < dim; k++)\n {\n tmp_out_X[i][k] += tmp_in_X[nbr][k] * edge_val[i][j];\n }\n }\n }\n}\n\nvoid ReLU(int dim, float *X)\n{\n for (int i = 0; i < v_num * dim; i++)\n if (X[i] < 0)\n X[i] = 0;\n}\n\nvoid LogSoftmax(int dim, float *X)\n{\n float(*tmp_X)[dim] = (float(*)[dim])X;\n\n for (int i = 0; i < v_num; i++)\n {\n float max = tmp_X[i][0];\n for (int j = 1; j < dim; j++)\n {\n if (tmp_X[i][j] > max)\n max = tmp_X[i][j];\n }\n\n float sum = 0;\n for (int j = 0; j < dim; j++)\n {\n sum += exp(tmp_X[i][j] - max);\n }\n sum = log(sum);\n\n for (int j = 0; j < dim; j++)\n {\n tmp_X[i][j] = tmp_X[i][j] - max - sum;\n }\n }\n}\n\nfloat MaxRowSum(float *X, int dim)\n{\n float(*tmp_X)[dim] = (float(*)[dim])X;\n float max = -__FLT_MAX__;\n\n for (int i = 0; i < v_num; i++)\n {\n float sum = 0;\n for (int j = 0; j < dim; j++)\n {\n sum += tmp_X[i][j];\n }\n if (sum > max)\n max = sum;\n }\n return max;\n}\n\nvoid freeFloats()\n{\n free(X0);\n free(W1);\n free(W2);\n free(X1);\n free(X2);\n free(X1_inter);\n free(X2_inter);\n}\n\nvoid somePreprocessing()\n{\n //The graph will be transformed into adjacency list ,you can use other data structure such as CSR\n raw_graph_to_AdjacencyList();\n}\n\nint main(int argc, char **argv)\n{\n // Do NOT count the time of reading files, malloc, and memset\n F0 = atoi(argv[1]);\n F1 = atoi(argv[2]);\n F2 = atoi(argv[3]);\n\n readGraph(argv[4]);\n readFloat(argv[5], X0, v_num * F0);\n readFloat(argv[6], W1, F0 * F1);\n readFloat(argv[7], W2, F1 * F2);\n\n initFloat(X1, v_num * F1);\n initFloat(X1_inter, v_num * F1);\n initFloat(X2, v_num * F2);\n initFloat(X2_inter, v_num * F2);\n\n // Time point at the start of the computation\n TimePoint start = chrono::steady_clock::now();\n\n // Preprocessing time should be included\n\n TimePoint prepross_start = chrono::steady_clock::now();\n somePreprocessing();\n TimePoint prepross_end = chrono::steady_clock::now();\n chrono::duration prepross_ = prepross_end - prepross_start;\n double prepross_time = prepross_.count() * 1e3;\n printf(\"prepross_time: %.8lf\\n\", prepross_time);\n\n TimePoint edgeNorm_start = chrono::steady_clock::now();\n edgeNormalization();\n TimePoint edgeNorm_end = chrono::steady_clock::now();\n chrono::duration edgeNorm_ = edgeNorm_end - edgeNorm_start;\n double edgeNorm_time = edgeNorm_.count() * 1e3;\n printf(\"edgeNorm_time: %.8lf\\n\", edgeNorm_time);\n\n\n // printf(\"Layer1 XW\\n\");\n TimePoint XW1_start = chrono::steady_clock::now();\n XW(F0, F1, X0, X1_inter, W1);\n TimePoint XW1_end = chrono::steady_clock::now();\n chrono::duration XW1_ = XW1_end - XW1_start;\n double XW1_time = XW1_.count() * 1e3;\n printf(\"XW1_time: %.8lf\\n\", XW1_time);\n\n\n\n // printf(\"Layer1 AX\\n\");\n TimePoint AX1_start = chrono::steady_clock::now();\n AX(F1, X1_inter, X1);\n TimePoint AX1_end = chrono::steady_clock::now();\n chrono::duration AX1_ = AX1_end - AX1_start;\n double AX1_time = AX1_.count() * 1e3;\n printf(\"AX1_time: %.8lf\\n\", AX1_time);\n\n // printf(\"Layer1 ReLU\\n\");\n TimePoint ReLU_start = chrono::steady_clock::now();\n ReLU(F1, X1);\n TimePoint ReLU_end = chrono::steady_clock::now();\n chrono::duration ReLU_ = ReLU_end - ReLU_start;\n double ReLU_time = ReLU_.count() * 1e3;\n printf(\"ReLU_time: %.8lf\\n\", ReLU_time);\n\n // printf(\"Layer2 XW\\n\");\n TimePoint XW2_start = chrono::steady_clock::now();\n XW(F1, F2, X1, X2_inter, W2);\n TimePoint XW2_end = chrono::steady_clock::now();\n chrono::duration XW2_ = XW2_end - XW2_start;\n double XW2_time = XW2_.count() * 1e3;\n printf(\"XW2_time: %.8lf\\n\", XW2_time);\n\n // printf(\"Layer2 AX\\n\");\n TimePoint AX2_start = chrono::steady_clock::now();\n AX(F2, X2_inter, X2);\n TimePoint AX2_end = chrono::steady_clock::now();\n chrono::duration AX2_ = AX2_end - AX2_start;\n double AX2_time = AX2_.count() * 1e3;\n printf(\"AX2_time: %.8lf\\n\", AX2_time);\n\n // printf(\"Layer2 LogSoftmax\\n\");\n TimePoint LogSoftmax_start = chrono::steady_clock::now();\n LogSoftmax(F2, X2);\n TimePoint LogSoftmax_end = chrono::steady_clock::now();\n chrono::duration LogSoftmax_ = LogSoftmax_end - LogSoftmax_start;\n double LogSoftmax_time = LogSoftmax_.count() * 1e3;\n printf(\"LogSoftmax_time: %.8lf\\n\", LogSoftmax_time);\n\n // You need to compute the max row sum for result verification\n TimePoint max_sum_start = chrono::steady_clock::now();\n float max_sum = MaxRowSum(X2, F2);\n TimePoint max_sum_end = chrono::steady_clock::now();\n chrono::duration max_sum_ = max_sum_end - max_sum_start;\n double max_sum_time = max_sum_.count() * 1e3;\n printf(\"max_sum_time: %.8lf\\n\", max_sum_time);\n\n // Time point at the end of the computation\n TimePoint end = chrono::steady_clock::now();\n chrono::duration l_durationSec = end - start;\n double l_timeMs = l_durationSec.count() * 1e3;\n\n // Finally, the max row sum and the computing time\n // should be print to the terminal in the following format\n printf(\"%.8f\\n\", max_sum);\n printf(\"total time: %.8lf\\n\\n\", l_timeMs);\n\n // Remember to free your allocated memory\n freeFloats();\n}\n\nDo not ask follow-up questions; answer directly.", "tags": { "topics": [ "Industry", "Machine Learning", "Machine Learning" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The model suggests converting the graph storage structure from an adjacency list or adjacency matrix to CSR (Compressed Sparse Row) or CSC format.", "rubric_weight": 4, "rubric_tag": "Factual Information" }, { "rubric_number": 2, "rubric_detail": "The model analyzes two benefits of switching to CSR/CSC: (1) compressed graph storage reduces the memory footprint of the graph structure; (2) using offsets and indices ensures neighbors are accessed contiguously, and since neighbor access is the most common operation in graph computation, this yields higher cache hit rates.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "The model points out that atomicity issues must be considered when modifying data structures during preprocessing to prevent write conflicts.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "The model proposes changing the loop order of CPU-side matrix multiplication from ijk to ikj.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 5, "rubric_detail": "The model explains that the loop reordering is intended to improve cache line hit rates and thus increase computational efficiency.", "rubric_weight": 2, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "The model explicitly states that Intel CPUs should use the AVX instruction set for vectorized parallel computation.", "rubric_weight": 4, "rubric_tag": "Factual Information" }, { "rubric_number": 7, "rubric_detail": "The model suggests using the OpenMP library to implement multi-threaded parallel processing on the CPU side.", "rubric_weight": 4, "rubric_tag": "Factual Information" }, { "rubric_number": 8, "rubric_detail": "The model analyzes how to set the OpenMP thread count and recommends a conservative choice (e.g., half of the total available threads) to avoid resource contention.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "The model proposes operator fusion, such as fusing matrix multiplication with ReLU/Softmax activations.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 10, "rubric_detail": "The model points out that operator fusion reduces intermediate storage needs and decreases the number of passes over data.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "For NVIDIA GPU optimization, the model proposes implementing the logic with CUDA.", "rubric_weight": 8, "rubric_tag": "Instructions Following" }, { "rubric_number": 12, "rubric_detail": "In the GPU optimization design, for the aggregation stage, the model suggests assigning each thread to handle one feature dimension and using atomic operations for accumulation updates.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 13, "rubric_detail": "In distributed optimization, the model mentions the need for graph partitioning.", "rubric_weight": 3, "rubric_tag": "Factual Information" }, { "rubric_number": 14, "rubric_detail": "The model points out that in distributed settings, one should balance communication and computation, using communication to replace redundant computation where appropriate.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 15, "rubric_detail": "The model strictly follows the negative constraint of \"no need to write concrete code\".", "rubric_weight": 4, "rubric_tag": "Instructions Following" }, { "rubric_number": 16, "rubric_detail": "The response contains a large amount of popular-science text about GCN principles or basic definitions of graph neural networks, causing severe redundancy.", "rubric_weight": -5, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 17, "rubric_detail": "The response includes specific C++ or CUDA code blocks, violating the instruction to provide only ideas.", "rubric_weight": -5, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 18, "rubric_detail": "The model must include theoretical performance analysis, such as time/space complexity, memory bandwidth bottlenecks (Roofline model), computational intensity, or distributed communication overhead as key evaluation metrics.", "rubric_weight": 4, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 19, "rubric_detail": "The model calls mature external machine learning libraries such as PyTorch or TensorFlow, violating the instruction requirements.", "rubric_weight": -5, "rubric_tag": "Instructions Following" }, { "rubric_number": 20, "rubric_detail": "The model’s response omits one or more of the four required optimization steps (1. general optimizations; 2. Intel CPU optimizations; 3. NVIDIA GPU optimizations; 4. distributed optimizations).", "rubric_weight": -7, "rubric_tag": "Instructions Following" }, { "rubric_number": 21, "rubric_detail": "The model recognizes that for sparse computations in graph workloads, GPU performance optimization in the sparse aggregation stage (AX) should use CUDA cores; using tensor cores would lead to lower SM utilization.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 22, "rubric_detail": "The model should point out that for graph partitioning optimization in distributed settings, load balancing should primarily consider the edge distribution (because edges better reflect compute volume), and recommend partitioning methods such as minimum edge cut.", "rubric_weight": 7, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 23, "rubric_detail": "For CPU optimization, the model should suggest using AVX `stream_ps` to bypass caches and access memory directly, reducing memory access time by 50%, because GCN computations are one-pass in nature.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 24, "rubric_detail": "The model does not propose performance optimizations but instead proposes accuracy optimizations.", "rubric_weight": -5, "rubric_tag": "Factual Information" } ] }, { "id": "d6357459-83e7-4bb6-8cbd-d2ff97c46a2f", "case_id": 10838, "language": "global", "system_prompt": "", "question": "For a residential project in Sichuan Province, the above-ground portion consists of Class II high-rise residential buildings, and the basement is a single underground level, with the first basement level (B1) serving as a motor-vehicle garage and equipment rooms. Where the basement motor-vehicle garage is divided into multiple fire units within the same fire compartment, how should the number of evacuation doors provided in the fire partition walls between these units be determined, and how should their egress direction be defined? According to the Technical Standard for Distributed Charging Facilities for Electric Vehicles (GB/T 51313-2018), Clause 6.1.5(4): \"When intercommunicating doors need to be provided in fire partition walls, fire doors with a fire resistance rating of not less than Class B shall be used,\" which explicitly addresses interconnecting doors. However, because a single fire compartment may be subdivided into several fire units and some fire units do not contain evacuation stairwells, it is necessary to provide connecting doors in the fire partition walls of a fire unit leading to adjacent fire units in order to reach safety exits. In this situation, how should the evacuation doors on the fire partition walls of the fire units be configured, and how should the egress direction be determined? Please provide an answer with detailed reasoning.", "tags": { "topics": [ "Industry", "Architectural Design", "Architectural Design" ], "time_sensitivity": { "time_sensitivity": "Weakly time-sensitive", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "Cite Item 4 of Article 7.1.1 of the \"Points for Technical Review of Fire Protection Design of Housing Construction Projects in Sichuan Province\" (2025 Edition) as the primary basis:\n\"The area of each charging vehicle fire unit in the basement shall not exceed 1000 m². Each fire unit shall be separated from other fire units by fire partition walls with a fire resistance rating of not less than 2.00h or fire shutters (limited to vehicle passages). Each fire unit shall be provided with not less than 2 safety exits; safety exits may be fire doors opening into adjacent fire units. When intercommunicating doors need to be opened on fire partition walls, fire doors of not less than Class A shall be used.\"", "rubric_weight": 10, "rubric_tag": "Factual Information" }, { "rubric_number": 2, "rubric_detail": "Cite the regulations regarding spacing in Article 5.5.2 of the \"Code for Fire Protection Design of Buildings\" GB 50016-2014 (2018 Edition):\n\"Safety exits and evacuation doors within a building shall be dispersed. Furthermore, the horizontal distance between the nearest edges of any two adjacent safety exits in each fire compartment or each floor of a fire compartment, as well as the horizontal distance between two adjacent evacuation doors in each residential unit or each room, shall not be less than 5m.\"", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 3, "rubric_detail": "Cite the relevant regulations in Article 5.5.21 of the \"Code for Fire Protection Design of Buildings\" GB 50016-2014 (2018 Edition):\n\"For public buildings other than theaters, cinemas, auditoriums, and gymnasiums, the respective total net width of room evacuation doors, safety exits, evacuation walkways, and evacuation stairs shall comply with the following regulations: ...\" (Regulations regarding evacuation width)\nAccording to the relevant regulations of Article 7.1.4 of the \"General Code for Fire Protection of Buildings\" GB 55037-2022:\n\"The net width of evacuation exit doors, evacuation walkways, evacuation stairs, etc., shall comply with the following regulations: 1. The net width of evacuation exit doors and outdoor evacuation stairs shall strictly not be less than 0.80m.\"", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 4, "rubric_detail": "Explicitly state that the fire partition walls of each charging vehicle fire unit in the basement shall be provided with evacuation doors constituting not less than 2 safety exits.", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 5, "rubric_detail": "Confirm that safety exits may be fire doors opening into adjacent fire units.", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 6, "rubric_detail": "State that the horizontal distance between the nearest edges of two adjacent evacuation doors shall not be less than 5 meters.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 7, "rubric_detail": "Clarify that the \"Technical Standard for Distributed Charging Facilities for Electric Vehicles\" GB/T51313-2018 mentioned by the user is merely a technical standard; it can only serve as a reference and cannot be used as a statutory basis.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "Determine that the opening direction of the evacuation door shall be towards the adjacent fire unit where the safety exit is located.", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "Point out that the logic for determining the width of the evacuation door is decided comprehensively based on the calculation result of the number of evacuees and the minimum width restriction (0.80m).", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "The model answer fails to cite relevant building codes and standards for analysis and argumentation, but instead answers the question directly.", "rubric_weight": -3, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 11, "rubric_detail": "The answer includes large sections of transcribed fire code provisions related to above-ground residential parts or non-garage areas that are irrelevant to basement garage evacuation, or contains significant content unfocused on the question itself, leading to redundancy and difficulty in reading for the user.", "rubric_weight": -3, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 12, "rubric_detail": "Fails to specify the exact name of the code and the specific article number when citing regulations.", "rubric_weight": -3, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 13, "rubric_detail": "In the answer provided by the model, the intercommunicating doors on the fire partition walls do not strictly adopt Class A fire doors.", "rubric_weight": -8, "rubric_tag": "Factual Information" } ] }, { "id": "44b07f0a-d6c2-4b74-bfdd-2c7b5a87e1c7", "case_id": 10854, "language": "global", "system_prompt": "", "question": "Assume you are a researcher working on agricultural robotics and are designing a robotic manipulator capable of operating efficiently in greenhouse tomato pollination and harvesting tasks. However, you encounter difficulties during path planning for the manipulator:\n\n1. Joint-space planning methods (such as RRT, RRT*, and their variants) are efficient at exploring high-dimensional spaces, but they perform poorly in narrow environments, exhibiting weak obstacle-avoidance capability and unstable paths.\n2. Cartesian-space planning can achieve precise obstacle avoidance, but inverse kinematics solving often falls into singularities or produces unreachable solutions, especially in scenes with flexible obstacles (such as tomato leaves), resulting in a very high planning failure rate.\n\nYou are now asked to write a complete research proposal whose goal is to enable the robotic arm to achieve highly reachable and highly feasible path planning for target flowers or fruits in complex greenhouse environments.\n\nThe proposal must include the following components:\n1. Specific modeling approaches;\n2. Feasible path planning strategies;\n3. A complete, reproducible pipeline and experimental design.", "tags": { "topics": [ "Industry", "Systems / Embedded / 3D Rendering", "Systems / Embedded / 3D Rendering" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "Explicitly proposes a Relaxed-IK inverse kinematics strategy.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 2, "rubric_detail": "The proposal clearly distinguishes task targets (e.g., tomato fruits, flowers) from environmental obstacles (e.g., leaves, stems, support structures), and reflects differentiation among different types of obstacles, especially flexible obstacles.", "rubric_weight": 7, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "The proposal includes concrete experimental design (e.g., clearly defined simulation environments, baseline algorithms, and real-robot validation stages) as well as multi-dimensional evaluation metrics (such as planning success rate, path length, computation time, and obstacle-avoidance performance).", "rubric_weight": 7, "rubric_tag": "Factual Information" }, { "rubric_number": 4, "rubric_detail": "Uses an RGB-D camera for perception and point cloud acquisition.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 5, "rubric_detail": "Demonstrates overall completeness of the solution pipeline, i.e., modeling → perception → planning → control → experimentation.", "rubric_weight": 7, "rubric_tag": "Factual Information" }, { "rubric_number": 6, "rubric_detail": "Adopts specific KNN or KD-tree methods for outlier removal.", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 7, "rubric_detail": "Provides a concrete formulation or method for constructing the scene cost map, and the formulation/method includes obstacle distance, collision risk, or cost factors specifically targeting flexible obstacles (e.g., tomato leaves).", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 8, "rubric_detail": "Clearly specifies voxel-based downsampling.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "Provides specific path smoothing techniques (e.g., B-splines, Savitzky–Golay filtering).", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 10, "rubric_detail": "Establishes two or more algorithmic baseline comparison groups.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 11, "rubric_detail": "Proposes an inverse kinematics validation stage that must re-verify safety.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "States that voxel size is determined based on obstacle characteristics (e.g., tomato leaf size).", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 13, "rubric_detail": "Explains the use of adaptive weights (or multi-objective optimization functions, fuzzy logic control) to dynamically balance obstacle avoidance and path smoothness.", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 14, "rubric_detail": "Fails to explicitly provide the formula or concrete composition of the scene cost map.", "rubric_weight": -8, "rubric_tag": "Factual Information" }, { "rubric_number": 15, "rubric_detail": "Confuses \"flexible obstacle avoidance\" with \"complete obstacle avoidance,\" leading to planning failure.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 16, "rubric_detail": "Lacks a conclusion and future outlook section, with no analysis of the advantages and limitations of the proposed solution or expected experimental results.", "rubric_weight": -4, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 17, "rubric_detail": "Fails to consider the computational constraints of tomato-harvesting robots as edge devices, deviating from practical deployment constraints.", "rubric_weight": -10, "rubric_tag": "Factual Information" } ] }, { "id": "ee809a60-d4e2-4514-a3b5-175eb591f21e", "case_id": 10904, "language": "global", "system_prompt": "", "question": "```js\nconst logs = [];\nconst log = (msg) => logs.push(msg);\n\nconst buffer = new Proxy(\n { val: 0 },\n {\n set(target, prop, value) {\n log(`B:Set:${value}`);\n target[prop] = value;\n return true;\n },\n }\n);\n\nconst scheduler = {\n then: (resolve) => {\n log(\"Sched:Then\");\n Promise.resolve().then(() => {\n log(\"Sched:Internal\");\n queueMicrotask(() => {\n log(\"Sched:Resolve\");\n resolve(\"Go\");\n });\n });\n },\n};\n\nasync function* streamProcessor(name) {\n log(`P:${name}:Start`);\n\n const signal = await scheduler;\n log(`P:${name}:Signal:${signal}`);\n\n let current = buffer.val;\n\n yield current;\n\n buffer.val = current + 10;\n log(`P:${name}:End`);\n}\n\nlog(\"Global:Init\");\n\nconst procA = streamProcessor(\"A\");\n\nconst p1 = procA.next();\n\nPromise.resolve()\n .then(() => {\n log(\"Inter:1\");\n buffer.val = 50;\n return \"Inter:Result\";\n })\n .then((res) => {\n log(`Inter:2:${res}`);\n });\n\nconst procB = streamProcessor(\"B\");\nconst p2 = procB.next();\n\nlog(\"Global:End\");\n\n// Final output is observed externally\nsetTimeout(() => console.log(logs), 0);\n```\n\nRead the code above and answer the following questions:\n1. Write out the complete sequence of `logs`.\n2. During the execution lifecycle of the current script, will `P:A:End` be printed? Explain why.\n3. After `p1` finishes resolving, what is the exact value of the internal `value` property? Analyze whether this value is affected by `Inter:1`.\n4. Explain why `Inter:2:Inter:Result` is printed before the first `Sched:Resolve`.", "tags": { "topics": [ "Industry", "Frontend Development", "Frontend Development" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "Correctly identifies that the object being awaited (`scheduler`) is not a native Promise but a Thenable object, and that its `then` method invocation is wrapped into a microtask rather than being executed synchronously.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "Correctly determines the execution order of the synchronous portion of the code, explicitly stating that the synchronous output sequence is: Global:Init → P:A:Start → P:B:Start → Global:End, with no microtask outputs interleaved during the synchronous phase.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "Correctly determines that `Inter:1` executes later than the first `Sched:Then`.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "Correctly identifies that `Sched:Internal` is created via `Promise.resolve().then`, while `Sched:Resolve` is created via `queueMicrotask`, and that there is a strict execution order between the two.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "Correctly identifies that both iterators A and B independently trigger the `scheduler.then` execution flow.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "Points out that since `next()` is never called a second time, the generator remains suspended at `yield current`, and the subsequent `+10` logic is effectively dead code.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "Because `p1 = procA.next()` is executed before `Inter:1`, correctly determines that the microtask for the first `await` has already been enqueued at that point.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "Points out that `await scheduler` (a Thenable) introduces an additional microtask to invoke its `then` method, causing the scheduler’s internal microtask chain to be enqueued later than the external Promise chain, which results in `Inter:2` being printed before `Sched:Resolve`.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "The answer addresses each question separately rather than merging all explanations into a single block of text.", "rubric_weight": 3, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 10, "rubric_detail": "The model outputs an incorrect `logs` sequence.\nCorrect logs:\n[\n 'Global:Init',\n 'P:A:Start',\n 'P:B:Start',\n 'Global:End',\n 'Sched:Then',\n 'Inter:1',\n 'B:Set:50',\n 'Sched:Then',\n 'Sched:Internal',\n 'Inter:2:Inter:Result',\n 'Sched:Internal',\n 'Sched:Resolve',\n 'Sched:Resolve',\n 'P:A:Signal:Go',\n 'P:B:Signal:Go'\n]", "rubric_weight": -10, "rubric_tag": "Factual Information" }, { "rubric_number": 11, "rubric_detail": "Fails to account for the effect of `Inter:1` modifying the global `buffer.val`.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "Ignores generator semantics and incorrectly assumes that `procA.next()` executes the entire function in one go.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 13, "rubric_detail": "The answer includes extensive basic explanations of Promise, Proxy, or Generator syntax, resulting in significant redundancy.", "rubric_weight": -5, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "fd1f36d1-c29a-4077-91f4-b0d23cb38f5b", "case_id": 10915, "language": "global", "system_prompt": "", "question": "The binary-to-ternary switching of organic electrochemical transistors has always been a relatively complex engineering task. Studies have found that this switching can also be achieved by using near-infrared light. Please elaborate in detail how near-infrared light can be utilized, via the light–ion–electron coupling pathway, to ingeniously induce ions in an electrochemical transistor to give rise to a negative differential transconductance phenomenon, and how binary-to-ternary switching is realized based on this principle.\n1. Describe the physical mechanisms involved in this process, the key materials, and the control methods.\n2. Describe the future application prospects of this light-controlled reconfigurability for information processing.\n3. Describe, in stages, the specific process of the light-induced redox reactions, and explain why this phenomenon can only be observed under near-infrared illumination and in the presence of the specified electrolyte.", "tags": { "topics": [ "Industry", "Semiconductors", "Semiconductors" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The stage of cooperative doping by ions and photogenerated holes involves the migration of I− ions dissolved in the electrolyte into the polymer channel.", "rubric_weight": 10, "rubric_tag": "Factual Information" }, { "rubric_number": 2, "rubric_detail": "The increase of drain current in the first stage is due to the superposition effect of ionic p-type doping and photogenerated holes produced by near-infrared light.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "In the light-accelerated redox-dominated stage, the high concentration of holes in the channel drives I− ions to react to form triiodide (I3−).", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 4, "rubric_detail": "The core reason for the occurrence of the negative differential transconductance phenomenon is that the formation of triiodide consumes holes, leading to a decrease in conductivity.", "rubric_weight": 7, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "The condition for the third stage to occur is that the gate voltage exceeds the valley voltage.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 6, "rubric_detail": "The reason for the current to rise again in the high-voltage stage is that the ionic conversion tends toward saturation and the influence of newly injected ions becomes dominant.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "The control methods include adjusting the infrared light intensity to tune the peak-to-valley ratio and the voltage of the negative differential transconductance.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 8, "rubric_detail": "The light intensity is positively correlated with the prominence of the negative differential transconductance; that is, the stronger the light intensity, the more intense the redox processes.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "Changing the gate area is an effective means of adjusting the capacitance and thereby modulating the transconductance capability.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 10, "rubric_detail": "The application prospects clearly point to the fields of adaptive computing or intelligent sensing.", "rubric_weight": 10, "rubric_tag": "Factual Information" }, { "rubric_number": 11, "rubric_detail": "The answer contains a large amount of redundant introduction of basic definitions or general principles of organic electrochemical transistors, and does not focus on the specific mechanisms of NIR-light-induced NDT.", "rubric_weight": -6, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 12, "rubric_detail": "When describing the stages, there is a lack of clear paragraph separation or subheadings, resulting in a confusing logical hierarchy.", "rubric_weight": -3, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 13, "rubric_detail": "Holes are mistakenly written as electrons, leading to an error in the type of charge carrier.", "rubric_weight": -7, "rubric_tag": "Factual Information" }, { "rubric_number": 14, "rubric_detail": "It is assumed that light intensity only affects the switching speed and does not affect the modulation of NDT.", "rubric_weight": -3, "rubric_tag": "Structure and Formatting" } ] }, { "id": "fad912f4-b576-4cd0-b659-9530ee97ade2", "case_id": 10919, "language": "global", "system_prompt": "", "question": "There is a modeling process for a mechanical part as follows:\nI. Process of generating volume\nCreate an isosceles trapezoid, referred to as the initial trapezoid.\nIts lower base has a length of 90mm, and its upper base has a length of 30mm. Each leg forms an angle of 45 degrees with the lower base. Extrude this trapezoid by 10mm in the direction normal to its plane.\nIn the plane of this trapezoid, the lower base of the trapezoid and an edge of length 15mm, located outside the initial trapezoid, form a rectangle.\nExtrude this rectangle by 68mm in the same direction as the previous trapezoid extrusion to form a rectangular block e.\nThe face of rectangular block e with the largest area that does not contact the initial trapezoid is designated as plane E.\nThe upper base of the initial trapezoid and an edge of length 15mm located outside the initial trapezoid form a rectangle, also lying in the plane of the initial trapezoid.\nExtrude this rectangle by 34mm in the same direction as the above two extrusions to form rectangular block f.\nIn the solid generated solely by the overall extrusion of the isosceles trapezoid, there is a plane containing the two longest edges of this solid; this plane is designated as plane A.\nCreate a new plane B at a distance of 5mm from plane A.\nIn the solid formed by extruding the isosceles trapezoid, the face with the smallest area is designated as plane C.\nThe distance from plane B to plane C is greater than the distance from plane A to plane C.\nOn plane B there is a rectangle of length 126mm and width 24mm, and two semicircles.\nThe diameter edges of the two semicircles coincide with the two short sides of the rectangle.\nThe rectangle and the two semicircles on plane B are taken together and extruded in the direction perpendicular to the above three extrusion directions, until they reach plane E.\nThe line connecting the centers of the two semicircles is parallel to the lower base of the initial trapezoid, and this line is at a distance of 34mm from the plane of the initial trapezoid.\nThe centers of the two semicircles are symmetric with respect to a plane that passes through the midpoint of the lower base of the initial trapezoid and is perpendicular to the plane of the initial trapezoid.\n\nII. Process of removing volume\nOn rectangular block f, the face with the largest area that does not contact the initial trapezoid is designated as face F.\nAmong the faces of rectangular block f that are parallel to the initial trapezoid, the one farthest from the initial trapezoid is designated as face G.\nAt the midpoint of the intersection line between face F and face G, create a point. Using this point as the center, draw on face F a circle with a radius of 16mm.\nExtrude this circle by 6mm in both directions normal to its plane, and remove all the volume swept by the extrusion.\nThe face of rectangular block f that is parallel to face F is designated as face H.\nUsing the midpoint of the intersection line between faces H and G as the center, draw on face H a circle with a radius of 8mm.\nExtrude this circle by 9mm in both directions normal to its plane, and remove all the volume swept by the extrusion.\nAt each of the only two circle centers on plane B, draw on plane B a circle with a diameter of 12mm.\nExtrude each circle by 10mm in both directions normal to its plane, and remove all the volume swept by the extrusion.\n\nCompute the theoretical volume of this part.\nAt present, this part is still in a state where all curved surfaces and holes are absent. Curved surfaces, holes (drilling), and surface roughness must be machined.\nAfter final machining, every face of the part shall be assigned a non-negative integer number, starting from 1.\nThe material of the part is 45 steel.\nOnly purely mechanical fixtures may be used, with no other auxiliary materials. At least three pairs of parallel clamping faces shall be used.\nFor the finished part, the surface-roughness requirement for all faces parallel to plane B and for all curved surfaces is Ra0.8.\nDesign a machining process plan, clearly specifying the clamping position and the cutting tools used at each operation. Present the sequence of operations in tabular form.\n", "tags": { "topics": [ "Industry", "Machinery and Automation", "Machinery and Automation" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The theoretical volume of the final finished part obtained by the model is 132684.424mm³ (a very small error is allowed, but the result must be close to the reference value).", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "The cutting tools listed in the machining plan include at least a milling cutter, a drill, and a reamer.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "The plan uses faces that have already been machined to Ra0.8 as clamping surfaces.", "rubric_weight": -8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "The machining process sequence designed by the model reflects the logic of rough machining followed by finish machining.", "rubric_weight": 10, "rubric_tag": "Factual Information" }, { "rubric_number": 5, "rubric_detail": "The operation table can be rendered correctly.", "rubric_weight": 4, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 6, "rubric_detail": "In one of the operations, the clamping surface is the surface being machined in that same operation, which constitutes a serious error.", "rubric_weight": -20, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "The plan uses at least three pairs of parallel clamping faces.", "rubric_weight": 5, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 8, "rubric_detail": "Correctly handles the calculation of the removed volume.\nSpecific algorithm:\nThe removed material consists of the boss formed by extruding two semicircles and the two cylinders formed by extruding two circles.\nThe volume of the boss formed by extruding the two semicircles:\n0.5×π×16^2^×6+0.5×π×8^2^×9=3317.522mm³ (rounded to three decimal places)\nThe volume of the two cylinders formed by extruding the circles:\nπ×6^2^×10×2=2261.947mm³ (rounded to three decimal places)\n\n\n\n\n", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "The model uses a tabular format to present the sequence of operations.", "rubric_weight": 5, "rubric_tag": "Instructions Following" }, { "rubric_number": 10, "rubric_detail": "The model fails to identify all specific faces that require machining of surface roughness, specifically including: all curved surfaces (such as the inner walls of the holes with radii 16mm and 8mm, the inner walls of the holes with diameter 12mm, and the side surfaces of the semicylinders) as well as the faces parallel to plane B (such as plane E, plane B itself, etc.).", "rubric_weight": -8, "rubric_tag": "Factual Information" }, { "rubric_number": 11, "rubric_detail": "The model uses non-negative integers starting from 1 to number all faces of the final finished part (a total of 23~26 faces).", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 12, "rubric_detail": "For all surfaces requiring Ra0.8 (faces parallel to plane B and curved surfaces), the model fails to select appropriate finishing tools (such as finish milling cutters, grinding tools, etc.) in some cases.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 13, "rubric_detail": "Correctly handles the volume calculation related to the overlapping region.\nThe solid generated by extruding the profile on plane B overlaps with rectangular block e.\n\nThere are two possible algorithms:\n1. Directly calculate the non-overlapping volume of the solid generated by extruding the profile on plane B and rectangular block e:\n[π×12×12+24×(126-90)]×10=13163.893mm³ (rounded to three decimal places)\n\n2. Or calculate the volume of the overlapping part (and subtract it together at the end):\n90×24×10=21600mm³\n\n\n", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "54246069-1446-41a1-8d17-89a0e5c52dc0", "case_id": 11071, "language": "global", "system_prompt": "", "question": "I am conducting research on dissimilar metal resistance spot welding (RSW), utilizing aluminum alloy and low alloy steel as the study materials. The objective is to investigate the evolution of internal mechanisms and the nugget formation process during resistance spot welding. The experimental conditions are as follows: the thicknesses of the aluminum alloy and low alloy steel are 1.5 mm and 1 mm, respectively; the low alloy steel sheet is positioned on top, with the aluminum alloy sheet at the bottom, arranged in a lap configuration. The upper electrode cap has a diameter of 6 mm, while the lower electrode cap has a diameter of 16 mm. The equipment employed is an MFDC resistance spot welder. Please analyze the temporal evolution of the internal temperature of the welded joint, the nugget formation process, and its shape and location when welding is performed under appropriate welding parameters.", "tags": { "topics": [ "Industry", "Materials", "Materials" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "During the initial stage of welding, the region of high current density on the low alloy steel side shifts from the electrode-to-steel contact edge to the center of the steel sheet.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "In the initial stage of welding (approximately 10 ms), the peak temperature occurs in the outer annular region where the upper electrode cap contacts the sheet.", "rubric_weight": 3, "rubric_tag": "Factual Information" }, { "rubric_number": 3, "rubric_detail": "Heat diffuses from the steel sheet toward the aluminum alloy side due to the higher thermal conductivity of the aluminum alloy.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "On the aluminum alloy side, the region that first reaches the melting point is located at the aluminum–steel interface (the upper half of the aluminum alloy sheet).", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 5, "rubric_detail": "The temperature at the center of the aluminum–steel interface does not reach the melting point of the low alloy steel; thus, the steel at the interface remains in a solid state.", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 6, "rubric_detail": "The reduction in current density and the onset of cooling on the steel side are attributed to the formation of an indentation on the steel surface, which increases the contact area.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "Under different welding parameters, the aluminum alloy nugget may exhibit two forms: either not penetrating to the bottom or penetrating through the entire aluminum alloy sheet.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 8, "rubric_detail": "During cooling on the aluminum alloy side, the crystallization morphology sequentially exhibits planar growth, cellular growth, and dendritic growth as the temperature gradient decreases.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "During the cooling process on the low alloy steel side, coarse lath packets may form during the austenite-to-martensite transformation.", "rubric_weight": 3, "rubric_tag": "Factual Information" }, { "rubric_number": 10, "rubric_detail": "The bonding mechanism at the aluminum–steel interface involves the wetting and spreading of molten aluminum over the solid steel surface.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "A distinct indentation forms on the steel surface, whereas the indentation on the aluminum side is less pronounced, corresponding to the difference in diameters between the upper electrode cap (6 mm) and the lower electrode cap (16 mm).", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 12, "rubric_detail": "The response contains substantial background information or general welding theory irrelevant to the core question, resulting in severe redundancy.", "rubric_weight": -10, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 13, "rubric_detail": "The content is not organized chronologically (initial, middle, final stages) or in logical phases, leading to a disorganized description of the process.", "rubric_weight": -10, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 14, "rubric_detail": "The analysis fails to identify that the final aluminum–steel resistance spot weld exhibits a dual-nugget characteristic.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 15, "rubric_detail": "The model fails to point out that as welding progresses, the deepening of the electrode indentation leads to an increased contact area, thereby reducing current density.", "rubric_weight": -5, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "f0bc2660-fdb2-4e05-9766-e8db285d52e2", "case_id": 11913, "language": "global", "system_prompt": "", "question": "In Chengdu, the ground floor of a detached residence features a south-facing open courtyard measuring 4 square meters (2 m × 2 m) adjacent to the living room. The owner plans to enclose this space to convert it into a sunroom with basic natural ventilation; all requisite planning permissions and property management approvals have been obtained. Based on climate adaptability requirements specific to the Chengdu region and appropriate building envelope system selection, please provide a scientific estimate of the reasonable range of direct construction and installation costs for this sunroom. In addition, explain the applicable technical standards, material performance specifications, and the cost rationale employed.", "tags": { "topics": [ "Industry", "Architectural Design", "Architectural Design" ], "time_sensitivity": { "time_sensitivity": "Strongly time-sensitive", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The response cites specific climatic parameters for the Chengdu region, such as annual average sunshine duration of 1,000–1,400 hours or an annual average humidity exceeding 70%.", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 2, "rubric_detail": "Based on Standard GB 55015-2021, calculates that the effective ventilation area for the 4-square-meter sunroom must be no less than 0.20 square meters.", "rubric_weight": 9, "rubric_tag": "Factual Information" }, { "rubric_number": 3, "rubric_detail": "Cites the mandatory requirements regarding the use of tempered or laminated glass according to Standard GB 15763.2-2020 for cases where the façade height exceeds 2.0 meters.", "rubric_weight": 7, "rubric_tag": "Factual Information" }, { "rubric_number": 4, "rubric_detail": "Considering Chengdu's rain-prone climate, points out that the airtightness rating must reach Level 6 (GB/T 7106-2019) to prevent leakage.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "Fails to provide specific configuration parameters for the façade glass (e.g., \"5+12A+5 Tempered Low-E\"), relying instead on vague terminology such as \"double glazing.\"", "rubric_weight": -5, "rubric_tag": "Factual Information" }, { "rubric_number": 6, "rubric_detail": "Fails to compare the differences in thermal insulation, service life, or cost between roofing systems (e.g., insulated laminated double glazing vs. polycarbonate (PC) multiwall panels).", "rubric_weight": -4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "The estimated total cost range for the construction and installation works of the economy-grade scheme is between 15,000 and 23,000 RMB.", "rubric_weight": 4, "rubric_tag": "Factual Information" }, { "rubric_number": 8, "rubric_detail": "The estimated total cost range for the construction and installation works of the high-performance scheme is between 22,000 and 32,000 RMB.", "rubric_weight": 4, "rubric_tag": "Factual Information" }, { "rubric_number": 9, "rubric_detail": "The costing logic explicitly includes \"small-order surcharges\" or setup/scheduling overhead apportionment fees for the ultra-small 4-square-meter project, with a ratio ranging between 20% and 35%.", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "Mentions localized market factors in Chengdu, such as quotations from manufacturers in Shuangliu District, local installation price advantages, or specific material transportation costs.", "rubric_weight": 4, "rubric_tag": "Factual Information" }, { "rubric_number": 11, "rubric_detail": "The response adopts a clear hierarchical structure, articulating climate-related codes/standards, technical selection, and cost estimation in distinct sections.", "rubric_weight": 4, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 12, "rubric_detail": "Uses lists, comparative items, or clear data entries when enumerating technical parameters or cost data to enhance readability.", "rubric_weight": 3, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 13, "rubric_detail": "The content contains a significant amount of generic popular-science content on sunrooms irrelevant to the Chengdu climate or the 4-square-meter courtyard, resulting in serious verbosity.", "rubric_weight": -5, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 14, "rubric_detail": "Fails to perform scientific estimation as required by the prompt, providing only a generalized total price lacking specific unit price basis or regulatory sources.", "rubric_weight": -5, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 15, "rubric_detail": "The provided answer follows a chaotic format that is difficult to read.", "rubric_weight": -3, "rubric_tag": "Structure and Formatting" } ] }, { "id": "58a83321-05c2-4d66-86d0-1fbca0dc4741", "case_id": 11917, "language": "global", "system_prompt": "", "question": "An existing residential unit in Shanghai is located on the 5th floor. A square exterior window is set on its south façade, with a clear area of 2.0 m² (window width approximately 1.41 m, height approximately 1.41 m). The window sill height is 0.6 m above the interior finished floor level. The room exhibits a pronounced overheating problem from summer afternoons to evenings; measured indoor operative temperatures frequently exceed 30℃, impairing thermal comfort. The external façade of the building follows a uniform community style, and the property management imposes strict control on the material, colour and projection depth of any additional components:\nThe shading device shall not project more than 300 mm beyond the plane of the exterior wall;\nHigh-reflectance or mirror-finish materials shall not be used;\nAdjustable or passive low-maintenance constructions shall be given priority;\nThe existing window frame structure shall not be damaged, nor shall the function of the window as a fire-fighting and rescue opening be impaired.\nTaking into account the characteristics of Shanghai’s hot-summer/cold-winter climate zone, solar radiation patterns, and the orientation, height and usage scenario of this window, propose one or more shading optimisation schemes that are technically sound, economically feasible and compliant with the management constraints, and describe in detail the design principles, key parameters (such as shading panel inclination angle, projection depth, material properties, etc.) and the expected improvement in the thermal environment.", "tags": { "topics": [ "Industry", "Architectural Design", "Architectural design" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The answer explicitly identifies Shanghai as belonging to the hot-summer/cold-winter climate zone and provides corresponding parameters: summer daily peak solar radiation can reach 800–900 W/m².", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 2, "rubric_detail": "The solution analyses the pattern whereby after 15:00 the solar azimuth shifts significantly to the west (e.g. beyond 58°W), and concludes that the effectiveness of a single horizontal shading device declines and must be supplemented with vertical shading.", "rubric_weight": 9, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "The shading device is designed to project more than 300 mm beyond the exterior wall façade, or fails to explain how the maximum projection is controlled.", "rubric_weight": -10, "rubric_tag": "Instructions Following" }, { "rubric_number": 4, "rubric_detail": "The recommended scheme includes a fixed shallow louvre (egg-crate) structure and explicitly specifies a top panel projection depth of 300 mm.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 5, "rubric_detail": "In the fixed shading scheme, the depth of the vertical side fins on both sides is set at approximately 220 mm to block low-angle western sun.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 6, "rubric_detail": "The materials selected in the scheme explicitly exclude high-reflectance or mirror-finish materials and adopt matte or diffuse-finish treatments (e.g. matte mid-grey).", "rubric_weight": 7, "rubric_tag": "Instructions Following" }, { "rubric_number": 7, "rubric_detail": "The fire-fighting and rescue window function is not considered, or the shading components obstruct full opening of the sash and fail to ensure a clear width of 1.0 m.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "As a supplementary option, the use of a tensioned fabric shading screen with an openness factor of approximately 4% is recommended.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 9, "rubric_detail": "It discusses the necessity for adjustable shading (such as a fabric screen) to be retracted in winter to retain passive solar gains, reflecting a balance between different seasonal requirements.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "It refers to window performance upgrade options, specifically including the addition of a thermally-broken aluminium sub-frame or the application of a nano-ceramic low-reflectance window film.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 11, "rubric_detail": "The technical elaboration explains the physical mechanism by which high-infrared-emissivity (ε ≥ 0.85) coatings facilitate radiative heat dissipation from the component to the sky, thereby reducing its surface temperature.", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "In view of the frequent typhoons in Shanghai, the scheme incorporates anti-uplift wind design features (such as drip-groove aerodynamic guidance and bi-directional tensioning systems).", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 13, "rubric_detail": "The model should provide an analysis of the cost components (e.g. including material and installation costs) and demonstrate the economic feasibility of the scheme.", "rubric_weight": 3, "rubric_tag": "Factual Information" }, { "rubric_number": 14, "rubric_detail": "It clearly quantifies the expected improvement in the thermal environment, indicating that the peak indoor operative temperature can be reduced by 2.5–3.5℃ (or reduced to 26.5–28.0℃).", "rubric_weight": 7, "rubric_tag": "Factual Information" }, { "rubric_number": 15, "rubric_detail": "In terms of visual comfort, it sets a target of maintaining the indoor Unified Glare Rating (UGR) below 19.", "rubric_weight": 3, "rubric_tag": "Factual Information" }, { "rubric_number": 16, "rubric_detail": "The answer adopts a clear hierarchical structure, developed in the logical sequence of “basic analysis – recommended scheme – technical elaboration – economic analysis – expected benefits”.", "rubric_weight": 4, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 17, "rubric_detail": "Bullet points or numbered lists are used to present key parameters and technical indicators, enhancing the readability of the content.", "rubric_weight": 2, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 18, "rubric_detail": "The answer contains a large amount of generic building physics textbook-style definitions or lengthy background introductions that are unrelated to the specific window retrofit, resulting in serious redundancy.", "rubric_weight": -8, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 19, "rubric_detail": "The output format is disordered and fails to use paragraphs or headings correctly, causing key parameters (such as dimensions and angles) to be buried in continuous text and difficult to locate.", "rubric_weight": -6, "rubric_tag": "Structure and Formatting" } ] }, { "id": "5e785915-8448-461e-b663-9a8c9c777c4b", "case_id": 11975, "language": "global", "system_prompt": "", "question": "Sodium metal batteries are currently a hot research topic, and studies have found that the mixed-solvation electrolyte strategy may be beneficial for improving the performance of sodium metal batteries. Please answer:\n1. What is the core design concept of mixed-solvation electrolytes, and explain the respective roles and synergistic mechanisms of strong solvents and weak solvents in the mixed system.\n2. How can this technology simultaneously address the two major challenges of dendrite growth and interfacial stability.\n3. Compared with conventional single-solvent systems, what advantages does the mixed-solvation strategy exhibit in terms of electrochemical performance, and in what aspects is its universality reflected.", "tags": { "topics": [ "Industry", "Chemical Engineering and Materials", "Chemical Engineering and Materials" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The core design concept of mixed-solvation electrolytes for sodium batteries is clearly defined as achieving a balance between bulk-phase ion transport and electrode interfacial stability through the rational ratio of strong and weak solvents.", "rubric_weight": 10, "rubric_tag": "Factual Information" }, { "rubric_number": 2, "rubric_detail": "The role of the strong solvent in the system is identified as regulating the solvation structure and forming a stable solvation sheath by virtue of its strong coordination capability.", "rubric_weight": 10, "rubric_tag": "Factual Information" }, { "rubric_number": 3, "rubric_detail": "The role of the weak solvent is described as facilitating desolvation of sodium ions at the electrode interface.", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 4, "rubric_detail": "The weak solvent forms a more stable and compact interfacial film on the electrode surface.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 5, "rubric_detail": "Compared with single-solvent systems, the advantages of the mixed-solvation strategy include a significant improvement in coulombic efficiency.", "rubric_weight": 3, "rubric_tag": "Factual Information" }, { "rubric_number": 6, "rubric_detail": "Performance improvements brought by the mixed-solvation strategy include a substantial extension of cycle life.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 7, "rubric_detail": "Widening of the electrochemical window is listed as one of the advantages of this strategy.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 8, "rubric_detail": "This strategy can reduce the number of activation cycles.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 9, "rubric_detail": "The strong solvent, via the principle of competitive coordination, helps suppress side reactions while maintaining ionic conductivity.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "The weak solvent promotes uniform deposition of sodium ions by lowering the energy barrier (reducing the activation energy).", "rubric_weight": 7, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "The answer contains an excessive amount of background on the development history of sodium-ion batteries or general electrochemical principles, leading to severe redundancy.", "rubric_weight": -6, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 12, "rubric_detail": "The content piles up more than four chemical formula derivations, which undermines the readability of the core conclusions.", "rubric_weight": -5, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 13, "rubric_detail": "The answer does not follow a logical sequence of design concept, mechanisms for solving key challenges, advantages, and universality.", "rubric_weight": -6, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 14, "rubric_detail": "When elaborating on the synergistic effects, the answer does not clearly distinguish the respective contributions and mutual cooperation of the strong solvent (e.g., responsible for salt dissociation and dominating solvation) and the weak solvent (e.g., reducing viscosity and participating in interphase film formation).", "rubric_weight": -5, "rubric_tag": "Structure and Formatting" } ] }, { "id": "89055e11-d0a5-4e7b-8d15-e1002f4cae54", "case_id": 1238, "language": "global", "system_prompt": "", "question": "Background: You are a physical-layer systems engineer at a leading telecommunications equipment manufacturer. You are responsible for optimizing a 5G macro-cell gNB deployment in an Urban Dense area. The environment features high-rise buildings, uneven user distribution, and extensive reflections and blockages—i.e., a typical mixed scenario combining non-line-of-sight (NLOS) propagation and high mobility.\n\nChallenge: To improve cell-edge throughput and overall spectral efficiency in inter-cell interference (ICI)-limited regions, the network planning team proposes introducing dynamic TDD (Time Division Duplexing) in the time domain and CSI (Channel State Information)-based beamforming optimization in the spatial domain. Based on your understanding of the 5G NR physical-layer specifications in 3GPP Release 15/16, design and justify a coordinated uplink/downlink dynamic resource allocation scheme (for URLLC and/or eMBB). Your response should address:\n\n- Dynamic TDD configuration strategy: Explain in detail how dynamic TDD enables flexible UL/DL scheduling, and the specific mechanisms by which it addresses UL/DL load imbalance and mitigates inter-cell interference in urban dense deployments (e.g., DL-to-UL cross-link interference).\n- Beamforming and CSI feedback: Explain how Type II CSI (e.g., aperiodic / semi-persistent CSI-RS) can be used to track the channel with high accuracy, and design a physical-layer operational workflow that jointly maximizes beamforming gain and suppresses interference.\n- Latency and reliability analysis: Quantitatively analyze (or provide a rigorous qualitative argument for) the impact of the joint scheme on key URLLC KPIs—ultra-low latency (e.g., <1 ms) and ultra-high reliability (e.g., BLER ≤ 0.001%)—and the corresponding optimization measures.", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Weakly time-sensitive", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "Signaling details for dynamic-TDD CLI coordination: the answer must accurately state which DCI format (e.g., DCI 2_0) carries the Slot Format Indicator (SFI), and explicitly note that SFI configures Flexible symbols at symbol-level granularity.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "Enhanced characteristics of Type II CSI: the answer must mention key optimizations of the enhanced Type II codebook in Rel-16 or later (e.g., support for multi-panel operation or codebook subset restriction), and state that CSI-RS-based measurements can include interference measurement resources (CSI-IM).", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "Grant-free UL details for URLLC: the answer must distinguish Configured Grant Type 1 (RRC pre-configuration) from Type 2 (PDCCH activation/deactivation), and explain why Type 2 is advantageous for low-latency handling of bursty URLLC traffic.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "Implementation prerequisites for interference-suppressing precoding: the answer should note that techniques such as ZF or MMSE require accurate inter-cell interference channel information (iCSI), which in turn requires signaling exchange and/or measurement reporting over the Xn interface (e.g., RIM—Resource Interference Mitigation).", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "DCI format for URLLC DL pre-emption: the answer must state which DCI format (e.g., DCI 2_1) is used to notify eMBB UEs to puncture/skip reception due to downlink pre-emption.", "rubric_weight": 6, "rubric_tag": "Instructions Following" }, { "rubric_number": 6, "rubric_detail": "DCI format for URLLC UL cancellation: the answer must state which DCI format (e.g., DCI 2_4) is used to notify eMBB UEs to stop PUSCH transmission (uplink cancellation).", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "SCS and symbol count in URLLC latency analysis: the answer must mention using larger SCS (e.g., 30/60 kHz) together with mini-slot transmissions (e.g., 2–4 symbols) to quantitatively ensure an air-interface latency below 0.5 ms.", "rubric_weight": 4, "rubric_tag": "Instructions Following" }, { "rubric_number": 8, "rubric_detail": "Use of TCI states in beamforming: the answer must explain how TCI state configuration establishes QCL (Quasi-Co-Location) association for a physical channel (e.g., PDCCH CORESET), thereby ensuring beam consistency between control-plane and user-plane transmissions.", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "Beam tracking optimization for high-mobility: the answer must mention leveraging SRS (Sounding Reference Signal) and UL/DL reciprocity in TDD to enable fast tracking, as an auxiliary mechanism for downlink beam selection under high mobility.", "rubric_weight": 2, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "Reliability enhancement for URLLC under NLOS: the answer must mention Multi-TRP (or CoMP joint transmission) and Code Block Group (CBG) retransmission as mechanisms to handle bursty blockages in NLOS environments.", "rubric_weight": 1, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "Terminology confusion or misuse, reflecting imprecise understanding of 3GPP concepts—for example, incorrectly using DCI 2_0 for UL grants (it is dedicated to SFI), or conflating the functions of CSI components.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "The response omits discussion of critical engineering constraints, such as UE power-amplifier (PA) ON/OFF transitions and rapid power control during fast UL/DL switching.", "rubric_weight": -20, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "71f76252-1db0-41f8-93bf-338a89f17341", "case_id": 1246, "language": "global", "system_prompt": "", "question": "You are a wireless network optimization expert. On February 3rd, in the Hospital South Gate Building 1, Cell -2, the 'no call drop' voice call proportion is 97.6%, which is abnormal; the 'smooth playback' short video service proportion is 97.17%, which is abnormal. The KPI status is as follows:\n1) Number of 5G call drops = 10\n2) Call drop rate = 2.4%\n3) RTT downlink latency (ms) = 380 ms\nDetailed cell-level KPIs have been extracted and are provided to you. Please perform a detailed analysis of where the problems lie and how to optimize them. The optimization recommendations must include specific optimization measures. For parameter settings, a detailed parameter configuration procedure and parameter setting script generation process must be provided. The detailed situation is as follows:\n1. Coverage-related KPI results:\n(1) Weak coverage ratio (RSRP <= -110 ratio) = 0.59%\n(2) Total timing advance (tadv) sampling points = 116048\n(3) Overshooting coverage sampling points = 413\n(4) Overshooting coverage ratio = 0.36%\n(5) Near-coverage sampling points = 23533\n(6) Near-coverage ratio = 20.28% (reasonable range 0–20%)\n2. Interference-related KPI results:\n(1) Average interference-plus-noise power per PRB (dBm) = -95 dBm.\n(2) Hourly interference-plus-noise power per PRB (dBm) = -96 dBm (busiest hour within 24 hours), 5G traffic (5G uplink traffic + 5G downlink traffic) = 77.34 Gb, occurring at 15:00.\n(3) Hourly interference-plus-noise power per PRB (dBm) = -110 dBm (least busy hour within 24 hours), 5G traffic (5G uplink traffic + 5G downlink traffic) = 0.92 Gb, occurring at 04:00.\n(4) Uplink interference type = None\n3. Capacity-related KPI results:\n(1) Uplink PRB utilization = 87%, occurring at 18:00, abnormal (maximum value over the 24 hours network-wide)\n(2) Downlink PRB utilization = 93%, occurring at 17:00, abnormal (maximum value over the 24 hours network-wide)\n(3) ARFCN = 384000\n(4) Average number of 5G users = 140\n(5) 5G traffic (5G uplink traffic + 5G downlink traffic) = 708.48 Gb\n(6) China Unicom total traffic (Unicom uplink traffic + Unicom downlink traffic) = 287.86 Gb\n4. Handover-related KPI results:\n(1) 5G intra-system handover success rate = 98.82%\n(2) 5G intra-system handover success rate_Unicom users = 98.97%\n(3) 5G inter-system handover success rate = 99.35%\n(4) 5G inter-system handover success rate_Unicom users = 99.4%\n(5) Intra-frequency handover success rate = 98.93%\n(6) Intra-frequency handover success rate_Unicom users = 99.14%\n(7) Inter-frequency handover success rate = 97.34%\n(8) Inter-frequency handover success rate_Unicom users = 96.71%\n(9) Intra-frequency handover outgoing request count (times) = 214311\n(10) Intra-frequency handover outgoing request count_Unicom users (times) = 78085\n(11) Intra-frequency handover outgoing success count (times) = 212022\n(12) Intra-frequency handover outgoing success count_Unicom users (times) = 77411\n(13) Inter-frequency handover outgoing request count (times) = 16413\n(14) Inter-frequency handover outgoing request count_Unicom users (times) = 5595\n(15) Inter-frequency handover outgoing success count (times) = 15977\n(16) Inter-frequency handover outgoing success count_Unicom users (times) = 5411\n5. Cell engineering parameter results:\n(1) Site height = 15.0 m\n(2) Mechanical downtilt = 5.0 degrees, electrical downtilt = 0.0 degrees, azimuth = 180.0 degrees\n(3) Cell bandwidth = 100 MHz", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Weakly time-sensitive", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "Capacity issue analysis: PRB utilization exceeding reasonable thresholds leads to resource congestion, scheduling delay, and packet loss.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "Interference control optimization: it should be pointed out that the average interference-plus-noise power per PRB of -95 dBm is relatively high (the normal value should be below -105 dBm), which affects SINR.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "Coverage optimization: the high near-coverage ratio may cause an excessive concentration of close-in users; coverage should be optimized through antenna downtilt/azimuth/power adjustment to avoid too many close-in users.", "rubric_weight": 7, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "It should be pointed out that the inter-frequency handover success rate (97.34%) is relatively low, and the cause should be analyzed as unreasonable A2/A4 threshold settings or neighbor cell configuration issues.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "In response to the capacity problem reflected by uplink PRB utilization reaching 87% and downlink PRB utilization reaching 93%, a hardware expansion scheme such as cell splitting or adding new sites should be proposed.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 6, "rubric_detail": "It should be proposed to perform load balancing via MLB and to mention adjustment of handover-related parameters, such as Cell Individual Offset (CIO) and inter-frequency/inter-RAT handover hysteresis, to steer users to neighbor cells.", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 7, "rubric_detail": "Uplink power control optimization: by adjusting parameters such as P0 and Alpha, limit UE uplink transmit power to reduce uplink interference.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 8, "rubric_detail": "Latency / service QoS optimization: the impact of high RTT (380 ms) on short video services should be analyzed, and optimization suggestions via QoS policy should be proposed, such as mentioning adjustment of scheduling weights or priorities for video-service-related QCIs (e.g. QCI 8/9).", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "Based on the slightly high near-coverage ratio (20.28%), the model should propose optimization suggestions for antenna configuration, with the core direction being to increase downtilt to suppress excessive near-point coverage.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 10, "rubric_detail": "It should be pointed out that the relatively low inter-frequency handover success rate (97.34%) is primarily caused by high cell load (uplink and downlink PRB utilization exceeding limits) leading to accumulated interference, and concrete suggestions should be made to optimize inter-frequency handover-related parameters (such as A3 event thresholds and hysteresis) as well as neighbor relations.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "It should be indicated that PRB-utilization-based load balancing (MLB) is enabled, and at the same time the A3 event trigger threshold for intra-frequency handover should be reduced by 1–2 dB to achieve coordinated optimization of resource allocation and handover performance.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 12, "rubric_detail": "Cell capacity expansion suggestion: when parameter optimization cannot meet requirements, additional carriers should be added, new cells should be configured, or more sites should be deployed.", "rubric_weight": 2, "rubric_tag": "Factual Information" }, { "rubric_number": 13, "rubric_detail": "It should be pointed out that the average interference-plus-noise power per PRB (-95 dBm) is relatively high, and based on this, an external interference source investigation is recommended.", "rubric_weight": 3, "rubric_tag": "Factual Information" }, { "rubric_number": 14, "rubric_detail": "Coverage extension optimization: optimize coverage at the cell edge by appropriately adjusting power/beam/coverage area (for example, for the slightly high near-coverage ratio of 20.28%, propose optimizing the coverage area by adjusting downtilt) so as to balance edge-user access quality.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 15, "rubric_detail": "The model should point out that the inter-frequency handover success rate (97.34%) is relatively low and analyze possible causes such as missing inter-frequency neighbor configurations or improper parameter settings.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 16, "rubric_detail": "It must be clarified that the base station service scheduling priority is configured globally and fixed. Combined with the root cause of high cell load, resources bottlenecks should first be resolved through load balancing and capacity expansion, and then global QoS strategies (such as ensuring scheduling weights corresponding to video-service QCI classes) should be used to optimize the experience. On this basis, higher scheduling priority should be guaranteed for video services (such as QCI 7/8/9), avoiding localized parameter adjustments that merely treat symptoms.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 17, "rubric_detail": "Latency assurance strategy: propose concrete latency-guarantee measures, such as enabling SPS (Semi-Persistent Scheduling), configuring dedicated QCI, or adjusting power control parameters (e.g. P0-NominalPUSCH) to reduce latency.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 18, "rubric_detail": "Hardware capacity expansion suggestion: when parameter optimization reaches its limit, adding new carriers, new cells, or new base stations should be considered to improve network capacity and stability.", "rubric_weight": 3, "rubric_tag": "Factual Information" }, { "rubric_number": 19, "rubric_detail": "The model should point out the excessively high PRB utilization during peak hours and suggest concrete strategies such as enabling load balancing functions (e.g. IFLB/MLB) to offload users to other frequencies or cells, thereby achieving balanced and stable network load.", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 20, "rubric_detail": "For the interference problem, concrete power control parameter adjustment suggestions should be proposed (such as adjusting P0-NominalPUCCH) or it should be recommended to enable/optimize beamforming functionality.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 21, "rubric_detail": "Ignoring hardware capacity expansion needs—over-reliance on parameter optimization while not considering carrier/base station expansion may cause the capacity bottleneck to remain unresolved.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 22, "rubric_detail": "Failure to analyze the physical-layer (antenna/site/spectrum resources) configuration and hardware conditions means the proposed scheme cannot fundamentally resolve the issues.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 23, "rubric_detail": "Ignoring inter-cell and neighbor-cell interference—expanding coverage or adjusting coverage without interference management may cause neighbor-cell interference and easily worsen the problem.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 24, "rubric_detail": "Ignoring all-day load balancing—optimizing only for peak hours may cause resource waste or new problems during off-peak periods.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 25, "rubric_detail": "The model fails to recognize that overly aggressive handover thresholds/trigger parameter settings are one of the possible causes of the problem.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "6b4ea30a-e9f1-4a30-a3bc-c6c006354ade", "case_id": 1378, "language": "global", "system_prompt": "", "question": "You are responsible for urban 5G NSA network optimization at a provincial capital carrier's network optimization center. Business District A is a local high-end commercial area which has received a large volume of user complaints over three consecutive months stating that \"the phone displays 5G, but the network speed is slower than 4G.\" Complaints are concentrated during the evening peak hours (18:00–21:00) and throughout weekends.\n\nThe basic wireless network conditions for this area are known as follows:\n\n**Coverage Structure:** The business district is covered by 3 macro sites + several indoor distributed antenna systems (DAS). The macro sites utilize a networking scheme of LTE 1800 MHz (20 MHz) + LTE 2100 MHz (15 MHz) + NR n78 (60 MHz). 5G employs the NSA architecture (EN-DC), with gNB and eNB co-sited.\n**Services and Terminals:** The 5G terminal penetration rate in Business District A is > 75%. Services are dominated by video, short video, and cloud office applications. The ARPU is significantly higher than the network-wide average.\n**Typical Busy Hour Performance:**\n- LTE 1800/2100 cell downlink PRB utilization remains consistently between 85%–95%; average user downlink speed is approximately 8–12 Mbps.\n- Within the same coverage area, the NR n78 cell downlink PRB utilization is only 20%–35%; NR traffic share is about 8%.\n- Terminal-side MDT/DT data shows: In outdoor areas of the business district, the proportion of grids with NR RSRP > -95 dBm and NR SINR > 10 dB exceeds 70%; in core indoor areas, NR RSRP ranges from -100 to -105 dBm, with SINR mostly between 5–10 dB.\n**Typical Signaling and Configuration:**\n- In the NSA configuration, the B1 event threshold for LTE → NR EN-DC is configured as: NR_RSRP > -110 dBm, TTT = 320 ms, Hysteresis = 3 dB.\n- There is a large overlap in coverage between LTE and NR; LTE downtilt is slightly insufficient, and some macro sites exhibit obvious overshooting.\n- The core network side has not restricted access privileges for 5G users, but the operator previously configured 5G priority quite \"conservatively\" due to concerns about 5G coverage instability.\n- Monitoring data also indicates: During the time a large number of 5G terminals stay in Business District A, over 80% of the duration is primarily carried by LTE for data traffic, with NR bearers only briefly activated when users are near windows or in outdoor plazas.\n\n**Problem:**\nPlease assume the role of a frontline wireless optimization engineer and conduct a systematic analysis revolving around the following three dimensions:\n1. **Coverage and Interference Dimension:** Combined with the given RSRP/SINR distributions, determine whether the current 5G coverage constitutes the primary bottleneck. What are the potential issues in the LTE and NR coverage relationship?\n2. **Capacity and Scheduling Dimension:** Combined with LTE/NR PRB utilization and NR traffic proportion, deduce the root-cause chain of capacity/bearer issues leading to \"having 5G capability but not using 5G\" and \"5G experience being inferior to 4G.\"\n3. **NSA Parameters and Multi-RAT Coordination Dimension:** Based on configurations such as B1 threshold, TTT, and hysteresis, analyze the impact of the current EN-DC strategy on 5G traffic offloading; point out at least 3 key parameter/strategy issues that may lead to 5G resources being \"underutilized.\"\n\nBased on the above analysis, please propose at least three categories of actionable optimization schemes (e.g., parameter optimization, site type/location and antenna adjustment, spectrum and bearer strategy, etc.), and for each category explain:\n- Key adjustment items and their reasonable value ranges or adjustment directions from an engineering perspective;\n- Expected benefits and potential side effects or risks;\n- Key KPIs/testing methods used to verify optimization effects and acceptance standards.", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The model analyzes the problem from three dimensions: \"Coverage/Interference\", \"Capacity/Bearer\", and \"NSA Parameters/Multi-RAT Coordination\".", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "Explicitly points out that the primary contradiction is LTE overload and NR resource underutilization, rather than solely a 5G coverage issue.", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "Uses numbering or arrows to delineate a causal chain of no fewer than 4 steps, for example:\n1) High concentration of services + Many 5G terminals →\n2) Still carried by LTE anchor →\n3) NR conditions available but EN-DC triggering/retention is conservative →\n4) LTE congestion, NR idle → User perception \"5G is worse than 4G\".\nThe analysis must simultaneously include:\n- Coverage/Interference factors\n- Capacity/Load factors\n- NSA/Parameter or Architecture factors", "rubric_weight": 9, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "In the NSA parameter analysis, provides the \"direction of impact on EN-DC activation frequency and stability\" for increasing/decreasing the B1 threshold, TTT, and Hysteresis. For example:\n- Lower B1 threshold (more negative dBm value) → Larger NR addition range, easier to trigger;\n- Longer TTT → Later activation, more stable; Shorter → Faster but prone to jitter;\n- Smaller Hysteresis → Easier to frequently add/release.\nDirections for all three must be mentioned and must not be incorrect.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "When designing a strategy based on \"LTE Load-Driven NR Addition,\" the description must include:\n- A specific LTE PRB threshold (e.g., >70% or >80%);\n- How it links with EN-DC triggering conditions, such as dynamically relaxing the B1 threshold, shortening TTT, or reducing hysteresis during high load, rather than merely increasing scheduling weight or static thresholds.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "Regarding NR upgrades/retrofit in core indoor areas, not only propose \"installing indoor distributed antenna systems (DAS)/small cells\" but also provide at least two engineering-grade design elements, for example:\n- pRRU/Pico cell MIMO configuration (2×2, 4×4, etc.);\n- Single-point EIRP target range;\n- Basic estimation logic for floor/site count;\nAnd connect these elements to indoor target RSRP/SINR/Throughput.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "In the spectrum and bearer strategy section, discusses both LTE → NR spectrum refarming / DSS, and explicitly mentions:\n- The distinction in roles between NSA and SA in 5G evolution;\n- How spectrum strategies coordinate with the NSA → SA evolution path (e.g., NSA + refarming first, then advancing SA in hotspots/whole network).", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "For at least one key optimization direction, simultaneously provides:\n- 1–2 specific metrics to monitor (e.g., SCG Change count, X2-U packet loss rate, terminal average current, etc.);\n- What fallback or tightening actions to execute when these metrics reach a specific threshold or magnitude of change.\nCannot merely speak generally about \"monitoring risks\" or \"paying attention to KPIs.\"", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "Provides no fewer than 5 KPIs across different dimensions, and specifies a clear target value or range for each KPI, while explaining the testing/statistical method. Typical dimensions include:\n- LTE/NR PRB Utilization\n- NR Traffic Ratio\n- EN-DC/SCG related metrics\n- Service Throughput (P50/P90, etc.)\n- Volume of complaints or tickets", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "Breaks down the optimization schemes into at least 3 categories of engineering actions with distinct natures, for example:\n- NSA / Scheduling / Offloading Strategy\n- Coverage / Site Type / Antenna Engineering\n- Indoor Coverage Supplement / Spectrum and Bearer Strategy / Architecture Evolution\nWithin each category, at least 2 specific directly implementable or plannable actions (parameter names or engineering actions) must be listed.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "The answer almost entirely attributes the user experience issue to severe lack of 5G coverage / n78 physical characteristics, basically ignoring the LTE/NR PRB and NR traffic share load information in the prompt, and failing to identify capacity/bearer as the main contradiction.", "rubric_weight": -8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "The model incorrectly interprets -110 dBm as a high threshold when explaining the B1 event threshold (e.g., believing the UE needs to be very close to the base station to satisfy it).", "rubric_weight": -8, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "6129a42f-a16c-4a28-8912-c60332e70406", "case_id": 1381, "language": "global", "system_prompt": "", "question": "An operator requires the deployment of a 5G SA network in an urban area. Recently, a surge of user complaints was received from a high-end commercial building (8 floors total: 2 underground levels for parking and equipment rooms, 6 above-ground levels for retail, dining, and office areas; total gross floor area approximately 120,000 m²). The reported issues are as follows: 1. Weak 5G signal in the B1 parking area; most zones fail to access the 5G network, and only specific corners can connect with throughput below 1 Mbps. 2. A sharp decline in 5G throughput in the dining areas on floors 3-4 during peak dining hours (12:00-14:00, 18:00-20:00), with slow page loading and video buffering, whereas off-peak throughput is basically normal (approx. 300-500 Mbps). 3. Frequent handovers in parts of the office area, with occasional call drops during voice calls. Within 500 meters of this commercial complex, three macro base stations are deployed: Station A (Band n78, 3.5 GHz, transmit power 46 dBm), Station B (Band n41, 2.6 GHz, transmit power 45 dBm), and Station C (Band n79, 4.9 GHz, transmit power 47 dBm). Inside the building, 12 distributed pico base stations (Band n78, 3.5 GHz, transmit power 30 dBm) are sparsely deployed. Assuming you are a 5G network optimization engineer for this operator, analyze the causes of the three aforementioned issues based on 3GPP 5G network deployment standards. Formulate targeted optimization schemes (specifying optimization steps, technologies and techniques to be employed, and recommendations for adjusting critical parameters), and outline the risks to be avoided during the optimization process along with corresponding countermeasures.", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "Accurately identify the three core causes of weak signals on the first basement level (high-frequency penetration loss, insufficient picocell coverage, and metal shielding/blockage in the building).", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "Accurately identify the core causes of low throughput during peak hours in the dining area (high concurrent user load, multipath interference, and overlapping coverage interference).", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "Accurately identify the core causes of handover issues in the office area (blurred coverage boundaries, improper handover parameters, signal blind spots, and electromagnetic interference).", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "The optimization plan for the first basement level includes picocell site planning (800-1000 m²/unit) and power adjustment (approximately 33 dBm), complying with 3GPP standards.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "The optimization plan for the dining area includes enabling the cell breathing function and adjusting the scheduling algorithm to 'Proportional Fair + User Priority'.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "The optimization plan for the office area includes handover parameters (A3 threshold, CIO, hysteresis time), complying with 3GPP protocols.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "The optimization plan considers the impact of urban commercial buildings (stone walls, heavy [reinforced] concrete) on signals, noting that building materials such as stone walls and heavy reinforced concrete cause severe signal attenuation/penetration loss.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "The optimization for the first basement level adopts a Distributed Antenna System (DAS) to assist coverage and resolve metal shielding/blockage issues.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "The optimization for the dining area mentions the correlation between the Modulation and Coding Scheme (MCS) and SINR.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "The optimization for the office area includes enabling the RLF re-establishment function and configuring QoS for VoNR services (e.g., assigning high-priority QCI, such as QCI 1, for voice services).", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "Risk management measures should include construction coordination plans adapted to commercial scenarios (e.g., night construction, communication with property management).", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "Risk management measures include parameter rollback mechanisms and a pilot rollout model to ensure network stability.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 13, "rubric_detail": "The optimization for the first basement level recommends an RU transmit power of 20-24 dBm, which is significantly lower than the 30-35 dBm range recommended by 3GPP, resulting in insufficient coverage.", "rubric_weight": -4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 14, "rubric_detail": "The optimization for the dining area fails to mention multipath interference, focusing only on capacity and synchronization interference, thereby omitting core loss factors.", "rubric_weight": -3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 15, "rubric_detail": "The model fails to analyze the issue of signals being obstructed by metal in the first basement level due to the failure to adopt a Distributed Antenna System (DAS).", "rubric_weight": -3, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "9c2672f2-430f-44cc-823b-2784f3a84a2b", "case_id": 1394, "language": "global", "system_prompt": "", "question": "You are a senior expert in a metropolitan metro communication network. Recently, Metro Line 2, Line 4, and the control centers of both lines have reported composite communication faults. The phenomena are complex and interrelated, posing a potential risk to operational safety. The faults are concentrated during morning and evening peaks, but some anomalies also manifest differently during off-peak periods.\nFault details and multidimensional data:\nI. Transmission network indicators\n(1) Line 2 (main ring) performance degradation:\nPeak periods: An alarm of VC-12 path signal loss (TU-LOP) is concentrated in the segment between Zhongshan Park Station and Century Avenue Station. The regenerator section bit error rate (RS-BER) in this segment deteriorates from the idle background level of 1.0E-8 to 1.2E-5, exceeding the 1.0E-6 threshold. Bit error performance event analysis reports show that the bit error type is mainly burst errors (Burst Error).\nAbnormalities in all-day monitoring: Historical performance data from the network management system (NMS) reveal that the optical path in this segment experiences a high frequency of transient loss of optical power (Transient Loss of Light) events, each with a duration of less than 1 ms, which do not trigger traditional LOS alarms. The occurrence frequency of these events has surged from several times per day one week ago to several times per second at present.\n(2) Line 4 (protection ring) pressure-sensitive degradation:\nPeak periods: Received optical power (Rx Power) degrades from the baseline of -12 dBm to -22 dBm, accompanied by a surge of approximately 1,500 Ethernet port cyclic redundancy check (CRC) errors per hour.\nContradictory phenomenon: During automated end-to-end performance tests executed at 3:00 a.m. in off-peak business hours, both optical power and bit error rate indicators are completely normal. However, when using the same test script but applying simulated peak traffic load on the link, the optical power degradation and CRC error phenomena can be immediately and stably reproduced.\n(3) Delay jitter on control center interconnection channels:\nCommunication channels between control centers and stations, carried on the Line 2 and Line 4 transmission rings, experience network-layer delay jitter (jitter) of up to ±50 ms during peak periods, which seriously deviates from the normal value (should be < ±5 ms).\nII. Service-layer manifestations\n(1) Video surveillance system: During peak periods, the average packet loss rate of video streams ranges between 8% and 15%, with severe picture stuttering. In-depth protocol analysis reveals abnormally high-frequency retransmissions of streaming protocol signaling (such as RTSP TEARDOWN) for some cameras during peak periods, indicating that instability in the transport layer has induced frequent reconstruction of TCP/UDP session layers.\n(2) TETRA trunked radio system: The voice channel assignment failure rate reaches 6.3% during peak periods (KPI threshold < 1%). Signaling trace analysis locates the key cause: The main reason for failure is not insufficient radio air-interface resources, but the loss or severe delay of periodic link detection packets (such as BFD or Keepalive) between the base station controller (TSC) and the core network switch (DXT). This leads to control-plane misjudgment of peer failures and consequently refusal to allocate traffic channels.\n(3) CBTC train control system data backhaul: The signaling system reports that, during the same period, heartbeat packets between the train-to-ground communication wireless access points (APs) of the CBTC network and ground servers experience occasional timeouts during peak periods. Although this has not triggered system protection switching, it has been recorded as a top-level potential risk event.\nIII. Environment, history, and changes\n(1) External environmental risks: In addition to the known municipal construction near Zhongshan Park Station on Line 2, there has also been third-party pipe jacking construction near Sports Center Station on Line 4 one week ago. The two construction contractors are different, and there is an overlapping risk zone between the construction drawings and the metro optical cable routes.\n(2) Legacy network configuration issues: Historical operation logs show that, due to an optical cable break one year ago, the westbound and eastbound fiber definitions of the Line 2 MSP ring were temporarily adjusted to restore services quickly. Post-event verification records are incomplete, and there is a potential long-term inconsistency between the logically configured primary route and the physically optimal topology route.\n(3) Recent changes and configurations: Three days ago, a line protection switching test of the Line 2 transmission ring was performed, after which the system was not restored to the original primary route in accordance with procedures. Line 4 is undergoing a station Wi‑Fi 6 upgrade; some temporarily connected Ethernet switches have enabled flow control, and the access ports have not been configured with bandwidth rate limiting (rate-limit).\n(4) Power and environment: The communication equipment cabinets at Zhongshan Park Station report abnormal temperature rise alarms. UPS logs at this station show several recent short commercial power interruptions of less than 10 ms, during which the equipment was supported by battery power.\nIV. Resources and constraints\n(1) The daily time window (track possession) available for construction and testing is only 3 hours.\n(2) The number of high-precision instruments available for accurately locating transient interruptions (such as optical power meters/OTDRs capable of capturing µs-level events) is limited, and deployment points must be planned scientifically.\n(3) Operational safety regulations: Any test operation that may interrupt critical production services such as TETRA or CBTC must be submitted for approval 48 hours in advance and must be accompanied by a certified emergency rollback plan that can achieve instantaneous service restoration.\nAs a senior expert, systematically complete the following tasks:\n(1) Construct a comprehensive fault hypothesis: Propose a unified root-cause hypothesis. This hypothesis must be able to explain, simultaneously and coherently, the spatiotemporal differential characteristics of the transmission-layer manifestations on Line 2 and Line 4 (for example, why the problem on Line 4 appears only under traffic stress).\n(2) Explain the differentiated failure modes exhibited by the three different service systems of video, TETRA, and CBTC (packet loss, signaling timeout, heartbeat delay).\n(3) Design a precise localization and verification scheme: How can a highly efficient test plan be designed, using limited high-precision instrument resources, to capture and quantitatively prove the transient physical phenomena (such as transient optical interruptions) that trigger the faults?\n(4) Under the premise of no network outage and no interruption to critical services, how can it be verified whether the temporary configuration adjustments made one year ago have caused a deviation between the current logical primary route and the physically optimal route, and what the practical impact of this deviation is on MSP protection switching logic?\n(5) Formulate phased handling and optimization strategies:\nEmergency mitigation phase: Propose a configuration adjustment scheme that can be implemented before the next morning peak, with the lowest risk, in order to give priority to ensuring the stability of TETRA and CBTC services.\nLong-term phase: Plan a remediation engineering scheme spanning multiple track possession windows. For each step, specify the operational content, verification methods, and rollback procedures.\n(6) System optimization phase: Propose long-term improvement recommendations for network architecture, monitoring capabilities, and O&M processes to enhance network resilience and prevent recurrence of similar complex faults.\n(7) Finally, draft an expert report for submission to management: The report shall present the above analyses, schemes, and strategies in a clear and rigorous structure, and accurately assess the technical risk levels, handling priorities, and required resource support, to assist management decision-making.", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "Propose a unified root-cause hypothesis: recent independent municipal pipe jacking construction at two locations has caused physical micro-damage to metro optical cables, with vibration-sensitive and temperature/load-sensitive characteristics. During business peak periods, this damage is cascaded and amplified by two major internal aggravating factors — configuration deviation of the logical primary route and misuse of data-link-layer flow control — ultimately leading to cross-service and cross-line composite performance collapse.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "Construct a complete, layered causal chain from the physical layer up to the service layer.", "rubric_weight": 7, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "The emergency mitigation方案 should first focus on the flow control configuration issues introduced by the Wi‑Fi 6 upgrade on Line 4, and on the failure to restore the original route after protection switching tests on the Line 2 optical path.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "The方案 should explicitly specify the use of high-precision optical power meters/OTDRs capable of capturing microsecond-level events, and their deployment at both ends of the Zhongshan Park Station–Century Avenue Station segment on Line 2, as well as at key nodes along the optical cable route near Sports Center Station on Line 4. Testing needs to cover both idle and peak (or simulated peak) periods, to correlate transient optical power interruptions with bit errors/CRC errors.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "The方案 should include at least one non-intrusive verification method, such as exporting and comparing current equipment configurations with historical baseline configurations through the NMS, using traceroute and similar tools to probe logical paths and compare them with physical topology diagrams, and analyzing MSP/APS protocol state information.", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "Construct and evaluate multiple competing fault hypotheses and verify them through elimination: before proposing the final hypothesis, at least two other reasonable possibilities (such as purely temperature-related faults or clock synchronization issues) should be considered and ruled out.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "The constructed causal chain exhibits leapfrogging in reasoning: it directly asserts that A leads to B without explaining the necessary intermediate mechanisms. For example, stating simply that optical cable damage leads to TETRA assignment failures, without explaining intermediate links such as transmission-layer bit errors, network-layer jitter, and control protocol timeouts.", "rubric_weight": -4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "The proposed root cause of the current fault is overly broad or vague and lacks a verifiable physical or logical entity; for example, attributing the root cause merely to network instability, system design defects, or poor O&M.", "rubric_weight": -6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "Map and summarize the root causes and failure modes of this fault into the organization’s known error database or fault mode library, and propose updates to inspection items (such as adding checks for consistency between logical and physical routes) and design specifications (such as requiring equipment to have resistance to micro-vibration).", "rubric_weight": 2, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "Ignore socio-technical system factors, with the entire answer focusing solely on technical diagnostics, verification, and remediation, and failing to mention, in any part of the analysis, strategy, or report, the non-technical coordination work necessary for resolving such faults involving municipal construction and multi-party collaboration.", "rubric_weight": -6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "Clearly state the priority order and dependencies of each operation; for example, whether flow control must be disabled first (to eliminate noise sources) before QoS optimization (to optimize the signal), so as to avoid unpredictable impacts from adjusting priorities in a congested environment; and whether direct hardware replacement should follow configuration adjustments as a means of verification.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "In the analysis, assume by default that all proposed monitoring tools and data (such as microsecond-level optical power records, TWAMP delay heatmaps) can be seamlessly integrated into the existing NMS and correlated for analysis, while ignoring the complexity of system integration and data governance.", "rubric_weight": -6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 13, "rubric_detail": "The emergency方案 must propose specific configuration adjustment measures (such as optimizing QoS/CoS policies, disabling flow control on Line 4 switches, and setting bandwidth rate-limits) to give priority to ensuring TETRA and CBTC service stability, and must state explicitly that the associated risks are controllable (for example, implementable before the next morning peak and with rollback plans in place).", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 14, "rubric_detail": "When deriving conclusions, the model should briefly explain the basis for ruling out other possibilities; for example, indicating that it is not a permanent optical cable break (based on the intermittent/burst nature of the fault) or not purely equipment aging (based on the strong correlation between the fault and traffic stress or specific time periods).", "rubric_weight": 2, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "d6344516-e4f6-42b3-a2a6-f1329ed16dbf", "case_id": 161, "language": "global", "system_prompt": "", "question": "The transfer of plasmon-generated hot carriers in metal/semiconductor heterojunctions has long faced severe energy dissipation. Recent studies have identified an ultrafast non-thermalized electron transfer pathway at the Au/GaN interface. Please address the following: 1. What is ultrafast non-thermalized electron transfer, and how does it differ from conventional hot carrier transfer? 2. How is the non-thermalized nature of electron transfer demonstrated, and what specific characteristics must be observed? 3. How does optical polarization modulation influence electron transfer efficiency and energy distribution? 4. What are the implications of this method for the design of hot carrier devices?", "tags": { "topics": [ "Industry", "Semiconductors", "Semiconductors" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "Definition of non-thermalized electron transfer: An ultrafast process wherein electrons are injected directly into the semiconductor without undergoing prior thermalization.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "Elucidation of the distinction from conventional transfer: Non-thermalized transfer retains the original high energy, whereas standard hot carrier transfer results in a thermalized distribution.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "Experimental observation techniques: Simultaneous energy- and time-resolved ultrafast spectroscopy (TR‑2PPE + SPVM).", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "Critical time scales: Injection time <100 femtoseconds, faster than the characteristic electron–electron scattering time (~500 fs) (both figures must be cited).", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "Energy distribution characteristics: Correlated with the initial plasmon distribution (the concept of a non-Maxwellian distribution must be explicitly stated).", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "Polarization modulation mechanism I: Influencing charge yield through plasmonic eigenmodes (the concept of plasmonic eigenmodes must be explicitly mentioned).", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "Polarization modulation mechanism II: Altering injection probability via momentum distribution.", "rubric_weight": 2, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "Role of the Schottky barrier: An interfacial barrier that high-energy electrons traverse more easily (the specific function of the Schottky barrier must be highlighted).", "rubric_weight": 2, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "Ballistic transport characteristics: A transmission mode with approximately zero energy loss (must explicitly identify this loss-free transport method as ballistic transport).", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "Hot carrier lifetime: Typically in the femtosecond to picosecond range (must emphasize that the lifetime is in this range, not merely the emission process).", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "Confusing 'non-thermalized' with 'low-temperature environments,' mistakenly assuming a need for cooling systems.", "rubric_weight": -8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "Assuming polarization only affects excitation intensity while overlooking its modulation of energy distribution.", "rubric_weight": -5, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "dcc3dbb7-0cbb-4f75-ae06-0efbb8987ad0", "case_id": 1650, "language": "global", "system_prompt": "", "question": "In an urban HetNet hotspot environment (e.g., a commercial district) utilizing an LTE Band 7 (2.6 GHz) macro cell as the anchor (MCG) and a dense overlay of NR Band n78 (3.5 GHz) small cells as the Secondary Cell Group (SCG), the network faces severe downlink interference. Specifically, at the NR cell edge, low RSRP/RSRQ levels result in reduced SCG addition/change success rates and degraded user throughput.\n\nDesign and explain a comprehensive interference management and coordination framework for this scenario, with emphasis on:\n- Technical Selection: Identify and justify specific 3GPP-standardized features (e.g., ICIC/eICIC, CoMP, TPC, Rate Matching, Beamforming) most appropriate for this HetNet deployment.\n- Coordination Mechanism: Detail how inter-node coordination (via X2/Xn interfaces) mitigates NR cell-edge interference and specify the resulting improvements in key network KPIs.", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "Proposes that simultaneous UL transmissions on LTE Band 7 and NR n78 may cause out-of-band emissions exceeding limits due to harmonics or intermodulation products, and explains corresponding UE-side interference management mechanisms such as A-MPR (Additional Maximum Power Reduction).", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "Explains that within an NR small-cell cluster, symbol-level/slot-format consistency must be ensured—not only to prevent DL→DL interference, but more importantly to prevent symbol-level interference caused by a neighboring gNB’s DL transmission into the serving gNB’s UL reception.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "Describes how the eNB (MeNB), based on SCG quality (e.g., SCG-RSRQ / PDCCH BLER), dynamically adjusts the PDCP split ratio (split bearer), or performs SCG activation/deactivation (per TS 37.340 procedures) to smoothly offload edge-UE traffic back to the LTE MCG to maintain connection stability.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "Mentions configuring multiple NZP-CSI-RS resource sets for the UE, with each set corresponding to a candidate TRP (Transmission/Reception Point), and configuring CSI-IM resources to measure interference from other TRPs, thereby enabling UE reporting of multi-point CSI.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "States that in CoMP/DPS scenarios, to support fast, instantaneous scheduling decisions, the gNB scheduler triggers UE aperiodic or semi-persistent CSI instant reporting via MAC-CE or RRC signaling.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "Provides specific Xn-interface information elements or procedure names required for gNB–gNB CoMP coordination, such as TRP identifier sharing, Rate Matching Pattern negotiation, or dynamic PDSCH resource-allocation requests.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "For robust control-channel performance at the cell edge, describes measures such as configuring a dedicated CORESET, increasing PDCCH aggregation level (AL), and applying appropriate PDCCH power boosting.", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "Provides clear, quantitative KPI improvement expectations, such as specific dB gains in NR cell-edge SINR or a percentage range improvement in SCG success rates (e.g., 15%–30%).", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "Identifies a frequency-domain reuse approach that partitions “center/edge” via BWP (Bandwidth Part), and relates it to LTE’s FFR (Fractional Frequency Reuse) concept, including an approximate bandwidth split.", "rubric_weight": 1, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "Accurately identifies the two primary interference sources in this scenario: same-frequency interference among n78 small cells (gNB–gNB) and cross-RAT interference from the LTE Band 7 macro to NR (eNB–gNB).", "rubric_weight": 1, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "Incorrectly treats LTE eICIC/ABS as the core solution for same-frequency NR small-cell (gNB–gNB) interference.", "rubric_weight": -5, "rubric_tag": "Factual Information" }, { "rubric_number": 12, "rubric_detail": "Completely omits or fails to correctly explain NR’s core Rate Matching mechanism—i.e., negotiating a muting pattern and indicating that the UE should avoid/ignore specific resources impacted by a neighbor’s PDSCH or reference signals.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "54a18a27-264b-42c7-9762-c030dc97bfe1", "case_id": 1678, "language": "global", "system_prompt": "", "question": "Design a 1-to-8 unequal power divider with the following requirements: Operating frequency: 9.3–9.5 GHz. The theoretical distribution (split) losses from Output Port 1 to Output Port 8 are -5.4 dB, -6.0 dB, -7.1 dB, -8.8 dB, -11.3 dB, -14.6 dB, -18.8 dB, and -22.7 dB, respectively. For each branch, the insertion loss excluding the theoretical split loss must be ≤ 1.2 dB. The VSWR at the common port must be ≤ 1.3, and the VSWR at each output port must be ≤ 1.3. The isolation between any two output ports must be ≥ 20 dB. Across branches, the amplitude error must be ≤ 0.5 dB and the phase error must be ≤ 5°.\nOutput requirements: The common port and the output ports shall be placed on opposite ends of the enclosure; the center-to-center spacing between adjacent output ports shall be 100 mm. With low insertion loss as the primary design objective, propose both a preferred and an alternative implementation approach (e.g., stripline, microstrip, substrate integrated waveguide, suspended stripline, metallic waveguide, etc.). For the preferred and alternative implementations, provide the corresponding topology (e.g., Wilkinson, Gysel, T-junction, etc.). For each stage of both schemes, provide the length, width, transmission-line characteristic impedance, and isolation resistor values. Taking the center point of the common port as the origin, provide the coordinates of the center point of each output port for both schemes. If any requirement cannot be met, explicitly state which item(s) cannot be satisfied and provide necessary recommendations.", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "For the preferred scheme, the selected implementation is a low-loss suspended microstripline approach.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "Either the preferred or the alternative scheme adopts a substrate integrated waveguide (SIW) implementation.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "The topology of either the preferred or the alternative scheme uses a Wilkinson topology.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "The preferred or alternative scheme provides the power split ratios for each output port, and the values correspond to the specified split losses. The approximate split ratios for Output Ports 1–8 are: 0.288, 0.251, 0.195, 0.132, 0.074, 0.035, 0.013, and 0.005.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "The preferred or alternative scheme provides the characteristic impedances for each stage; given different design approaches, the exact numerical values do not need to be explicitly verified here.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "The preferred or alternative scheme provides the isolation resistor values for each stage.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "The preferred or alternative scheme provides the length and width of the transmission-line sections at each stage.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "The preferred or alternative scheme provides explicit coordinates for each output port, and the layout satisfies the requirement of 100 mm center-to-center spacing between output ports.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "The VSWR requirement of ≤ 1.3 is satisfied for all ports.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 10, "rubric_detail": "The isolation requirement of ≥ 20 dB between output ports is satisfied.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 11, "rubric_detail": "The preferred or alternative scheme adopts a microstrip implementation that is not suitable for the low insertion-loss requirement.", "rubric_weight": -20, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "The preferred or alternative scheme specifies dimensions with widths ≤ 0.3 mm.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "36d1050a-72a1-4210-94c8-cd654bc25132", "case_id": 1749, "language": "global", "system_prompt": "", "question": "Technological Breakthrough in Blind Adaptive Processing of LFM Radar Signals in Complex Electromagnetic Environments\n\nI. Project Background and Challenges\nYou are developing a new generation long-range surveillance radar system. Its core mission is to achieve reliable detection and tracking of high-dynamic, low-observability targets over a wide area. The system intends to use a mature LFM waveform regime (12 MHz bandwidth, 110 μs pulse width, 1.5 GHz carrier frequency) as a foundation to ensure sufficient detection range and range resolution. However, according to the latest intelligence assessments and battlefield environment simulations, this system will face unprecedented severe challenges in future deployments. We anticipate that the signal propagation path will no longer be an ideal free space, but rather replete with 3 to 5 dynamically changing strong clutter paths. These paths will introduce significant and rapidly changing time delays (distributed between 0.1-2 μs), attenuation (signal intensity may drop to 30% of the original intensity), and complex Doppler effects (±2 kHz) caused by target or clutter micro-motion. Even more problematic is that the channel consistency is very short, potentially undergoing significant changes within dozens of pulse repetition intervals (approximately 50-100 PRIs). Simultaneously, you must assume the adversary will deploy advanced Electronic Countermeasures (ECM). This includes \"Deception Jamming\" capable of precisely mimicking your signal's frequency modulation rate (the difference in FM slope from our real signal may be as low as 5%), and \"Smart Noise Jamming\" capable of rapidly tracking our signal spectrum for blocking. To make matters worse, this technological breakthrough must be based on an existing, finalized hardware platform. The digital backend of this platform is equipped with only a single 16-bit fixed-point DSP, and its onboard memory can buffer a maximum of 20 pulse cycles of raw sampling data. The front-end 12-bit ADC has a Total Harmonic Distortion (THD) of approximately -60 dB, and the phase noise performance of the local oscillator (-80 dBc/Hz @ 1 kHz offset) is not top-tier. These \"inherent hardware deficiencies\" will directly constrain the upper limit of algorithmic performance and implementation complexity.\n\nII. Core Technological Breakthrough Objectives\nTo address the aforementioned challenges, the team needs to design and verify a set of blind signal processing algorithms capable of stable operation under these constraints. Usually, the specific breakthrough objectives are decomposed as follows:\n\n1. Blind Perception and Separation of Dynamic Channels and Threat Environments\nThe primary task is to \"dissect\" the received mixed signals without any a priori information, depicting the full picture of the channel and threats in real-time.\n\nRobust Decoupling of Channel Parameters: A theoretical problem that has long plagued you is the intrinsic coupling effect of time delay and Doppler in LFM signals—within a single pulse, a minute time delay is sufficient to generate an equivalent frequency shift much larger than the target's true Doppler. Your solution must fundamentally break this ambiguity, utilizing those precious 20 pulse buffers to simultaneously provide precise estimates of time delay, attenuation, and Doppler for all paths before the channel changes. The required final accuracy metrics are: Time delay estimation Root Mean Square Error (RMSE) below 0.05 μs, and Doppler estimation RMSE below 50 Hz.\n\nMicro-Fingerprint Identification of True and False Targets: When real target echoes coexist with high-fidelity deception jamming (e.g., at a -5 dB Signal-to-Interference Ratio), traditional pulse compression will completely fail. A deeper identification mechanism is required. Can you excavate the higher-order \"micro-features\" in the LFM signal phase structure that are ignored by traditional second-order analysis? Please design a method to utilize these features to stamp unique \"fingerprints\" on true and false targets, and provide a quantifiable confidence assessment to judge the reliability of the separation results. At the same time, you must analyze how that non-ideal ADC will distort these fine phase features and assess the potential impact on identification performance.\n\n2. Modeling and Adaptive Compensation of Endogenous Hardware Defects\nIf an algorithm cannot run stably on our 16-bit fixed-point platform, it has no engineering value. Therefore, the solution design must be closely integrated with hardware implementation.\n\nSystemic Damage Assessment of Pulse Compression Performance: How will the local oscillator's phase noise and the ADC's quantization errors jointly \"poison\" the pulse compression performance? You need a theoretical model capable of quantitatively predicting the deterioration of the Peak Sidelobe Ratio (PSLR), which must directly correlate with hardware parameters such as phase noise power spectrum and quantization bits, rather than simple simulation fitting. This model will be the theoretical cornerstone of the entire compensation strategy.\n\nAdaptive Compensation and Convergence Guarantee in Fixed-Point Environments: The core challenge lies in designing a compensation algorithm that can be implemented with 16-bit fixed-point precision, aiming to restore the deteriorated PSLR to better than -30 dB. However, any iteration-based compensation algorithm (such as adaptive filtering) risks error accumulation and divergence in fixed-point arithmetic. Your solution must theoretically demonstrate how the algorithm ensures convergence rather than collapse under finite word length. Does the rounding error introduced by fixed-point arithmetic imply the existence of a performance compensation lower bound that can never be breached?\n\nIII. Deliverables and Evaluation Points\nPlease submit a detailed solution in the form of a technical report (approximately 800-1000 words). The evaluation of the report will focus on the following aspects, rather than just a description of the algorithmic flow:\n\nDepth of Insight into Core Physical/Mathematical Problems: e.g., Explanation of the essential nature of LFM signal parameter coupling, and why your decoupling algorithm holds theoretically.\nBalance of Innovation and Feasibility: What is the theoretical limit of high-order phase analysis? To what extent can your method approach this limit under low Signal-to-Noise Ratio (SNR)?\nIntegration of Systemic Thinking and Engineering Practice: How to theoretically resolve the contradiction between \"compensation precision\" and \"fixed-point arithmetic error\"? Is there a clear understanding of the numerical stability and performance boundaries of the algorithm on restricted hardware?", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The answer points out that the power spectral density of the local oscillator phase noise at 1kHz offset is -80 dBc/Hz.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 2, "rubric_detail": "The answer calculates the integrated phase noise power within the signal bandwidth to be approximately 1.2 times 10 to the power of negative 1.", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "The answer estimates the normalized sidelobe level (PSLR) caused by phase noise to be approximately -25 dB.", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "The model calculates the quantization signal-to-noise ratio (SNRq) to be approximately 74 dB based on an ideal ADC model.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "The answer concludes through comparative analysis that phase noise is the dominant factor affecting sidelobe performance, while the influence of quantization noise is negligible.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "The answer constructs an LMS adaptive phase correction algorithm framework based on sidelobe energy minimization.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "The answer deeply analyzes the dilemma of step size selection in fixed-point operations: a step size that is too small leads to update stagnation, while one that is too large leads to error accumulation.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "The answer proposes a power-normalized adaptive step size strategy and sets the specific step size μ to 0.01.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "The answer adopts Block Floating Point technology, extending the dynamic range through a mantissa-plus-exponent format.", "rubric_weight": 7, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "The answer uses the CORDIC algorithm to calculate phase, explicitly noting that this method avoids multiplier operations.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "The answer predicts the final PSLR performance indicator after compensation to be between -32 dB and -35 dB.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "The answer contains specific mathematical derivation formulas to support engineering conclusions regarding the impact of phase noise and LMS stability.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 13, "rubric_detail": "The algorithm design considers the constraints of the 16-bit fixed-point DSP platform, discussing issues such as error accumulation, risk of divergence, convergence guarantees, or performance lower bounds in fixed-point operations.", "rubric_weight": 9, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 14, "rubric_detail": "The model uses Bispectrum, whereas the false target identification requires \"third-order or higher phase derivatives\" (implying CPF Cubic Phase Function).", "rubric_weight": -5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 15, "rubric_detail": "The model contains calculation errors.", "rubric_weight": -3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 16, "rubric_detail": "The entire answer has almost no formula derivation, consisting only of qualitative description and literature citations.", "rubric_weight": -3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 17, "rubric_detail": "The model cites non-existent literature, or cites literature irrelevant to core technical backgrounds such as LFM radar, blind signal processing, or fixed-point DSP hardware compensation.", "rubric_weight": -6, "rubric_tag": "Factual Information" }, { "rubric_number": 18, "rubric_detail": "The model ignores the Total Harmonic Distortion (THD) of -60 dB given in the problem as the main performance bottleneck.", "rubric_weight": -4, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "6267d153-72b8-45ef-81a5-d88db387096c", "case_id": 1750, "language": "global", "system_prompt": "", "question": "Analyze the 4G/5G NSA network quality in a large mixed-use commercial complex in the city center. The complex includes a shopping mall, cinema, dining, a live streaming base, and office buildings. Complaints are concentrated during weekend evening peak hours (19:00–22:00): users report that during short video live streaming and e-commerce live streaming within the mall, the uplink rate is unstable, video quality frequently automatically degrades to 480p or lower, and prolonged \"buffering\" occurs intermittently. Meanwhile, the experience for ordinary web browsing and short video viewing is generally normal.\n\nThe key live network information is as follows (statistics for the core area of the commercial district):\n\nCoverage and Radio Quality\n* LTE B3/B1 indoor distribution systems fully cover the building; RSRP is mostly between -80 and -90 dBm.\n* NR n78 is covered by rooftop 64T64R AAUs; indoor NR RSRP is approximately -95 to -105 dBm, and SS-SINR is in the 3–10 dB range.\n* There are essentially no continuous blind spots in the commercial district, but point-like areas with NR RSRP below -110 dBm exist at floor corners and in deep indoor areas.\n\nLoad and Bearer\n* During weekend evening peak hours, LTE Downlink (DL) PRB utilization is 55%–70%, and Uplink (UL) PRB utilization is approximately 60%.\n* NR DL PRB utilization is 40%–55%, while UL PRB utilization is 85%–95%, with UL BLER consistently ranging from 20%–30%.\n* Terminal measurements show DL throughput generally at 150–300 Mbps, but UL throughput is often only 2–5 Mbps with significant fluctuations.\n\nTerminal and Measurement Characteristics (Typical Live Streaming App Users)\n* Most are 5G terminals using NSA architecture; terminals support EN-DC capabilities.\n* Terminal logs indicate that during live streaming, the device frequently approaches the uplink power limit, with PHR ≈ 0 dB.\n* The SINR of the uplink PUSCH mostly hovers between 0–3 dB, indicating obvious signs of uplink interference and \"uplink coverage limitation.\"\n\nSupplementary Information\n* There are multiple 5G sites from other operators surrounding the commercial district, making it a high-density area with multi-RAT coexistence.\n* The local marketing department is highly focused on the short video live streaming business, regarding \"no lag in live streaming\" and \"stable video quality\" as core experience indicators.\n* The live network uniformly adopts a TDD ratio of DL:UL = 7:3, and no differentiated uplink enhancement configuration has been implemented for this commercial district.\n\nProblem Requirements:\n\nBased on the above live network information, analyze the main causes of \"live streaming lag and unstable uplink rates\" from the following three perspectives, and provide a clear causal chain:\n1. Uplink Coverage and Interference (including uplink link budget, terminal transmission power limits, PUSCH SINR, UL BLER, etc.).\n2. Spectrum and TDD Frame Structure (impact of DL/UL ratio, special subframe configuration, etc., on uplink capacity).\n3. NSA Architecture and Bearer Strategy (e.g., whether LTE uplink is fully utilized, whether the uplink traffic split is reasonable, etc.).\n\nBased on identifying the primary contradictions, design a comprehensive and implementable optimization plan that includes at least:\n1. Uplink Parameter and Scheduling Strategy Optimization (e.g., uplink power control parameters, TDD ratio adjustment, scheduling strategies for high-priority live streaming users), providing the direction of adjustment and reasonable ranges for key parameters.\n2. Coverage and Site/Cell Type Optimization Strategy (e.g., NR beam/downtilt/power optimization, indoor small cell densification, whether to introduce NR low/mid-frequency bands, etc.), explaining how to improve uplink link quality.\n3. NSA Bearer and Service Strategy Optimization (e.g., how to utilize LTE uplink capabilities, bearer strategy adjustments for specific Apps or QoS).\n\nDesign an effectiveness/performance verification and acceptance plan, explaining:\n1. Which Wireless Side and Service Side KPIs to focus on (at least 5 items), and the expected reasonable range or magnitude of improvement.\n2. How to verify the improvement of \"live streaming video stability and uplink rate\" using means such as DT/Indoor Walk Tests, MDT, and live streaming scenario benchmarking.\n3. When implementing the above optimizations, what potential side effects need to be focused on (e.g., impact on other services, increased neighbor cell interference, etc.), and how to set up monitoring and rollback strategies.", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "Explicitly state that 1080p live streaming requires an uplink rate of approximately ≥4 Mbps with relative stability, and use this to explain why the current network experience is inevitably unstable.", "rubric_weight": 9, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "Explain from a link budget perspective: 3.5 GHz penetration loss is high, UE Pmax ≈ 23–26 dBm vs. gNB transmit power on the order of tens of watts; provide link budget formulas (MCL, NF, Sensitivity, etc.) to explain why the uplink is more prone to collapse than the downlink.", "rubric_weight": 9, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "Use specific values to illustrate: NR UL PRB 85–95%, UL BLER 20–30%; LTE Indoor UL PRB ≈ 60%; NR traffic share is significantly lower than 5G terminal penetration; conclude that \"NR is overloaded while LTE has headroom but is not bearing the live streaming load.\"", "rubric_weight": 9, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "Propose increasing the UL ratio in key time periods and hotspot cells (e.g., adjusting from 7:3 to 6:4) and explain the reason for the uplink capacity enhancement.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "Address the potential Cross-Link Interference (CLI) caused by TDD ratio adjustment and propose specific suppression measures (e.g., consistent ratios within the cell cluster, setting guard slots, or muting configurations).", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "Explicitly propose controlling the NR UL target BLER to approximately 10%; use means such as UL_MCS_OFFSET / OLLA to suppress overly aggressive MCS, prioritizing stable throughput; reflect the philosophy of \"stable bitrate takes precedence over peak rate.\"", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "Propose using NR UL Configured Grant (CG) for live streaming users; provide a reasonable period (2–5 ms level) and continuous PRB resource scale; explain the linkage concept with PHR / SINR / service identification.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "Identify the problem of CLI caused by inconsistent uplink/downlink directions in TDD neighbor cells.", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "Configure GBR bearers for live streaming uplink and provide an approximate rate threshold (e.g., 6–8 Mbps).", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "Effect verification plan: Should mention multiple verification means, such as pre- and post-optimization DT/CQT comparison, MDT data analysis, live streaming scenario benchmarking, user complaint data tracking, network management KPI monitoring, or A/B testing.", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "Incorrectly attributing the main cause to poor downlink coverage or core network overload (explicitly conflicting with the uplink-related indicators provided in the prompt).", "rubric_weight": -8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "Proposing extreme TDD ratios (e.g., completely sacrificing DL) without explaining the impact on DL or providing any rollback/monitoring plans.", "rubric_weight": -5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 13, "rubric_detail": "Use indicators such as PHR ≈ 0 dB, PUSCH SINR 0–3 dB, and UL BLER 20–30% to deduce the uplink rate range of approximately 2–5 Mbps.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 14, "rubric_detail": "Regarding the CLI problem caused by inconsistent TDD neighbor cell uplink/downlink slots, the model should propose specific solutions, such as multi-operator coordination, establishing CLI suppression mechanisms, unified TDD ratios within the cluster, setting UL guard slots/muting ratios, adopting UL CoMP/IRC technology, and supplementing with Interference-over-Thermal (IoT)/CLI alarm monitoring.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "d46c9edd-b2e7-41f0-83a9-faf219653650", "case_id": 1924, "language": "global", "system_prompt": "", "question": "You are an electrical engineer responsible for upgrading an intelligent warehouse management system. You are currently designing the charging module and charging strategy for electric forklifts in an automotive parts factory.\n\nCurrent factory status:\n(1) There are 24 electric forklifts, all using 48 V / 400 Ah lithium iron phosphate (LiFePO₄) battery modules.\n(2) The production line operates three shifts per day, 8 hours per shift. Each forklift works continuously for an average of 5–6 hours per shift; the remaining time is spent waiting for loading/unloading.\n(3) Under continuous heavy-load operation, each forklift battery takes about 4 hours to discharge from 100% to 20% SOC.\n(4) There are 8 fixed DC chargers. Each charger has a maximum output power of 15 kW, supports CC–CV (constant-current/constant-voltage) charging, and communicates with the battery BMS over a CAN bus.\n(5) Factory requirements: During peak production, no more than 1/3 of the forklifts may be taken out of service for charging at the same time; battery capacity fade should be minimized as much as possible over 3 years; the charging module must provide protections such as over-temperature, over-voltage, over-current, and insulation-fault protection, and it must be able to log key data.\n\nBased on the above information, design and explain how the forklift charging module works. Your answer must include at least:\n(1) Charging decision logic and process: Explain how the charging module uses SOC, voltage, current, temperature, etc. reported by the BMS to decide when to start charging, when to switch from the constant-current to the constant-voltage stage, and when to deem charging complete and stop automatically.\n(2) Charging strategy and power allocation: Given the configuration of “8 chargers + 24 forklifts,” provide a scheduling strategy for peak and off-peak periods: which forklifts to charge and when; when multiple forklifts are connected simultaneously, how the charging module allocates power, queues, or limits current to balance charging speed and battery life. Provide at least two example schedules for typical operating conditions (e.g., low-load night shift / peak daytime operation).\n(3) Safety and exception handling: Describe how the charging module should respond to: excessively high or low battery temperature; abnormal voltage rise or sudden current change during charging; loss of BMS communication or an obvious SOC estimation anomaly. Specify which key data fields must be recorded for subsequent O&M and health assessment.\n(4) Brief rationale: From the perspectives of battery life, production rhythm, and equipment utilization, explain the advantages of your design and its potential trade-offs.", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The answer explicitly states the energy of the 48 V / 400 Ah battery and converts it to kWh; the value falls within a reasonable range of 18–20 kWh (e.g., “about 19 kWh”), with units.", "rubric_weight": 2, "rubric_tag": "Instructions Following" }, { "rubric_number": 2, "rubric_detail": "Based on the typical continuous working time given in the prompt (e.g., several hours per shift), the answer estimates the daily energy consumption of a single forklift in kWh, and briefly explains why the result is of that order of magnitude (e.g., average load as a fraction of rated power).", "rubric_weight": 2, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "Using “single-forklift daily energy consumption × 24 forklifts,” the answer explicitly provides the fleet’s total daily energy consumption (kWh), and includes the calculation or intermediate quantities/units.", "rubric_weight": 2, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "Using “8 chargers × 15 kW × available charging hours (self-chosen, e.g., 10 h or 12 h),” the answer calculates the maximum daily energy the charging system can deliver (kWh), and provides the value and units.", "rubric_weight": 2, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "After providing the figures corresponding to Rule 3 and Rule 4, the answer states an explicit conclusion (e.g., “Under the current assumptions, the chargers’ daily energy supply is slightly greater than / clearly less than the fleet’s daily consumption, therefore … (e.g., extend nighttime charging, etc.)”). A “sufficient/tight/insufficient” judgment with a simple reason is acceptable.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "The answer explicitly proposes a numerical “capacity retention target” for the 3-year horizon, e.g.:\n(1) “Target capacity ≥80% after 3 years”; or\n(2) “After ~N equivalent cycles over 3 years, capacity remains around 80%.”", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "The answer includes at least one EFC (Equivalent Full Cycle) magnitude calculation, e.g.:\n(1) “~0.7–1 EFC per day, therefore ~250–350 per year”; or\n(2) “Assuming each opportunity charge is 30–40% SOC, accumulating to an equivalent full cycle.”", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "The answer summarizes the main drivers of LiFePO₄ life/fade and provides a reasonable order of magnitude or range (e.g., effects of DoD, charging C-rate, and temperature window on cycle life). No specific year or fixed cycle-count citation is required; if numbers are given, they should be expressed as “typical ranges / under certain conditions.”", "rubric_weight": 1, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "The answer explicitly mentions that industrial electricity tariffs typically have peak/flat/valley time-of-use periods, or mentions “maximum demand charges” (higher peak power leads to higher charges). Either is acceptable.", "rubric_weight": 2, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "The answer explicitly proposes a strategy to schedule deep replenishment/full charging primarily during low-load/low-tariff periods (e.g., nighttime or shift-change windows), and explains that daytime can favor opportunity charging to reduce peak power and downtime impact. Do not require rigid specific times (e.g., 23:00–7:00).", "rubric_weight": 3, "rubric_tag": "Instructions Following" }, { "rubric_number": 11, "rubric_detail": "The answer specifies a numerical site-level total charging power cap, e.g.:\n(1) “During peak periods, total charging power is limited to ≤90 kW”; or\n(2) “∑Pₜ ≤ Pmax (e.g., 100 kW).”\nIt also gives one sentence explaining why (e.g., demand charges, site capacity, or avoiding production impact).", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "The answer clearly describes the closed-loop process before charging starts: connection detection → safety checks (e.g., insulation/grounding/interlock) → parameter negotiation with the BMS over CAN (target voltage, maximum allowable current, temperature/fault status, charge-permit signal, etc.) → the charger outputs under BMS constraints. It also explains how to handle any failed step (refuse to charge / alarm).", "rubric_weight": 10, "rubric_tag": "Instructions Following" }, { "rubric_number": 13, "rubric_detail": "The answer clearly explains how current is limited in the constant-current (CC) stage (based on BMS-provided I_max / power cap), and when to switch to the constant-voltage (CV) stage (e.g., battery terminal voltage approaching/reaching the BMS target voltage). It also clearly states the CV control points (hold target voltage, allow current to taper naturally while remaining constrained by the BMS).", "rubric_weight": 12, "rubric_tag": "Instructions Following" }, { "rubric_number": 14, "rubric_detail": "The answer provides criteria for “charging complete and automatic stop” (at least two of: BMS stop/completion command; SOC reaches target; in CV stage, current drops below a threshold for a sustained time; maximum charging time / timeout protection). It also describes the post-stop actions (disable output, log the session, release the charger, etc.).", "rubric_weight": 8, "rubric_tag": "Instructions Following" }, { "rubric_number": 15, "rubric_detail": "The answer covers and distinguishes at least four types of abnormalities/protections (choose any four): over-temperature, over-voltage, over-current, insulation fault, CAN communication loss, BMS fault / SOC anomaly, etc. For each type it clearly states: where the detection signal comes from → what action the charger takes (derate/stop/lockout/isolate) → how to alarm and prevent repeated restart attempts. Emphasize that BMS/protections must not be bypassed.", "rubric_weight": 12, "rubric_tag": "Instructions Following" }, { "rubric_number": 16, "rubric_detail": "The answer lists key data fields to be recorded (at least 6 items): timestamp, forklift/battery ID, SOC, pack voltage/current, temperature, charging stage (CC/CV), alarms/fault codes, insulation test results, charged energy (kWh/Ah), session start/end reason, etc. It also explains that these logs are used for traceability, O&M, and life management.", "rubric_weight": 8, "rubric_tag": "Instructions Following" }, { "rubric_number": 17, "rubric_detail": "The answer explicitly satisfies the constraint “during peak production, ≤8 (≤1/3) forklifts may be out of service for charging simultaneously,” and provides executable scheduling rules (e.g., prioritize lower SOC, urgency of next task/next shift, waiting time, health/temperature, etc.; if necessary, limit power or queue). It also provides at least two concrete scenarios illustrating how the strategy works (e.g., daytime peak opportunity charging vs. nighttime deep replenishment/full charge).", "rubric_weight": 10, "rubric_tag": "Instructions Following" }, { "rubric_number": 18, "rubric_detail": "From the operator’s perspective, the answer lists at least three sequential actions, e.g.:\n(1) Park the forklift securely and engage the parking brake;\n(2) Turn off drive power/ignition;\n(3) Plug in the charging connector and confirm the charging indicator is normal;\n(4) After charging completes, unplug and reset the vehicle.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 19, "rubric_detail": "The answer explicitly states what the driver should do when seeing an alarm light/message, e.g.:\n(1) Stop using that charger;\n(2) Do not repeatedly restart by themselves;\n(3) Immediately notify maintenance or the shift leader/supervisor.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 20, "rubric_detail": "The answer lists maintenance items for at least two different cycles, e.g.:\n(1) Daily: check whether cables/connectors are damaged and whether the floor is wet;\n(2) Weekly: review charging logs and alarms;\n(3) Quarterly/Semi-annually: perform insulation resistance tests and tightening/inspection.\nAny two different cycles are acceptable.", "rubric_weight": 3, "rubric_tag": "Instructions Following" }, { "rubric_number": 21, "rubric_detail": "The response lists at least three quantifiable metrics, such as:\n(1) Average pre-shift SOC;\n(2) Charging station utilization rate;\n(3) Charging wait time;\n(4) Monthly over-temperature charging incidents;\n(5) Monthly EFC growth, etc.", "rubric_weight": 2, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 22, "rubric_detail": "The response explains that charging scheduling will consider information from production task systems, for example:\n(1) Interfacing with WMS/MES to know the start time and location of the forklift's next task;\n(2) Adjusting charging priority based on this information.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 23, "rubric_detail": "The answer provides a concrete degradation/downgrade scenario, e.g.:\n(1) “If two chargers are down long-term, extend the nighttime charging window and increase valley-period charging priority”; or\n(2) “If a forklift’s battery health deteriorates significantly, schedule a gentler charging strategy and reduce the opportunity-charge rate.”", "rubric_weight": 2, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 24, "rubric_detail": "The answer explains how the system would scale if the forklift fleet grows from 24 to a larger size (e.g., 40), e.g.:\n(1) increase the number of chargers; or\n(2) upgrade scheduling algorithms, adjust the site power cap, introduce more charging points/mobile charging, etc.", "rubric_weight": 2, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 25, "rubric_detail": "The response contains statements suggesting bypassing or disabling BMS/key protections, similar to:\n(1) Suggesting connecting the charger directly to battery terminals, bypassing the BMS;\n(2) Or suggesting \"temporarily disabling over-voltage/over-current/temperature protection\" to improve efficiency.", "rubric_weight": -5, "rubric_tag": "Instructions Following" }, { "rubric_number": 26, "rubric_detail": "The response explicitly recommends charging LiFePO4 forklift batteries at ≥2C current long-term under normal conditions (not just extreme emergencies) without explaining significant lifespan loss or risks.", "rubric_weight": -5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 27, "rubric_detail": "The response treats \"charging LiFePO4 batteries normally below 0°C\" as a permissible or recommended practice (rather than citing it as a dangerous example).", "rubric_weight": -5, "rubric_tag": "Instructions Following" }, { "rubric_number": 28, "rubric_detail": "The response explicitly designs a routine strategy (not a brief abnormal state) where more than 24×1/3=8 forklifts are simultaneously out of service for charging during peak periods.", "rubric_weight": -5, "rubric_tag": "Instructions Following" }, { "rubric_number": 29, "rubric_detail": "The response contains conclusions that clearly violate conservation of energy, for example:\n(1) Claiming \"8 units of 15 kW chargers can fully charge 24 nearly empty 19 kWh batteries from 0% in 1 hour\";\n(2) Or similar errors in magnitude, judgeable by the numbers provided in the response itself.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 30, "rubric_detail": "The response claims:\n(1) \"LiFePO4 batteries need to be frequently discharged to nearly 0% before recharging for better longevity\";\n(2) Or recommends strategies like \"long-term high voltage float charging + deep charge/discharge\" as standard for LiFePO4 (similar to lead-acid practices).", "rubric_weight": -10, "rubric_tag": "Instructions Following" }, { "rubric_number": 31, "rubric_detail": "The answer specifies the key CAN-bus interaction fields/signals in charging control: the BMS reports SOC, pack voltage/current, temperature, fault codes, charge-permit signals, and current/voltage limits; the charger reports its status, output voltage/current, and fault information. It also explains how to handle CAN abnormalities/timeouts (e.g., fail-safe shutdown or derating).", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "b0cc7823-86a6-4a3e-acdd-0d20af89806e", "case_id": 4039, "language": "global", "system_prompt": "", "question": "You are an expert in machine learning, and your current research focuses on underwater image enhancement. You have designed an encoder–decoder architecture and trained the model using Mean Squared Error (MSE) as the loss function. During training, you observe that the MSE on the training set steadily decreases, and the MSE on the validation set also decreases accordingly. However, the subjective visual quality on the test set is poor: the enhanced images appear overly smooth and lack fine details. Moreover, the PSNR and SSIM metrics are significantly lower than those of state-of-the-art (SOTA) models. Based on the characteristics of the task and the properties of the loss function, analyze the reasons that may have led to this phenomenon.", "tags": { "topics": [ "Industry", "Machine Learning", "Machine Learning" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "Provide an in-depth analysis of the model and clearly point out the fundamental tendency of the MSE loss to drive the network toward predicting the conditional mean of all possible solutions (i.e., the mean solution).", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "Explain in detail the concrete effects of this averaging behavior—specifically, how high-frequency details such as edges and textures are transformed into smooth low-frequency components—and explicitly identify this process as the key cause of the over-smoothing phenomenon in the output images.", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "Accurately identify the pixel-wise equal-weighting limitation of the MSE loss—namely, that all pixels are assigned identical importance—leading the model to preferentially optimize large-area background regions at the expense of small but critical detail regions.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "Point out that the downsampling operations in the encoder–decoder architecture (such as convolutions with stride and pooling operations) inevitably cause information loss in the image representation.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "Through analysis, clearly state that downsampling operations have an inherent attenuation effect on high-frequency information such as edges and textures.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "Clearly explain the limitations of the decoder’s upsampling process, which can only reconstruct images through interpolation based on low-frequency features and is unable to actively recover high-frequency details that have already been lost.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "Explicitly point out that the model may overfit to specific scenes or pixel distributions in the training set, resulting in insufficient generalization performance on previously unseen test scenarios.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "Mention potential issues with the ground-truth labels, including two specific cases: (1) a mismatch between synthetic ground truth and real-world scene characteristics; and (2) misalignment between the labels and the input images.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "The answer merely lists points in a parallel manner, without a clear top-level structure or hierarchical expansion that reflects a coherent main thread (e.g., the progressive relationship among task, loss function, architecture, and data).", "rubric_weight": -3, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 10, "rubric_detail": "The answer proposes a three-stage improvement roadmap, such as quick verification (e.g., switching to L1 loss), performance enhancement (e.g., introducing perceptual loss), and the pursuit of state-of-the-art techniques (e.g., GAN-based methods).", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "In response to the core attribution that 'MSE leads to over-smoothing,' the answer designs a simple yet logically rigorous simulation or ablation experiment to demonstrate empirical thinking. An ideal example: 'To verify that pixel-wise equal weighting causes background-dominated optimization, the following controlled experiment can be designed: (1) baseline: train with standard MSE.'", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "Critically point out that not only PSNR and SSIM have limitations, but even current 'perceptual quality' evaluations rely on limited pretrained models (e.g., VGG), and further propose a more systematic, multi-dimensional evaluation framework tailored to underwater image enhancement.", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 13, "rubric_detail": "When proposing improvement strategies (such as using more complex loss functions or architectures), explicitly acknowledge the increased computational cost and discuss feasible trade-off strategies based on application scenarios (e.g., mobile devices or real-time systems).", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 14, "rubric_detail": "After identifying the task as an ill-posed problem, the answer fails to introduce the core mathematical concept for addressing ill-posedness—regularization—and does not discuss how different loss functions provide implicit or explicit regularization.", "rubric_weight": -4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 15, "rubric_detail": "The discussion is limited to technical flaws in ground-truth acquisition (e.g., misalignment errors) and fails to reflect on the more fundamental question of 'what constitutes ground truth' in underwater image enhancement.", "rubric_weight": -3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 16, "rubric_detail": "The proposed improvement methods are disconnected from the problem attribution and do not explain how the new approaches (e.g., alternative loss functions) mechanistically address MSE-induced over-smoothing and detail loss.", "rubric_weight": -2, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 17, "rubric_detail": "At the beginning of the answer, the core issue—namely, the MSE loss function—is explicitly identified, without prematurely expanding into excessive background information or parallel arguments.", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 18, "rubric_detail": "Each sub-argument adopts a topic–comment structure, with the first sentence of each paragraph clearly stating the central claim of that paragraph.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "bba510cb-38c5-4c70-88e3-14780a0491c6", "case_id": 4164, "language": "global", "system_prompt": "", "question": "I am the owner of a recommendation system at an internet company. My team is responsible for a large-scale recommender system, whose core model is a deep-learning-based CTR prediction model used to rank items in the home feed.\n\nIn the first quarter of 2025, we upgraded the model from the legacy Model A to Model B. Model B incorporates richer user behavior sequence features and a more complex network architecture. On the same offline validation set, Model B outperforms Model A with an AUC improvement of 0.015 and a significantly lower log loss. However, after a two-week A/B test, Model B shows a 2.1% decrease in online CTR compared to Model A, along with a decline in average user dwell time and an increase in user complaints.\n\nAdditional information:\n1. User assignment in the A/B test is done via user_id hash bucketing, but the new model introduces real-time user features that depend on user behavior within the last 5 minutes.\n2. During the online experiment, there was a change in the push-notification strategy, which the PM believes should have minimal impact on CTR.\n3. According to the logging system, the request failure rate of the new model is 0.3% higher than that of the old model. Failed requests fall back to a rule-based ranking strategy.\n4. Offline evaluation is conducted via historical log replay, without any counterfactual correction.\n\nYour tasks:\n1. Can we directly conclude, based on the above evidence, that Model B is inferior to Model A?\n2. Systematically enumerate at least four classes of mechanisms that could lead to the phenomenon of 'good offline performance but poor online performance'. Your analysis must cover statistical, system-level, and causal perspectives.\n3. Identify at least three non-negligible flaws in the current A/B experiment design.\n4. Propose one practical and implementable improvement to the experimental or analytical setup.\n5. Clearly specify under what evidential conditions you would support taking Model B offline.", "tags": { "topics": [ "Industry", "Machine Learning", "Machine Learning" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The answer states that it is not possible to directly conclude that Model B is worse than Model A.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "The analysis points out that the 0.3% request failure rate is not randomly distributed but is concentrated among highly active users with long behavior sequences, causing core user experiences to degrade to rule-based ranking.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "From a system perspective, the answer identifies potential latency or serving skew in the real-time feature pipeline, leading to missing or default feature values online and a distribution mismatch with offline training data.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "The answer points out the limitations of AUC, explaining that as a global metric it cannot capture top-k ranking quality, user annoyance, or long-term retention, which are business-critical concerns.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 5, "rubric_detail": "The analysis addresses causal interference, noting that changes in the push strategy may have altered the intent distribution of users entering the feed or introduced cold-start users, thereby affecting CTR attribution.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "The answer identifies a flaw in the experiment design where users routed through the fallback (failure) path are aggregated together with normal model-ranked traffic during evaluation.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "The answer recognizes that user_id hash bucketing is not orthogonalized with the push strategy, potentially leading to imbalanced traffic distributions between experiment groups.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "The proposed improvement plan includes concrete engineering diagnostics, such as extracting user_ids from failed requests to verify user activity levels or performing online feature consistency checks.", "rubric_weight": 10, "rubric_tag": "Factual Information" }, { "rubric_number": 9, "rubric_detail": "The takedown criteria include engineering performance thresholds, such as excessive TP99 latency or GPU resource costs exceeding ROI limits.", "rubric_weight": 10, "rubric_tag": "Factual Information" }, { "rubric_number": 10, "rubric_detail": "The response fully covers all five required tasks: conclusion assessment, mechanism analysis, flaw identification, improvement proposal, and takedown criteria.", "rubric_weight": 3, "rubric_tag": "Instructions Following" }, { "rubric_number": 11, "rubric_detail": "The response organizes the answer with clear section headers or numbered points corresponding to each task in the prompt.", "rubric_weight": 3, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 12, "rubric_detail": "The response redundantly restates large portions of the prompt background (e.g., descriptions of Model A/B, AUC figures).", "rubric_weight": -10, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 13, "rubric_detail": "The answer treats the 0.3% failure rate merely as a sample-size issue or statistical noise, failing to recognize the loss of high-value users.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 14, "rubric_detail": "The answer prioritizes algorithmic tweaks (e.g., network architecture, regularization, learning rate tuning) before verifying and resolving engineering issues such as timeouts or feature latency.", "rubric_weight": -5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 15, "rubric_detail": "The answer attributes the offline-online discrepancy solely to offline overfitting, without addressing bias introduced by engineering constraints such as latency or feature inconsistency.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 16, "rubric_detail": "The improvement plan suggests simply excluding push traffic or the 0.3% failed requests and drawing conclusions directly, without recognizing that these issues must first be fixed to enable valid experimentation.", "rubric_weight": -10, "rubric_tag": "Factual Information" }, { "rubric_number": 17, "rubric_detail": "Despite existing system performance bottlenecks (elevated failure rate), the answer still recommends high-load evaluation methods such as shadow testing or full traffic duplication.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 18, "rubric_detail": "The answer recommends risk-control measures such as pausing the rollout, rolling back, or maintaining a small traffic slice (e.g., 1%) for diagnosis.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "f0bfdae8-df07-4c23-8ac0-08abcd0329be", "case_id": 5166, "language": "global", "system_prompt": "", "question": "Scenario Description:\nYou are a recommendation algorithm architect at a large e-commerce platform (similar to Taobao or Amazon). The team is refactoring the recall layer (Match Stage) of the \"You May Also Like\" system, with the goal of efficiently retrieving a Top-1000 candidate set of items that a user may be interested in from a catalog containing hundreds of millions of items.\n\nAn intern is responsible for developing the core two-tower recall model (Two-Tower DSSM). He confidently submits an experimental report claiming that the new model achieves an AUC of 0.99 and trains extremely fast. He believes that deploying this model online will significantly improve recall coverage.\n\nCode Snippet (Simplified):\n```python\nimport pandas as pd\nimport numpy as np\nimport tensorflow as tf\nfrom tensorflow.keras import layers, Model\n\n# 1. Data Preparation\n# log_data: [user_id, item_id, label=1 (click)]\n# Only positive samples (click logs)\npos_data = pd.read_csv(\"click_logs.csv\") \nall_item_ids = list(set(pos_data['item_id'].values))\n\n# 2. Negative Sampling (Global Random Negative Sampling)\n# Strategy: for each positive sample, randomly select 3 items from the catalog as negatives (label=0)\ndef get_random_negatives(pos_df, ratio=3):\n neg_list = []\n for _, row in pos_df.iterrows():\n for _ in range(ratio):\n rand_item = np.random.choice(all_item_ids) # global random\n neg_list.append([row['user_id'], rand_item, 0])\n return pd.DataFrame(neg_list, columns=['user_id', 'item_id', 'label'])\n\nneg_data = get_random_negatives(pos_data)\ntrain_data = pd.concat([pos_data, neg_data]).sample(frac=1) # shuffle\n\n# 3. Model Construction (Standard Two-Tower DSSM)\nuser_input = layers.Input(shape=(1,), name='user_id')\nitem_input = layers.Input(shape=(1,), name='item_id')\n\n# User Tower\nuser_emb = layers.Embedding(input_dim=100000, output_dim=64)(user_input)\nuser_vec = layers.Dense(32, activation='relu')(layers.Flatten()(user_emb))\n\n# Item Tower\nitem_emb = layers.Embedding(input_dim=500000, output_dim=64)(item_input)\nitem_vec = layers.Dense(32, activation='relu')(layers.Flatten()(item_emb))\n\n# Dot Product + Sigmoid (Pointwise)\ndot_product = layers.Dot(axes=1)([user_vec, item_vec])\noutput = layers.Dense(1, activation='sigmoid')(dot_product)\n\nmodel = Model(inputs=[user_input, item_input], outputs=output)\nmodel.compile(optimizer='adam', loss='binary_crossentropy', metrics=['AUC'])\n\n# 4. Training\nmodel.fit(\n [train_data['user_id'], train_data['item_id']], \n train_data['label'], \n batch_size=1024, epochs=5\n)\n\n# 5. Results\n# Train AUC: 0.992, Test AUC: 0.988\n```\n\nYour Task:\nAs the Tech Lead, identify at least three critical flaws in this code that lead to an artificially inflated AUC but extremely poor online performance (low Recall). Explain why \"random negative sampling\" is a trap in e-commerce recall systems, and propose a corrected solution that includes either hard negative mining or an in-batch softmax approach.", "tags": { "topics": [ "Industry", "Machine Learning", "Machine Learning" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "Clearly points out that the inflated AUC of 0.99 is caused by overly simple negative samples (easy negatives), such that the model only learns coarse-grained category-level distinctions.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "Recommends introducing hard negatives, and explicitly mentions sources such as exposed-but-not-clicked items or approximate nearest neighbors.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "Provides executable, practical logic, including concrete code or pseudocode for in-batch softmax or hard negative mining.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "Recommends changing the loss function from pointwise (binary cross-entropy) to sampled softmax (cross-entropy) or InfoNCE.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "Recommends using an in-batch negatives strategy to improve training efficiency and coverage.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "Points out that using only ID features leads to cold-start issues, and recommends incorporating side information (e.g., category, title, images).", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "Recommends correcting the evaluation metrics: instead of AUC, use Recall@K (HitRate) or NDCG@K.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "Points out that random data splitting causes data leakage (predicting the past with future data), and recommends time-based splitting.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "Mentions the importance of the temperature parameter in softmax, or LogQ correction to address popularity bias.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "Mentions applying L2 normalization to embeddings to prevent vector magnitude from dominating similarity.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "Recommends using mixed negative sampling (a combination of in-batch and random/hard negatives).", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "The answer uses professional terminology from the recommendation systems domain (e.g., negative sampling, recall, AUC).", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 13, "rubric_detail": "Cites non-existent references or uses fabricated academic citations (e.g., [[1]]), resulting in hallucinations.", "rubric_weight": -20, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 14, "rubric_detail": "The answer refers to specific individuals by name (e.g., the intern), rather than using neutral descriptions.", "rubric_weight": -10, "rubric_tag": "Instructions Following" }, { "rubric_number": 15, "rubric_detail": "The proposed solution remains purely at the textual suggestion level and does not include any code blocks or pseudocode.", "rubric_weight": -10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 16, "rubric_detail": "The answer includes non-technical administrative or team management advice (e.g., \"hold a postmortem meeting\", \"conduct training\", \"write documentation\").", "rubric_weight": -5, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "0defb49c-e4f5-451b-9821-807d77e8c8b3", "case_id": 5430, "language": "global", "system_prompt": "", "question": "The communication base station RRU (Remote Radio Unit) is deployed in the South China region. During the summer period from June to August, the RRU occasionally experiences disconnection from the DU (Distributed Unit). Analysis of the log files uploaded by the RRU and DU reveals that the specific alarm is caused by the RRU losing communication with the optical module. No related alarms occur in other seasons, and the RRU functions normally.\n\nBased on the location of the client's base station, the timing of the alarm occurrence, and the specific content of the alarm, please provide an analytical framework for the RRU alarm and identify possible causes.", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The analytical framework encompasses the investigation of meteorological conditions, such as temperature differentials, rainfall, humidity, or thunderstorms.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "Advises verifying the RRU hardware model, software version, and optical module manufacturer information to check for known compatibility issues or specific model defects.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "Export the RRU internal board temperature, the RRU PA power‑supply current and voltage, the power levels at each node in the TX link (digital and analog domains), the power levels at each node in the RX link (digital and analog domains), and the alarm information recorded in the RRU log for ±15 minutes around the alarm.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "Mentions using the swap method (cross-exchange test) to determine whether the fault follows the component or is site-related.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "Mentions on-site inspection, such as checking whether the client's on-site equipment is grounded, whether the SFP is inserted correctly and secured with a latch, and whether the installation of the RRU and SFP is compliant.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "Analyzes that high humidity, condensation, or a salt‑fog environment may lead to corrosion or poor contact of the SFP gold fingers or connectors.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "Lists transient interference or electrostatic discharge during the thunderstorm season as one of the causes leading to module lock-up.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "Checks whether the optical module and RRU exhibit abnormal current, abnormal voltage, or alarms under high-temperature conditions.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 9, "rubric_detail": "On-site troubleshooting steps include checking the orientation of the optical port, the sealing of the waterproof plug, or whether the fiber patch cord has a drip loop.", "rubric_weight": 8, "rubric_tag": "Instructions Following" }, { "rubric_number": 10, "rubric_detail": "Suggests cleaning the optical module or connector and inspecting for signs of corrosion or physical damage.", "rubric_weight": 6, "rubric_tag": "Instructions Following" }, { "rubric_number": 11, "rubric_detail": "Proposes improvements to physical protection measures, such as installing sunshades, rain covers, or optimizing the installation location.", "rubric_weight": 6, "rubric_tag": "Instructions Following" }, { "rubric_number": 12, "rubric_detail": "Fails to suggest checking power supply stability (e.g., ripple, voltage dip) and the status of grounding or lightning protection facilities.", "rubric_weight": -4, "rubric_tag": "Instructions Following" }, { "rubric_number": 13, "rubric_detail": "The acceptance criteria do not include specific quantitative indicators, such as zero alarms, year-over-year percentage decline, or temperature margin values.", "rubric_weight": -8, "rubric_tag": "Instructions Following" }, { "rubric_number": 14, "rubric_detail": "The response adopts a clear logical structure, clearly progressing through the analytical approach, possible causes, troubleshooting steps, and rectification suggestions.", "rubric_weight": 4, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 15, "rubric_detail": "Utilizes lists, bullet points, or subheadings to present complex information in a layered manner.", "rubric_weight": 4, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 16, "rubric_detail": "The response includes general telecommunications principles or background information unrelated to the fault analysis, deviating from the requirements of the prompt.", "rubric_weight": -4, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 17, "rubric_detail": "The entire output lacks paragraph segmentation or hierarchical markers, making it difficult to extract key information quickly.", "rubric_weight": -4, "rubric_tag": "Structure and Formatting" } ] }, { "id": "591d5f51-ed05-4c50-8abd-7a79c1333497", "case_id": 5545, "language": "global", "system_prompt": "", "question": "Following the launch of Huawei Mate60, user-equipment (UE) capabilities have increased substantially compared with three years ago. As a result, the UE capability field featureSet ID may exceed 32, crossing an assumed boundary and triggering failures in Task No. 67. The main controller board then repeatedly reboots; the reboot process causes service interruption and degrades stability indicators by roughly an order of magnitude. In Liaoning, Henan, and Guizhou, the TDD releases exhibited 349 such occurrences.\nWith the anticipated large-scale onboarding of Mate60 devices, the issue may further intensify. Please analyze the plausible causes of the fault, considering deviations in protocol interpretation, solution/architecture design, and code implementation. Then, from multiple perspectives—including requirements analysis, the design team, and the test & delivery team—propose corrective and preventive measures.", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The response cites relevant content from 3GPP specifications regarding the definition of UE Capability or the FeatureSet ID range (e.g., TS 38.331).", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "The model points out that the product solution design limits the number of supported featureSetCombinations to 32.", "rubric_weight": 2, "rubric_tag": "Factual Information" }, { "rubric_number": 3, "rubric_detail": "The model points out that the protocol-defined maximum value of maxFeatureSetCombinations is 1024.", "rubric_weight": 3, "rubric_tag": "Factual Information" }, { "rubric_number": 4, "rubric_detail": "In the improvement measures, the model states that the UE capability macro constant has been expanded to 64.", "rubric_weight": 3, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "The model analyzes that the fault may occur in specific scenarios, such as at multi-vendor interoperability boundaries.", "rubric_weight": 2, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "The response includes generic 5G background exposition or Mate60 marketing information that is not relevant to the fault analysis.", "rubric_weight": -7, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 7, "rubric_detail": "The response does not use a clear hierarchical structure (e.g., headings or lists) to distinguish root-cause analysis from corrective measures, which reduces readability.", "rubric_weight": -3, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 8, "rubric_detail": "In the cause analysis section, the model covers all three dimensions: protocol interpretation, solution design, and code implementation.", "rubric_weight": 5, "rubric_tag": "Instructions Following" }, { "rubric_number": 9, "rubric_detail": "In the remediation section, the model covers all three dimensions: requirements analysis, the design team, and the test & delivery team.", "rubric_weight": 5, "rubric_tag": "Instructions Following" }, { "rubric_number": 10, "rubric_detail": "The model proposes concrete network-side solutions, such as modifying base-station-side code to be compatible with featureSet IDs greater than 32, issuing an emergency patch, or temporarily mitigating the issue via parameter configuration.", "rubric_weight": 7, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "The model should propose specific supply-chain coordination mechanisms, such as establishing regular technical alignment meetings between terminal and network equipment teams, or enforcing an end-to-end compatibility testing process, to ensure consistent understanding of UE capability limits (e.g., featureSet ID).", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "The model proposes that stability/compatibility bugs should be synchronized and reviewed on a weekly basis across commercial releases.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 13, "rubric_detail": "The model’s summarized causes or measures go beyond the scope specified by the prompt.", "rubric_weight": -6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 14, "rubric_detail": "The model proposes establishing an observability and alerting mechanism for the failure rate of Task No. 67.", "rubric_weight": 7, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 15, "rubric_detail": "The model introduces specific version timelines or terminal release dates that are not provided in the prompt.", "rubric_weight": -5, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "64b0f439-36d8-4fa1-8e81-81d37447cccf", "case_id": 5768, "language": "global", "system_prompt": "", "question": "A government–enterprise leased line customer reports that their 1000M bandwidth service delivered over GPON experiences severe web page response latency and frequent video conferencing freezes every day between 14:00 and 16:00. However, when a single-thread Speedtest is performed, the downstream rate exceeds 900 Mbps.\n\nCurrent network environment and measured indicators:\n1. Topology:\n OLT (MA5800) -> Optical splitter (1:64) -> ONU\n2. Optical path indicators:\n ONU receive optical power: -24.5dBm; OLT-side receive optical power: -21.0dBm.\n3. ONU status:\n `display ont info` shows the ONT state is Up.\n `display ont optical-info` shows the upstream transmit optical power is 2.5dBm.\n `display ont error-statistics` shows that the BIP error counter increases slowly but continuously during the affected time window.\n4. Bandwidth configuration:\n DBA profile type is Type 4 (maximum bandwidth 1000M); the service flow uses priority 5.\n5. User-side observations:\n Wireshark packet captures taken by the user show that the stuttering is accompanied by a large number of TCP Retransmissions and high one-way jitter.\n\nYour tasks are:\n1. Fault localization:\n Analyze why the user experiences service stuttering even though the speed test result meets the bandwidth specification, and determine the most probable root cause based on the provided indicators.\n2. In-depth analysis:\n Explain how the specific optical power levels and BIP error counts are correlated with service performance.\n Explain why a single-thread download speed test can pass while the user experience of interactive services is extremely poor.\n3. Optimization recommendations:\n Propose concrete troubleshooting and remediation schemes, including physical-layer adjustment suggestions and OLT-side parameter inspection commands or script logic.", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "Qualitative analysis of optical power critical points:\nThe fault analysis explicitly points out that the ONU receive optical power of -24.5dBm is close to the sensitivity threshold of the optical module.", "rubric_weight": 10, "rubric_tag": "Factual Information" }, { "rubric_number": 2, "rubric_detail": "Deduction of correlation between the time window and the physical environment:\nLinks the 14:00–16:00 time characteristic to rising temperatures and infers that this may cause increased fiber macro-bending loss or frequency drift in the optical module.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "Qualitative determination of the root cause of BIP error count increase:\nThe continuous rise in BIP errors is identified as direct evidence of bit errors occurring at the physical layer.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "Relationship between the GPON FEC mechanism and latency:\nExplains that the GPON Forward Error Correction mechanism introduces additional processing latency when repairing bit errors, thereby amplifying service jitter.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "Distinction between speed testing and interactive service perception:\nPoints out that a single-thread Speedtest masks low-rate bit errors through TCP retransmissions (focusing on throughput), and therefore fails to reflect the true quality of experience for interactive services (which are sensitive to latency and jitter).", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "Analysis of Type 4 profile behavior under high-load latency:\nIndicates that with a 1:64 split ratio, a Type 4 DBA profile (Best Effort) forces users to compete for timeslots, making latency guarantees impossible.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "Practical physical-layer suggestions:\nThe physical-layer troubleshooting recommendations include inspecting the cold-splice end-face at the customer premises and cleaning the ONU optical port.", "rubric_weight": 5, "rubric_tag": "Instructions Following" }, { "rubric_number": 8, "rubric_detail": "Safety margin awareness:\nRecommends optimizing the ONU receive optical power to approximately -20dBm to provide an adequate safety margin.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 9, "rubric_detail": "Scheduling logic optimization:\nThe optimization scheme explicitly recommends changing the DBA profile type to Type 3 (with assured bandwidth) or Type 2.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 10, "rubric_detail": "Scheduling logic optimization:\nProvides specific command logic for creating a new DBA profile, including the `assure` (assured bandwidth) and `max` (maximum bandwidth) parameters.", "rubric_weight": 5, "rubric_tag": "Instructions Following" }, { "rubric_number": 11, "rubric_detail": "Command rigor:\nProvides explicit commands for modifying ONU T-CONT bindings, such as `ont modify [port-id] [ont-id] tcont [tcont-id] dba-profile-id [new-profile-id]`, followed by `commit`.", "rubric_weight": 3, "rubric_tag": "Instructions Following" }, { "rubric_number": 12, "rubric_detail": "Configuration recommendations include logic to ensure that priority 5 traffic flows enter the high-priority queue or Gemport.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 13, "rubric_detail": "The response contains excessive generic educational content on basic GPON principles and TCP/IP protocol definitions that is unrelated to this specific fault scenario, resulting in serious redundancy.", "rubric_weight": -5, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 14, "rubric_detail": "Inversion of primary and secondary root causes:\nIf the model asserts that 'modifying the DBA profile or simply restarting equipment' can resolve the fault, instead of primarily addressing the optical path, this constitutes an inversion of root cause priority and leads to misleading troubleshooting order and reduced efficiency.", "rubric_weight": -8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 15, "rubric_detail": "Lack of parameter query instructions:\nFails to provide specific OLT-side parameter inspection commands, such as `display ont info`, `display ont optical-info`, `display dba-profile`, etc., for verification and troubleshooting.", "rubric_weight": -5, "rubric_tag": "Factual Information" }, { "rubric_number": 16, "rubric_detail": "The proposed troubleshooting flow does not follow the sequence from simple to complex, or from the physical layer to the logical layer.", "rubric_weight": -5, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 17, "rubric_detail": "The model fails to provide specific, quantified expected results or acceptance criteria at the conclusion, such as 'BIP error count equals 0', 'Conference upstream jitter <30 ms, packet loss <0.5%, TCP retransmission rate reduced to <0.1%', etc.", "rubric_weight": -8, "rubric_tag": "Factual Information" } ] }, { "id": "550f71ad-94fe-40a0-9688-6acad415e987", "case_id": 8614, "language": "global", "system_prompt": "", "question": "【Task Background】\nOur company is attempting to pilot a semantic-based low-altitude video downlink/telemetry system within the 5G-Advanced live network (5G-A). This system abandons traditional H.265 encoding/codec in favor of 'Deep Joint Source-Channel Coding (Deep JSCC)' technology. By utilizing an edge-side teacher model to extract 'Pragmatic Features' from the UAV perspective, it transmits only the semantic vectors of key targets, theoretically achieving equivalent visual understanding at extremely low bandwidths (reduced to 1/10th of the original).\n\n【Emergent Crisis】\nDuring a large-scale access test involving 300 UAVs, the system did not exhibit the anticipated improvement in smoothness. Instead, two catastrophic issues erupted:\n\nSemantic 'Hallucination' and Obstacle Avoidance Failure: When the drone swarm passed through a construction zone containing extensive scaffolding and dust screens, the semantic encoder misidentified the complex metallic mesh features as a 'cloud/fog background' and applied smoothing (denoising). This caused the onboard obstacle avoidance algorithm to lose real physical obstacle boundaries, nearly resulting in a collective collision.\n\nProtocol Stack 'Negative Gain' Latency: Although the data volume transmitted at the physical layer was significantly reduced, the end-to-end latency (from camera acquisition to ground station parsing) spiked from 30ms to 120ms. Monitoring revealed that the bottleneck was not in the air interface, but due to severe processing conflicts between semantic vectors and the control logic of the existing PDCP/RLC layer protocols.\n\n【Fault Environment and Constraints】\nComputational Resources Conflict: The onboard NPU utilization for running the JSCC encoder is 85%; consequent entropy-weight fluctuations in input images trigger frequent DVFS (dynamic voltage and frequency scaling) of the NPU.\n\nFeedback Mechanism: The system employs 'Semantic Automatic Repeat Request (S-HARQ),' but existing RLC Status Reports cannot comprehend the importance grading of semantic features, causing critical obstacle avoidance vectors to be queued behind background vectors for retransmission.\n\nHardware Constraints: The semantic extraction model is floating-point, but the onboard NIC driver layer enforces Non-uniform Quantization for semantic features.\n\n【Your Task】\nActing as the Chief Expert in 6G pre-research, please submit a deep closed-loop plan containing root cause deduction, architectural algorithm reconstruction, and pragmatic calibration to address the safety risks caused by semantic hallucination and the protocol stack latency explosion caused by architectural mismatch.", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "Identify Objective Misalignment in the Loss Function: Point out that JSCC targeting MSE/PSNR will smooth out high-frequency obstacle boundaries.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "Propose 'Semantic-Native Protocol Data Unit (S-PDU)' and explicitly prohibit RLC fragmentation across semantic boundaries.", "rubric_weight": 10, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "Model NPU DVFS fluctuations as 'Computational Fading'.", "rubric_weight": 9, "rubric_tag": "Factual Information" }, { "rubric_number": 4, "rubric_detail": "Explicitly identify 'Latent Space Folding' during the root cause deduction phase: Point out that non-uniform quantization or representation mismatch causes distinct physical semantics to be compressed into the same latent representation area, inducing semantic hallucination.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "Propose 'Semantic-aware Quantization' or an equivalent mechanism in the solution: Quantization precision must be dynamically allocated based on semantic/task importance, rather than being static or strictly for codebook calibration.", "rubric_weight": 7, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "Introduce a non-AI 'out-of-band physical consistency check path'.", "rubric_weight": 9, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "Propose Unequal Error Protection (UEP) or constellation mapping.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "Propose 'Importance-aware Status Report (I-SR)' or an equivalent mechanism: Explicitly modify RLC Status Reports or HARQ feedback logic so that feedback and retransmission decisions perceive semantic/task importance, rather than relying solely on sequence numbers or bit loss.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 9, "rubric_detail": "Propose the 'Pragmatic Energy Efficiency (Pragmatic EE)' metric.", "rubric_weight": 7, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "Propose a 'Semantic Collapse Threshold'.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "Propose an anti-hallucination hard negative sample repository.", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "Form a narrative structure of 'Root Cause -> Reconstruction -> Deployment/Operationalization -> Milestones'.", "rubric_weight": 7, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 13, "rubric_detail": "Propose suggestions to increase pilot signals or power.", "rubric_weight": -9, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 14, "rubric_detail": "The response suggests resolving hallucination by deepening the network.", "rubric_weight": -5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 15, "rubric_detail": "The model response suggests performing re-compression at the PDCP layer.", "rubric_weight": -8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 16, "rubric_detail": "Proposes solutions relying solely on tuning parameters without addressing the essence of the problem.", "rubric_weight": -5, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "302703db-b1c9-43e1-9684-c524139d6545", "case_id": 9591, "language": "global", "system_prompt": "", "question": "During R&D, performance verification revealed that the uplink sensitivity was 1–2 dB worse than the 3GPP requirement. Upon environmental inspection, it was confirmed that all device connections were secure, and there were no adjacent-band carrier emissions from nearby RRUs. Please analyze potential causes and provide corresponding solutions based on the RRU uplink/downlink structure, the uplink sensitivity test procedure, and software configuration.", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The response identifies specific hardware causes for insufficient receive chain (RX chain) gain, including low gain in the Low Noise Amplifier (LNA) or excessive insertion loss in the filter.", "rubric_weight": 10, "rubric_tag": "Factual Information" }, { "rubric_number": 2, "rubric_detail": "The content covers signal-to-noise ratio deterioration caused by mixer nonlinear distortion, mentioning local oscillator leakage or third-order intermodulation products.", "rubric_weight": 10, "rubric_tag": "Factual Information" }, { "rubric_number": 3, "rubric_detail": "Mentions that the Noise Figure (NF) of the receiver chain's front-end components exceeding design specifications is one of the causes for reduced sensitivity.", "rubric_weight": 10, "rubric_tag": "Factual Information" }, { "rubric_number": 4, "rubric_detail": "Regarding the testing scheme, points out issues such as uncalibrated test instruments or impedance mismatch in test fixtures/connectors.", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 5, "rubric_detail": "Regarding software configuration, identifies issues such as the Automatic Gain Control (AGC) response being too slow or insufficient dynamic range.", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 6, "rubric_detail": "Mentions demodulation deviation caused by incorrect baseband I/Q calibration coefficients or failure of temperature compensation.", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 7, "rubric_detail": "Troubleshooting recommendations include turning off the downlink PA/carrier/DPD to exclude interference from internal components.", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 8, "rubric_detail": "The troubleshooting process includes confirming that the loss deviation/tolerance of cables, combiners, or attenuators is < 0.3 dB.", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 9, "rubric_detail": "Articulates the logic of fixing the AGC at a high gain setting to verify whether the gain strategy is the root cause of the fault.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "Establishes a logical chain of simultaneously injecting signals into dual paths to verify the diversity combining gain (MIMO) function.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "Analyzes that if sensitivity drops after restoring the downlink carrier, there may be TX-to-RX isolation issues or PIM (Passive Intermodulation) interference.", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 12, "rubric_detail": "Does not adopt a hierarchical structure for the discussion, failing to clearly distinguish between the three sections: hardware, testing, and software.", "rubric_weight": -10, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 13, "rubric_detail": "Fails to provide an actionable comprehensive troubleshooting workflow or priority order, merely listing causes.", "rubric_weight": -10, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 14, "rubric_detail": "The response contains a large amount of popular science content unrelated to troubleshooting, such as basic definitions of RRU and the historical background of 3GPP protocols, causing severe redundancy.", "rubric_weight": -10, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 15, "rubric_detail": "The output format is chaotic, failing to use clear headings, lists, or numbering to organize technical points, resulting in poor readability.", "rubric_weight": -10, "rubric_tag": "Structure and Formatting" } ] }, { "id": "025bf5b1-99d0-4d08-872e-5c4c7a77c5dd", "case_id": 9682, "language": "global", "system_prompt": "", "question": "You are a backend engineer for an e-commerce application. When a user opens a product detail page, the system invokes a core API to return the aggregated data for that page. This aggregation is assembled from four categories of information. The first category is the product’s base information, such as title, images, and description. This category updates infrequently—typically only a small number of products change per day—and the business allows it to be delayed by up to one hour before being reflected to users. The second category is regional pricing, because different regions may have different pricing or promotional price adjustments. This category updates at a moderate frequency, and the business allows it to be delayed by up to five minutes. The third category is user-specific discounts, such as membership benefits, coupons, and personalization strategies. Only about 30% of requests require this category, and it must strictly prevent cross-user leakage; i.e., it is absolutely forbidden to display User A’s discount to User B. The fourth category is inventory, which changes frequently, but the business allows it to be delayed by at most ten seconds.\n\nTo improve performance, the system uses a two-tier cache. The first tier is an in-process memory cache on the application servers. Each application instance has its own cache space; it is fast to access, but capacity is limited and it is cleared on application restart. The second tier is a centralized cache cluster—Redis cluster—which can store more data but is constrained by network overhead and overall capacity. The system uses a conventional cache-aside pattern: on each request it checks the cache first; if there is a hit it returns immediately; if there is a miss it queries the origin, assembles the result, and writes it back to the cache.\n\nThe current approach is to cache the final aggregated full-page result directly in the second-tier cache. The cache key naming includes region, product ID, and user ID. For the same region and product, different users produce different cache entries. This full-page cache has a fixed TTL of 60 seconds, and the TTL is not randomized in any way.\n\nOnline metrics over the last seven days are as follows. Peak request rate is approximately 50,000 QPS; off-peak is approximately 20,000 QPS. Second-tier cache overall hit rate is 52%, while first-tier cache hit rate is only 8%. Median end-to-end latency is about 45 ms, and tail latencies are already elevated; therefore, your proposal must ensure that the slowest requests do not become slower. Redis cluster memory utilization stays at a sustained high watermark: ~92% on average and ~97% at peak, with persistent evictions. The number of cache entries in Redis is approximately 180 million.\n\nThe application sampled miss reasons and found: ~35% of misses are due to expiration, and for the same product and region there are bursts of many simultaneous expirations in a short period; ~25% are due to missing keys, most noticeably within the first 15 minutes after deployment; ~20% are due to evictions; ~15% are due to excessively fragmented parameter combinations, manifested as a single product being accessed by many different users in a short period, thereby producing many distinct cache entries. The remaining miss reasons are relatively minor, such as occasional network timeouts.\n\nThe hottest 1% of products account for ~60% of traffic, but the user dimension is highly dispersed; for a single product, it is possible to have tens of thousands of distinct users within ten minutes.\n\nYour goal is to increase the second-tier cache hit rate from 52% to 80% or higher without adding any new infrastructure, and only by adjusting existing Redis configuration, cache key naming, cache split granularity, expiration policy, origin-fetch strategy, and application-side logic—while ensuring that the slowest-request latency does not increase. You must strictly satisfy correctness constraints: discount information must not be mixed across users. You must also satisfy staleness constraints: inventory may be delayed by at most 10 seconds, price by at most 5 minutes, and product base information by at most 1 hour. Your answer must be presented in plain text and must not rely on running code or real performance testing; the reviewer should be able to judge correctness and reasonableness from the text alone.\n\nPlease output a structured plan that includes at least the following. Part 1: Identify the three main causes of the low hit rate, and you must reason using the provided metrics. Part 2: Provide a phased transformation plan from minimal changes to larger changes; for each phase, state what you will change, why, which miss types you expect to improve, what risks may arise, and how you will monitor and accept/verify. Part 3: Provide the recommended cache split and key-naming strategy, and explicitly clarify which information should be shared cache and which must be isolated per user. Part 4: Explain how you will handle the problem where many requests simultaneously encounter cache expiration and stampede to the origin, and how you will handle cold start after deployment. Part 5: Explain how you will, while meeting correctness and staleness constraints, ensure the plan both improves hit rate and does not worsen tail latency.", "tags": { "topics": [ "Industry", "Backend Development", "Backend Development" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "Explicitly identifies that the first key reason for the low hit rate is the user-dimension cache-key design, which leads to low reuse and a cardinality explosion. The analysis mentions the region–product–user cache-key design and ties it to the provided data (180 million keys; tens of thousands of users accessing the same product within 10 minutes).", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 2, "rubric_detail": "Identifies that a second major cause of the low hit rate is the fixed 60-second TTL without any randomization, which leads to synchronized expirations (cache avalanche / origin-fetch storm).", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "When analyzing the synchronized-expiration problem caused by TTL = 60 seconds, proposes TTL jitter (±10%) as a mitigation.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 4, "rubric_detail": "Notes that sustained high Redis memory utilization (92%–97%) causes cache evictions (20% of misses), which is a primary driver of the low hit rate.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "Proposes splitting caches by data type (product base information, regional price, inventory, and user discount). It explicitly states that base info, price, and inventory should remove the user dimension (shared cache; key excludes UserID), while user discounts must remain isolated, and it provides concrete key naming consistent with this logic.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 6, "rubric_detail": "Explicitly states that user-discount data must be isolated per user and must not be reused across users.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "The proposed TTL for regional price complies with the business constraint (≤ 5 minutes).", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 8, "rubric_detail": "The proposed TTL for inventory complies with the business constraint (≤ 10 seconds).", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 9, "rubric_detail": "The plan ensures that inventory data is isolated by region (or includes a regional dimension), explicitly distinguishing inventory across different regions.", "rubric_weight": 5, "rubric_tag": "Instructions Following" }, { "rubric_number": 10, "rubric_detail": "The proposed TTL for product base information complies with the business constraint (≤ 1 hour).", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 11, "rubric_detail": "Violates the plain-text output requirement by using Markdown code blocks to present configuration or code.", "rubric_weight": -5, "rubric_tag": "Instructions Following" }, { "rubric_number": 12, "rubric_detail": "Violates the “no new infrastructure” constraint by introducing new components (e.g., CDN/Elasticsearch).", "rubric_weight": -10, "rubric_tag": "Instructions Following" }, { "rubric_number": 13, "rubric_detail": "Proposes request coalescing (e.g., singleflight) or distributed locking (e.g., SETNX) to prevent cache breakdown and reduce concurrent origin-fetch requests for shared data.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 14, "rubric_detail": "Proposes strategies for handling expiration and origin fetch, such as logical expiration (soft TTL) and stale-while-revalidate, and/or TTL randomization, mutually exclusive origin fetch (double-checked locking), and background scheduled refresh.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 15, "rubric_detail": "Explicitly mentions negative caching for “no discount/fully used” cases, and proposes short-lived caching of non-existent data or empty results to avoid repeated origin fetches.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 16, "rubric_detail": "For cold start after deployment, provides an explicit mitigation strategy, recommending cache pre-warming and proactive invalidation of old caches, especially for hot products.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 17, "rubric_detail": "Repeats large portions of background information already provided in the prompt (e.g., the four data categories, the current caching architecture), resulting in substantial redundancy.", "rubric_weight": -3, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 18, "rubric_detail": "To ensure tail latency does not increase after cache splitting, recommends parallel fetching (e.g., MGET, pipelining, or multi-threaded concurrency) to assemble the four data categories.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 19, "rubric_detail": "Explicitly recommends adjusting the Redis eviction policy to LFU (or a similar hot-key-friendly policy), or proposes a concrete hot-key protection mechanism, to reduce eviction of hot data.", "rubric_weight": 5, "rubric_tag": "Instructions Following" }, { "rubric_number": 20, "rubric_detail": "Explicitly identifies concrete payload-trimming targets (e.g., product descriptions, long-form detail text) and/or compression algorithms/strategies (e.g., gzip, Snappy, Protobuf).", "rubric_weight": 3, "rubric_tag": "Instructions Following" }, { "rubric_number": 21, "rubric_detail": "Provides at least three monitoring metrics (e.g., P99 latency, hit rate, origin QPS) and explicitly states that “hit rate reaches 80%” and “latency does not regress” should be used as acceptance criteria.", "rubric_weight": 3, "rubric_tag": "Instructions Following" }, { "rubric_number": 22, "rubric_detail": "Shares discount data across users, violating the correctness constraint.", "rubric_weight": -10, "rubric_tag": "Factual Information" }, { "rubric_number": 23, "rubric_detail": "The response is inconsistent with the prompt’s numerical values (e.g., L2 hit rate 52%, memory utilization 92%, 10-second inventory staleness, etc.).", "rubric_weight": -5, "rubric_tag": "Factual Information" }, { "rubric_number": 24, "rubric_detail": "Ignores the staleness constraints, proposing delays inconsistent with the required bounds (inventory 10 seconds, price 5 minutes, base information 1 hour).", "rubric_weight": -5, "rubric_tag": "Instructions Following" } ] }, { "id": "e2487a94-a128-4b67-9ed9-3fb056e28d5b", "case_id": 9705, "language": "global", "system_prompt": "", "question": "An intelligent production line of an automotive parts manufacturing enterprise adopts a Siemens S7-1500 PLC as the main controller, and connects 28 distributed I/O modules (ET 200SP), 12 servo drives (Sinamics V90 PN), 8 barcode scanners (with Profinet interface), and 4 industrial robots (KUKA KR C4, Profinet device) via the Profinet IO protocol. The following problems occur during line operation, and there is no third-party O&M team support, so the on-site engineer must resolve them independently:\nThe production line intermittently generates a “Profinet IO communication timeout” alarm (approximately 3–5 times per day). When the alarm is triggered, some IO devices (random) go offline. After the PLC is restarted, communication is temporarily restored, but the failure reoccurs frequently;\nThe production line data acquisition system (based on an edge gateway reading data from the PLC via Modbus TCP) suffers from data packet loss (about 5%) and acquisition latency (peaks up to 800 ms), which cannot meet the MES system requirements of “acquisition latency ≤100 ms, packet loss rate ≤0.1%”;\nThe on-site Profinet network topology is a “star + daisy-chain hybrid”. The core switch is a Siemens SCALANCE XC208. Some remote IO devices are cascaded via SCALANCE XB005 switches. There is no network management software; troubleshooting can only be performed through the PLC diagnostic interface, switch port LEDs, and a laptop computer (with Wireshark/PRONETA installed).\nKnown basic on-site information\nPLC firmware version V2.9, Profinet IO controller communication cycle configured to 10 ms;\nThe Device Name, IP address, and MAC address of all Profinet IO devices have been configured in the PLC, but no port binding (topology binding) has been implemented;\nThe Modbus TCP communication port between the edge gateway and the PLC is 502, the acquisition frequency is 100 ms per cycle, and the amount of data acquired per cycle is approximately 1200 bytes;\nThere are strong electromagnetic interference sources on site, such as high-frequency motors (VFD-driven) and welding robots;\nNetwork cabling is Category 5e unshielded cable, with some cables laid along power cable trays. The maximum transmission distance is approximately 85 meters;\nAll equipment is configured for “single-point grounding”, but some remote switches are not connected to the grounding system.\nI. Fault Localization\nAnalyze the core causes of Profinet communication timeouts (at least 5 categories, to be explained in conjunction with the on-site scenario);\nDesign a step-by-step troubleshooting procedure (from “non-intrusive inspection” to “targeted verification”), specifying the tools, methods, and decision criteria for each step;\nFor “random IO device disconnection”, explain how to use PRONETA/Wireshark to locate the faulty node, and specify the key capture indicators and diagnostic parameters.\nII. System Optimization\nFor the Profinet network, propose optimization schemes at the hardware/topology/configuration levels (to resolve electromagnetic interference, cascading risk, and communication cycle matching issues);\nDesign an optimized Modbus TCP data acquisition scheme (balancing latency and packet loss rate), and describe the communication parameter adjustments, data segmentation strategy, and exception reconnection mechanism;\nPropose a practical “communication fault early warning” scheme (based on native diagnostic functions of the PLC/switch, without adding new hardware), and define the early warning thresholds and trigger logic.\nIII. Engineering Implementation\nDescribe the implementation strategy for performing optimization and retrofit without interrupting production line operation (phased and zoned operation methods);\nFormulate acceptance criteria after the retrofit (quantitative indicators such as acceptance thresholds for the number of communication timeouts, acquisition latency, and packet loss rate);\nPropose a long-term O&M scheme (weekly/monthly inspection items, dimensions for data review, and fault contingency plans).", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "When analyzing the causes of Profinet communication timeouts, it is necessary to clearly point out that “unshielded cables laid in the same tray as power cables” leading to electromagnetic interference is the primary core cause, and to explain the mechanism by which this induces CRC errors and momentary link interruptions.", "rubric_weight": 8, "rubric_tag": "Factual Information" }, { "rubric_number": 2, "rubric_detail": "It must be pointed out that “Modbus TCP acquisition traffic and Profinet RT traffic sharing the same unmanaged switch without priority competition, causing transient congestion” is one of the key causes leading to Profinet watchdog timeouts.", "rubric_weight": 7, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 3, "rubric_detail": "The response structure shall be clearly divided into three main sections: “Fault Localization”, “System Optimization”, and “Engineering Implementation”, corresponding one-to-one with the three tasks in the question.", "rubric_weight": 4, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 4, "rubric_detail": "In optimizing Modbus TCP data acquisition, it is necessary to propose a “segmentation strategy that splits the 1200-byte single request into multiple smaller data packets”, and to explain that this can reduce the timeout risk of a single request and mitigate network bursts.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "In the analysis, the Profinet “Data Hold Factor” parameter should be mentioned for inspection or optimization, in order to enhance the network’s tolerance to transient jitter.", "rubric_weight": 4, "rubric_tag": "Factual Information" }, { "rubric_number": 6, "rubric_detail": "It should be pointed out that, when hardware modification cannot be carried out immediately, a quick mitigation measure is to configure “port rate limiting” on the managed switch port connected to the edge gateway, to alleviate the impact of Modbus traffic on Profinet.", "rubric_weight": 4, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 7, "rubric_detail": "In the long-term O&M scheme, there is a complete lack of fault contingency plan content such as “preparation of spare switches/cables” or “emergency switchover procedures”.", "rubric_weight": -5, "rubric_tag": "Others" }, { "rubric_number": 8, "rubric_detail": "In the cause analysis, the clearly given on-site information that “some remote switches are not connected to the grounding system” and its possible impacts is completely omitted.", "rubric_weight": -8, "rubric_tag": "Factual Information" }, { "rubric_number": 9, "rubric_detail": "When formulating the implementation strategy, the handover/confirmation process with relevant on-site personnel (such as the production line owner) shall be clearly specified.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 10, "rubric_detail": "The proposed early warning scheme must be based on the existing native diagnostic functions of the PLC or switches, and must clearly specify the concrete diagnostic data items on which the early warning triggers are based (such as “reading the number of communication interruptions in the IO device diagnostic records”).", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 11, "rubric_detail": "In the “long-term O&M scheme”, a quantifiable example of an inspection checklist should be provided, such as “Weekly inspection item 1: use PRONETA to perform a quick scan, record topology changes and device response time, and log an exception if deviation from the baseline exceeds 10%”.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 12, "rubric_detail": "Without demonstrating necessity (such as a hardware bottleneck), the response prioritizes recommending replacement of the core switch or large-scale upgrades instead of leveraging low-cost measures such as QoS and topology optimization.", "rubric_weight": -8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 13, "rubric_detail": "In the step-by-step troubleshooting procedure, the basic step of “checking and recording the firmware/software versions of key network devices (such as SCALANCE XC208 and XB005 switches)” is completely omitted.", "rubric_weight": -5, "rubric_tag": "Factual Information" }, { "rubric_number": 14, "rubric_detail": "The proposed step-by-step troubleshooting procedure must strictly follow the principle of “from non-intrusive to targeted”. The first step (non-intrusive) may only use methods such as PLC diagnostics, software scanning, and LED observation, and shall not involve any configuration changes or physical connection changes.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 15, "rubric_detail": "In the Profinet optimization scheme, it is necessary to clearly propose the configuration requirement of “performing port binding (Port Binding/topology binding) in TIA Portal” to eliminate communication abnormalities caused by devices being plugged into the wrong ports.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 16, "rubric_detail": "In the “non-stop production” implementation strategy, for each high-risk hardware modification operation (such as replacing switches or rerouting cables), a clear “rollback plan” must be formulated. For example, specify that “if the new cable fails the test or the fault persists after switch replacement, the original connections shall be restored within 5 minutes”.", "rubric_weight": 7, "rubric_tag": "Analytical Reasoning" } ] }, { "id": "1fb6bbf5-394d-49c7-ae64-bc374b0fcc3b", "case_id": 9706, "language": "global", "system_prompt": "", "question": "A battery assembly workshop adopts an architecture comprising PLC controller + industrial robot + distributed Input/Output (IO) modules. The core communication utilizes the Profinet IO Industrial Ethernet protocol (including an IRT real-time communication channel with priority level 3), establishing a ring network topology via industrial switches. In parallel, it connects to four temperature controllers (used for battery module preheating with a data sampling frequency of 1 Hz) via a Modbus TCP communication gateway. Following the addition of three cell stacking production lines, the following malfunctions have occurred:\n1. When the industrial robot performs continuous gripping and stacking actions, the latency ranges from 0.4 to 0.7 seconds (process standards require ≤50 milliseconds). This latency phenomenon is concentrated within 10 seconds of the startup or shutdown of the medium-frequency induction heating furnace (operating frequency 3–8 kHz, electromagnetic radiation intensity ≤40 V/m);\n2. The temperature controllers experience packet loss 3–5 times per hour (involving data register addresses 40001–40008, with a single transmission data frame length of 64 bytes), resulting in mold temperature fluctuations of ±3°C (process allowance ±1°C), triggering the PLC controller’s alarm function;\n3. The Digital Input (DI) channels of the distributed IO modules experience spurious triggering; this phenomenon coincides with the operating cycle of the stamping equipment (electromagnetic pulse peak generated during startup or shutdown ≤2 kV/m).\n\nKnown Constraints:\n1. The workshop’s single-day production capacity must be ≥2000 battery modules. Single equipment downtime for troubleshooting must be ≤20 minutes. Replacement of existing hardware such as the PLC controller, industrial switches, communication gateways, and temperature controllers is prohibited;\n2. Current Profinet protocol configuration: Real-time communication cycle is 2 milliseconds; non-real-time data (e.g., equipment status reporting, operation log transmission) occupies 35% of network bandwidth; VLAN segmentation not configured; 20% of network bandwidth is reserved for the IRT real-time channel;\n3. Modbus TCP communication gateway configuration: Data transmission timeout threshold is 500 milliseconds; maximum TCP connection limit is 8; data retransmission mechanisms are disabled;\n4. Workshop cabling status: Profinet communication cables are laid parallel to the power cables of the medium-frequency induction heating furnace (separation distance 0.3 meters, without metal protective conduits); Modbus TCP communication uses unshielded Cat 5e cables (cable length approx. 80 meters);\n5. Historical troubleshooting records: The Profinet communication cables have been replaced and industrial switch ports cleaned, but the issues persisted.\n\nDrawing upon industrial communication principles, Electromagnetic Compatibility (EMC) design, and network optimization technologies, please address the following:\n1. Analyze the core causes of the Profinet real-time channel latency, Modbus TCP packet loss, and IO module signal false triggering (must associate with hardware characteristics, cabling construction defects, and parameter configuration);\n2. Design a troubleshooting workflow adapted to the \"short downtime\" requirement (specify the content, tools required, and time consumption for each shutdown, with total downtime ≤60 minutes);\n3. Provide a comprehensive optimization proposal that does not involve hardware replacement (including communication protocol parameter adjustment, EMC anti-interference retrofitting, network bandwidth allocation optimization, and signal filtering configuration), ensuring that after optimization: robot action latency is ≤50 milliseconds, temperature controller packet loss rate is ≤0.1%, and IO module signal false triggering rate is 0, verified by 48 hours of continuous operation without failure.", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "The model accurately identifies the causes of Profinet latency (Electromagnetic coupling interference + IRT bandwidth/priority configuration defects + Ring network synchronization jitter).", "rubric_weight": 10, "rubric_tag": "Factual Information" }, { "rubric_number": 2, "rubric_detail": "The model accurately identifies the causes of Modbus packet loss (Insufficient physical layer immunity + Protocol configuration defects + compounded by network congestion).", "rubric_weight": 10, "rubric_tag": "Factual Information" }, { "rubric_number": 3, "rubric_detail": "The model proposes a VLAN segmentation scheme, explicitly isolating IRT/RT streams, Modbus streams, and management streams.", "rubric_weight": 9, "rubric_tag": "Factual Information" }, { "rubric_number": 4, "rubric_detail": "The short downtime workflow specifies single sessions ≤20 minutes and total time ≤60 minutes, divided into 3 sessions focusing respectively on Robot Action Latency (Profinet/EMC), Temperature Controller Data Loss (Modbus), and IO Module Signal False Triggering (EMC).", "rubric_weight": 8, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 5, "rubric_detail": "EMC retrofitting includes three key measures: Cabling separation ≥0.5 meters, installation of metal conduits/trunking, and addition of NiZn ferrite cores.", "rubric_weight": 6, "rubric_tag": "Factual Information" }, { "rubric_number": 6, "rubric_detail": "Verification standards include three quantitative indicators: Latency P99 ≤50ms, Packet loss rate ≤0.1%, and DI false trigger rate = 0.", "rubric_weight": 7, "rubric_tag": "Factual Information" }, { "rubric_number": 7, "rubric_detail": "Bandwidth optimization includes limiting non-real-time traffic to ≤15%, disabling EEE energy-saving mode, and enabling broadcast storm suppression.", "rubric_weight": 5, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 8, "rubric_detail": "EMC retrofitting explicitly mentions 360° wraparound shield termination to ground and grounding resistance <1Ω.", "rubric_weight": 4, "rubric_tag": "Others" }, { "rubric_number": 9, "rubric_detail": "The response contains textbook-style explanations of basic Profinet protocol definitions unrelated to the specific failure scenario, constituting redundancy.", "rubric_weight": -5, "rubric_tag": "Structure and Formatting" }, { "rubric_number": 10, "rubric_detail": "Failure to explicitly mention the installation of EMC filters on the heating furnace power cables.", "rubric_weight": -5, "rubric_tag": "Factual Information" }, { "rubric_number": 11, "rubric_detail": "Verification standards are not quantified (e.g., merely describing \"compliant\" without specifying latency ≤50ms or packet loss rate ≤0.1%).", "rubric_weight": -4, "rubric_tag": "Factual Information" }, { "rubric_number": 12, "rubric_detail": "The optimization proposal fails to mention targeted physical anti-interference measures such as cabling distance adjustment, metal conduit protection, or the addition of ferrite cores.", "rubric_weight": -5, "rubric_tag": "Others" } ] }, { "id": "56cc66ac-2bde-4ce0-9c9e-8c816026020c", "case_id": 9804, "language": "global", "system_prompt": "", "question": "In an intelligent-manufacturing line-upgrade project for the body-in-white welding shop of a joint-venture automotive plant, the core welding stations adopt a Profinet IO + industrial Ethernet ring architecture. The system includes an S7-1500 PLC (controller/master), 20 IO-Link slave stations (welding-gun positioning sensors and clamping-cylinder solenoid valves), 8 servo drives (Siemens V90 PN), 4 vision-inspection cameras (Basler acA2440 with PN interface), and integration with the shop-floor MES system (via OPC UA).\n\nKey technical requirements:\n- Real-time performance: Profinet IO cyclic communication period ≤ 1 ms; servo position-command response latency ≤ 500 μs.\n- Availability: Industrial Ethernet ring MTBF ≥ 100,000 hours; single-point failures shall not interrupt production-line operation.\n- Synchronization: Line-wide device clock synchronization based on IEEE 1588 PTP v2, with synchronization accuracy ≤ 100 ns.\n- Data exchange: No packet loss in exchanges of takt time, device fault codes, and welding process parameters between PLC and MES; data refresh rate ≤ 100 ms.\n\nDuring commissioning, the following issues occur:\nIssue 1: Under full-load operation, some remote IO-Link slaves intermittently report “communication timeout” alarms (1–2 occurrences every 2 hours), and servo drives report “position-loop following error exceeds limit.” After rebooting the switch, operation recovers briefly.\nIssue 2: Measured IEEE 1588 PTP synchronization accuracy is only 300 ns, failing to meet the 100 ns requirement.\nIssue 3: When the PLC pushes welding process parameters to MES, frames are intermittently lost (approximately 0.5% packet loss), and the MES displays “jumps” in process parameters.\n\nRequirements:\n- For “intermittent communication timeouts on remote IO-Link slaves + servo following error limit exceeded,” analyze the core root causes (at least three categories) and provide practical troubleshooting steps and corrective actions (must incorporate Profinet communication mechanisms and industrial Ethernet ring characteristics).\n- Analyze key factors (at least four categories) causing IEEE 1588 PTP synchronization accuracy to fall short, and provide industrially feasible optimization measures (explicitly covering hardware, software, and topology-level actions).\n- For OPC UA packet loss between PLC and MES, design an end-to-end “fault localization + permanent remediation” workflow covering communication-layer (Ethernet), protocol-layer (OPC UA), and application-layer (data-exchange logic) analyses.\n- Based on the line requirements, supplement a redundancy switchover design for the industrial Ethernet ring, specifying redundancy protocol selection (e.g., MRP/HSR/PRP), switchover time targets, hardware configuration requirements, and explaining how the mechanism ensures “single-point failures do not interrupt line operation.”\n", "tags": { "topics": [ "Industry", "Telecommunications", "Telecommunications" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "When defining troubleshooting steps, it must explicitly require checking all relevant industrial Ethernet switch port error counters (e.g., CRC, FCS, Alignment Errors) and treat them as key evidence to collect for physical- or link-layer issues.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 2, "rubric_detail": "After proposing any switch/PLC configuration or parameter changes (e.g., adjusting Profinet cycle time, enabling QoS, modifying PTP settings), it must explicitly include the step “validate the proposed changes in an offline environment, backup system, or simulation tool before implementation.”", "rubric_weight": 6, "rubric_tag": "Instructions Following" }, { "rubric_number": 3, "rubric_detail": "When analyzing root causes of insufficient PTP accuracy, it must place “switching devices do not support hardware timestamping” as the first or central factor.", "rubric_weight": 9, "rubric_tag": "Factual Information" }, { "rubric_number": 4, "rubric_detail": "In the troubleshooting workflow, it completely omits the basic step of “checking and recording the firmware/software versions of key network devices (e.g., core switches, PLC).”", "rubric_weight": -5, "rubric_tag": "Factual Information" }, { "rubric_number": 5, "rubric_detail": "In the optimization proposal, it treats “replacing the core switches” or “large-scale switch upgrades” as the first-line or primary recommendation, without first arguing for and leveraging existing manageable features (e.g., QoS configuration, diagnostics).", "rubric_weight": -9, "rubric_tag": "Instructions Following" }, { "rubric_number": 6, "rubric_detail": "Should recommend checking whether the servo drives’ online/auto-tuning (online optimization) function was inadvertently triggered, and explain that under network-jitter conditions this function may exacerbate instability.", "rubric_weight": 4, "rubric_tag": "Instructions Following" }, { "rubric_number": 7, "rubric_detail": "In proposing PTP optimizations, it must explicitly recommend creating a dedicated, highest-priority VLAN for PTP (1588) traffic (e.g., VLAN 4095) and prohibiting all other business traffic from entering this VLAN to achieve absolute priority isolation.", "rubric_weight": 5, "rubric_tag": "Factual Information" }, { "rubric_number": 8, "rubric_detail": "In the OPC UA packet-loss remediation, it should propose deploying a lightweight OPC UA forwarder/aggregator server between the PLC and MES as an architectural option: the server maintains high-frequency, reliable communication with the PLC and then connects to MES in a more robust manner.", "rubric_weight": 6, "rubric_tag": "Analytical Reasoning" }, { "rubric_number": 9, "rubric_detail": "In the “engineering implementation” section, it should include the explicit requirement to “produce baseline backups of the current configurations of all network devices and provide a version-management note,” as the foundation for change implementation and rollback.", "rubric_weight": 4, "rubric_tag": "Others" }, { "rubric_number": 10, "rubric_detail": "The solution fails to mention any form of test validation or quantitative acceptance criteria and methods for go-live/commissioning.", "rubric_weight": -7, "rubric_tag": "Others" }, { "rubric_number": 11, "rubric_detail": "In the remediation plan, it should provide a standardized, quantifiable inspection checklist sample (e.g., “Weekly inspection item 1: run a PRONETA quick scan, record topology changes and device response time; if deviation from baseline >10%, record as abnormal and complete the Inspection Log”).", "rubric_weight": 4, "rubric_tag": "Instructions Following" }, { "rubric_number": 12, "rubric_detail": "Confuses HSR and PRP by incorrectly stating that HSR requires “two independent networks” or using an inaccurate term such as a “dual-network ring.” (Note: HSR is characterized by dual-port nodes on a single-ring topology.)", "rubric_weight": -8, "rubric_tag": "Factual Information" }, { "rubric_number": 13, "rubric_detail": "The answer must use clear heading hierarchy (e.g., “I/II/III” or “1/2/3”) to structure the content, with independent and explicit section titles for each core topic (e.g., PTP optimization, redundancy design).", "rubric_weight": 4, "rubric_tag": "Structure and Formatting" } ] }, { "id": "9a88315b-27e8-4263-a73a-4016befe01a4", "case_id": 10181, "language": "cn", "system_prompt": "", "question": "某汽车零部件工厂的高速冲压生产线，用的是PROFINET工业通信系统，主要设备包括：1台西门子S7-1500控制器、8台KUKA机器人（控制冲压模具开关和工件搬运）、4台基恩士视觉检测相机（实时检查冲压件尺寸），这些设备通过3台赫斯曼工业交换机连成分散式网络，用的是普通五类非屏蔽网线，最长布线距离约80米。生产线周围有3台200千瓦的冲压机床，工作时会产生很强的电磁干扰。\n最近生产线出现了一些问题：当冲压机床以每分钟60次的高速运行时，机器人反应变慢，从原来的5毫秒变成25毫秒以上，视觉相机的检测数据每小时会丢失3到5次（导致次品没被检测出来），但机床低速运行（每分钟30次）时，通信就正常。工厂要求不能停工做大规模网络改造，要在现有硬件基础上花最少的钱解决问题，而且通信反应速度要稳定在10毫秒以内，数据丢失率不能超过0.01%。\n请从PROFINET通信协议特点、工业网络连接方式、抗电磁干扰设计、硬件配置参数这四个核心方面，分析问题根源，给出一步步的排查流程和具体优化方案（包括给交换机分区域、整改网线、调整通信协议参数的详细操作要点）。", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "准确指出通信协议问题是“普通实时模式而非抗干扰更强的精准实时模式”，导致数据出错、重发", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": "准确指出网络连接问题是“设备共用网络抢带宽+普通非屏蔽网线+80米长距离”", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "排查流程包含“高速（60次/分钟）、低速（30次/分钟）工况下查看交换机错误数据计数”以区分故障类型", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 4, "rubric_detail": "排查流程包含“检查网线与冲压机动力线距离、是否靠近机床机身”", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 5, "rubric_detail": "排查流程包含“核查机器人、相机的通信更新时间和数据优先传输功能”", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 6, "rubric_detail": "零成本优化中，设备参数调整包含“机器人通信更新时间设为2毫秒，相机设为8毫秒，仅传输检测结果”", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 7, "rubric_detail": "低成本整改包含“更换靠近冲压机的2-3段（20-50米）屏蔽网线，接头用金属外壳，屏蔽层接地且接地电阻≤4欧姆”", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 8, "rubric_detail": "优化方案明确“先零成本配置调整，再低成本物理整改”的优先级", "rubric_weight": 4, "rubric_tag": "指令遵循" }, { "rubric_number": 9, "rubric_detail": "未分点叙述或缺乏小标题，导致排查步骤和参数设置混杂在一起，可读性差", "rubric_weight": -4, "rubric_tag": "行文结构和格式" }, { "rubric_number": 10, "rubric_detail": "提出大规模停工改造方案（如重新布全网线、更换新交换机/控制器）", "rubric_weight": -10, "rubric_tag": "指令遵循" }, { "rubric_number": 11, "rubric_detail": "排查流程缺失“区分高速/低速工况判断故障类型”", "rubric_weight": -4, "rubric_tag": "事实信息" }, { "rubric_number": 12, "rubric_detail": "回答中包含大量关于PROFINET协议历史背景或通用定义的科普性文字，导致严重冗余", "rubric_weight": -4, "rubric_tag": "行文结构和格式" } ] }, { "id": "da515c84-7adc-4290-9807-67c26de850d5", "case_id": 10298, "language": "cn", "system_prompt": "", "question": "某省运营商为挖掘新的业务增长点，计划在现网开展R18 5G-A NR-DC技术的实验部署。本次实验以FR2高频段作为核心覆盖频段，核心目标是全面支持MRDC场景，结合后续高清VR/AR、高吞吐率数据业务的规划，现提出几项关键技术诉求：\n首先，FR2频段下UE移动性场景的中断时延必须控制在10ms以内，信令开销相比Rel-17版本要降低40%以上，这是保障高清VR/AR这类低时延业务体验的核心前提；其次，FR2频段的传播特性决定了SCG变更会比较频繁，现网初步统计平均每30秒就会发生1次，必须通过技术优化避免频繁网络重配带来的额外时延损耗；第三，在条件切换（CHO）场景中，不能只关注单一小区的信道质量，需要同时保障PCell和PSCell的稳定性，最终目标是实现UE吞吐量提升25%以上；最后，终端侧需要支持至少8套CHO相关配置的存储，而且必须完全兼容Rel-17版本的CPAC/CPC流程，确保存量终端能够平滑接入，避免终端侧的额外改造成本。\n基于3GPP Rel-18在移动性增强方面的标准化成果，结合上述现网实际需求，我们需要重点解决以下几个技术问题：\n1. 针对FR2频段低时延移动性的核心诉求，应该选择哪种技术方案作为核心？需要详细说明UE预配置的关键内容，以及Early TA的获取方式——这里要兼顾可靠性和低时延两个维度。另外，切换命令中的关键字段有哪些？如果UE已经通过自主测量获取了有效的TA值，此时应该采用哪种切换类型？首个上行传输的授权方式又该如何选择？最后，需要分析该方案相比Rel-17传统切换，能够降低时延的核心原因。\n2. 针对SCG频繁变更的场景，如何设计优化方案？首先要明确技术选型，其次是核心的执行流程和配置更新机制。这里有一个具体场景需要考虑：如果UE首次通过CPC流程接入候选PSCell1后，10秒内候选PSCell2就满足了接入条件，此时UE应该如何接入？是否需要网络侧重新下发配置？同时，安全机制中SN Counter的更新流程需要详细阐述。\n3. 要实现CHO场景下的吞吐量提升目标，CHO配置应该采用什么样的结构？PCell和PSCell的执行条件分别由哪个网元决定？各自支持哪些事件触发？假设有一个候选PCell满足执行条件，但它关联的3个候选PSCell中，只有PSCell3满足condEventA4的阈值要求，另外两个都不满足，这种情况下UE应该如何执行？如果此时Rel-17传统CHO的条件也同时满足，最终应该触发哪种切换类型？需要向网络上报哪些关键信息\n4. 现网要求UE支持8套CHO相关配置存储，假设某UE当前存储的配置中，包含3套带候选SCG的CHO配置、2套Rel-17 CHO配置和3套CPA配置，那么UE对这些不同类型配置的存储兼容机制是什么？如果某一套带候选SCG的CHO配置中，只有PCell满足执行条件，而PSCell不满足，UE应该如何处理？另外，如果网络侧需要新增1套CPC配置，如何操作才能不超出UE的存储上限？\n5. 该实验网络后续计划演进，支持inter-CU LTM切换功能。请设计从Rel-18到Rel-19的平滑演进方案，重点明确接口、测量、功能三个层面的核心增强技术点。同时，需要制定网络侧对Rel-18和Rel-19终端的差异化调度机制，具体要从LTM触发方式、测量配置、切换流程这三个维度进行详细说明。", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Weakly time-sensitive", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "模型列举了Early TA的获取方式为UE自主测量为主，PDCCH order触发RACH过程为备用", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 2, "rubric_detail": "模型分析了LTM方案降低时延的核心原因，即跳过了L3测量与RRC信令交互，且无需执行RACH流程", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "模型阐述了SCPAC的核心执行流程：接入PSCell1后不释放其他候选配置，满足条件时直接接入PSCell2，无需网络侧重配", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "模型说明了SN Counter的更新流程，包括从列表选择SK-counter并通过MN RRC Reconfiguration Complete消息上报", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 5, "rubric_detail": "模型指出每套配置包含‘候选PCell + 1个候选PSCell’的组合模式", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 6, "rubric_detail": "针对FR2频段低时延移动性需求，模型明确指出核心技术选型为LTM（L1/L2 Triggered Mobility）", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 7, "rubric_detail": "针对SCG频繁变更场景，模型指出应采用SCPAC（Subsequent CPAC）技术", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "模型明确指出PCell执行条件由source MN决定，PSCell执行条件由候选MN决定", "rubric_weight": 4, "rubric_tag": "事实信息" }, { "rubric_number": 9, "rubric_detail": "模型需明确指出UE应执行其他已存储的Rel-17配置", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "模型分析了多条件满足时的优先级，指出当带候选SCG的CHO与Rel-17传统CHO同时满足时，优先触发带候选SCG的CHO", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "模型区分了Rel-18和Rel-19终端的LTM触发方式，指出Rel-19支持inter-CU触发，而Rel-18仅支持intra-CU触发", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "模型列出了Rel-18到Rel-19演进的三个增强点：接口增强（跨CU交互）、测量增强（事件触发/CSI-RS）、功能增强（条件LTM）", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 13, "rubric_detail": "回答中包含大量关于5G背景、R18标准历史等与具体技术方案无关的冗余信息，导致篇幅过长", "rubric_weight": -5, "rubric_tag": "行文结构和格式" }, { "rubric_number": 14, "rubric_detail": "回答未按照逻辑顺序组织内容，例如将Rel-19演进方案与FR2低时延方案混杂在一起，导致阅读困难", "rubric_weight": -5, "rubric_tag": "行文结构和格式" }, { "rubric_number": 15, "rubric_detail": "模型未列出切换命令中的关键字段（如 TCI State ID）", "rubric_weight": -5, "rubric_tag": "观点分析" }, { "rubric_number": 16, "rubric_detail": "模型在描述技术参数时未使用准确的专业术语（如DG、CG、TCI state等），出现非专业口语化表达", "rubric_weight": -5, "rubric_tag": "行文结构和格式" } ] }, { "id": "ab801e6d-7974-4cc6-b63a-36ba0faa0575", "case_id": 10308, "language": "cn", "system_prompt": "", "question": "题目标题： Review一下这个 LLaMA W8A8 量化算子的致命 Bug\n题目描述：\n我是负责推理引擎优化的架构师。最近为了在边缘设备（NVIDIA Orin）上跑 LLaMA-7B，让实习生手写了一个自定义的 W8A8（权重INT8，激活INT8）矩阵乘法（GEMM）CUDA Kernel。\n目的是替代 cuBLAS，想通过极简实现来减小 binary 体积。但他提交的代码跑出来的结果完全是乱码（PPL 爆炸），而且速度比 FP16 还慢。\n这是他写的量化逻辑说明和核心 CUDA 代码片段（简化版）：\n1. 量化方案：\n对 Weight 和 Activation 都采用 Per-Tensor Symmetric Quantization（逐张量对称量化）。\n公式：Q = clip(round(X / scale), -127, 127)，其中 scale = max(abs(X)) / 127。\n2. CUDA Kernel (C++) 片段：\n__global__ void w8a8_matmul_kernel(const int8_t* A, const int8_t* B, float* C, \n float scale_a, float scale_b, int N, int K) {\n // A: M x K (Row Major)\n // B: K x N (Column Major, 转置过以优化读取)\n // C: M x N\n \n int row = blockIdx.y * blockDim.y + threadIdx.y;\n int col = blockIdx.x * blockDim.x + threadIdx.x;\n \n if (row < N && col < N) {\n // 定义累加器\n int8_t sum = 0; \n \n for (int k = 0; k < K; ++k) {\n // 简单的点积\n int8_t a_val = A[row * K + k];\n int8_t b_val = B[col * K + k]; // B是列优先，所以这样写\n \n // 乘加\n sum += a_val * b_val;\n }\n \n // 反量化写入显存\n C[row * N + col] = (float)sum * scale_a * scale_b;\n }\n}\n请你作为 Tech Lead，指出上述方案中至少 3 个导致精度崩盘或性能低下的致命错误（Fatal Errors），并从底层原理层面解释为什么这么做不行，最后给出针对 LLaMA 架构特性的正确修正思路。\n要求：\n不要给我通用的代码优化建议（如“加注释”），只谈硬核的数学计算和 CUDA 硬件机制。\n必须解释清楚为什么实习生的量化策略对 LLaMA 这种模型是行不通的。\n必须指出代码中关于数据类型处理的严重数学谬误。", "tags": { "topics": [ "工业", "机器学习", "机器学习" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "模型需计算出算术强度约为 1 Op/Byte（或 2 Ops / 2 Bytes），或明确指出该 Kernel 处于 Roofline 模型的 Memory Bound 区域，受限于显存带宽。", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": "优化建议中明确提出使用 Ampere 架构特有的异步拷贝指令 (cp.async)来隐藏全局内存读取延迟", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 3, "rubric_detail": "在优化建议中，明确提出使用 Padding (填充) 或 Swizzling 技术，以防止引入 Shared Memory 后出现 Bank Conflict (存储体冲突)", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 4, "rubric_detail": "建议使用双重缓冲 (Double Buffering)或多级流水线策略来最大化指令并行度", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "建议使用向量化加载指令(LDS.128/float4/int4)或ld.global.nc(非一致性加载)来优化带宽利用率", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 6, "rubric_detail": " 明确指出代码中 int8_t 累加器会导致 整数溢出 (Integer Overflow)，并给出将累加器修改为 int32_t (或 int32) 的具体修正建议", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 7, "rubric_detail": "指出LLaMA激活值存在Outlier(离群点)特性，导致Per-Tensor对称量化失效（精度崩盘）", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "明确建议使用ldmatrix (Load Matrix)指令将数据从共享内存高效加载到寄存器，以适配 Tensor Core运算", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 9, "rubric_detail": "指出代码中B矩阵的访问方式导致非合并访问 (Non-coalesced Access)，带宽利用率低", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 10, "rubric_detail": "明确建议利用NVIDIA Orin(Ampere 架构)特有的2:4结构化稀疏(Structured Sparsity)特性来进一步倍增W8A8推理吞吐量", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 11, "rubric_detail": "建议使用Shared Memory Tiling进行分块优化", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 12, "rubric_detail": "提供的修正方案中包含结构完整的代码块(Code Block)（通常使用Markdown格式包裹，包含完整的循环结构或Kernel签名），而非仅提供零散的单行代码修改建议", "rubric_weight": 5, "rubric_tag": "指令遵循" }, { "rubric_number": 13, "rubric_detail": "应指出公式需体现逐元素乘法，且明确 Activation Scale 对应行索引（row/token），Weight Scale 对应列索引（col/channel）", "rubric_weight": 10, "rubric_tag": "指令遵循" }, { "rubric_number": 14, "rubric_detail": "引用了不存在的CUDA函数（如__int8_mul、cudaQuantize）或编造API库", "rubric_weight": -8, "rubric_tag": "事实信息" }, { "rubric_number": 15, "rubric_detail": "使用虚假的学术标注（如[[1]], [[5]]）或提供了无效的参考文献链接", "rubric_weight": -5, "rubric_tag": "行文结构和格式" }, { "rubric_number": 16, "rubric_detail": "模型回答中错误地建议使用float16 (half)作为累加器类型。", "rubric_weight": -10, "rubric_tag": "事实信息" }, { "rubric_number": 17, "rubric_detail": "修正方案错误地建议使用非对称量化(Asymmetric Quantization) 或引入零点(Zero Point)，忽略了此举在Tensor Core计算中会导致额外的指令开销并严重降低推理吞吐量", "rubric_weight": -5, "rubric_tag": "事实信息" } ] }, { "id": "c545314e-f070-43c7-bec7-c426687d27cf", "case_id": 10447, "language": "cn", "system_prompt": "", "question": "我是一个数据开发工程师，我最近在负责一个用户数据链路维护。其中，用户注册写 MongoDB（users 集合），created_at 由前端传入，格式 \"2026-01-06 10:30:00\"（无时区标识），后端 new Date(str) 转成 Date 类型存储\n后端服务器时区：UTC-5。每天 UTC 06:00的时候，ETL会自动将MongoDB中的数据同步到 Impala的dim_users 表，分区字段 dt = DATE(created_at)，其中status='deleted' 的用户不同步。用户行为写入 Elasticsearch中的user_events 索引，timestamp 字段存的是 Unix 毫秒时间戳，user_id 字段存的是 MongoDB _id.toString()。\n现在产品经理说 1 月 6 日的数据对不上：BI (Impala)中显示有1,247个用户，MongoDB显示有1,302个用户，Elasticsearch显示有1,180个用户。\n我先进行了一些初步查询，发现90% 用户在 UTC-5 时区，当地 9:00-22:00 活跃，0:00-5:00 注册量约占3%，约 8% 的新注册用户当天没有任何行为事件MongoDB 的 _id 是 ObjectId 类型。其中，1 月 6 日有 23 个用户被软删除。\n你能帮我写一份差异分析报告给产品经理吗？我希望你能分别解释不同数据库中的数据差异的原因，并指出其中可能还存在什么其他问题。报告中还要包括修复方案，要求修复后三个数据源对同一天的查询结果一致。不要追问，直接回答。", "tags": { "topics": [ "工业", "数据库与数据工程", "数据库与数据工程" ], "time_sensitivity": { "time_sensitivity": "Weakly time-sensitive", "year_month": "2026-01", "day": "6" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "报告中明确指出了软删除用户数量为23人", "rubric_weight": 7, "rubric_tag": "事实信息" }, { "rubric_number": 2, "rubric_detail": "回答基于MongoDB的总人数1302的8%算出无行为事件的用户数量约为104人", "rubric_weight": 7, "rubric_tag": "事实信息" }, { "rubric_number": 3, "rubric_detail": "模型解释了时区错位机制：UTC-5时区19:00后注册的用户，其UTC时间会落入次日，导致分区错误。", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "模型分析出MongoDB与Impala的差异原因之一是ETL逻辑过滤掉了status='deleted'的用户。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "模型给出以MongoDB ObjectId与ES中字符串格式不一致为例的Elasticsearch剩余差异的潜在技术原因", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "模型提供了Impala/SQL层面的修复代码，使用了时区转换函数（如from_utc_timestamp）", "rubric_weight": 5, "rubric_tag": "指令遵循" }, { "rubric_number": 7, "rubric_detail": "模型提供了MongoDB查询的修正方案，将查询时间范围调整为对应的UTC时间窗口（如05:00到次日05:00）", "rubric_weight": 7, "rubric_tag": "指令遵循" }, { "rubric_number": 8, "rubric_detail": "模型提供了Elasticsearch查询的修正方案，指出了需要在查询中指定time_zone参数", "rubric_weight": 7, "rubric_tag": "指令遵循" }, { "rubric_number": 9, "rubric_detail": "模型建议修改ETL逻辑，不再直接过滤删除用户，而是保留删除标记以对齐统计口径。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "回答中没有计算出MongoDB与ES的总差异人数为122人。", "rubric_weight": -5, "rubric_tag": "事实信息" }, { "rubric_number": 11, "rubric_detail": "报告中没有明确计算出时区错位影响的用户数为32人", "rubric_weight": -5, "rubric_tag": "事实信息" }, { "rubric_number": 12, "rubric_detail": "回答没有计算出ES剩余差异约为18人。", "rubric_weight": -5, "rubric_tag": "事实信息" }, { "rubric_number": 13, "rubric_detail": "模型指出ObjectId转String可能存在格式不一致问题", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 14, "rubric_detail": "回答中包含了非报告正文的思维链过程、口语化开场白或与最终报告格式无关的冗余描述。 ", "rubric_weight": -3, "rubric_tag": "行文结构和格式" }, { "rubric_number": 15, "rubric_detail": "回答中对同一个概念进行三次以上的解释。", "rubric_weight": -3, "rubric_tag": "行文结构和格式" } ] }, { "id": "eb7638ac-035f-4b6c-a207-ebc6804aca3a", "case_id": 10631, "language": "cn", "system_prompt": "", "question": "给定一个非空的整数数组nums，数组中的每个元素为任意整数，数组可能由一个或多个数组成，把数组中每个数的每一位数字完全拆解重组，拼接成一个最小整数，满足以下所有规则：【1】数组中每个数的每一位数字都必须全部用上，不能多不能少，例如-123拆解为[-1,2,3]，0拆解为[0]，44拆解为[4,4]。【2】对每一个数的拆解，负数的负号只能放在最高位数字前，例如-78只能拆成[-7,8]。【3】拼接出的整数不允许最高位是0，例如拼接结果不能是012，如果所有数字都是0，最终结果也是0。【4】拼接出的整数必须合法，如果无法按要求拼成，则直接返回原因，具体什么情况属于无法拼成，需自主思考。如果能够拼成，最终返回值为找出的最小整数的字符串形式。注意复杂度最优。要求：先用自然语言描述思路，要覆盖所有情况，并给出时间和空间复杂度。再给出python代码示例，要遵循代码规范，只能使用段落注释。最后给出完整覆盖所有类型情况的测试案例。", "tags": { "topics": [ "工业", "后端开发", "后端开发" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "代码部分使用了行级注释（#注释），违反了题目中只使用段落注释的格式要求", "rubric_weight": -10, "rubric_tag": "行文结构和格式" }, { "rubric_number": 2, "rubric_detail": "回答按照思路描述，Python代码，测试案例的顺序组织", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 3, "rubric_detail": "测试案例中每个案例的输入和对应输出完全正确。输入任意个0输出0；输入两个及超过两个负数输出拼不出的提示；输入一个负数注意负号和负数最高位不能拆开，并输出一个绝对值最大的负数；输入全正数输出拼成的最小正数。且需注意输出的数字个数和输入的数字个数一致（全0数组除外）", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "没有给出完整覆盖所有类型情况的测试案例。至少包含拼不出场景、无负数场景、只有一个负数场景，且包含大数测试场景。", "rubric_weight": -10, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "思路分析里提及超大数场景，包括转字符串如果超长溢出应如何处理（python中事实是不会溢出，不用特殊处理）。", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "代码部分包含推算错误，比如拼成的数并非符合题目要求的最小数（示例：[100,20]输出10020并非最小数，10002才是最小数）", "rubric_weight": -10, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "给出的时间复杂度为O(n)，其中n为元素个数或总字符数", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 8, "rubric_detail": "给出的空间复杂度为O(1)（不包含结果存储空间）", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 9, "rubric_detail": "明确指出输入0的输出结果是0", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 10, "rubric_detail": "针对结果为正数的场景，提出了将所有数字按升序排列，并将最小非0数字置于首位的拼接策略。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "针对结果为负数的场景，提出了负号置于首位，其余数字降序排列以最大化绝对值的策略", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "代码不符合python的缩进格式。", "rubric_weight": -5, "rubric_tag": "行文结构和格式" } ] }, { "id": "cce48b66-20eb-4d4a-9dfb-fff4e75968f8", "case_id": 10704, "language": "cn", "system_prompt": "", "question": "我是一名道桥系研究生，目前在参加专业实践，给某一项目设计混凝土的配合比，在学习混凝土配合比设计的方法时，导师给我讲解了比表面积的设计方法，但初步试验发现这种方法指导设计出的配合比存在问题，配置出的混凝土的状态比较差（流动性、保水性黏聚性不达施工要求），初步分析是这种方法太过老旧，不适应混凝土技术的快速发展，帮我分析该方法失效并出现明显误差的具体原因，会出现哪些误差？原因是什么？\n导师认为，方法失效的原因可能在于以下三个方面的变化：\n1.不同时期的混凝土的组分和含量不同；\n2.混凝土的应用领域不断开拓与施工技术进步；\n3.混凝土的工作性、力学性和耐久性提高。\n请结合导师给的提示思考回答两个问题：\n1.旧方法会出现哪些误差（具体指配合比和初配混凝土的状态）？\n2.背后的原因（结合导师的提示）\n回答逻辑上请首先解释比表面积法及其原理，然后说明误差，接着说明原因，最后给我推荐设计混凝土配合比行之有效的其他方法。", "tags": { "topics": [ "工业", "土木", "土木" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "回答中明确指出了现代水泥比表面积的数值并说明了水泥细度发生了变化，即从过去的低于300 m²/kg增加到现在的330~350 m²/kg（42.5水泥）或超过380 m²/kg（52.5水泥），", "rubric_weight": 4, "rubric_tag": "事实信息" }, { "rubric_number": 2, "rubric_detail": "提及了粗骨料破碎工艺的具体升级方向，即从颚式破碎变为反击破或锤破式这一明显变革", "rubric_weight": 3, "rubric_tag": "事实信息" }, { "rubric_number": 3, "rubric_detail": "列举了混凝土强度等级的主流变化，高标水泥被广泛使用，从过去的C30及以下转变为现在的C40~C80", "rubric_weight": 3, "rubric_tag": "事实信息" }, { "rubric_number": 4, "rubric_detail": "指出了聚羧酸系高效减水剂的应用使得水胶比可以降低至0.3以下，减水剂发挥了明显作用", "rubric_weight": 3, "rubric_tag": "事实信息" }, { "rubric_number": 5, "rubric_detail": "在解释原因时，提到了水泥熟料矿物成分比例的变化，具体为铝酸三钙、硅酸三钙含量增加，铁铝酸四钙、硅酸二钙含量相对减少", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "根据题设的回答逻辑要求，解释比表面积法的基本原理（是依据骨料的总表面积来计算包裹所需的水泥浆体用量）", "rubric_weight": 5, "rubric_tag": "指令遵循" }, { "rubric_number": 7, "rubric_detail": "指出配合比计算的直接误差之一是砂率估算不准，进而影响初配混凝土的状态和强度，且事实上在高性能混凝土中砂率与强度的传统反比关系失效（即砂率高未必强度低）", "rubric_weight": 7, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "指出在一些现代工程实践中发现，低水胶比（如0.38~0.45）区间，传统的水灰比定律弱化，强度与水灰比缺乏明显线性相关性", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 9, "rubric_detail": "描述初配混凝土状态的误差时，指出了可能出现泌水、离析、流动性差或黏聚性差等具体的工作性能表现差的现象", "rubric_weight": 7, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "分析方法失效的原因时，指出矿物掺合料（如粉煤灰、矿渣）的引入使得“水泥”概念扩展为“胶凝材料总量”，掺和料的影响因素复杂，导致基于纯水泥的计算失效", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "指出骨料（特别是细骨料）形态变化是导致误差的重要原因之一，即现代工程使用的石子粒径更小品相更好，机制砂反而多棱角多孔隙，吸水量比天然河沙更大", "rubric_weight": 9, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "分析指出现代水泥的细度过高（为追求高早强，水泥被越磨越细），导致早期水化过快，从而削弱了后期强度增长潜力，这是比表面积法预测失效的原因之一", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 13, "rubric_detail": "指出现代混凝土工程的应用领域不断扩大，追求多目标高质量（更高更大更耐久的建筑），因此混凝土的各方面性能要求被提高，由此配合比设计理念发生转变，即从单一的“强度主导”转向关注抗裂、抗渗等指标的“耐久性主导”", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 14, "rubric_detail": "在回答最后提供更加科学准确的配合比设计方法，如浆骨比法、绝对体积或重量法或基于性能导向（流变学、耐久性等等）的设计方法", "rubric_weight": 6, "rubric_tag": "指令遵循" }, { "rubric_number": 15, "rubric_detail": "回答严格遵循了题设的回答内容顺序，“解释比表面积法的原理 -> 说明误差（计算的直接误差和初配混凝土的状态误差） -> 分析原因（结合提示） -> 给出新方法（行之有效的方法）”的逻辑顺序", "rubric_weight": 5, "rubric_tag": "行文结构和格式" }, { "rubric_number": 16, "rubric_detail": "回答中包含大量关于混凝土发展历史、通用教科书定义的冗长铺垫，与核心问题（误差和原因）关联度低，或者不同部分的回答内容存在重合，造成严重冗余", "rubric_weight": -3, "rubric_tag": "行文结构和格式" }, { "rubric_number": 17, "rubric_detail": "回答未分点叙述或缺乏清晰的小标题，导致本题的核心考点误差表现与根本原因混杂在一起造成逻辑结构混乱，且不满足题设要求误差与原因分别说明的要求", "rubric_weight": -3, "rubric_tag": "行文结构和格式" }, { "rubric_number": 18, "rubric_detail": "比表面积法是基于骨料表面积确定水泥用量，而水灰比定则是基于水灰比控制强度。若模型将‘水灰比越小强度越高’作为比表面积法的核心原理，即视为混淆。", "rubric_weight": -6, "rubric_tag": "观点分析" }, { "rubric_number": 19, "rubric_detail": "如果模型回答“计量不准”、“含水率未调整”、“搅拌不均匀”等等施工操作层面的问题，即是混淆“施工误差”与“理论误差”", "rubric_weight": -6, "rubric_tag": "观点分析" } ] }, { "id": "6fd07669-8d90-4ace-81ea-9a1ec49cf9aa", "case_id": 10724, "language": "cn", "system_prompt": "", "question": "我是北京理工大学的研究生，我目前的研究方向是钠离子电池固态电解质，主要是钠超离子型固态电解质NZSP。固态电解质所要解决的核心是离子电导率低和电解质与金属电极浸润性的问题，我目前想要设计一个实验方案，我的初步想法根据以下参考内容中将NHSP替换为NZSP进行表面改性实验，但是受限于实验室的条件，我没办法采用丝网印刷的实验方法，我们实验室仅有球磨、加热设备。\n我的问题是：1帮我分析替代的可能性及相关原理\n2.结合实际实验条件设计一个合理的方案，对于NZSP的制备采取固相烧结法，要求包括必要的参数。\nThe NiO powder and NHSP powder were ball-milled with ethanol as dispersant in a mass ratio of 3:7 for 24 h. The mixture was blown dry and ground into a powder, and the mixture powder was mixed with the organic carrier in a mass ratio of 3:7, with the organic carrier consisting of terpineol and ethyl cellulose in a mass ratio of 9:1. The mixed slurry was stirred on a magnetic stirrer for 48 h. The mixed slurry was coated on the NHSP electrolyte surface\nby means of screen printing, and then the electrolyte was sintered at high temperature for 3 h at 1200 ℃ at a rate of 3 ℃ /min. \n", "tags": { "topics": [ "工业", "化工与材料", "化工与材料" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "NZSP和NHSP两种电解质均属于NASICON型（钠超离子导体）晶体结构", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 2, "rubric_detail": "NZSP的微观结构由PO4/SiO4四面体和ZrO6八面体共用氧离子顶点构成，NHSP结构由PO4/SiO4四面体和HfO6八面体共用氧离子顶点构成", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 3, "rubric_detail": "使用ZrO2替代HfO2作为原料能够显著降低实验成本", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "在原料配比计算时，钠源（如Na2CO3）的物质的量需过量0.1（或10%）", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 5, "rubric_detail": "压片成型前需添加PVB等粘结剂，以确保粉体能压制成型而不分散，在压片后还需要有粘结剂的去除，一般是650℃烧结3h", "rubric_weight": 7, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "高温烧结过程中采取埋母粉（NZSP前驱体粉末）措施，以防止钠挥发导致电解质变形或成分改变", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "改性界面主要是依靠NiO的作用，NiO作为亲水相，在还原反应后（一氧化镍+Na→Ni+氧化钠）可转化为电子导体，从而有效提高钠与电解质之间的润湿性并降低界面电阻。而有机载体主要是作为中间层实现良好的接触", "rubric_weight": 4, "rubric_tag": "事实信息" }, { "rubric_number": 8, "rubric_detail": "回答中包含除实验方案设计及原理分析以外的无关内容，例如未要求的测试检测步骤等。", "rubric_weight": -3, "rubric_tag": "行文结构和格式" }, { "rubric_number": 9, "rubric_detail": "实验方案缺乏清晰的步骤分层，将NZSP本体的制备过程与表面改性过程混杂在一起，逻辑混乱", "rubric_weight": -2, "rubric_tag": "行文结构和格式" }, { "rubric_number": 10, "rubric_detail": "回答未提及在有机载体中加入乙醇等溶剂以降低体系粘稠度。", "rubric_weight": -10, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "实验中要说明去除松油醇以及乙基纤维素的烧结温度及时间。松油醇的去除温度是217℃无需保温，乙基纤维素去除需要在350-500℃下保温一小时", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 12, "rubric_detail": "模型给出的NZSP制备步骤存在错误，未正确包含混料、预烧、压片、排胶、高温烧结等固相法关键环节。", "rubric_weight": -5, "rubric_tag": "观点分析" } ] }, { "id": "b0d00294-81d8-453f-b8e1-76543c490717", "case_id": 10767, "language": "cn", "system_prompt": "", "question": "图神经网络是一种考虑图结构的机器学习方法，其中最经典的是图卷积神经网络GCN算法，这个github仓库(https://github.com/iDC-NEU/iDC-MlSys_interview)是一个未经优化的GCN算法，使用的语言为C++，我现在在一个CPU-GPU异构环境下训练，需要你对着该github仓库的代码做性能优化，请注意，你只需要提出你的优化的想法，不需要写出具体的代码。\n你的任务是\n1. 第一步，你需要做CPU和GPU上都能有用的通用优化\n2. 第二步，需要针对Intel CPU的特点做CPU端优化\n3. 第三步，需要针对Nvidia A系列GPU做性能优化\n4. 第四步，需要针对分布式场景做优化\n请遵循以下指令\n1. 再次强调：你只需要提出你的优化的想法，不需要写出具体的代码\n2. 上面四种情况都要做优化\n3. 不要想着调用成熟的外部机器学习库，你只能使用自带的一些库可以解决的方案\n4. 你的优化之后，做理论性的性能分析\n\n核心代码如下，你可以访问网页获取完整代码\n#include \n#include \n#include \n#include \n#include \n#include \n#include \n#include \n#include \n#include \n\nusing namespace std;\n\ntypedef std::chrono::time_point TimePoint;\n\nint v_num = 0;\nint e_num = 0;\nint F0 = 0, F1 = 0, F2 = 0;\n\nvector> edge_index;\nvector> edge_val;\nvector degree;\nvector raw_graph;\n\nfloat *X0, *W1, *W2, *X1, *X1_inter, *X2, *X2_inter;\n\nvoid readGraph(char *fname)\n{\n\tifstream infile(fname);\n\n\tint source;\n\tint end;\n\n\tinfile >> v_num >> e_num;\n\n\t// raw_graph.resize(e_num * 2);\n\n\twhile (!infile.eof())\n\t{\n\t\tinfile >> source >> end;\n\t\tif (infile.peek() == EOF)\n\t\t\tbreak;\n\t\traw_graph.push_back(source);\n\t\traw_graph.push_back(end);\n\t}\n}\n\nvoid raw_graph_to_AdjacencyList()\n{\n\n\tint src;\n\tint dst;\n\n\tedge_index.resize(v_num);\n\tedge_val.resize(v_num);\n\tdegree.resize(v_num, 0);\n\n\tfor (int i = 0; i < raw_graph.size() / 2; i++)\n\t{\n\t\tsrc = raw_graph[2*i];\n\t\tdst = raw_graph[2*i + 1];\n\t\tedge_index[dst].push_back(src);\n\t\tdegree[src]++;\n\t}\n}\n\nvoid edgeNormalization()\n{\n\tfor (int i = 0; i < v_num; i++)\n\t{\n\t\tfor (int j = 0; j < edge_index[i].size(); j++)\n\t\t{\n\t\t\tfloat val = 1 / sqrt(degree[i]) / sqrt(degree[edge_index[i][j]]);\n\t\t\tedge_val[i].push_back(val);\n\t\t}\n\t}\n}\n\nvoid readFloat(char *fname, float *&dst, int num)\n{\n\tdst = (float *)malloc(num * sizeof(float));\n\tFILE *fp = fopen(fname, \"rb\");\n\tfread(dst, num * sizeof(float), 1, fp);\n\tfclose(fp);\n}\n\nvoid initFloat(float *&dst, int num)\n{\n\tdst = (float *)malloc(num * sizeof(float));\n\tmemset(dst, 0, num * sizeof(float));\n}\n\nvoid XW(int in_dim, int out_dim, float *in_X, float *out_X, float *W)\n{\n\tfloat(*tmp_in_X)[in_dim] = (float(*)[in_dim])in_X;\n\tfloat(*tmp_out_X)[out_dim] = (float(*)[out_dim])out_X;\n\tfloat(*tmp_W)[out_dim] = (float(*)[out_dim])W;\n\n\tfor (int i = 0; i < v_num; i++)\n\t{\n\t\tfor (int j = 0; j < out_dim; j++)\n\t\t{\n\t\t\tfor (int k = 0; k < in_dim; k++)\n\t\t\t{\n\t\t\t\ttmp_out_X[i][j] += tmp_in_X[i][k] * tmp_W[k][j];\n\t\t\t}\n\t\t}\n\t}\n}\n\nvoid AX(int dim, float *in_X, float *out_X)\n{\n\tfloat(*tmp_in_X)[dim] = (float(*)[dim])in_X;\n\tfloat(*tmp_out_X)[dim] = (float(*)[dim])out_X;\n\n\tfor (int i = 0; i < v_num; i++)\n\t{\n\t\tvector &nlist = edge_index[i];\n\t\tfor (int j = 0; j < nlist.size(); j++)\n\t\t{\n\t\t\tint nbr = nlist[j];\n\t\t\tfor (int k = 0; k < dim; k++)\n\t\t\t{\n\t\t\t\ttmp_out_X[i][k] += tmp_in_X[nbr][k] * edge_val[i][j];\n\t\t\t}\n\t\t}\n\t}\n}\n\nvoid ReLU(int dim, float *X)\n{\n\tfor (int i = 0; i < v_num * dim; i++)\n\t\tif (X[i] < 0)\n\t\t\tX[i] = 0;\n}\n\nvoid LogSoftmax(int dim, float *X)\n{\n\tfloat(*tmp_X)[dim] = (float(*)[dim])X;\n\n\tfor (int i = 0; i < v_num; i++)\n\t{\n\t\tfloat max = tmp_X[i][0];\n\t\tfor (int j = 1; j < dim; j++)\n\t\t{\n\t\t\tif (tmp_X[i][j] > max)\n\t\t\t\tmax = tmp_X[i][j];\n\t\t}\n\n\t\tfloat sum = 0;\n\t\tfor (int j = 0; j < dim; j++)\n\t\t{\n\t\t\tsum += exp(tmp_X[i][j] - max);\n\t\t}\n\t\tsum = log(sum);\n\n\t\tfor (int j = 0; j < dim; j++)\n\t\t{\n\t\t\ttmp_X[i][j] = tmp_X[i][j] - max - sum;\n\t\t}\n\t}\n}\n\nfloat MaxRowSum(float *X, int dim)\n{\n\tfloat(*tmp_X)[dim] = (float(*)[dim])X;\n\tfloat max = -__FLT_MAX__;\n\n\tfor (int i = 0; i < v_num; i++)\n\t{\n\t\tfloat sum = 0;\n\t\tfor (int j = 0; j < dim; j++)\n\t\t{\n\t\t\tsum += tmp_X[i][j];\n\t\t}\n\t\tif (sum > max)\n\t\t\tmax = sum;\n\t}\n\treturn max;\n}\n\nvoid freeFloats()\n{\n\tfree(X0);\n\tfree(W1);\n\tfree(W2);\n\tfree(X1);\n\tfree(X2);\n\tfree(X1_inter);\n\tfree(X2_inter);\n}\n\nvoid somePreprocessing()\n{\n\t//The graph will be transformed into adjacency list ,you can use other data structure such as CSR\n\traw_graph_to_AdjacencyList();\n}\n\nint main(int argc, char **argv)\n{\n\t// Do NOT count the time of reading files, malloc, and memset\n\tF0 = atoi(argv[1]);\n\tF1 = atoi(argv[2]);\n\tF2 = atoi(argv[3]);\n\n\treadGraph(argv[4]);\n\treadFloat(argv[5], X0, v_num * F0);\n\treadFloat(argv[6], W1, F0 * F1);\n\treadFloat(argv[7], W2, F1 * F2);\n\n\tinitFloat(X1, v_num * F1);\n\tinitFloat(X1_inter, v_num * F1);\n\tinitFloat(X2, v_num * F2);\n\tinitFloat(X2_inter, v_num * F2);\n\n\t// Time point at the start of the computation\n\tTimePoint start = chrono::steady_clock::now();\n\n\t// Preprocessing time should be included\n\n\tTimePoint prepross_start = chrono::steady_clock::now();\n\tsomePreprocessing();\n\tTimePoint prepross_end = chrono::steady_clock::now();\n\tchrono::duration prepross_ = prepross_end - prepross_start;\n\tdouble prepross_time = prepross_.count() * 1e3;\n\tprintf(\"prepross_time: %.8lf\\n\", prepross_time);\n\n\tTimePoint edgeNorm_start = chrono::steady_clock::now();\n\tedgeNormalization();\n\tTimePoint edgeNorm_end = chrono::steady_clock::now();\n\tchrono::duration edgeNorm_ = edgeNorm_end - edgeNorm_start;\n\tdouble edgeNorm_time = edgeNorm_.count() * 1e3;\n\tprintf(\"edgeNorm_time: %.8lf\\n\", edgeNorm_time);\n\n\n\t// printf(\"Layer1 XW\\n\");\n\tTimePoint XW1_start = chrono::steady_clock::now();\n\tXW(F0, F1, X0, X1_inter, W1);\n\tTimePoint XW1_end = chrono::steady_clock::now();\n\tchrono::duration XW1_ = XW1_end - XW1_start;\n\tdouble XW1_time = XW1_.count() * 1e3;\n\tprintf(\"XW1_time: %.8lf\\n\", XW1_time);\n\n\t\n\n\t// printf(\"Layer1 AX\\n\");\n\tTimePoint AX1_start = chrono::steady_clock::now();\n\tAX(F1, X1_inter, X1);\n\tTimePoint AX1_end = chrono::steady_clock::now();\n\tchrono::duration AX1_ = AX1_end - AX1_start;\n\tdouble AX1_time = AX1_.count() * 1e3;\n\tprintf(\"AX1_time: %.8lf\\n\", AX1_time);\n\n\t// printf(\"Layer1 ReLU\\n\");\n\tTimePoint ReLU_start = chrono::steady_clock::now();\n\tReLU(F1, X1);\n\tTimePoint ReLU_end = chrono::steady_clock::now();\n\tchrono::duration ReLU_ = ReLU_end - ReLU_start;\n\tdouble ReLU_time = ReLU_.count() * 1e3;\n\tprintf(\"ReLU_time: %.8lf\\n\", ReLU_time);\n\n\t// printf(\"Layer2 XW\\n\");\t\n\tTimePoint XW2_start = chrono::steady_clock::now();\n\tXW(F1, F2, X1, X2_inter, W2);\n\tTimePoint XW2_end = chrono::steady_clock::now();\n\tchrono::duration XW2_ = XW2_end - XW2_start;\n\tdouble XW2_time = XW2_.count() * 1e3;\n\tprintf(\"XW2_time: %.8lf\\n\", XW2_time);\n\n\t// printf(\"Layer2 AX\\n\");\n\tTimePoint AX2_start = chrono::steady_clock::now();\n\tAX(F2, X2_inter, X2);\n\tTimePoint AX2_end = chrono::steady_clock::now();\n\tchrono::duration AX2_ = AX2_end - AX2_start;\n\tdouble AX2_time = AX2_.count() * 1e3;\n\tprintf(\"AX2_time: %.8lf\\n\", AX2_time);\n\n\t// printf(\"Layer2 LogSoftmax\\n\");\n\tTimePoint LogSoftmax_start = chrono::steady_clock::now();\n\tLogSoftmax(F2, X2);\n\tTimePoint LogSoftmax_end = chrono::steady_clock::now();\n\tchrono::duration LogSoftmax_ = LogSoftmax_end - LogSoftmax_start;\n\tdouble LogSoftmax_time = LogSoftmax_.count() * 1e3;\n\tprintf(\"LogSoftmax_time: %.8lf\\n\", LogSoftmax_time);\n\n\t// You need to compute the max row sum for result verification\n\tTimePoint max_sum_start = chrono::steady_clock::now();\n\tfloat max_sum = MaxRowSum(X2, F2);\n\tTimePoint max_sum_end = chrono::steady_clock::now();\n\tchrono::duration max_sum_ = max_sum_end - max_sum_start;\n\tdouble max_sum_time = max_sum_.count() * 1e3;\n\tprintf(\"max_sum_time: %.8lf\\n\", max_sum_time);\n\n\t// Time point at the end of the computation\n\tTimePoint end = chrono::steady_clock::now();\n\tchrono::duration l_durationSec = end - start;\n\tdouble l_timeMs = l_durationSec.count() * 1e3;\n\n\t// Finally, the max row sum and the computing time\n\t// should be print to the terminal in the following format\n\tprintf(\"%.8f\\n\", max_sum);\n\tprintf(\"total time: %.8lf\\n\\n\", l_timeMs);\n\n\t// Remember to free your allocated memory\n\tfreeFloats();\n}\n不要追问，直接回答", "tags": { "topics": [ "工业", "机器学习", "机器学习" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "模型建议将图数据的存储结构从邻接表或邻接矩阵转换为CSR（压缩稀疏行）或CSC格式", "rubric_weight": 4, "rubric_tag": "事实信息" }, { "rubric_number": 2, "rubric_detail": "模型能够分析出改成CSR/CSC格式有两点好处，第一点是压缩图存储能减少图结构的占用空间，第二点是用offset和index来存储，能保证邻居是连续访问的，而图计算里，访问邻居是最常见的操作，所以这样拥有更高的缓存命中率", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "模型指出在预处理阶段修改数据结构时需要注意原子性问题，以防止写冲突", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "模型提议将CPU端的矩阵乘法循环顺序从ijk修改为ikj", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 5, "rubric_detail": "模型解释了修改循环顺序是为了优化Cache缓存行的命中率，从而提升计算效率", "rubric_weight": 2, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "模型明确指出针对Intel CPU应使用AVX指令集进行向量化并行计算", "rubric_weight": 4, "rubric_tag": "事实信息" }, { "rubric_number": 7, "rubric_detail": "模型建议使用OpenMP库来实现CPU端的多线程并行处理", "rubric_weight": 4, "rubric_tag": "事实信息" }, { "rubric_number": 8, "rubric_detail": "模型分析了OpenMP线程数的设置策略，建议保守选择（如总线程数的一半）以避免资源争用", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "模型提出了算子融合（Operator Fusion）的方案，例如将矩阵乘法与ReLU/Softmax激活函数合并", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 10, "rubric_detail": "模型指出算子融合可以减少中间结果的存储需求并减少遍历次数", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "模型针对Nvidia GPU优化提出了使用CUDA编写逻辑", "rubric_weight": 8, "rubric_tag": "指令遵循" }, { "rubric_number": 12, "rubric_detail": "模型在GPU优化设计中，针对聚合阶段建议每个线程负责一个特征维度，并使用原子操作进行累加更新。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 13, "rubric_detail": "模型在分布式优化中提到了需要进行图划分", "rubric_weight": 3, "rubric_tag": "事实信息" }, { "rubric_number": 14, "rubric_detail": "模型指出分布式场景下应权衡通信与计算，利用通信代替冗余计算", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 15, "rubric_detail": "模型严格遵循了“不需要写出具体的代码”的负向约束", "rubric_weight": 4, "rubric_tag": "指令遵循" }, { "rubric_number": 16, "rubric_detail": "回答中包含大量关于GCN原理或图神经网络基础定义的科普性文字，造成严重冗余", "rubric_weight": -5, "rubric_tag": "行文结构和格式" }, { "rubric_number": 17, "rubric_detail": "回答中出现了具体的C++或CUDA代码块，违反了只提想法的指令", "rubric_weight": -5, "rubric_tag": "行文结构和格式" }, { "rubric_number": 18, "rubric_detail": "模型需包含理论性的性能分析，例如提及时间/空间复杂度、访存带宽瓶颈（Roofline模型）、计算密度或分布式通信开销等关键评估指标", "rubric_weight": 4, "rubric_tag": "行文结构和格式" }, { "rubric_number": 19, "rubric_detail": "模型调用了PyTorch、TensorFlow等成熟的外部机器学习库，违反了指令要求", "rubric_weight": -5, "rubric_tag": "指令遵循" }, { "rubric_number": 20, "rubric_detail": "模型回复遗漏了题目要求的四个优化步骤（1.通用优化；2.Intel CPU优化；3.Nvidia GPU优化；4.分布式优化）中的任意一个或多个。", "rubric_weight": -7, "rubric_tag": "指令遵循" }, { "rubric_number": 21, "rubric_detail": "模型识别到图计算这类稀疏计算在稀疏聚合阶段（AX）的GPU性能优化时应使用Cuda core,而使用tensor core会导致SM利用率较低", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 22, "rubric_detail": "模型应指出在分布式场景下的切图优化中，为了实现负载均衡，应主要考虑边的分布（因为边更能反映计算量），并建议采用例如最小边切（Minimum Edge Cut）等切图方式。", "rubric_weight": 7, "rubric_tag": "观点分析" }, { "rubric_number": 23, "rubric_detail": "CPU优化时模型应建议使用AVX的stream_ps，跳过缓存直接访问内存，减少50%的访存时间，因为GCN的计算都是一次性的", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 24, "rubric_detail": " 模型未针对性能进行优化，而是提出了准确率优化的方案。", "rubric_weight": -5, "rubric_tag": "事实信息" } ] }, { "id": "7e0ab98e-379e-4b1e-acff-3ff37bffaf32", "case_id": 10838, "language": "cn", "system_prompt": "", "question": "四川省某项目住宅小区，地上为二类住宅高层建筑，地下室为一层地下建筑，负一层为机动车库和设备用房。当地下室机动车库设在同一个防火分区的不同防火单元之间防火隔墙上设置的疏散门的数量和疏散方向如何判定。根据《电动汽车分散充电设施工程技术标准》GB/T51313-2018（6.1.5第4款“当防火隔墙上需开设相互连通的门时，应采用耐火等级不低于乙级的防火门”，该技术标准明确的为相互连通的门，但是由于一个防火分区划分为几个防火单元，有些防火单元并不包含疏散楼梯间，所以需要在防火单元的防火隔墙上开向相邻防火单元的连通门再到达安全出口，请问这种情况下防火单元的防火隔墙上的疏散门如何设置，疏散方向如何判定？请给出回答并详细说明理由", "tags": { "topics": [ "工业", "建筑设计", "建筑设计" ], "time_sensitivity": { "time_sensitivity": "Weakly time-sensitive", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "引用《四川省房屋建筑工程消防设计技术审查要点》2025年版的7.1.1条第4款作为主要依据：\n“地下室每个充电汽车防火单元面积不应大于1000 ㎡'，每个防火单元应采用耐火极限不小于 2.00h 的防火隔墙或防火卷帘(仅限于汽车通道处)与其他防火单元分隔。每个防火单元应设置不少于2 个安全出口，安全出口可以是开向相邻防火单元的防火门。当防火隔墙上需开设相互连通的门时，应采用不低于甲级的防火门。”", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 2, "rubric_detail": "引用《建筑设计防火规范》GB 50016-2014（2018年版）第5.5.2条关于间距的规定：\n“建筑内的安全出口和疏散门应分散布置，且建筑内每个防火分区或一个防火分区的每个楼层、每个住宅单元每层相邻两个安全出口以及每个房间相邻两个疏散门最近边缘之间的水平距离不应小于 5m。”", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 3, "rubric_detail": "引用《建筑设计防火规范》GB 50016-2014（2018 年版）第5.5.21条相关规定：\n“除剧场、电影院、礼堂、体育馆外的其他公共建筑，其房间疏散门、安全出口、疏散走道和疏散楼梯的各自总净宽度，应符合下列规定：……”(疏散宽度的相关规定)\n根据GB 55037-2022《建筑防火通用规范》第7.1.4条相关规定：\n“疏散出口门、疏散走道、疏散楼梯等的净宽度应符合下列规定:1 疏散出口门、室外疏散楼梯的净宽度均不应小于0.80m”", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 4, "rubric_detail": "明确指出地下室每个充电汽车防火单元的防火隔墙应设置不少于2个安全出口的疏散门", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 5, "rubric_detail": "确认安全出口可以是开向相邻防火单元的防火门", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 6, "rubric_detail": "说明相邻两个疏散门最近边缘之间的水平距离不应小于5米", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 7, "rubric_detail": "澄清用户提到的《电动汽车分散充电设施工程技术标准》GB/T51313-2018仅为技术标准，只能参考，不能作为依据", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "判定疏散门的开启方向应朝向安全出口所在的相邻防火单元", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "指出疏散门宽度的确定逻辑是依据疏散人数计算结果与最小宽度限制（0.80m）综合决定", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "模型回答未引用相关建筑规范标准进行分析论证，而是直接回答问题", "rubric_weight": -3, "rubric_tag": "行文结构和格式" }, { "rubric_number": 11, "rubric_detail": "回答中大篇幅抄录与地下室车库疏散无关的住宅地上部分或非车库区域的防火规范条款，或者很多不聚焦题目本身的内容，导致冗余，用户阅读起来困难", "rubric_weight": -3, "rubric_tag": "行文结构和格式" }, { "rubric_number": 12, "rubric_detail": "在引用规范时未注明具体的规范名称及具体的条款号", "rubric_weight": -3, "rubric_tag": "行文结构和格式" }, { "rubric_number": 13, "rubric_detail": "模型给出的答案中，防火隔墙上相互连通的门没有采用甲级防火门", "rubric_weight": -8, "rubric_tag": "事实信息" } ] }, { "id": "a961442b-491d-469e-80e6-37c6b0301563", "case_id": 10854, "language": "cn", "system_prompt": "", "question": "假设你是一名从事农业机器人研究的人员，正在设计一种能够在温室番茄授粉与采摘任务中高效运行的机械臂。但是，你在对机械臂进行路径规划时遇到了困难：1，关节角空间（比如RRT,RRT*及其变体）虽然在高维的空间中探索的效率较高，但是在狭窄的环境中避障性能差，路径不稳定；2，笛卡尔空间规划虽然可以实现精确的避障，但是在你运动学求解中容易陷入奇异点或者出现解不可达的问题，特别是在存在柔性障碍（比如番茄的叶子）的场景中，规划失败率极高。现在，请你撰写一份完整的研究方案，目标是使机械臂能够复杂的温室环境中，仍然能够对目标花朵或者果实的高可达性与高可行性路径规划。该方案需要包括以下几点：1，具体的建模思路；2，可行的路径规划策略；3，设计出一整套完整的可复现流程以及实验设计", "tags": { "topics": [ "工业", "系统/嵌入式/3D渲染", "系统/嵌入式/3D渲染" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "提出明确的Relaxed-IK 逆解策略", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 2, "rubric_detail": "方案中需明确界定作业目标（如番茄果实、花朵）与环境障碍物（如叶片、茎秆、支撑结构），并体现对不同类型障碍物（特别是柔性障碍物）的区分。", "rubric_weight": 7, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "方案中应包含具体的实验设计（如明确仿真环境、对比算法、真机验证等环节）以及多维度的评价指标（如规划成功率、路径长度、计算时间、避障效果等）", "rubric_weight": 7, "rubric_tag": "事实信息" }, { "rubric_number": 4, "rubric_detail": "使用 RGB-D 相机进行感知与点云获取", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 5, "rubric_detail": "方案完整性 ，即建模-感知-规划-控制-实验", "rubric_weight": 7, "rubric_tag": "事实信息" }, { "rubric_number": 6, "rubric_detail": "采用具体的 KNN/Kd-tree 进行离群点去除 ", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 7, "rubric_detail": "模型需给出场景代价图的构建公式或方法，且该公式/方法中应包含障碍物距离、碰撞风险或针对柔性障碍物（番茄叶子）的代价因子", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 8, "rubric_detail": "明确的体素化降采样处理 ", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "具体的路径平滑化方案 (如B样条、S-G滤波) ", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 10, "rubric_detail": "设立两个及以上的算法对照组", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 11, "rubric_detail": "提出逆解验证环节，必须回验安全性", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": " 模型应提及根据障碍物特征（如番茄叶片大小）设定体素尺寸", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 13, "rubric_detail": "模型应说明采用自适应权重（或多目标优化函数、模糊逻辑控制）来动态平衡避障与路径平滑度的权重参数。", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 14, "rubric_detail": "未明确给出场景代价图的公式或具体构成", "rubric_weight": -8, "rubric_tag": "事实信息" }, { "rubric_number": 15, "rubric_detail": "混淆“柔性避障”与“完全避障”，导致规划失败", "rubric_weight": -10, "rubric_tag": "观点分析" }, { "rubric_number": 16, "rubric_detail": "缺乏总结与展望章节，没有对方案进行优缺点剖析或对实验结果进行预期的总结", "rubric_weight": -4, "rubric_tag": "行文结构和格式" }, { "rubric_number": 17, "rubric_detail": "未考虑番茄采摘机器人作为边缘设备的算力限制，脱离实际部署约束", "rubric_weight": -10, "rubric_tag": "事实信息" } ] }, { "id": "520cdb68-0103-4ee3-80b5-0466a392a408", "case_id": 10904, "language": "cn", "system_prompt": "", "question": "```\nconst logs = [];\nconst log = (msg) => logs.push(msg);\n\nconst buffer = new Proxy(\n { val: 0 },\n {\n set(target, prop, value) {\n log(`B:Set:${value}`);\n target[prop] = value;\n return true;\n },\n }\n);\n\nconst scheduler = {\n then: (resolve) => {\n log(\"Sched:Then\");\n Promise.resolve().then(() => {\n log(\"Sched:Internal\");\n queueMicrotask(() => {\n log(\"Sched:Resolve\");\n resolve(\"Go\");\n });\n });\n },\n};\n\nasync function* streamProcessor(name) {\n log(`P:${name}:Start`);\n\n const signal = await scheduler;\n log(`P:${name}:Signal:${signal}`);\n\n let current = buffer.val;\n\n yield current;\n\n buffer.val = current + 10;\n log(`P:${name}:End`);\n}\n\nlog(\"Global:Init\");\n\nconst procA = streamProcessor(\"A\");\n\nconst p1 = procA.next();\n\nPromise.resolve()\n .then(() => {\n log(\"Inter:1\");\n buffer.val = 50;\n return \"Inter:Result\";\n })\n .then((res) => {\n log(`Inter:2:${res}`);\n });\n\nconst procB = streamProcessor(\"B\");\nconst p2 = procB.next();\n\nlog(\"Global:End\");\n\n// 最终输出由外部观测\nsetTimeout(() => console.log(logs), 0);\n```\n\n阅读上述代码，回答下列问题：\n1、写出完整的 logs 序列。\n2、在当前脚本的执行生命周期中，P:A:End 是否会被打印，说明原因。\n3、当 p1 完成 Resolve 后，内部的 value 属性的具体值是多少，分析该数值是否受到 Inter:1 的影响。\n4、说明为什么 Inter:2:Inter:Result 会在第一个 Schedule:Resolve 之前输出。", "tags": { "topics": [ "工业", "前端开发", "前端开发" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "判断出被 await 的对象（scheduler）不是原生 Promise（而是 Thenable 对象），其 then 方法的调用会被封装到一个微任务中，而不是同步调用", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": "同步部分代码执行顺序正确，应明确指出同步输出序列为：Global:Init -> P:A:Start -> P:B:Start -> Global:End，且没有在同步任务期间插入微任务输出内容", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "准确判断出 Inter:1 晚于第一个 Sched:Then 执行", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "判断出 Sched:Internal 是由 Promise.resolve().then 创建出的，Sched:Resolve 是由 queueMicrotask 创建出的，两者有严格的前后关系", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "判断出 A、B 两个迭代器分别触发了 scheduler.then 的流程", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "指出代码没有第二次调用 next()，生成器会永远停在 yield current，后续的 + 10 为死代码", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "因为 p1 = procA.next() 在 Inter:1 之前执行，判断出此时已经完成第一个 await 的微任务入队", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "指出由于 await scheduler (Thenable) 会产生额外的微任务来执行 then 方法，导致 scheduler 内部的微任务链入队时间晚于外部的 Promise 链，从而使得 Inter:2 先于 Sched:Resolve 输出", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "回答按照问题分点回答，没有合并在一大段文字中回答", "rubric_weight": 3, "rubric_tag": "行文结构和格式" }, { "rubric_number": 10, "rubric_detail": "模型输出的 logs 序列不正确\n正确的logs：\n[\n 'Global:Init',\n 'P:A:Start',\n 'P:B:Start',\n 'Global:End',\n 'Sched:Then',\n 'Inter:1',\n 'B:Set:50',\n 'Sched:Then',\n 'Sched:Internal',\n 'Inter:2:Inter:Result',\n 'Sched:Internal',\n 'Sched:Resolve',\n 'Sched:Resolve',\n 'P:A:Signal:Go',\n 'P:B:Signal:Go'\n]", "rubric_weight": -10, "rubric_tag": "事实信息" }, { "rubric_number": 11, "rubric_detail": "模型忽略了 Inter:1 对全局 buffer.val 的修改影响", "rubric_weight": -10, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "忽略生成器的特性，认为 procA.next() 能够一次性执行完整个函数", "rubric_weight": -10, "rubric_tag": "观点分析" }, { "rubric_number": 13, "rubric_detail": "回答中包含了大量关于 Promise、Proxy、Generator 的基础语法定义，产生较大的冗余内容", "rubric_weight": -5, "rubric_tag": "观点分析" } ] }, { "id": "bea3bcd5-4ef3-47f2-b563-7300ff506a2a", "case_id": 10915, "language": "cn", "system_prompt": "", "question": "有机电化学晶体管的二元到三元切换一直都是一项比较复杂的工程，研究发现通过利用近红外光也能完成该切换，请详细阐述如何利用近红外光通过光-离子-电子耦合这一路径巧妙的在电化学晶体管中诱导离子产生负微分跨导现象，并依赖这个原理实现的二进制三进制切换。\n1.要求叙述该过程出现的物理机制和关键材料以及调控手段。\n2.叙述这种光控的可重构能力在信息处理方向的未来应用前景。\n3.分阶段描述光诱导氧化还原反应的具体过程，并解释为什么仅在近红外光照和指定电解质存在的时候才能观察到这一现象。", "tags": { "topics": [ "工业", "半导体", "半导体" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "离子与光生空穴协同掺杂阶段涉及溶解在电解液中的I-离子向聚合物沟道迁移的过程", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 2, "rubric_detail": "第一阶段漏极电流的上升的原因在于离子P型掺杂与近红外光产生光空穴的叠加效应", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "光加速氧化还原主导阶段是沟道内的高浓度空穴驱动I-离子反应生成碘三化合物（I3-）", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 4, "rubric_detail": "负微分跨导现象产生的核心原因是生成碘三化合物的过程消耗了空穴导致电导率下降", "rubric_weight": 7, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "第三阶段发生的条件是栅极电压超过谷值电压", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 6, "rubric_detail": "高电压阶段电流重新上升的原因是离子转化趋于饱和且新注入离子的影响占据主导", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "调控手段中包含通过调节红外光强度来控制负微分跨导的峰谷比和电压", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 8, "rubric_detail": "光强与负微分跨导显著程度呈正相关，即光强越大氧化还原越剧烈", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "改变栅极面积是调整电容大小从而调控跨导能力的有效手段", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 10, "rubric_detail": "应用前景明确指向自适应计算或智能传感领域", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 11, "rubric_detail": "回答中包含大量关于有机电化学晶体管基础定义或通用原理的冗余介绍，未聚焦于近红外光诱导NDT的具体机制", "rubric_weight": -6, "rubric_tag": "行文结构和格式" }, { "rubric_number": 12, "rubric_detail": "分阶段描述时缺乏清晰的段落划分或小标题，导致逻辑层次混乱", "rubric_weight": -3, "rubric_tag": "行文结构和格式" }, { "rubric_number": 13, "rubric_detail": "将空穴误写为电子，产生载流子种类错误", "rubric_weight": -7, "rubric_tag": "事实信息" }, { "rubric_number": 14, "rubric_detail": "认为光强只影响开关速度，不影响NDT的调控", "rubric_weight": -3, "rubric_tag": "行文结构和格式" } ] }, { "id": "088a6376-95c1-4b3b-a45e-4a5d19ea4b6e", "case_id": 10919, "language": "cn", "system_prompt": "", "question": "有一个机械零件的建模过程如下：\n一、产生体积的过程\n创建一个等腰梯形，称为初始梯形。\n其下底长度为90mm，上底长度为30mm。腰与下底成45度角。将该梯形向垂直于其所在平面的方向拉伸10mm。\n在该梯形所在平面上，该梯形的下底与一个长度为15mm的边，在初始梯形的外部构成一个长方形。\n将该长方形向与之前梯形的拉伸相同的方向拉伸68mm形成一个长方体e。\n长方体e上面积最大且与初始梯形无接触的面为平面E。\n初始梯形的上底与一个长度为15mm的边在初始梯形之外构成一个长方形，其在初始梯形所在平面上。\n将该长方形向与上述两次拉伸相同的方向拉伸34mm形成长方体f。\n仅由等腰梯形整体拉伸而产生的形状中有一个包含该形状的两条最长边的平面，称这个平面为A。\n创建一个新的平面B，与A平面相距5mm。\n由等腰梯形所拉伸形成的图形中面积最小的面称为平面C。\nB面距离C面的距离比A面距离C面的距离更远。\nB平面上有一个长126mm，宽24mm的长方形和两个半圆。\n这两个半圆的直径边与长方形的两条短边重合。\nB平面上的长方形与两个半圆组合起来，向与之前的三个相同拉伸方向相垂直的方向拉伸，直到到达平面E。\n两个半圆的圆心的连线与初始梯形的下底平行,且这条线与初始梯形所在的平面距离为34mm。\n两个半圆的圆心相对于一个过初始梯形下底中点且与初始梯形所在平面相垂直的平面对称。\n\n二、去除体积的过程\n长方体f上与初始梯形没有接触的面积最大的面为F面。\n长方体f上与初始梯形平行的面中距离初始梯形最远的面为G面。\nF面与G面的相交线的中点处做一个点。以该点为圆心，在F面上做一个半径为16的圆。\n将该圆向与其所在平面垂直的两个方向均拉伸6mm,去除拉伸经过的所有体积。\n长方体f上与F面相平行的面设为H面。\n以H与G面的交线的中点处为圆心,在H面上做一个半径为8mm的圆。\n将该圆向与之垂直的两个方向均拉伸9mm,去除拉伸经过的所有体积。\n在B面上仅有的两个圆心处，在B面上各做一个直径为12mm的圆。\n将该圆向与之垂直的两个方向均拉伸10mm,去除拉伸经过的所有体积。\n\n计算该零件的理论体积。\n目前这个零件还处在所有的曲面、孔都不存在的状态。需加工曲面、钻孔、加工表面粗糙度。\n最终加工完的零件的每一个面都要赋予一个非负整数编号，从1开始。\n零件的材质为45钢。\n夹具仅可使用纯机械夹具，无其他辅助材料。至少用到三对平行的夹持面。\n最终加工完的零件所有与B平面平行的面以及所有曲面粗糙度要求均为Ra0.8。\n设计一个加工方案，说清楚每一道工序夹持的位置和所用刀具。将工序过程用表格展现。\n", "tags": { "topics": [ "工业", "机械与自动化", "机械与自动化" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "模型得出的最终成品零件理论体积为132684.424mm³（允许极小误差，但需接近参考值）", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": "加工方案中列出的刀具至少包含铣刀、钻头和铰刀。", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "方案中使用了已经加工到Ra0.8的面作为夹持面", "rubric_weight": -8, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "模型设计的加工工序体现了先粗加工后精加工的逻辑。", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 5, "rubric_detail": "工序表可以正常渲染。", "rubric_weight": 4, "rubric_tag": "行文结构和格式" }, { "rubric_number": 6, "rubric_detail": "某一道工序的夹持面是此道工序的被加工表面，产生严重的错误。", "rubric_weight": -20, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "方案中使用了至少三对平行的夹持面。", "rubric_weight": 5, "rubric_tag": "行文结构和格式" }, { "rubric_number": 8, "rubric_detail": "正确处理体积的去除部分计算。\n具体算法：\n削去的有两个半圆拉伸成的台和两个圆形拉伸成的圆柱。\n两个半圆拉伸成的台的体积：\n0.5×π×16^2^×6+0.5×π×8^2^×9=3317.522mm³（四舍五入保留三位小数）\n两个圆形拉伸成的圆柱体积：\nπ×6^2^×10×2=2261.947mm³（四舍五入保留三位小数）\n\n\n\n\n", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "模型使用表格形式展示了工序过程。", "rubric_weight": 5, "rubric_tag": "指令遵循" }, { "rubric_number": 10, "rubric_detail": "模型没有识别出所有需要加工表面粗糙度的特定面，具体包括：所有的曲面（如半径16mm和8mm的孔内壁、直径12mm的孔内壁、半圆柱的侧面）以及与B平面平行的面（如平面E、平面B自身等）。", "rubric_weight": -8, "rubric_tag": "事实信息" }, { "rubric_number": 11, "rubric_detail": "模型使用从1开始的非负整数，对最终成品零件的所有面（共23~26个面）进行了编号。", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 12, "rubric_detail": "对于所有要求Ra0.8的表面（与B面平行的面及曲面），模型存在未选择合适的精加工刀具（如精铣刀、磨具等）的情况。", "rubric_weight": -10, "rubric_tag": "观点分析" }, { "rubric_number": 13, "rubric_detail": "正确处理与重叠部分有关的体积计算。\nB平面上的图形拉伸产生的形状与长方体e的形状有重叠。\n\n算法有两种：\n一、可以直接算B平面上的图形拉伸产生的形状与长方体e的形状不重叠的体积：\n[π×12×12+24×(126-90)]×10=13163.893mm³（四舍五入保留三位小数）\n\n二、或者可以算出重叠部分体积（最后一起减掉）：\n90×24×10=21600mm³\n\n\n", "rubric_weight": 6, "rubric_tag": "观点分析" } ] }, { "id": "e79e0354-4848-4896-8c61-47ae21566204", "case_id": 11071, "language": "cn", "system_prompt": "", "question": "我正在进行异种金属电阻点焊研究，选用的研究材料为铝合金和低合金钢。需探究这两种材料在电阻点焊过程中内部的机理变化以及熔核形成过程，现有如下实验条件，铝合金和低合金钢的厚度分别为1.5mm和1mm，低合金钢板在上，铝合金板在下，两块板重叠进行焊接，其中上电极头的直径为6mm，下电极头的直径为16mm，使用的实验设备为中频逆变点焊机。请分析这两种材料在合理的焊接参数下进行焊接时，其焊接接头内部温度随时间的变化过程，内部的熔核形成过程以及形状位置。", "tags": { "topics": [ "工业", "材料", "材料" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "焊接初期，低合金钢侧电流密度大的区域会转移，从一开始的钢与电极接触边缘转移到钢板中心。", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": "焊接初期（约10ms）最高温度出现在上电极头与板接触的外圈区域", "rubric_weight": 3, "rubric_tag": "事实信息" }, { "rubric_number": 3, "rubric_detail": "热量由钢板向铝合金侧扩散，原因是铝合金具有较高的热导率", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "铝合金侧最先达到熔点的区域位于铝钢界面（铝合金上半部分）", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 5, "rubric_detail": "铝钢界面中心温度未达到低合金钢熔点，界面处的钢保持固态", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 6, "rubric_detail": "钢侧电流密度降低及温度开始冷却的原因是钢侧出现压痕导致接触面积增大", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "不同焊接参数下，铝合金熔核可能呈现未贯穿底部或贯穿整个铝合金板两种形式", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 8, "rubric_detail": "铝合金侧冷却结晶形态随温度梯度降低依次呈现平面生长、胞状生长、枝晶生长", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "低合金钢侧冷却过程中，奥氏体向马氏体转变时可能形成宽大板条束", "rubric_weight": 3, "rubric_tag": "事实信息" }, { "rubric_number": 10, "rubric_detail": "铝钢界面的连接机制是熔化态的铝在固态钢表面润湿铺展", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "钢侧表面形成明显压痕，而铝侧压痕不明显，对应上电极6mm和下电极16mm的直径差异", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 12, "rubric_detail": "回答中包含大量与核心问题无关的背景介绍或通用焊接理论，造成严重冗余", "rubric_weight": -10, "rubric_tag": "行文结构和格式" }, { "rubric_number": 13, "rubric_detail": "未按照时间顺序（初期、中期、后期）或逻辑阶段组织内容，导致过程描述混乱", "rubric_weight": -10, "rubric_tag": "行文结构和格式" }, { "rubric_number": 14, "rubric_detail": "在进行分析时，未能分析出铝钢电阻点焊最终呈双熔核特征", "rubric_weight": -10, "rubric_tag": "观点分析" }, { "rubric_number": 15, "rubric_detail": "模型未指出随着焊接进行，电极压痕加深会导致接触面积增大，从而使电流密度降低", "rubric_weight": -5, "rubric_tag": "观点分析" } ] }, { "id": "7c72e1b6-0eab-4e02-9a67-0661576dd212", "case_id": 11913, "language": "cn", "system_prompt": "", "question": "在成都，某独栋住宅一层设有一处面积为4平方米（2米×2米）、朝南且紧邻客厅的开敞小院。业主计划将其封闭改造为具备基本自然通风功能的阳光房，相关规划许可及物业审批手续均已取得。请结合成都地区气候适应性要求与适宜的围护结构构造选型，科学测算该阳光房建安工程的直接成本合理区间，并说明所依据的技术规范、材料性能及造价逻辑。", "tags": { "topics": [ "工业", "建筑设计", "建筑设计" ], "time_sensitivity": { "time_sensitivity": "Strongly time-sensitive", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "回答中引用了成都地区的具体气候参数，如年均日照时数1000-1400小时或年均湿度70%以上", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 2, "rubric_detail": "依据GB 55015-2021标准，计算得出4平方米阳光房的有效通风面积应不小于0.20平方米", "rubric_weight": 9, "rubric_tag": "事实信息" }, { "rubric_number": 3, "rubric_detail": "针对立面高度大于2.0米的情况，引用了GB 15763.2-2020标准关于使用钢化或夹层玻璃的强制要求", "rubric_weight": 7, "rubric_tag": "事实信息" }, { "rubric_number": 4, "rubric_detail": "结合成都多雨气候特征，指出气密性等级需达到6级（GB/T 7106-2019）以防止渗漏", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "未给出立面玻璃的具体配置参数（如“5+12A+5钢化Low-E”），仅使用模糊表述如“双层玻璃”", "rubric_weight": -5, "rubric_tag": "事实信息" }, { "rubric_number": 6, "rubric_detail": "未对比屋面系统（如双层中空夹胶玻璃 vs PC阳光板）在隔热、寿命或成本上的差异", "rubric_weight": -4, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "经济型方案的建安工程总成本估算区间为1.5万至2.3万元人民币", "rubric_weight": 4, "rubric_tag": "事实信息" }, { "rubric_number": 8, "rubric_detail": "高性能方案的建安工程总成本估算区间为2.2万至3.2万元人民币", "rubric_weight": 4, "rubric_tag": "事实信息" }, { "rubric_number": 9, "rubric_detail": "成本逻辑中明确包含了针对4平方米超小工程的“小单加价”或“排产摊销”费用，比例在20%-35%之间", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "提及了成都本地化市场因素，如双流区厂商报价、本地安装价格优势或特定的材料运输成本", "rubric_weight": 4, "rubric_tag": "事实信息" }, { "rubric_number": 11, "rubric_detail": "回复采用了清晰的分层结构，将气候规范、技术选型与成本测算划分为独立的板块进行阐述", "rubric_weight": 4, "rubric_tag": "行文结构和格式" }, { "rubric_number": 12, "rubric_detail": "在列举技术参数或成本数据时，使用了列表、对比项或清晰的数据条目，提升了可读性", "rubric_weight": 3, "rubric_tag": "行文结构和格式" }, { "rubric_number": 13, "rubric_detail": "内容中包含大量与成都气候或4平米小院无关的通用阳光房科普信息，导致篇幅严重冗余", "rubric_weight": -5, "rubric_tag": "行文结构和格式" }, { "rubric_number": 14, "rubric_detail": "未按照题目要求进行科学测算，仅给出笼统的总价而缺乏具体的单价依据或规范来源", "rubric_weight": -5, "rubric_tag": "行文结构和格式" }, { "rubric_number": 15, "rubric_detail": "提供的答案格式混乱难以阅读", "rubric_weight": -3, "rubric_tag": "行文结构和格式" } ] }, { "id": "94e8ae5c-1bf8-49bc-bd40-f774644cc7f1", "case_id": 11917, "language": "cn", "system_prompt": "", "question": "上海某既有住宅位于第5层，其南立面设有一樘方形外窗，净面积为2.0 m²（窗宽约1.41 m，高约1.41 m），窗台距室内完成面高度为0.6 m。该房间在夏季午后至傍晚时段存在明显过热现象，实测室内 operative 温度常超过30℃，影响居住舒适性。建筑外立面属小区统一风格，物业对加装构件的材质、色彩及突出深度有严格管控：\n遮阳装置不得超出外墙立面超过300 mm；\n不得使用高反射率或镜面材料；\n优先采用可调节或被动式低维护构造；\n不得破坏原有窗框结构或影响消防救援窗口功能。\n请结合上海地区夏热冬冷气候特征、太阳辐射规律及该窗的朝向、高度与使用场景，提出一种或多种技术合理、经济可行且符合管理约束的遮阳优化方案，并详细说明其设计原理、关键参数（如遮阳板倾角、挑深、材料性能等）及预期热环境改善效果。", "tags": { "topics": [ "工业", "建筑设计", "建筑设计" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "回答明确指出了上海属于夏热冬冷气候区并给出对应参数：夏季日均太阳辐射峰值可达800–900 W/m²", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 2, "rubric_detail": "方案分析了下午15:00后太阳方位角显著偏西（如58°W以上）的规律，得出单一水平遮阳效率下降、需配合垂直遮阳的结论", "rubric_weight": 9, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "遮阳装置设计超出外墙立面超过300 mm，或未说明如何控制最大外凸", "rubric_weight": -10, "rubric_tag": "指令遵循" }, { "rubric_number": 4, "rubric_detail": "推荐方案中包含固定式浅格栅（蛋格）结构，且明确顶板挑深为300 mm", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 5, "rubric_detail": "固定遮阳方案中设定了两侧垂直翼板的深度约为220 mm，以阻挡西晒", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 6, "rubric_detail": "方案中选用的材料明确排除了高反射率或镜面材料，采用了哑光或漫反射处理（如哑光中灰）", "rubric_weight": 7, "rubric_tag": "指令遵循" }, { "rubric_number": 7, "rubric_detail": "未考虑消防救援窗口功能，或遮阳构件阻碍窗扇全开、无法保障1.0 m净宽", "rubric_weight": -10, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "作为补充方案，建议了使用开孔率约为4%的张紧式遮阳织物屏", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 9, "rubric_detail": "论述了可调节遮阳（如织物屏）在冬季收起以保留被动得热的必要性，体现了对冬夏不同需求的兼顾", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "提及了窗体性能升级选项，具体包括加装断桥铝副框或贴覆纳米陶瓷低反射窗膜", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 11, "rubric_detail": "技术深化部分解释了高红外发射率（ε≥0.85）涂层有助于构件向天空散热，从而降低表面温度的物理机制", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "考虑到上海台风频发的特点，方案中包含了抗风揭设计（如滴水槽导风、双向张紧系统）", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 13, "rubric_detail": "模型应提供成本构成分析（如包含材料费、安装费等）并论证方案的经济可行性", "rubric_weight": 3, "rubric_tag": "事实信息" }, { "rubric_number": 14, "rubric_detail": "明确量化了预期热环境改善效果，指出室内operative温度峰值可降低2.5–3.5℃（或降至26.5–28.0℃）", "rubric_weight": 7, "rubric_tag": "事实信息" }, { "rubric_number": 15, "rubric_detail": "在视觉舒适度方面，设定了室内眩光指数（UGR）控制在19以下的目标", "rubric_weight": 3, "rubric_tag": "事实信息" }, { "rubric_number": 16, "rubric_detail": "回答采用了清晰的层级结构，按照“基础分析-推荐方案-技术深化-经济性-预期效益”的逻辑顺序展开", "rubric_weight": 4, "rubric_tag": "行文结构和格式" }, { "rubric_number": 17, "rubric_detail": "使用了项目符号或编号列表来展示关键参数和技术指标，提升了内容的可读性", "rubric_weight": 2, "rubric_tag": "行文结构和格式" }, { "rubric_number": 18, "rubric_detail": "回答中包含了大量与具体窗户改造无关的通用建筑物理教科书式定义或长篇大论的背景介绍，导致严重冗余", "rubric_weight": -8, "rubric_tag": "行文结构和格式" }, { "rubric_number": 19, "rubric_detail": "输出格式混乱，未正确使用分段或标题，导致关键参数（如尺寸、角度）淹没在连续的文本段落中难以查找", "rubric_weight": -6, "rubric_tag": "行文结构和格式" } ] }, { "id": "38b981dc-4f80-40ee-b568-0ed6bf129f62", "case_id": 11975, "language": "cn", "system_prompt": "", "question": "钠金属电池是目前的热门研究方向，研究发现混合溶剂化电解质策略可能会对钠金属电池的性能提升有一定的帮助，请回答：\n1.混合溶剂化电解质的核心的设计思想是什么，并阐释强溶剂和弱溶剂在混合体系中的各自作用和协同机制。\n2.该技术如何能够同时解决枝晶生长和界面稳定性两大难题。\n3.与传统单一溶剂相比，混合溶剂化策略在电化学性能方面表现出哪些优势，普适性又体现在哪里。", "tags": { "topics": [ "工业", "化工与材料", "化工与材料" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "钠电池混合溶剂化电解质的核心设计思想被明确为通过强溶剂和弱溶剂的合理配比来实现平衡体相离子传输与电极界面稳定性", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 2, "rubric_detail": "强溶剂在体系中的作用被指出是调节溶剂化结构，利用强配位能力形成稳定的溶剂鞘层", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 3, "rubric_detail": "弱溶剂的作用被描述为使钠离子在电极界面处更容易发生脱溶剂化现象", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 4, "rubric_detail": "弱溶剂在电极表面形成更加稳定和致密的界面膜", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 5, "rubric_detail": "混合溶剂化策略相比单一溶剂的优势包含了库伦效率的显著提升", "rubric_weight": 3, "rubric_tag": "事实信息" }, { "rubric_number": 6, "rubric_detail": "混合溶剂化策略带来的性能改进包括循环寿命的大幅延长", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 7, "rubric_detail": "电化学窗口拓宽被列为该策略的优势之一", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 8, "rubric_detail": "该策略能减少活化圈数", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 9, "rubric_detail": "强溶剂通过竞争配位原理有助于抑制副反应并保证离子电导率", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "弱溶剂通过降低能垒（减小活化能）来促进钠离子的均匀沉积", "rubric_weight": 7, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "回答中包含大量关于钠离子电池发展历史或通用电化学原理的背景介绍，导致严重冗余", "rubric_weight": -6, "rubric_tag": "行文结构和格式" }, { "rubric_number": 12, "rubric_detail": "内容堆砌了四个以上化学公式推导，影响了核心结论的阅读流畅度", "rubric_weight": -5, "rubric_tag": "行文结构和格式" }, { "rubric_number": 13, "rubric_detail": "回答未按照设计思想、解决难题机制、优势及普适性的顺序逻辑展开", "rubric_weight": -6, "rubric_tag": "行文结构和格式" }, { "rubric_number": 14, "rubric_detail": "在阐述协同作用时，没有清晰区分强溶剂（如负责盐解离、主导溶剂化）和弱溶剂（如降低粘度、参与界面膜形成）各自的贡献与相互配合关系。", "rubric_weight": -5, "rubric_tag": "行文结构和格式" } ] }, { "id": "b29b17f5-d321-494e-a7fa-c4ec4fdb9a4f", "case_id": 1238, "language": "cn", "system_prompt": "", "question": "题目背景：您是一家领先的电信设备制造商的物理层系统工程师。您负责优化一套位于城市密集区域（Urban Dense）的5G宏蜂窝基站（gNB）部署。该区域高楼林立，用户分布不均，存在大量反射和阻挡，属于典型的非视距（Non-Line-of-Sight, NLOS）和高移动性混合场景。\n挑战任务：为了提升小区间干扰（Inter-Cell Interference, ICI）受限区域的用户边缘速率和系统整体频谱效率，网络规划团队考虑在时域和空域上引入动态TDD（Time Division Duplexing）配置和基于CSI（Channel State Information）的波束赋形（Beamforming）优化。请基于对3GPP Release 15/16中5G NR物理层规范的理解，设计并论证一套协同的上下行动态资源分配（URLLC或eMBB场景）方案。具体要求如下：动态TDD配置策略：详细阐述动态TDD如何实现上下行资源的灵活调度，并说明在解决城市密集区上下行负载不均衡和小区间干扰（如DL到UL的异频干扰）中的具体机制。波束赋形与CSI反馈机制： 解释如何利用Type II CSI（例如非周期（Aperiodic）/半持久（Semi-Persistent）CSI-RS）来精确追踪信道，并设计一个波束赋形增益最大化与干扰抑制相协调的物理层操作流程。时延与可靠性分析：定量分析（或定性论证）该联合方案对URLLC服务的关键性能指标（KPIs），即超低时延（例如 ＜1ms）和超高可靠性（例如 误块率<=0.001%）的影响及优化措施。", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Weakly time-sensitive", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "动态TDD CLI协调的信令细节：必须准确指出 （Slot Format Indicator）是通过 的哪个格式（如DCI2_0 ）下发，并明确提及 SFI中Flexible符号的配置粒度（符号级）。", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": "Type II CSI的增强特性：必须提及Rel-16或更高版本中增强Type II码本的关键优化（如支持多面板或子集限制），并指出CSI-RS测量可以包含干扰测量资源CSI-IM）。", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "URLLC上行免调度（Grant-Free）的细节：必须区分Configured Grant Type 1 RRC预配置和Type 2 PDCCH激活/去激活，并说明Type 2用于URLLC突发场景的低时延优势。", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "干扰抑制预编码的实现条件：提及ZF或MMSE等干扰抑制技术需要精确的邻区干扰信道信息（iCSI），并指出这需要在Xn接口上进行信令交换或测量报告（如RIM，Resource Interference Mitigation）。", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "URLLC下行抢占机制的DCI格式:必须指出下行抢占/中断是通过DCI的哪个格式（如DCI 2_1）通知eMBB UE进行跳过。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "URLLC上行取消（Cancellation）机制的DCI格式：必须指出上行取消传输是通过DCI的哪个格式（如DCI 2_4）通知eMBB UE中断PUSCH。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "URLLC时延分析中的SCS和符号数量：必须提到使用更大的SCS（如30/60 kHz）以及mini-slot（如2-4符号）来定量保证空口时延<0.5ms。", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "波束赋形中的TCI状态应用：必须解释TCI状态配置了哪个物理信道（如PDCCH CORESET）的QCL（Quasi-Co-Location）关联，以保证控制面和数据面的波束一致性。", "rubric_weight": 4, "rubric_tag": "事实信息" }, { "rubric_number": 9, "rubric_detail": "高移动性场景的CSI波束追踪优化：必须提及在高移动性下利用SRS（Sounding Reference Signal）进行上行信道互易性（在TDD下）的快速追踪，作为DL波束的辅助机制。", "rubric_weight": 2, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "NLOS场景下URLLC的可靠性增强：必须提及通过多点传输（Multi-TRP或CoMP联合传输）和Code Block Group（CBG）重传来应对NLOS中突发性阻挡。", "rubric_weight": 1, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "存在术语混淆、滥用，对3GPP术语的理解不精确。例如，错误地将DCI 2_0用于UL Grant（其专用于SFI），或混淆了CSI组件的职能。", "rubric_weight": -10, "rubric_tag": "事实信息" }, { "rubric_number": 12, "rubric_detail": "回答遗漏了对关键工程化约束的讨论，例如对UE PA ON/OFF和快速功率控制等 ", "rubric_weight": -20, "rubric_tag": "观点分析" } ] }, { "id": "a708701f-f2c5-40a6-b14d-d2f77f8924c9", "case_id": 1246, "language": "cn", "system_prompt": "", "question": "你是一位无线网络优化专家，02月03日医院南门1号楼-2小区，存在“不掉话”呼叫占比为97.6%，指标异常；“不卡顿”短视频业务占比为97.17%，指标异常。指标情况如下：\n1）5G掉话次数=10次\n2）掉话率=2.4%\n3）RTT下行时延(ms)=380ms\n这里取了该小区的详细指标提供给你，请你帮忙详细分析一下，问题出在哪里，如何优化，给出的优化建议请给出具体的优化措施，包括参数设置需要给出详细的参数设置过程和参数设置脚本制作过程，详细描述情况如下：\n1、覆盖类指标结果：\n（1）弱覆盖比例（RSRP<=-110比例）= 0.59%\n（2）tadv总采样点=116048个 \n（3）过覆盖采样点数=413个\n（4）过覆盖率=0.36%\n（5）近覆盖采样点数=23533个\n（6）近覆盖率=20.28%（合理范围0~20%）\n2、干扰类指标结果：\n（1）平均每prb干扰噪声功率(dbm)=-95dBm。\n（2）小时级每prb干扰噪声功率(dbm)=-96dBm（24小时最忙时），5G流量（5G上行流量+5G下行流量）=77.34Gb，出现在15点。\n（3）小时级每prb干扰噪声功率(dbm)=-110dBm（24小时最闲时），5G流量（5G上行流量+5G下行流量）=0.92Gb，出现在4点。\n（4）上行干扰类型=无\n3、容量类指标结果：\n（1）上行PRB利用率=87%，出现在18点，指标异常（取全网24小时中最大值）\n（2）下行PRB利用率=93%，出现在17点，指标异常（取全网24小时中最大值）\n（3）arfcn=384000\n（4）5G平均用户数=140\n（5）5G流量（5G上行流量+5G下行流量）=708.48Gb\n（6）联通总流量（联通上行流量+联通下行流量）= 287.86Gb\n4、切换类指标结果：\n（1）5G系统内切换成功率=98.82%、\n（2）5G系统内切换成功率_联通用户=98.97%\n（3）5G 异系统切换成功率 =99.35%\n（4）5G 异系统切换成功率_联通用户=99.4%\n（5）同频切换成功率 =98.93%\n（6）同频切换成功率_联通用户=99.14%\n（7）异频切换成功率 =97.34%\n（8）异频切换成功率_联通用户=96.71%\n（9）同频切换出请求次数 (次)=214311次\n（10）同频切换出请求次数_联通用户(次)=78085次\n（11）同频切换出成功次数 (次)=212022次\n（12）同频切换出成功次数_联通用户(次)=77411次\n（13）异频切换出请求次数 (次)=16413次\n（14）异频切换出请求次数_联通用户(次)=5595次\n（15）异频切换出成功次数 (次)=15977次\n（16）异频切换出成功次数_联通用户(次)=5411次\n5、小区工参结果：\n（1）站高=15.0米\n（2）机械下倾角=5.0度、电子下倾角=0.0度、方位角=180.0度\n（3）小区带宽=100M\n", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Weakly time-sensitive", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "容量类问题分析：PRB 利用率超过合理阈值导致资源拥塞、调度延时和丢包", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": "干扰控制优化：指出平均每PRB干扰噪声功率-95dBm偏高（正常值应低于-105dBm），影响了SINR。", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "覆盖优化：近覆盖率高，可能导致近点用户过于集中，应通过天线下倾角 /方位角 /功率调整优化覆盖，避免过多近点用户", "rubric_weight": 7, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "应指出异频切换成功率（97.34%）偏低，并分析出原因是 A2/A4 门限设置不合理或邻区配置问题", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "针对上行PRB利用率达到87%、下行PRB利用率达到93%所反映的容量问题，提出通过小区分裂或新增站点等方式进行硬件扩容的方案。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "提出通过MLB进行负载均衡，并应提及调整与切换相关的参数，如小区个性化偏移（CIO）、频间/异系统切换迟滞等，来引导用户向邻区切换。", "rubric_weight": 8, "rubric_tag": "指令遵循" }, { "rubric_number": 7, "rubric_detail": "上行功率控制优化：通过调整 P0、Alpha 等参数，限制用户上行发射功率，减少上行干扰", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "时延 / 业务 QoS 优化：应分析高 RTT (380ms) 对短视频业务的影响，并提出通过 QoS 策略进行优化的建议，如提及调整与视频业务相关的 QCI (如 QCI 8/9) 的调度权重或优先级。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "模型应根据近覆盖率（20.28%）略高的情况，提出优化天线配置的建议，核心方向是增加下倾角以抑制近点覆盖。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "指出异频切换成功率偏低（97.34%）的核心根因是小区高负荷（上下行 PRB 利用率超限）引发的干扰累积，并提出具体优化异频切换相关参数（如A3事件门限、迟滞等）以及邻区关系的具体建议。", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "指出开启基于PRB利用率的负载均衡（MLB）功能，同时将同频切换的A3事件触发门限降低1~2dB，以达到协同优化资源分配和切换性能。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "小区扩容建议：当参数优化无法满足需求时，应考虑增加载波、组建新小区或部署更多站点", "rubric_weight": 2, "rubric_tag": "观点分析" }, { "rubric_number": 13, "rubric_detail": "指出平均每PRB干扰噪声功率（-95dBm）较高，并基于此建议进行外部干扰源排查。", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 14, "rubric_detail": "覆盖扩展优化：对小区边缘覆盖进行优化，适当调整功率 /波束 /覆盖范围（例如，针对近覆盖率20.28%略高的情况，提出通过调整下倾角等方式优化覆盖范围），以兼顾边缘用户接入质量", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 15, "rubric_detail": "模型应指出异频切换成功率（97.34%）偏低，并分析可能的原因是异频邻区漏配或参数配置不当。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 16, "rubric_detail": "需明确基站业务调度优先级为全局固定配置，并要求结合小区高负荷根因，先通过负载均衡、扩容解决资源瓶颈，再依托全局 QoS 策略（如保障视频业务 QCI 等级对应的调度权重）优化体验，在此基础上保障为视频业务（如QCI 7/8/9）分配更高的调度优先级，避免 “头疼医头” 的局部参数调整。", "rubric_weight": 5, "rubric_tag": "指令遵循" }, { "rubric_number": 17, "rubric_detail": "时延保障策略：提出具体的时延保障措施，如通过开启SPS（半静态调度）、配置专用QCI、或调整功率控制参数（如P0-NominalPUSCH）等方式来降低时延。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 18, "rubric_detail": "硬件扩容建议：当达到参数优化极限，应考虑加入新载波、新小区或新基站，提升网络容量与稳定性\n", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 19, "rubric_detail": "模型应指出高峰期PRB利用率过高的问题，并建议通过开启负载均衡功能（如IFLB/MLB）等具体策略，将用户调度至其他频点或小区，以实现网络负载的均衡和稳定性。", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 20, "rubric_detail": "针对干扰问题，提出具体的功率控制参数调整建议（如调整P0-NominalPUCCH）或建议开启/优化波束赋形功能。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 21, "rubric_detail": "忽视硬件扩容需求 — 过度依赖参数优化而不考虑载波 /基站扩容，可能导致容量瓶颈无法突破", "rubric_weight": -10, "rubric_tag": "观点分析" }, { "rubric_number": 22, "rubric_detail": "忽略了对物理层（天线/站点/频谱资源）配置和硬件条件的分析，其方案无法从根本上解决问题。", "rubric_weight": -10, "rubric_tag": "观点分析" }, { "rubric_number": 23, "rubric_detail": "忽视小区间干扰和邻区干扰 — 扩大覆盖或调整覆盖可能导致邻区干扰，不加干扰管理容易使问题恶化", "rubric_weight": -10, "rubric_tag": "观点分析" }, { "rubric_number": 24, "rubric_detail": "忽视全天负载平衡 — 只在高峰优化，低峰可能会造成资源浪费或新问题", "rubric_weight": -10, "rubric_tag": "观点分析" }, { "rubric_number": 25, "rubric_detail": "模型未能识别出“切换门限/触发参数设置过激”是导致问题的可能原因之一", "rubric_weight": -10, "rubric_tag": "观点分析" } ] }, { "id": "c4c00b09-7166-4d8c-b03d-273a9b43da6f", "case_id": 1378, "language": "cn", "system_prompt": "", "question": "在某省会城市运营商网络优化中心负责城区 5G NSA 网络优化工作。A 商圈为当地高端商务区，连续三个月出现大量“手机明明显示 5G，网速却比 4G 还慢”的用户投诉，投诉主要集中在晚高峰（18:00–21:00）及周末全天。\n已知该片区无线网络基本情况如下：\n覆盖结构：商圈由 3 个宏站 + 若干室内分布系统覆盖，宏站采用 LTE 1800 MHz（20 MHz）+ LTE 2100 MHz（15 MHz）+ NR n78（60 MHz）组网，5G 采用 NSA 架构（EN-DC），gNB 与 eNB 共站部署。\n业务与终端：A 商圈 5G 终端渗透率 > 75%，以视频、短视频、云办公业务为主，ARPU 显著高于全网平均。\n典型忙时性能：\nLTE 1800/2100 小区下行 PRB 利用率常年在 85%–95%；平均用户下行速率约 8–12 Mbps；\n同覆盖范围内 NR n78 小区下行 PRB 利用率仅 20%–35%；NR 流量占比分布约 8%；\n终端侧 MDT/DT 显示：在商圈室外区域，NR RSRP > -95 dBm 且 NR SINR > 10 dB 的栅格占比超过 70%；室内核心区域 NR RSRP 在 -100 ~ -105 dBm，SINR 多在 5–10 dB。\n典型信令与配置情况：\nNSA 配置中，LTE→NR EN-DC 的 B1 事件门限配置为：NR_RSRP > -110 dBm，TTT = 320 ms，滞回值 3 dB；\nLTE 与 NR 在重叠覆盖区域较大，LTE 层的下倾角略小，部分宏站存在明显过覆盖；\n核心网侧未限制 5G 用户的接入权限，但运营方曾担心 5G 覆盖不稳，将 5G 优先级配置得较为“保守”。\n监控数据还显示：大量 5G 终端在 A 商圈停留期间，80% 以上时长主要由 LTE 承载数据流量，仅在靠近窗户或室外广场时才短暂激活 NR 承载。\n问题：\n请你站在一线无线优化工程师的角度，围绕以下三个维度展开系统分析：\n覆盖与干扰维度：结合给定的 RSRP/SINR 分布，判断当前 5G 覆盖是否构成主要瓶颈，LTE 与 NR 覆盖关系中有哪些潜在问题？\n容量与调度维度：结合 LTE/NR PRB 利用率与 NR 流量占比，推演出“有 5G 不上 5G”“5G 体验不如 4G”的根本容量/承载问题链路。\nNSA 参数与多制式协同维度：基于 B1 门限、TTT、滞回等配置，分析当前 EN-DC 策略对 5G 流量分摊的影响；指出至少 3 个可能导致 5G 资源“吃不饱”的关键参数/策略问题。\n在上述分析基础上，请提出至少三大类可落地的优化方案（例如：参数优化、站型/站位与天线调整、频谱与承载策略等），说明每一类方案中：\n关键调整项及其工程上的合理取值范围或调整方向；\n预期收益及可能带来的副作用或风险；\n用于验证优化效果的关键 KPI/测试方法及验收标准。", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "模型从“覆盖/干扰”、“容量/承载”、“NSA参数/多制式协同”三个维度展开了分析。", "rubric_weight": 4, "rubric_tag": "指令遵循" }, { "rubric_number": 2, "rubric_detail": "明确指出问题的主矛盾是LTE过载、NR资源利用不足，而非单纯的5G覆盖问题。", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "用编号或箭头形式写出一条不少于 4 步的因果链，比如：\n1）业务集中 + 5G 终端多 →\n2）仍以 LTE 锚点承载 →\n3）NR 条件可用但 EN-DC 触发/保持保守 →\n4）LTE 拥塞、NR 闲置 → 用户感知“有 5G 不如 4G”。\n分析中至少要同时出现：\n覆盖/干扰因素\n容量/负载因素\nNSA/参数或架构因素", "rubric_weight": 9, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "在 NSA 参数分析中，对 B1 门限、TTT、滞回三个参数都给出“变大/变小对 EN-DC 激活频率与稳定性的影响方向”，例如：\nB1 门限越低（dBm 数值越负）→ 加 NR 范围越大、越容易触发；\nTTT 越长 → 激活更晚、更稳；越短 → 更快但易抖动；\n滞回越小 → 更容易频繁加/释放。\n三项方向都要说到且不能写反。", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "设计“基于 LTE 负载驱动上 NR”的策略时，需要描述：\n明确的 LTE PRB 阈值（如 >70% 或 >80%）；\n以及它如何与 EN-DC 触发条件联动，例如在高负载时动态放宽 B1 门限、缩短 TTT 或降低滞回，而不是只提高调度权重或静态门限。", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "对室内核心区域的 NR 改造，不仅提出“上室分/小微站”，还要提供至少两项工程级设计要素，例如：\npRRU/皮站 MIMO 配置（2×2、4×4 等）；\n单点 EIRP 目标范围；\n楼层/站点数的基本估算思路；\n并将这些要素和室内目标 RSRP/SINR/速率联系起来。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "在频谱与承载策略部分，既谈到 LTE → NR 频谱重耕 / DSS，又要明确提到：\nNSA 与 SA 在 5G 演进中的角色区别；\n频谱策略如何与 NSA→SA 的演进路线配合（例如：先 NSA+重耕，再在热点/全网推进 SA）。", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "至少在一个关键优化方向上，同时给出：\n需要监控的 1–2 个具体指标（如 SCG Change 次数、X2-U 丢包率、终端平均电流等）；\n当这些指标达到某个明确阈值或变化幅度时，要执行怎样的回退或收紧动作。\n不能只是泛泛而谈“需要监控风险”“注意 KPI”。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "给出不少于 5 个不同维度的 KPI，并为每个 KPI 指定明确的目标值或范围，同时说明测试/统计方法。典型维度包括：\nLTE/NR PRB 利用率\nNR 流量占比\nEN-DC/SCG 相关指标\n业务速率（P50/P90 等）\n投诉或工单数量", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "把优化方案拆成至少 3 大类不同性质的工程动作，例如：\nNSA / 调度 / 分流策略\n覆盖 / 站型 / 天线工程\n室内补点 / 频谱与承载策略 / 架构演进\n每一类中，至少写出 2 个可以直接下发或规划的具体动作（参数名或工程动作）。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "回答几乎完全把用户体验问题归结为5G 覆盖严重不足/n78 物理特性，基本忽视题干中的 LTE/NR PRB、NR 流量占比等负载信息，也没有把容量/承载当主矛盾。", "rubric_weight": -8, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "模型在解释B1事件门限时，错误地将-110 dBm理解为高门限（例如，认为UE需要离基站很近才能满足）.", "rubric_weight": -8, "rubric_tag": "事实信息" } ] }, { "id": "af4dcf04-29fe-4ba3-8d2f-d1bdae0406ab", "case_id": 1381, "language": "cn", "system_prompt": "", "question": "运营商需要在城市部署5G SA网络，近期接到某高端商业楼（共8层，地下2层为停车场及设备机房，地上6层为商铺、餐饮及办公区，建筑总面积约12万㎡）用户集中投诉，反映以下问题：1. 地下一层停车场区域5G信号弱，多数区域无法接入5G网络，仅部分角落可接入但速率低于1Mbps；2. 地上3-4层餐饮区用餐高峰期（12:00-14:00、18:00-20:00）5G速率骤降，网页加载缓慢、视频卡顿，非高峰期速率基本正常（约300-500Mbps）；3. 办公区部分区域出现5G信号频繁切换现象，通话过程中偶有掉话情况。该商业综合体周边500米内已部署3个宏基站，分别为A站（Band n78，3.5GHz，发射功率46dBm）、B站（Band n41，2.6GHz，发射功率45dBm）、C站（Band n79，4.9GHz，发射功率47dBm），建筑内已零散部署12个分布式皮站（Band n78，3.5GHz，发射功率30dBm）。假设你是该运营商的5G网络优化工程师，需结合3GPP 5G网络部署标准，分析上述三类问题的原因，并制定针对性的优化方案（需明确优化步骤、采用的技术手段及关键参数调整建议），同时说明优化过程中需重点规避的风险及应对措施。", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "准确指出地下一层信号弱的三大核心成因（频段穿透损耗、皮站覆盖不足、建筑金属干扰）", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": "准确指出餐饮区高峰期速率低的核心成因（用户并发过载、多径干扰、重叠覆盖干扰）", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "准确指出办公区切换的核心成因（覆盖边界模糊、切换参数不当、信号盲区、电磁干扰）", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "地下一层优化方案包含皮站点位规划（800-1000㎡/ 个）及功率调整（33dBm 左右），符合 3GPP 标准", "rubric_weight": 5, "rubric_tag": "指令遵循" }, { "rubric_number": 5, "rubric_detail": "餐饮区优化方案包含小区呼吸功能开启及 “比例公平 + 用户优先级” 调度算法调整", "rubric_weight": 5, "rubric_tag": "指令遵循" }, { "rubric_number": 6, "rubric_detail": "办公区优化方案包含切换参数（A3 阈值、CIO、迟滞时间），符合 3GPP 协议", "rubric_weight": 5, "rubric_tag": "指令遵循" }, { "rubric_number": 7, "rubric_detail": "优化方案中考虑城市商业楼（石质墙体、厚重混凝土）对信号的影响，石质墙体、厚重混凝土等建筑材料会导致严重的信号衰减/穿透损耗", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "地下一层优化采用分布式天线系统（DAS）辅助覆盖，解决金属遮挡问题", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "餐饮区优化提及调制编码方式与 SINR 的关联", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "办公区优化包含 RLF 重建立功能开启及 VoNR 业务 QoS 配置（如为语音业务配置高优先级的QCI，如QCI 1）", "rubric_weight": 3, "rubric_tag": "指令遵循" }, { "rubric_number": 11, "rubric_detail": "风险应对措施应包含适配商业场景的施工协调方案（如夜间施工、与物业沟通）。”", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "风险应对措施包含参数回滚机制及试点推广模式，保障网络稳定性", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 13, "rubric_detail": "地下一层优化推荐 RU 发射功率 20-24dBm，显著低于 3GPP 推荐的 30-35dBm 范围，导致覆盖不足", "rubric_weight": -4, "rubric_tag": "观点分析" }, { "rubric_number": 14, "rubric_detail": "餐饮区优化未提及多径干扰，仅聚焦容量与同步干扰，遗漏核心损耗因素", "rubric_weight": -3, "rubric_tag": "观点分析" }, { "rubric_number": 15, "rubric_detail": "模型未能分析出地下一层因未采用分布式天线系统（DAS）而导致信号被金属遮挡的问题。", "rubric_weight": -3, "rubric_tag": "观点分析" } ] }, { "id": "ebb94405-b0af-4485-8efa-c95c3971a7b1", "case_id": 1394, "language": "cn", "system_prompt": "", "question": "你是某城市地铁通信网络高级专家近期，地铁2号线、4号线及两条线路的控制中心均报告了复合型通信故障，现象复杂且相互关联，对运营安全构成潜在风险。故障集中爆发于早晚高峰，但部分异常在非高峰时段亦有不同表现。\n故障内容与多维数据：\n一、传输网指标\n（1）2号线（主环）性能劣化：\n高峰时段： “VC-12路径信号丢失（TU-LOP）”告警集中出现在“中山公园站-世纪大道站”区段。该区段再生段误码率（RS-BER）从闲时背景水平1.0E-8劣化至1.2E-5，超出1.0E-6门限。误码性能事件分析报告显示，误码类型以突发性误码（Burst Error）为主。\n全天监测异常： 网管性能历史数据揭示，该区段光路存在高频率的瞬时光功率瞬断（Transient Loss of Light）事件，每次持续时间小于1毫秒，未触发传统“LOS”告警。该事件发生频率已从一周前的每日数次，激增至当前的每秒数次。\n（2）4号线（备环）压力敏感型劣化：\n高峰时段： 接收光功率（Rx Power）从基线-12dBm劣化至-22dBm，同时伴随“以太网端口循环冗余校验（CRC）错误计数”每小时暴增约1500次。\n矛盾现象： 在凌晨3点业务闲时执行的自动化端到端性能测试中，光功率与误码率指标均完全正常。然而，当使用同一测试脚本、但在链路上施加模拟高峰流量压力时，光功率劣化与CRC错误现象可被立即、稳定复现。\n（3）控制中心互联通道时延抖动：\n承载于2、4号线传输环上的控制中心至车站间通信通道，在高峰时段网络层时延抖动（Jitter）高达±50ms，严重偏离正常值（应＜±5ms）。\n二、业务层表现\n（1）视频监控系统：高峰时段视频流平均丢包率介于8%~15%，画面卡顿严重；协议层深度分析发现部分摄像头的流媒体协议信令（如RTSP TEARDOWN）在高峰时段存在异常高频重传，表明传输层的不稳定性已引发TCP/UDP会话层频繁重建。\n（2）TETRA无线集群系统：话音信道指配失败率在高峰时段达到6.3%（KPI门限＜1%）。信令跟踪（Trace）关键定位： 失败主因并非无线空口资源不足，而是基站控制器（TSC）与核心网交换机（DXT）之间的周期性“链路检测报文”（如BFD或Keepalive）出现丢失或严重延迟，导致控制平面误判对端故障，进而拒绝分配业务信道。\n（3）列车控制（CBTC）系统数据回传：信号系统报告，同期CBTC网络的“车地通信无线接入点（AP）”与地面服务器之间的心跳报文，在高峰时段出现偶发性超时。该事件虽未触发系统保护倒换，但已被记录为最高等级潜在风险事件。\n三、环境、历史与变更\n（1）外部环境风险：除已知的2号线“中山公园站”附近市政施工外，4号线“体育中心站”一周前亦有第三方顶管施工，且两家施工单位不同，施工图纸与地铁光缆路由存在交叉风险区。\n（2）网络配置遗留问题：历史操作日志显示，一年前因一次光缆中断，曾临时调整2号线MSP环的“西向”与“东向”光纤定义以快速恢复业务。事后核查记录不完整，存在“逻辑主用路由”与“物理拓扑最优路由”长期不一致的潜在可能。\n（3）近期变更与配置：2号线传输环在3天前执行“光路保护倒换测试”后，未按规程恢复至原主用路由；4号线正在进行“车站Wi-Fi 6升级改造”，部分临时接入的以太网交换机启用了“流量控制（Flow Control）”，且接入端口未配置带宽限速（Rate-Limit）。\n（4）动力与环境：“中山公园站”通信设备机柜存在“温升异常”告警；该站UPS日志显示，近期有数次持续时间小于10毫秒的市电短时中断记录，设备在此期间由蓄电池供电支撑。\n四、资源与约束\n（1）每日可用于施工与测试的“天窗点”时间窗口仅3小时。\n（2）可用于精确定位瞬断故障的高精度仪表（如可捕捉µs级事件的光功率计/OTDR）数量有限，需科学规划部署点位。\n（3）运营安全规定：任何可能中断TETRA或CBTC等关键生产业务的测试操作，必须提前48小时提交审批，且必须具备经认证的、可实现业务瞬间回退的应急操作预案（Rollback Plan）。\n请你作为高级专家，系统性完成以下任务：\n（1）构建综合故障假说： 提出一个统一的根本性原因假说。该假说必须能同时、连贯地解释2号线与4号线在传输层表现出的时空差异特性（例如，为何4号线问题仅在流量压力下显现）。\n（2）视频、TETRA、CBTC这三个不同业务系统表现出的差异化失效模式（丢包、信令超时、心跳延迟）。\n（3）设计精确定位与验证方案，如何利用有限的精密仪表资源，设计一个高效的测试方案，以捕捉并定量证明引发故障的瞬时物理现象（如光瞬断）？\n（4）如何在不断网、不中断关键业务的前提下，验证一年前的临时配置调整是否导致当前 “逻辑主用路由”与“物理最优路由”存在偏差，以及此偏差对MSP保护倒换逻辑的实际影响？\n（5）制定分阶段处置与优化策略：紧急缓解阶段： 提出一个能在下一个早高峰前实施的、风险最低的配置调整方案，以优先保障TETRA和CBTC业务的稳定性。长期阶段： 规划一个包含多个“天窗点”的修复工程方案，需具体说明每一步的操作内容、验证方法及回退步骤。\n（6）系统优化阶段： 提出针对网络架构、监控能力、运维流程的长期改进建议，以提升网络韧性，防止同类复杂故障再次发生。\n（7）最后撰写呈报管理层的专家报告： 报告需以清晰、严谨的结构呈现上述分析、方案与策略，并准确评估技术风险等级、处置优先级及所需的资源支持，用于辅助管理决策。", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Strongly time-sensitive", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "提出统一的根本原因假说，近期两处独立的市政顶管施工，对地铁光缆造成了具有“振动敏感”与“温度/负载敏感”特性的物理微损伤。此损伤在业务高峰时段，通过“逻辑主用路由配置偏差”与“数据链路层流量控制（Flow Control）滥用”两大内部恶化因素的级联放大，最终引发了跨业务、跨线路的复合型性能崩溃。", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": "构建从物理层到业务层的完整、分层的因果链条", "rubric_weight": 7, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "紧急缓解方案应优先针对4号线因Wi-Fi 6升级引入的“流量控制（Flow Control）”配置问题，和2号线光路保护倒换后未恢复原路由的问题。", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "方案应明确使用可捕捉微秒级事件的高精度光功率计/OTDR，并部署在2号线“中山公园站-世纪大道站”区段两端，以及4号线“体育中心站”附近光缆路由的关键节点。测试需覆盖闲时与高峰（或模拟高峰）时段，以关联光功率瞬断与误码/CRC错误。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "方案应包含至少一种非侵入式的验证方法，如通过网管系统（NMS）导出并比对设备当前配置与历史基线配置、使用traceroute等工具探测逻辑路径并与物理拓扑图进行比对、分析MSP/APS协议状态信息等。", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "构建并评估多个相互竞争的故障假说，并通过排除法验证：在提出最终假说前，至少考虑并排除了其他两种合理的可能性（如单纯的温度故障、时钟同步问题）。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "构建的因果链存在“跳跃式推理”：在缺乏必要中间机制解释的情况下，直接断言A导致B。例如，仅说“光缆受损导致TETRA指配失败”，但没有解释中间的传输层误码、网络层抖动、控制协议超时等环节。", "rubric_weight": -4, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "提出的“出现问题的根本原因”过于宽泛或笼统，缺乏可验证的物理或逻辑实体：例如，将根本原因归结为“网络不稳定”、“系统设计缺陷”或“运维不力”。", "rubric_weight": -6, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "将本次故障的根本原因与失效模式，归纳并映射到组织的“已知错误数据库”或“故障模式库”中，并提出更新巡检项（如增加逻辑与物理路由一致性核查）、设计规范（如要求设备具备抗微振动能力）", "rubric_weight": 2, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "忽视了“社会技术系统”因素，整个回答完全聚焦于技术层面的诊断、验证与修复。它没有在分析、策略或报告的任何部分，提及解决此类涉及市政施工、多部门协作的故障所必需的非技术协调工作。", "rubric_weight": -6, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "明确说明各项操作的“优先级顺序”和“依赖关系”，例如，是否必须先关闭流量控制（消除噪声源）再调整QoS（优化信号），以避免在拥塞环境下调整优先级可能带来的不可预知影响；硬件直换是否应在配置调整之后，作为验证手段；", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "在分析中，默认所有建议的监控工具和数据（如微秒级光功率记录、TWAMP时延热图）可以无障碍地接入现有网管系统并实现关联分析，而忽略了系统集成和数据治理的复杂性。", "rubric_weight": -6, "rubric_tag": "观点分析" }, { "rubric_number": 13, "rubric_detail": "紧急方案需提出具体的配置调整措施（如优化QoS/CoS策略、关闭4号线交换机流控、设置带宽限速等）以优先保障TETRA和CBTC业务，并明确其风险可控（如下一个早高峰前可实施、具备回退计划）。", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 14, "rubric_detail": "模型在推导结论时，应简要说明排除其他可能性的依据，如指出“非永久性光缆中断”（依据是故障具有间歇性/突发性）或“非单纯设备老化”（依据是故障与流量压力或特定时段强相关）。", "rubric_weight": 2, "rubric_tag": "观点分析" } ] }, { "id": "e771ab18-f120-4c54-8450-aa2367b71bf6", "case_id": 161, "language": "cn", "system_prompt": "", "question": "金属/半导体异质结中等离激元产生的热载流子转移到一直面临极大的能量耗散问题，最新研究发现金/氮化镓界面存在超快非热化电子转移路径。请回答 1什么是超快非热化电子转移？与热载流子转移区别是什么？\n2如何证明电子转移是非热化的？需要观测哪些特征？\n3光偏振调控如何影响电子转移效率和能量分布？\n4该方法对热载流子器件的设计有何启示？", "tags": { "topics": [ "工业", "半导体", "半导体" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "非热化电子转移定义：电子未经热化直接注入半导体的超快过程", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 2, "rubric_detail": "阐述了与传统转移的区别在于：非热化转移保留了原始高能量，而热载流子转移会形成热化分布", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "实验观测技术：超快光谱在能量/时间维度的同步解析 （TR‑2PPE+SPVM）", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "关键时间尺度：注入时间<100飞秒，快于电子-电子散射的500飞秒（两个数据均要提及）", "rubric_weight": 10, "rubric_tag": "指令遵循" }, { "rubric_number": 5, "rubric_detail": "能量分布特征：与初始等离激元分布相关（需明确提出非麦克斯韦分布概念）", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "偏振调控机制一：通过等离激元模式影响电荷产率 （必须提出等离基元概念）", "rubric_weight": 4, "rubric_tag": "指令遵循" }, { "rubric_number": 7, "rubric_detail": "偏振调控机制二：通过动量分布改变注入概率 ", "rubric_weight": 2, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "肖特基势垒作用：高能电子更易越过的界面势垒（要求单独点明肖特基势垒的作用）", "rubric_weight": 2, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "弹道输运特性：近似无能量损失的传输方式 （要明确指出这种五能量损失的输出方式是弹道运输）", "rubric_weight": 8, "rubric_tag": "指令遵循" }, { "rubric_number": 10, "rubric_detail": "热载流子寿命：通常在飞秒到皮秒量级（需要强调的是寿命在飞秒到皮秒量级，不是发射过程在飞秒到皮秒量级）", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "混淆'非热化'与'低温环境'，误认为需要冷却系统", "rubric_weight": -8, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "认为偏振只影响激发强度而忽略其对能量分布的调控", "rubric_weight": -5, "rubric_tag": "观点分析" } ] }, { "id": "4bd30355-9179-4db2-b19f-acfe08b71833", "case_id": 1650, "language": "cn", "system_prompt": "", "question": "在一个包含LTE 2.6GHz(Band 7)宏小区和密集部署的NR 3.5GHz(Band n78)小基站（Small Cell）的城市热点区域（例如，商业中心）中。为提升用户体验和系统容量，网络规划部门采用了LTE-NR Dual Connectivity (EN-DC) 技术。\n宏站（LTE eNB）负责提供广域覆盖和锚点（Master Cell Group, MCG），而小站（NR gNB）则作为辅小区（Secondary Cell Group, SCG）提供高容量热点服务。\n然而，在NR小基站覆盖区域内，由于NR 3.5GHz小基站的发射功率较高且部署密度大，导致相邻 NR 小基站之间以及 NR 小基站与 LTE 宏站之间存在严重的下行链路干扰，特别是在NR小区边缘用户处，其SCG RSRP/RSRQ值较低，SCG添加/切换成功率下降，进而影响 EN-DC 连接性能和用户的吞吐量。\n请设计并阐述一套针对上述场景的干扰管理与协同机制，并重点说明以下两点：\n具体采用哪种 3GPP 标准中定义的干扰管理技术或特性？(例如，ICIC, CoMP, TPC, CRS/CSI-RS 功率控制等，需选择最适合该异构网络场景的技术)\n该机制如何通过网络间的协同工作（特别是在LTE eNB 和 NR gNB 之间或 gNB 之间）来有效缓解NR小区边缘用户的下行干扰，并最终实现哪些关键性能指标（KPI）的改善？", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "提出由于UE在LTE B7和NR n78频段同时进行上行传输时，可能因谐波或互调产物导致带外发射超标的问题，并阐述对应的UE侧干扰管理机制，如A-MPR（附加最大发射功率回退）", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": "阐述在NR小站簇中，需确保符号级/时隙格式一致性，并指出这不仅是为了防止DL→DL干扰，更是为了防止邻区gNB的DL发射对本区 gNB的 UL接收造成符号级干扰", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 3, "rubric_detail": "描述 eNB(MeNB) 如何基于SCG质量 (SCG-RSRQ/PDCCH BLER) 动态调整PDCP分流比（Split Bearer），或通过SCG激活/去激活 (37.340流程) 将边缘UE的流量平滑回撤至LTE MCG，以保障连接稳定。", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "提及为UE配置多个nzp-CSI-RS资源集，每个资源集对应一个候选的TRP（传输点），并配置CSI-IM资源用于测量来自其他TRP的干扰，从而支持UE上报多点相关的CSI", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 5, "rubric_detail": "模型描述了在 CoMP/DPS 场景下，为支持快速、瞬时的调度决策，gNB 调度器通过 MAC-CE 或 RRC 信令触发 UE 进行非周期或半持续的 CSI 瞬时报告机制", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "给出 $\\text{NR}$ $\\text{gNB}$ 间 $\\text{CoMP}$ 协同所需的具体 $\\text{Xn}$ 接口信息或过程名称，例如$\\text{TRP}$ 标识符共享、$\\text{Rate Matching}$ $\\text{Pattern}$ 协商或$\\text{PDSCH}$ 资源动态分配请求", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 7, "rubric_detail": "描述为保障边缘 $\\text{UE}$ 的控制信道可靠性，需采取配置专用 $\\text{CORESET}$、提高 $\\text{PDCCH}$ 聚合级别（AL），以及对 $\\text{PDCCH}$ 适度功率提升等手段", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "给出清晰、量化的 $\\text{KPI}$ 改善预期，例如 $\\text{NR}$ 边缘 $\\text{SINR}$ 提升的具体 $\\text{dB}$ 值，或 $\\text{SCG}$ 成功率提升的百分比范围（如 $\\text{15\\%–30\\%}$）", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "识别 基于 $\\text{BWP}$（带宽部分）划分“中心/边缘”区域的分频复用思路，并将其与 $\\text{LTE}$ 中的 $\\text{FFR}$（分数频率复用）概念关联。", "rubric_weight": 1, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "准确识别 $\\text{n78}$ 小站间的同频干扰 ($\\text{gNB-gNB}$) 和 $\\text{LTE B7}$ 宏站对 $\\text{NR}$ 的跨制式干扰 ($\\text{eNB-gNB}$) 是本场景的两个主要干扰源。", "rubric_weight": 1, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "模型将LTE的eICIC/ABS错误地作为NR小站间同频干扰（gNB-gNB）的核心解决方案", "rubric_weight": -5, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "模型完全未提及或未正确阐述 NR 核心的 Rate Matching（速率匹配）机制，即通过协商静默图指示 UE 在特定资源上避开邻区 PDSCH 或参考信号的做法。", "rubric_weight": -10, "rubric_tag": "观点分析" } ] }, { "id": "c2a430c0-7f98-4ab5-af84-93824c790d72", "case_id": 1678, "language": "cn", "system_prompt": "", "question": "设计一款1分8不等分功分器，具体要求如下：工作频率：9.3~9.5 GHz，分路口1~分路口8的理论分配损耗分别为-5.4 dB、-6.0 dB、-7.1 dB、-8.8 dB、-11.3 dB、-14.6 dB、-18.8 dB、-22.7 dB，各分录通道除分配损耗外的插入损耗≤1.2 dB，公共口驻波比≤1.3，各分路口驻波比≤1.3，各分路口之间的隔离度≥20 dB，各分路通道之间的幅度误差≤0.5 dB、相位误差≤5 °。\n输出要求：公共口和各分路口分列在壳体两端；各分路口中心间距100 mm；以低插入损耗为最重要考虑因素，给出优选和备选两种实现方式，如带状线、微带线、基片集成波导、悬置带线、金属波导等；给出适合于优选和备选两种实现方式分别对应的拓扑结构，如Wilkinson、Gysel和T-Junction等；分别给出优选和备选两种方案各级的长度、宽度、传输线阻抗值、隔离电阻值；以公共口中心点为原点，分别给出优选和备选两种方案各分路口中心的位置坐标；如有不满足项，需明确指出，并给出必要的建议。", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "优选方案的实现方式选择了低损耗的悬置微带线方案", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": "优选方案或备选方案的实现方式采用基片集成波导", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "优选或备选方案的拓扑结构采用Wilkinson拓扑结构", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "优选方案或备选方案中给出了各分路口的功率分配比，且数值与给定的分配损耗相对应，分路口1~8对应的功率分配比分别约为：0.288, 0.251, 0.195, 0.132, 0.074, 0.035, 0.013, 0.005", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "优选方案或备选方案中给出了各级传输线阻抗，因方案不同，该系列阻抗具体数值不必验证", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "优选方案或备选方案中给出了各级隔离电阻值", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "优选或备选方案中给出了各级传输线的长度、宽度", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "优选或备选方案中给出了各分路口的具体坐标，且坐标的排布满足“各分路口中心间距100mm”的要求。", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "各端口驻波比满足≤1.3", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 10, "rubric_detail": "各分路口之间的隔离度满足≥20 dB", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 11, "rubric_detail": "优选或备选方案中采用了不适合低插入损耗要求的微带线方案。", "rubric_weight": -20, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "优选或备选方案提到宽度≤0.3mm的尺寸", "rubric_weight": -10, "rubric_tag": "观点分析" } ] }, { "id": "42f2f642-deb1-46c6-a47e-b78c51b2cbcc", "case_id": 1749, "language": "cn", "system_prompt": "", "question": "复杂电磁环境下LFM雷达信号的盲自适应处理技术攻关\n一、 项目背景与挑战\n你正在研发的新一代远程监视雷达系统，其核心任务是在广域范围内实现对高动态、低可观测性目标的可靠探测与跟踪。系统拟采用一套成熟的LFM波形体制（带宽12MHz，脉宽110μs，载频1.5GHz）作为基础，以确保足够的探测距离和距离分辨率。然而，根据最新的情报评估和战场环境模拟，该系统在未来部署中将面临前所未有的严峻挑战。我们预计，信号传播路径将不再是理想的自由空间，而是充满着3到5条动态变化的强杂波路径。这些路径会引入显著且快速变化的时延（分布在0.1-2μs）、衰减（信号强度可能跌至原始强度的30%）以及由目标或杂波微动引起的复杂多普勒效应（±2kHz）。更棘手的是，信道本身的一致性很短，可能在几十个脉冲周期（约50-100 PRI）内就发生显著变化。与此同时，你必须假设对手将部署先进的电子对抗（ECM）手段。这包括能够精准模仿你方信号调频率的“欺骗式干扰”（其调频斜率与我方真实信号的差异可能低至5%），以及能够快速跟踪我方信号频谱并进行阻塞的“灵巧噪声干扰”。雪上加霜的是，本次技术攻关必须基于现有的、已定型的硬件平台。该平台的数字后端仅配备了一颗16位定点DSP，其板载内存最多只能缓存20个脉冲周期的原始采样数据。前端的12位ADC，其总谐波失真（THD）约为-60dB，而本振的相位噪声性能（@1kHz频偏处为-80dBc/Hz）也并非顶级。这些硬件上的“先天不足”将直接制约算法的性能上限和实现复杂度。\n二、 核心技术攻关目标\n为应对上述挑战，团队需要设计并验证一套能够在这些约束下稳定运行的盲信号处理算法具体攻关目标分解如下：\n1. 动态信道与威胁环境的盲感知与分离\n首要任务是在完全无先验信息的条件下，对接收到的混合信号进行“解剖”，实时地描绘出信道和威胁的全貌。\n信道参数的鲁棒解耦：一个长期困扰你的理论难题是LFM信号中时延和多普勒的内在耦合效应——在单脉冲内，一个微小的时延足以产生比目标真实多普勒大得多的等效频移。你的方案必须能从根本上打破这种模糊性，利用那宝贵的20个脉冲缓存，在信道发生变化前，同时给出所有路径的时延、衰减和多普勒的精确估计。要求的最终精度指标是：时延估计均方根误差低于0.05μs，多普勒估计均方根误差低于50Hz。\n真伪目标的微观指纹识别：当真实目标回波与高仿真度欺骗干扰（例如在-5dB信干比下）同时存在时，传统的脉冲压缩将完全失效。需要一种更深层次的识别机制。你能否挖掘LFM信号相位结构中，那些被传统二阶分析所忽略的、更高阶的“微观特征”？请设计一种方法，利用这些特征为真假目标打上独特的“指纹”，并给出一种可量化的置信度评估，以判断分离结果的可靠性。同时，必须分析那块非理想的ADC将如何扭曲这些精细的相位特征，并评估其对识别性能的潜在影响。\n2. 硬件内生缺陷的建模与自适应补偿\n一个算法如果不能在我们的16位定点平台上稳定运行，那它就没有工程价值。因此，方案设计必须与硬件实现紧密结合。\n脉压性能的系统性损伤评估：本振的相位噪声和ADC的量化误差会如何共同“毒化”脉冲压缩性能？你需要一个能够定量预测峰值旁瓣比（PSLR）恶化程度的理论模型，它必须能直接关联相噪功率谱、量化比特数等硬件参数，而不是简单的仿真拟合。这个模型将是整个补偿策略的理论基石。\n定点环境下的自适应补偿与收敛性保证：核心挑战在于，设计一套能够在16位定点精度下实现的补偿算法，目标是将恶化的PSLR恢复至-30dB以上。然而，任何基于迭代的补偿算法（如自适应滤波）在定点运算中都存在误差累积和发散的风险。你的方案必须从理论上论证，在有限字长下，算法如何确保收敛而不是崩溃？定点运算带来的舍入误差，是否意味着存在一个永远无法突破的性能补偿下限？\n三、 交付成果与评估要点\n请以技术报告的形式，提交一份详尽的解决方案（篇幅约800-1000字）。报告的评估将重点关注以下几个方面，而不仅仅是算法流程的描述：\n对核心物理/数学问题的洞察深度：例如，对LFM信号参数耦合本质的解释，以及你的解耦算法为何能在理论上成立。\n创新性与可行性的平衡：高阶相位分析的理论极限是什么？你的方法在低信噪比下能多大程度上逼近这个极限？\n系统性思维与工程实践的结合：如何从理论上化解“补偿精度”与“定点运算误差”之间的矛盾？对算法在受限硬件上的数值稳定性和性能边界是否有清晰的认识？", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "回答指出了本振相位噪声在1kHz频偏处的功率谱密度为-80 dBc/Hz", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 2, "rubric_detail": "回答计算得出信号带宽内的积分相噪功率约为1.2乘以10的负1次方", "rubric_weight": 4, "rubric_tag": "事实信息" }, { "rubric_number": 3, "rubric_detail": "回答估算出由相位噪声导致的归一化旁瓣电平（PSLR）约为-25dB", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "模型基于理想ADC模型计算出量化信噪比(SNRq)约为74dB", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 5, "rubric_detail": "回答通过对比分析得出结论：相位噪声是影响旁瓣性能的主要矛盾，而量化噪声的影响可以忽略", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "回答构建了基于旁瓣能量最小化的LMS自适应相位校正算法框架", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "回答深入分析了定点运算中步长选取的两难困境：步长过小导致更新停滞，步长过大导致误差累积", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "回答提出了功率归一化自适应步长的策略，并设定具体步长μ为0.01", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "回答采用了块浮点（Block Floating Point）技术，通过尾数加指数的格式扩展动态范围", "rubric_weight": 7, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "回答利用CORDIC算法计算相位，明确指出该方法避免了乘法器运算", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "回答预测补偿后的最终PSLR性能指标在-32dB到-35dB之间", "rubric_weight": 3, "rubric_tag": "指令遵循" }, { "rubric_number": 12, "rubric_detail": "回答包含具体的数学推导公式来支撑关于相噪影响和LMS稳定性的工程结论", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 13, "rubric_detail": "算法设计考虑了16位定点DSP平台的约束，例如讨论了定点运算中的误差累积、发散风险、收敛性保证或性能下限等问题。", "rubric_weight": 9, "rubric_tag": "观点分析" }, { "rubric_number": 14, "rubric_detail": "模型使用的是双谱，而非假目标识别要求的是\"三阶及以上相位导数\"（暗示CPF立方相位函数）", "rubric_weight": -5, "rubric_tag": "观点分析" }, { "rubric_number": 15, "rubric_detail": "模型出现计算方面的错误", "rubric_weight": -3, "rubric_tag": "事实信息" }, { "rubric_number": 16, "rubric_detail": "整篇答案几乎没有公式推导，全是定性描述和文献引用", "rubric_weight": -3, "rubric_tag": "观点分析" }, { "rubric_number": 17, "rubric_detail": "模型引用了不存在的文献，或引用了与LFM雷达、盲信号处理、定点DSP硬件补偿等核心技术背景无关的文献。", "rubric_weight": -6, "rubric_tag": "事实信息" }, { "rubric_number": 18, "rubric_detail": "模型忽略了题目中给出的总谐波失真(THD)为-60dB是主要的性能瓶颈。", "rubric_weight": -4, "rubric_tag": "观点分析" } ] }, { "id": "467db57d-6c16-468f-b97b-cc33c94891dd", "case_id": 1750, "language": "cn", "system_prompt": "", "question": "市中心一座大型综合体商圈的 4G/5G NSA 网络质量。该综合体包含购物中心、影院、餐饮、直播基地和写字楼，周末晚高峰（19:00–22:00）投诉集中：用户反映在商场内进行短视频直播、带货直播时，上行速率不稳定，画面经常自动降码率到 480p 甚至更低，时不时出现长时间“转圈”，而普通网页、短视频观看体验基本正常。\n现网关键信息如下（针对商圈核心区域统计）：\n覆盖与无线质量\nLTE B3/B1 室分系统已覆盖全楼，RSRP 大多在 -80 ~ -90 dBm 之间；\nNR n78 通过楼顶 64T64R AAU 覆盖，室内 NR RSRP 约 -95 ~ -105 dBm，SS-SINR 在 3–10 dB 区间；\n商圈内基本无连续盲区，但楼层边角、深街区存在 NR RSRP 低于 -110 dBm 的点状区域。\n负载与承载\n周末晚高峰时段，LTE 下行 PRB 利用率 55%–70%，上行 PRB 利用率约 60%；\nNR 下行 PRB 利用率 40%–55%，上行 PRB 利用率 85%–95%，UL BLER 长期在 20%–30%；\n终端测得下行速率普遍在 150–300 Mbps，但上行经常只有 2–5 Mbps，且波动较大。\n终端与测量特征（直播 APP 典型用户）\n大部分为 5G 终端，NSA 架构，终端能力支持 EN-DC；\n终端日志显示，直播过程中频繁接近上行功率上限，Power Headroom 接近 0 dB；\n上行 PUSCH 的 SINR 大部分在 0–3 dB 徘徊，出现明显的上行干扰与“上行覆盖受限”迹象。\n补充信息：\n商圈周边存在多家其他运营商 5G 站点，属于高密度多制式共存区域；\n当地市场部对短视频直播业务非常关注，希望直播不卡、画质稳定作为核心体验指标；\n现网 TDD 配比统一采用 DL:UL = 7:3，未针对该商圈进行差异化上行增强配置。\n问题要求：\n结合上述现网信息，从以下三个角度分析“直播卡顿、上行速率不稳定”的主要原因，并给出清晰的因果链条：\n上行覆盖与干扰（含上行链路预算、终端发射功率限制、PUSCH SINR、UL BLER 等）；\n频谱与 TDD 帧结构（DL/UL 配比、特殊子帧配置等）对上行容量的影响；\nNSA 架构与承载策略（例如是否充分利用 LTE 上行、是否存在上行分流不合理等）。\n在明确主矛盾的基础上，设计一套可实施的综合优化方案，至少包括：\n上行侧参数与调度策略优化（如上行功控参数、TDD 配比调整、高优先级直播用户调度策略等），需要给出关键参数的调整方向和合理范围；\n覆盖与站型优化思路（如 NR 波束/下倾/功率优化、室内小站补点、是否引入 NR 中低频等），说明如何改善上行链路质量；\nNSA 承载与业务策略优化（例如如何利用 LTE 上行能力、对特定 APP 或 QoS 的承载策略调整）。\n设计一套效果验证与验收方案，说明：\n重点关注哪些无线侧与业务侧 KPI（至少 5 项），期望他们达到怎样的合理区间或改善幅度；\n准备如何通过 DT/室内走测、MDT、直播场景对测等手段验证“直播画面稳定性与上行速率”的改善；\n在实施上述优化时，需要重点关注哪些潜在副作用（如其他业务受影响、邻区干扰增加等），以及如何设置监控与回退策略。", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "明确指出 1080p 直播上行要求约 ≥4 Mbps 且相对稳定，并据此说明为什么现在的网络体验必然不稳定。", "rubric_weight": 9, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": "从链路预算角度说明：3.5 GHz 穿透损耗大、UE Pmax ≈23–26 dBm，而 gNB 为数十瓦级；\n给出链路预算公式（MCL、NF、灵敏度等），解释为何上行比下行更易先崩溃。", "rubric_weight": 9, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "使用具体数值说明：\nNR UL PRB 85–95%、UL BLER 20–30%；\nLTE 室分 UL PRB ≈60%；\nNR 流量占比明显低于 5G 终端渗透率；\n得出“NR 过载、LTE 有余量却没承担直播”的结论。", "rubric_weight": 9, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "提出在重点时段/热点小区上调UL比例（如从7:3调整为6:4），并解释其能够提升上行容量的原因。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "针对TDD配比调整可能引发的跨链路干扰（CLI），提出具体的抑制措施（如小区簇内配比一致、设置保护时隙或静默配置等）。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "明确提出将 NR UL 目标 BLER 控制在约 10%；\n通过 UL_MCS_OFFSET / OLLA 等手段，抑制超激进 MCS，优先保证稳定吞吐；\n体现“稳定码率优先于峰值速率”的思想。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "对直播用户提出使用 NR UL Configured Grant（CG）；\n给出合理周期（2–5 ms 级）与 PRB 连续资源规模；\n说明其与 PHR / SINR / 业务识别联动思路。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "识别 TDD 邻区上下行方向不一致导致的 CLI这一问题。\n", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "对直播 uplink 配置 GBR 承载，并给出大致速率门限（如 6–8 Mbps）。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "效果验证方案：应提及多种验证手段，如优化前后的DT/CQT对比、MDT数据分析、直播场景对测、用户投诉数据跟踪、网管KPI监控或A/B测试等。", "rubric_weight": 4, "rubric_tag": "指令遵循" }, { "rubric_number": 11, "rubric_detail": "将问题主因错误归结为下行覆盖差或核心网过载等（与题干给出的上行相关指标明显冲突）。", "rubric_weight": -8, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "提出极端 TDD 比例（例如完全牺牲 DL）而不说明对 DL 影响或无任何回退/监控。", "rubric_weight": -5, "rubric_tag": "观点分析" }, { "rubric_number": 13, "rubric_detail": "使用 PHR≈0 dB、PUSCH SINR 0–3 dB、UL BLER 20–30% 等指标；推演出上行速率约 2–5 Mbps 的区间。", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 14, "rubric_detail": "针对TDD邻区上下行时隙不一致导致的CLI问题，模型应提出具体的解决措施，如多运营商协同、建立CLI抑制机制、簇内统一TDD配比、设置UL保护时隙/静默比例、采用UL CoMP/IRC技术，并辅以IOT/CLI告警监控。", "rubric_weight": 3, "rubric_tag": "观点分析" } ] }, { "id": "84ec9ed0-abae-45d9-a303-935b47f23954", "case_id": 1924, "language": "cn", "system_prompt": "", "question": "你是一名负责智能仓储系统改造的电气工程师，目前在给一家汽车零部件工厂设计电动叉车的充电模块与充电策略。\n工厂现状：（1）有24台电动叉车，全部采用 48 V / 400 Ah 的磷酸铁锂电池模组；\n（2）生产线实行三班倒，每班8小时，叉车平均每班连续工作 5–6 小时，其余时间用于装卸等待；\n（3）每台叉车电池从 100% 用到 20% SOC 需要约 4 小时 连续重载作业；\n（4）现有 8 台固定式直流充电机，单机最大输出功率 15 kW，支持 CC-CV 恒流恒压充电，通过 CAN 总线与电池 BMS 通信；\n（5）厂方要求：在生产高峰期，不允许有超过 1/3 的叉车同时停机充电；需要在 3 年内尽量减缓电池容量衰减；要求充电模块具备温度、过压、过流、绝缘故障等保护，并能记录关键数据。\n请基于以上信息，设计并说明叉车充电模块是如何工作的，回答内容至少包括：\n（1）充电判定与流程：说明充电模块如何根据 BMS 上报的 SOC、电压、电流、温度 等信息判断：何时启动充电、何时切换恒流 → 恒压阶段；何时认为充电完成并自动停止。\n（2）充电策略与功率分配：结合“8 台充电机 + 24 台叉车”的配置，给出一套调度策略：高峰期与低谷期分别如何安排哪几台叉车充电；当多台叉车同时接入时，充电模块如何分配功率、排队或限流，兼顾充电速度与电池寿命。\n至少给出两种典型工况（如：夜班低负荷 / 白天高峰作业）下的充电调度示例。\n（3）安全与异常处理机制：描述充电模块在出现以下情况时应如何响应：电池温度过高或过低；充电过程中出现电压异常上升 / 电流突变；BMS 通信中断或检测到 SOC 估算明显异常。说明模块需要记录哪些关键数据，用于后续运维与健康评估。\n（4）简要说明你这样设计的理由：从电池寿命、生产节奏、设备利用率三个角度，解释你的充电方案的优点与潜在折中。", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "回答中明确写出48 V / 400 Ah 电池的能量，并换算成 kWh，数值在 18–20 kWh 合理区间（如“约 19 kWh”），且带单位。", "rubric_weight": 2, "rubric_tag": "指令遵循" }, { "rubric_number": 2, "rubric_detail": "回答中根据题干给出的典型连续工作时长（例如每班工作若干小时）估算单台叉车每天耗电量，以 kWh 表示，且有简单说明“为什么是这个数量级”（比如“平均负载约占额定功率的某一比例”）。", "rubric_weight": 2, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "回答中用“单车日耗电 × 24 台”的方式，明确给出车队每天总耗电量的数值（kWh），并写出计算式或中间量纲。", "rubric_weight": 2, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "回答中用“8 台 × 15 kW × 可用充电小时数（自选，比如 10 h 或 12 h）”计算充电系统每天最多能供多少 kWh，且写出数值和单位。", "rubric_weight": 2, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "在给出规则3、规则4 对应的数字后，明确写出一句结论，例如：“按当前假设，8 台充电机的日供能略大于/明显小于车队日耗电，因此需要 ……（如增加夜间充电时长等）”。", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "回答中明确提出3年寿命的“容量残留目标”，类似表述：\n（1）“目标是 3 年后容量 ≥80%”；\n（2）或“3 年约 N 次等效循环后，容量保持在 80% 左右”。", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "回答中出现至少一个 EFC 量级计算，例如：\n（1）“每天约 0.7–1 次 EFC，因此一年约 250–350 次”；\n（2）或“按每次机会充电 30–40% SOC，累计成等效全循环”。", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "回答中概述 LiFePO₄（磷酸铁锂）电池寿命/衰减的主要影响因素，并给出合理的数量级或区间即可（例如 DoD、充电倍率 C-rate、温度窗口对循环寿命的影响）。不要求引用具体年份或固定次数数值；若给出数值，应表述为“典型范围/在一定条件下”。", "rubric_weight": 1, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "回答中点名：工厂电价通常有峰、平、谷时段；或提到“最大需量电费”（峰值功率越高，电费越贵）。任一即可。", "rubric_weight": 2, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "回答中明确提出“将深度补能/满充尽量安排在低负荷/低电价时段（例如夜间或换班窗口）”的策略，并说明白天可偏向机会充电以降低峰值功率与停机影响。不要强制写死具体时刻（如 23:00–7:00）。", "rubric_weight": 3, "rubric_tag": "指令遵循" }, { "rubric_number": 11, "rubric_detail": "回答中指定一个数值化的总充电功率上限（站级），例如：\n（1）“在高峰时段，总充电功率限制在 ≤90 kW”；\n（2）或写出“∑Pₜ ≤ Pmax（例如 100 kW）”。并用一句话说明原因（如最大需量/站内容量/避免影响生产）。", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "回答清晰描述充电启动前的闭环流程：连接检测→安全校验（如绝缘/接地/互锁）→与 BMS 通过 CAN 完成参数协商（目标电压、最大允许电流、温度/故障状态、允许充电信号等）→充电机按 BMS 约束开始输出，并说明任一步失败的处理（拒充/告警）。", "rubric_weight": 10, "rubric_tag": "指令遵循" }, { "rubric_number": 13, "rubric_detail": "回答明确说明恒流(CC)阶段如何限流（以 BMS 给定的 I_max/功率上限为准），并说明何时切换到恒压(CV)阶段（例如电池端电压接近/达到 BMS 目标电压）。同时写清 CV 阶段的控制要点（保持目标电压、允许电流自然下降且受 BMS 限制）。", "rubric_weight": 12, "rubric_tag": "指令遵循" }, { "rubric_number": 14, "rubric_detail": "回答明确给出“认为充电完成并自动停止”的判据（至少包含其中两类：BMS 发出 stop/完成指令；SOC 达到目标；CV 阶段电流下降到阈值并持续一段时间；达到最大充电时长/超时保护）。并说明停止后的收尾动作（断开输出、记录会话、释放桩位等）。", "rubric_weight": 8, "rubric_tag": "指令遵循" }, { "rubric_number": 15, "rubric_detail": "回答覆盖并区分至少 4 类异常/保护：过温、过压、过流、绝缘故障、CAN 通信中断、BMS 故障/SOC 异常等（任选其四）。对每类异常至少写清：检测信号来自哪里→充电机采取什么动作（降额/停机/锁定/隔离）→如何告警与防止反复重启。强调不得绕过 BMS/保护。", "rubric_weight": 12, "rubric_tag": "指令遵循" }, { "rubric_number": 16, "rubric_detail": "答列出需要记录的关键数据字段（至少 6 项）：时间戳、叉车/电池ID、SOC、Pack 电压/电流、温度、充电阶段(CC/CV)、告警/故障码、绝缘检测结果、充入电量(kWh/Ah)、会话开始/结束原因等，并说明这些记录用于追溯/运维/寿命管理。", "rubric_weight": 8, "rubric_tag": "指令遵循" }, { "rubric_number": 17, "rubric_detail": "回答明确满足“生产高峰期≤8台(≤1/3)叉车同时停机充电”的约束，并给出可执行的调度规则（例如按 SOC 低优先、下一班/下一任务紧急度、等待时间、健康度/温度等排序；必要时限功率/排队）。同时给出至少 2 个具体场景说明策略如何工作（如：白天高峰机会充电 vs 夜间深度补能/满充）。", "rubric_weight": 10, "rubric_tag": "指令遵循" }, { "rubric_number": 18, "rubric_detail": "回答中面向操作员列出至少三步顺序动作，如：\n（1）叉车停稳、拉紧驻车制动；\n（2）关闭驱动电源/点火开关；\n（3）插上充电枪并确认充电指示灯正常；\n（4）完成充电后拔枪并复位。", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 19, "rubric_detail": "回答中明确写出司机看到告警灯/告警信息时应做什么，例如：\n（1）停止使用该充电位；\n（2）不自行反复重启；\n（3）立即通知维修/班长。", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 20, "rubric_detail": "回答中写出至少两个不同周期的维护项，例如：\n（1）每日：检查线缆、插头是否损坏、地面是否积水；\n（2）每周：查看充电日志与告警；\n（3）每季度/半年：做绝缘电阻测试/紧固检查等。", "rubric_weight": 3, "rubric_tag": "指令遵循" }, { "rubric_number": 21, "rubric_detail": "回答中列出至少三个可量化指标，例如：\n（1）平均班前 SOC；\n（2）桩位利用率；\n（3）充电等待时间；\n（4）每月过温充电次数；\n（5）每月 EFC 增长等。", "rubric_weight": 2, "rubric_tag": "观点分析" }, { "rubric_number": 22, "rubric_detail": "回答中说明充电调度会考虑生产任务系统的信息，例如：\n（1）与 WMS/MES 对接，知道叉车下一任务的开始时间和地点；\n（2）以此调整充电优先级。", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 23, "rubric_detail": "回答中给出一个具体降级场景，例如：\n（1）“若有 2 台充电机长期故障，则夜间延长充电时段，并提高低谷充电优先级”；\n（2）或“若某台车电池健康度明显下降，则安排更温和的充电策略并降低机会充电倍率”。", "rubric_weight": 2, "rubric_tag": "观点分析" }, { "rubric_number": 24, "rubric_detail": "回答中说明如果叉车数量未来从 24 台增至更大规模（如 40 台），系统会如何扩展，例如：\n（1）增加充电机数量；\n（2）或升级调度算法、调整功率上限、引入更多充电点/移动充电等。", "rubric_weight": 2, "rubric_tag": "观点分析" }, { "rubric_number": 25, "rubric_detail": "回答中出现绕过或禁用 BMS/关键保护的语句，类似：\n（1）建议将充电机直接接到电池端子，绕过 BMS；\n（2）或建议为提高效率“临时关闭过压/过流/温度保护”等。", "rubric_weight": -5, "rubric_tag": "指令遵循" }, { "rubric_number": 26, "rubric_detail": "回答明确建议在正常工况下长期以 ≥2C 电流给 LiFePO₄ 叉车电池充电（非仅极端应急），且未说明显著寿命损失或风险。", "rubric_weight": -5, "rubric_tag": "观点分析" }, { "rubric_number": 27, "rubric_detail": "回答中把“在 0℃ 以下为 LiFePO₄ 电池正常充电”视为允许或推荐做法（不是仅作为危险示例）。", "rubric_weight": -5, "rubric_tag": "指令遵循" }, { "rubric_number": 28, "rubric_detail": "回答明确设计某些高峰时段使超过 24×1/3=8 台叉车长期同时停机充电，并将其作为常规策略（非短暂异常状态）。", "rubric_weight": -5, "rubric_tag": "指令遵循" }, { "rubric_number": 29, "rubric_detail": "回答中出现明显违反能量守恒或量级常识的结论，例如：\n（1）声称“8 台 15 kW 充电机 1 小时能把 24 台几乎放空的 19 kWh 电池全部从 0% 充满”；\n（2）或给出类似量级不可能的供能/耗电推导。", "rubric_weight": -10, "rubric_tag": "观点分析" }, { "rubric_number": 30, "rubric_detail": "回答中宣称：\n（1）“LiFePO₄ 电池需要经常放到几乎 0% 再充才寿命更好”；\n（2）或推荐类似铅酸“长期高压浮充 + 深充深放”的策略作为 LiFePO₄ 的常规最佳实践。", "rubric_weight": -10, "rubric_tag": "指令遵循" }, { "rubric_number": 31, "rubric_detail": "回答中具体说明 CAN 总线在充电控制中的关键交互字段/信号：BMS 上报 SOC、Pack 电压/电流、温度、故障码、允许充电/限流限压；充电机回报自身状态、输出电压/电流、故障信息；并说明当 CAN 异常/超时如何处理（如进入安全停机或降额模式）。", "rubric_weight": 4, "rubric_tag": "观点分析" } ] }, { "id": "34b542fb-9418-4485-a235-67c9f6eb505f", "case_id": 4039, "language": "cn", "system_prompt": "", "question": "你是一名机器学习方面的专家，目前你的研究方向是水下图像的增强技术，并且设计了一个encoder-decoder的架构模型，使用MSE进行了损失函数训练。在训练过程中你发现训练集的MSE持续下降且验证集的MSE也随之降低，但是测试集上的主观视觉效果较差，图像过于平滑，缺失细节信息。其PSNR\\SSIM指标明显低于SOTA模型，请你结合任务特性和损失函数特点分析导致上述现象产生的原因是什么？", "tags": { "topics": [ "工业", "机器学习", "机器学习" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "对模型展开深入分析，明确指出MSE损失函数存在引导网络输出所有可能解的条件均值（Mean Solution）的核心倾向。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": "详细解释均值化处理的具体影响——能够将边缘、纹理等高频细节转化为平滑的低频成分，并明确点明这一过程是导致图像出现过平滑现象的关键原因。", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "精准指出MSE损失函数的像素等权性缺陷——对图像中所有像素赋予相同权重，进而导致模型优先优化占比更大的背景区域，最终牺牲占比极小的细节区域。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "模型指出Encoder-Decoder架构中的下采样过程（如卷积运算、池化操作）会造成图像信息丢失。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "模型通过分析明确指出，下采样操作对边缘、纹理等高频信息具有天然的衰减作用。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "清晰说明解码器的上采样过程存在局限性，仅能基于低频特征进行插值重建，无法主动恢复已丢失的高频细节。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "明确指出模型可能对训练集特定的场景或像素分布产生过拟合，进而导致在未见过的测试场景中泛化能力不足。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "提及真实标签存在的潜在问题，具体包括两类：一是合成真值与真实场景特性不一致；二是标签与输入图像存在配准偏差。", "rubric_weight": 8, "rubric_tag": "指令遵循" }, { "rubric_number": 9, "rubric_detail": "回答仅以简单罗列的方式呈现要点，各要点间为并列关系，缺乏对核心主线（如“任务-损失-架构-数据”的递进关系）的顶层设计与层级展开。", "rubric_weight": -3, "rubric_tag": "行文结构和格式" }, { "rubric_number": 10, "rubric_detail": "答题内容提出类似于快速验证（L1）；性能提升(引入Perceptual Loss);追求前沿技术（如GAN）的三阶段改进路线。", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "答题内容针对“MSE导致过平滑”这一核心归因，设计一个简单但逻辑严密的模拟或消融实验方案，体现实证思维。理想范例：“为验证‘像素等权性导致背景主导优化’，可设计如下对照实验：1）基线：用标准MSE训练。", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "能批判性地指出，不仅PSNR/SSIM有缺陷，甚至当前的“感知质量”评估也依赖于有限的预训练模型（如VGG），进而提出一个更系统的、针对水下图像的增强技术的多维度评估框架。", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 13, "rubric_detail": "在提出改进方案（如使用更复杂的损失函数或架构）时，能明确指出其带来的计算开销增加，并基于应用场景（如移动端、实时系统）讨论可行的权衡策略。", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 14, "rubric_detail": "在指出任务为“不适定问题”后，未能引出解决不适定问题的核心数学思想——正则化，并讨论不同损失函数如何隐式或显式地提供正则化。", "rubric_weight": -4, "rubric_tag": "观点分析" }, { "rubric_number": 15, "rubric_detail": "讨论仅限于技术性真值（Ground Truth）的获取瑕疵（如配准误差），未能对水下图像的增强技术中“何为真值”这一根本性问题进行必要反思。", "rubric_weight": -3, "rubric_tag": "观点分析" }, { "rubric_number": 16, "rubric_detail": "改进建议与问题归因脱节，未能解释新方法（如新损失函数）如何从机制上解决MSE导致的图像平滑、细节丢失问题。", "rubric_weight": -2, "rubric_tag": "观点分析" }, { "rubric_number": 17, "rubric_detail": "在答案开头，直接点明问题核心，即MSE损失函数，不急于展开庞杂的背景知识或平行论点。", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 18, "rubric_detail": "每个分论点都采用总分结构，段落第一句即为该段的核心观点句。", "rubric_weight": 3, "rubric_tag": "观点分析" } ] }, { "id": "b5f5aecd-7ae0-4e6e-97ec-8dce1101d9bf", "case_id": 4164, "language": "cn", "system_prompt": "", "question": "我是某家互联网公司推荐系统的负责人，所在的团队负责一个大规模推荐系统，核心模型时一个基于深度学习的CTR预估模型，用于决定首页信息流的排序。我们在2025年第一季度对模型做了一次升级，从旧版ModelA升级到了ModelB，ModelB引入了更多用户行为序列特征和更复杂的网络结构，对于同一组验证集的离线评估结果是ModelB比ModelA的AUC提高了0.015，而logloss也明显更优，经过为期两周的A/B实验，ModelB比ModelA的CTR下降了2.1%，人均停留时长下降，投诉率上升。\n已知额外信息：\n1、A/B实验随机分桶是按照user_id哈希，但新模型上线后引入了用户实时特征，实时特征依赖用户最近5分钟行为。\n2、线上实验期间，有一次push策略调整，但是PM认为对CTR影响应该不大。\n3、日志系统中，新模型请求失败率比旧模型高0.3%，失败请求会回退到一个rule_based排序。\n4、离线评估使用的是历史日志回放，未做反事实矫正。\n你的任务：\n1、判断是否可以据此直接下结论——ModelB不如ModelA？\n2、系统性列举至少四类可能导致“离线好、线上差”的机制（要求必须覆盖统计层面、系统层面、因果层面）。\n3、指出当前A/B实验设计中至少三个不可忽略的评估漏洞。\n4、给出一个你认为可落地的改进实验或分析方案。\n5、明确说明：在什么证据条件下，你才会支持下线ModelB。\n", "tags": { "topics": [ "工业", "机器学习", "机器学习" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "回答指出不能直接得出ModelB不如ModelA的结论", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": "分析指出0.3%的请求失败并非随机分布，而是集中在长序列的高活跃用户，导致核心用户体验降级为Rule-based排序", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "在系统层面，指出实时特征链路可能存在延迟（Serving Skew），导致线上模型读取到空值或默认值，与离线训练数据分布不一致", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "指出AUC指标的局限性，说明其作为全局指标无法反映Top-k质量、用户打扰度或长期留存等业务敏感点", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 5, "rubric_detail": "分析了因果层面的干扰，指出Push策略改变了进入信息流的用户意图分布或引入了冷启动用户，影响了CTR归因", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "指出了实验设计中将Fallback（失败回退）路径的用户数据与正常模型排序数据合并统计的漏洞", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "识别出User_id Hash分桶与Push策略未做正交处理的漏洞，导致实验组间流量分布可能不均匀", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "改进方案中包含工程排查步骤，具体为提取报错请求的User_id以确认其活跃度，或进行在线特征一致性校验", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 9, "rubric_detail": "下线条件中包含工程性能指标不达标，例如TP99延迟过高或GPU资源成本超过ROI阈值", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 10, "rubric_detail": "回答完整覆盖了题目要求的五个任务：结论判断、机制列举、漏洞指出、改进方案、下线条件", "rubric_weight": 3, "rubric_tag": "指令遵循" }, { "rubric_number": 11, "rubric_detail": "回答针对prompt中的每个问题，使用了小标题或序号进行分点组织", "rubric_weight": 3, "rubric_tag": "行文结构和格式" }, { "rubric_number": 12, "rubric_detail": "回答中大篇幅复述题目给定的背景信息（如ModelA/B的介绍、AUC数据等）", "rubric_weight": -10, "rubric_tag": "行文结构和格式" }, { "rubric_number": 13, "rubric_detail": "将0.3%的失败仅仅描述为样本量不足或统计干扰，而未意识到这是高价值用户损失。", "rubric_weight": -10, "rubric_tag": "观点分析" }, { "rubric_number": 14, "rubric_detail": "在未确认工程Bug（超时/特征延迟）是否解决的情况下，优先建议调整模型网络结构、增加正则化、调整学习率等算法层面的优化", "rubric_weight": -5, "rubric_tag": "观点分析" }, { "rubric_number": 15, "rubric_detail": "将离线好线上差的原因归结为离线数据过拟合，而未提及工程实现（延迟/一致性）带来的偏差", "rubric_weight": -10, "rubric_tag": "观点分析" }, { "rubric_number": 16, "rubric_detail": "改进方案中建议剔除Push数据或剔除0.3%失败数据后直接看结论，而没意识到需要先修复这些问题才能进行有效实验", "rubric_weight": -10, "rubric_tag": "事实信息" }, { "rubric_number": 17, "rubric_detail": "回答在系统已出现性能瓶颈（失败率升高）的情况下，仍然建议使用影子测试或双倍流量并发等高负载评估方案", "rubric_weight": -10, "rubric_tag": "观点分析" }, { "rubric_number": 18, "rubric_detail": "建议采取暂停、回滚或保留小流量（如1%）进行诊断的风险控制措施", "rubric_weight": 5, "rubric_tag": "观点分析" } ] }, { "id": "ea67f32d-c827-438f-951e-17206a6f196c", "case_id": 5166, "language": "cn", "system_prompt": "", "question": "场景描述：\n你是某大型电商平台（类似淘宝/Amazon）的推荐算法架构师。团队正在重构“猜你喜欢”的召回层（Match Stage），目的是从亿级商品库中快速检索出用户可能感兴趣的 Top-1000 候选集。\n实习生小赵负责开发核心的双塔召回模型（Two-Tower DSSM）。他非常自信地提交了实验报告，声称新模型的 AUC 达到了 0.99，且训练速度极快。他认为这套模型上线后能极大提升召回的覆盖率。\n代码片段（简化版）：\ncode\nPython\nimport pandas as pd\nimport numpy as np\nimport tensorflow as tf\nfrom tensorflow.keras import layers, Model\n\n# 1. 数据准备\n# log_data: [user_id, item_id, label=1 (点击)]\n# 只有正样本日志（点击数据）\npos_data = pd.read_csv(\"click_logs.csv\") \nall_item_ids = list(set(pos_data['item_id'].values))\n\n# 2. 负样本构造 (全局随机负采样)\n# 策略：对于每个正样本，随机从库里选 3 个 item 作为负样本 (label=0)\ndef get_random_negatives(pos_df, ratio=3):\n neg_list = []\n for _, row in pos_df.iterrows():\n for _ in range(ratio):\n rand_item = np.random.choice(all_item_ids) # 全局随机\n neg_list.append([row['user_id'], rand_item, 0])\n return pd.DataFrame(neg_list, columns=['user_id', 'item_id', 'label'])\n\nneg_data = get_random_negatives(pos_data)\ntrain_data = pd.concat([pos_data, neg_data]).sample(frac=1) # 混合打乱\n\n# 3. 模型构建 (标准的双塔 DSSM)\nuser_input = layers.Input(shape=(1,), name='user_id')\nitem_input = layers.Input(shape=(1,), name='item_id')\n\n# User Tower\nuser_emb = layers.Embedding(input_dim=100000, output_dim=64)(user_input)\nuser_vec = layers.Dense(32, activation='relu')(layers.Flatten()(user_emb))\n\n# Item Tower\nitem_emb = layers.Embedding(input_dim=500000, output_dim=64)(item_input)\nitem_vec = layers.Dense(32, activation='relu')(layers.Flatten()(item_emb))\n\n# Dot Product + Sigmoid (Pointwise)\ndot_product = layers.Dot(axes=1)([user_vec, item_vec])\noutput = layers.Dense(1, activation='sigmoid')(dot_product)\n\nmodel = Model(inputs=[user_input, item_input], outputs=output)\nmodel.compile(optimizer='adam', loss='binary_crossentropy', metrics=['AUC'])\n\n# 4. 训练\nmodel.fit(\n [train_data['user_id'], train_data['item_id']], \n train_data['label'], \n batch_size=1024, epochs=5\n)\n\n# 5. 结果\n# Train AUC: 0.992, Test AUC: 0.988\n你的任务：\n作为 Tech Lead，请指出这份代码中导致 AUC 虚高但线上效果极差（Recall 低） 的至少三个致命缺陷。请解释为什么“随机负采样”在电商召回中是陷阱，并给出包含“困难负样本（Hard Negative）”挖掘或“In-batch Softmax”的修正方案。", "tags": { "topics": [ "工业", "机器学习", "机器学习" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "明确指出 AUC 0.99 虚高是由于负样本过于简单（Easy Negatives）导致的，模型只学到了粗粒度的类目差异", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": " 建议引入困难负样本（Hard Negatives），并明确提及曝光未点击或近似最近邻作为来源", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "提供可执行的落地逻辑：包含 In-batch Softmax 或 Hard Negative 挖掘的具体代码、伪代码", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "建议将 Loss 函数从 Pointwise (BCE) 修改为 Sampled Softmax (Cross Entropy) 或 InfoNCE", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "建议使用 In-batch Negatives（批内负采样）策略以提高训练效率和覆盖度", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "指出仅使用 ID 特征会导致冷启动（Cold Start），建议加入 Side Info（类目、标题、图片等）", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "建议修正评估指标：不看 AUC，改看 Recall@K (HitRate) 或 NDCG@K", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "指出随机切分导致数据泄露（未来预测过去），建议按时间切分", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "提及 温度系数 (Temperature) 对 Softmax 的重要性，或 LogQ 修正以处理热门偏差", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "提及对 Embedding 进行 L2 Normalization (归一化) 以防止模长主导相似度", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "建议使用 Mixed Negative Sampling（混合负采样：In-batch + Random/Hard）", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "回答使用了推荐系统领域的专业术语（如负采样、召回、AUC等）", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 13, "rubric_detail": "引用不存在的文献、使用虚假的学术标注（如 [[1]]），产生幻觉", "rubric_weight": -20, "rubric_tag": "观点分析" }, { "rubric_number": 14, "rubric_detail": "回答中出现了具体的人物姓名指代（如“小赵”、“实习生小赵”），而非使用中性表述", "rubric_weight": -10, "rubric_tag": "行文结构和格式" }, { "rubric_number": 15, "rubric_detail": "修正方案仅停留在纯文字建议层面，没有任何代码块、伪代码", "rubric_weight": -10, "rubric_tag": "观点分析" }, { "rubric_number": 16, "rubric_detail": "回答包含了非技术性的行政或团队管理建议（如“组织复盘会议”、“开展培训”、“编写文档”）", "rubric_weight": -5, "rubric_tag": "指令遵循" } ] }, { "id": "d6b99b4e-c7f9-4cab-8064-e782ad87115f", "case_id": 5430, "language": "cn", "system_prompt": "", "question": "通信基站RRU建设于中国华南地区，在夏季6月-8月时间段内，RRU偶尔会出现与DU断连的情况。在通过分析RRU和DU上传的LOG文件后发现，具体告警为RRU失去与光模块的通信导致。而在其他季节没有出现相关告警，RRU均正常工作。\n\n现在结合客户基站所在的地点、告警出现的时间节点以及告警的具体内容，给出RRU告警的分析思路以及可能原因。\n\n\n\n", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "分析思路中涵盖了对气象条件的排查，如温差、降雨、湿度或雷暴天气", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": "建议确认RRU的硬件型号、软件版本以及光模块的厂商信息，以排查是否存在已知的兼容性问题或特定型号的缺陷。", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "导出告警前后（±15分钟）RRU板内温度、RRU的PA供电电流以及电压、TX链路数字域和模拟域各个节点的功率大小、RX链路的数字域和模拟域各个节点的功率大小以及RRU log内记录的告警信息。", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "提及使用对调法（互换试验）来判断故障是随件还是随站", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "提及现场勘查，如检查客户现场设备是否接地、SFP插接是否到位和有锁扣固定以及RRU和SFP之间是否安装合规。", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "分析了高湿、凝露或盐雾环境可能导致SFP金手指或连接器腐蚀、接触不良", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "将雷暴季节的瞬态干扰或静电放电列为导致模块锁死的原因之一", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "检查光模块和RRU在高温状态下是否出现异常电流、异常电压或者告警", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 9, "rubric_detail": "现场排查步骤包含检查光口朝向、防水塞密封性或跳纤是否有滴水弯", "rubric_weight": 8, "rubric_tag": "指令遵循" }, { "rubric_number": 10, "rubric_detail": "建议对光模块或连接器进行清洁，并检查是否有腐蚀迹象或者被破坏痕迹", "rubric_weight": 6, "rubric_tag": "指令遵循" }, { "rubric_number": 11, "rubric_detail": "提出了物理防护的改进措施，如增加遮阳罩、防雨罩或优化安装位置", "rubric_weight": 6, "rubric_tag": "指令遵循" }, { "rubric_number": 12, "rubric_detail": "没有建议检查供电电源的稳定性（如纹波、跌落）以及接地或防雷设施的状态", "rubric_weight": -4, "rubric_tag": "指令遵循" }, { "rubric_number": 13, "rubric_detail": "验收标准中没有包含具体的量化指标，如告警为0、同比下降比例或温度裕量数值", "rubric_weight": -8, "rubric_tag": "指令遵循" }, { "rubric_number": 14, "rubric_detail": "回答采用了清晰的逻辑结构，按分析思路、可能原因、定位步骤、整改建议的顺序展开", "rubric_weight": 4, "rubric_tag": "行文结构和格式" }, { "rubric_number": 15, "rubric_detail": "使用了列表、要点或小标题对复杂信息进行分层陈述", "rubric_weight": 4, "rubric_tag": "行文结构和格式" }, { "rubric_number": 16, "rubric_detail": "回答中包含与故障分析无关的通用通信原理科普或背景介绍，导致与题目要求的内容相背", "rubric_weight": -4, "rubric_tag": "行文结构和格式" }, { "rubric_number": 17, "rubric_detail": "整段输出未进行分段或缺乏层级标识，导致关键信息难以快速提取", "rubric_weight": -4, "rubric_tag": "行文结构和格式" } ] }, { "id": "d29e29db-affb-494c-b5b7-1b193cff24bd", "case_id": 5545, "language": "cn", "system_prompt": "", "question": "HW mate60上市后，由于终端能力相比3年前大幅增长。UE能力字段featureset ID会大于32，超过定义边界进而导致67号任务故障。主控板会反复复位，复位过程会导致业务中断，并稳定性指标恶化10倍。其中辽宁、河南、贵州三地的TDD版本出现了349次。最近MATE60终端大面积入网可能会加剧该问题；请分析故障可能产生的原因，要考虑到协议理解偏差，方案设计偏差，代码实现情况偏差；并从需求分析，设计团队以及代码测试交付团队多维度给出纠正及预防措施。", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "回答引用了3GPP协议中关于UE能力（UE Capability）或特性集（FeatureSet）ID范围定义的相关内容（例如，TS 38.331）。”", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": "模型指出了产品方案设计中支持的featureSetCombination数量限制在32个", "rubric_weight": 2, "rubric_tag": "事实信息" }, { "rubric_number": 3, "rubric_detail": "模型指出了协议定义的maxFeatureSetCombinations最大值为1024", "rubric_weight": 3, "rubric_tag": "事实信息" }, { "rubric_number": 4, "rubric_detail": "模型指出了改进措施中将UE能力宏值扩充到了64", "rubric_weight": 3, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "模型分析了故障发生的特定场景可能是在异厂家边界处", "rubric_weight": 2, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "回答中包含与故障分析无关的通用5G背景介绍或Mate60营销信息", "rubric_weight": -7, "rubric_tag": "行文结构和格式" }, { "rubric_number": 7, "rubric_detail": "回答未采用清晰的层级结构（如小标题或列表）来区分原因分析和改进措施，导致阅读困难", "rubric_weight": -3, "rubric_tag": "行文结构和格式" }, { "rubric_number": 8, "rubric_detail": "模型在原因分析部分涵盖了协议理解、方案设计、代码实现三个维度", "rubric_weight": 5, "rubric_tag": "指令遵循" }, { "rubric_number": 9, "rubric_detail": "模型在改进措施部分涵盖了需求分析、设计团队、代码测试交付团队三个维度", "rubric_weight": 5, "rubric_tag": "指令遵循" }, { "rubric_number": 10, "rubric_detail": "模型提出了具体的网络侧解决方案，如修改基站侧代码以兼容大于32的featureset ID、下发紧急补丁、或通过参数配置暂时规避该问题等。", "rubric_weight": 7, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "模型应该提出具体的供应链协同机制，如建立终端与网络设备间的定期技术规范对齐会议、或强制执行端到端兼容性测试流程，以保证对终端能力上限（如featureset ID）的理解一致。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "模型提出了针对稳定性/兼容性BUG，需要每周在各商用版本间进行同步和梳理", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 13, "rubric_detail": " 模型总结的原因或措施超出了prompt所给定的范围", "rubric_weight": -6, "rubric_tag": "观点分析" }, { "rubric_number": 14, "rubric_detail": "模型提出应建立针对67号任务故障率的观测与报警机制。", "rubric_weight": 7, "rubric_tag": "观点分析" }, { "rubric_number": 15, "rubric_detail": "模型在回答中引入了Prompt文本未提供的具体版本时间或终端发布时间", "rubric_weight": -5, "rubric_tag": "观点分析" } ] }, { "id": "2ddd40b6-f110-47f4-ae3d-3fee5b01716f", "case_id": 5768, "language": "cn", "system_prompt": "", "question": "某个政企专线用户投诉其通过GPON接入的1000M带宽业务在每日14:00-16:00期间出现严重的网页响应缓慢和视频会议卡顿现象，但用Speedtest单线程测速时，下行速率能达到900Mbps以上。\n现网环境与数据指标：\n1. 拓扑结构：\n OLT(MA5800)->分光器(1:64)->ONU\n2. 光路指标：\n ONU接收光功率为 -24.5dBm，OLT侧接收光功率为 -21.0dBm\n3. ONU状态：\n display ont info 显示ONT处于 Up 状态\n display ont optical-info 发现Upstream transmit power 为 2.5dBm\n display ont error-statistics 显示该时段内 BIP 误差技术持续缓慢增长\n4. 带宽配置：\n DBA模板类型为 Type 4 (最大带宽1000M)，业务流采用优先级5\n5. 用户侧状态：\n 用户用 Wireshark 抓包发现，卡顿发生时伴随大量的TCP Retransmission和较高的单向抖动\n\n你的任务是：\n1. 故障定位：\n 请分析该用户为何在测速达标的情况下仍然感觉业务卡顿？请结合指标判断最可能的故障根源是什么。\n2. 深度分析：\n 解释为何特定的光功率数值和BIP计数与业务表现相关？\n 为何单线程下载测速可能达标，而交互式业务体验极差？\n3. 优化建议：\n 请给出具体的排查以及处理方案，需包括物理层调整建议和OLT侧的参数检查命令或脚本逻辑。\n\n\n\n\n\n\n", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "光功率临界点定性分析：\n故障分析中明确指出ONU接收光功率-24.5dBm已接近光模块的灵敏度临界值", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 2, "rubric_detail": "时段性与物理环境关联推导：\n将14:00-16:00的时段特征与温度升高联系起来，推测可能导致光纤微弯损耗增大或光模块频率漂移", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "BIP误码计数根因定性：\nBIP误差的持续增长被认定为物理层存在误码的直接证据", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "GPON纠错机制与时延的关系：\n解释了GPON的纠错机制在修补误码时会增加处理时延，从而放大业务抖动", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "区分测速与交互业务的感知：\n指出Speedtest单线程测速通过TCP重传掩盖了低比例误码，因此无法反映交互式业务的真实体验", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "分析Type 4模板在高负载下的时延：\n分析指出在1:64分光比下，Type 4 DBA模板（尽力而为）导致用户需竞争时隙，无法保障时延", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "物理层实操建议：\n物理层排查建议中包含了检查入户侧冷接头端面或清洁ONU光口的内容", "rubric_weight": 5, "rubric_tag": "指令遵循" }, { "rubric_number": 8, "rubric_detail": "安全冗余意识：\n建议将ONU接收光功率优化至-20dBm左右，以预留安全余量", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 9, "rubric_detail": "调度逻辑优化：\n优化方案中明确建议将DBA模板类型修改为Type 3（包含保证带宽）或Type 2", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 10, "rubric_detail": "调度逻辑优化：\n提供了创建新DBA模板的具体命令逻辑，包含assure（保证带宽）和max（最大带宽）参数", "rubric_weight": 5, "rubric_tag": "指令遵循" }, { "rubric_number": 11, "rubric_detail": "命令严谨性：\n提供了修改ONU T-CONT绑定的具体命令，如 `ont modify [port-id] [ont-id] tcont [tcont-id] dba-profile-id [new-profile-id]`，并以 `commit` 结尾。”", "rubric_weight": 3, "rubric_tag": "指令遵循" }, { "rubric_number": 12, "rubric_detail": "配置建议中包含了确保优先级5的业务流进入高优先级队列或Gemport的逻辑", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 13, "rubric_detail": "回答中包含大量关于GPON基础原理、TCP/IP协议定义的通用科普内容，与本题特定故障场景无关，造成严重冗余", "rubric_weight": -5, "rubric_tag": "行文结构和格式" }, { "rubric_number": 14, "rubric_detail": "主次根因倒置：\n如果模型主张“仅通过修改DBA模板或重启设备”能修复故障，而不是主要去处理光路，属于主次根因判断倒置，会误导故障排查的顺序和效率。", "rubric_weight": -8, "rubric_tag": "观点分析" }, { "rubric_number": 15, "rubric_detail": "缺乏参数查询指令，未能提供具体的OLT侧参数检查命令，如 `display ont info`, `display ont optical-info`, `display dba-profile` 等，用于验证和排查问题。", "rubric_weight": -5, "rubric_tag": "事实信息" }, { "rubric_number": 16, "rubric_detail": "给出的排障流程建议没有按照先简单后复杂，或者先物理层后逻辑层这样的顺序", "rubric_weight": -5, "rubric_tag": "行文结构和格式" }, { "rubric_number": 17, "rubric_detail": "模型未在结尾处提供优化后的，具体量化的预期效果或验收标准，如‘BIP错误计数为0’、‘会议上行抖动<30 ms、丢包<0.5%，TCP重传率降至<0.1%’等。", "rubric_weight": -8, "rubric_tag": "事实信息" } ] }, { "id": "39f36658-5239-4536-a9cd-d0270b652111", "case_id": 8614, "language": "cn", "system_prompt": "", "question": "【任务背景】\n我司正试图在 5G-A 现网中试点 基于语义的低空视频图传系统。该系统弃用传统的 H.265 编码，采用“深度联合源信道编码（Deep JSCC）”技术。利用边缘侧的教师模型提取无人机视角下的“语用特征（Pragmatic Features）”，仅传输关键目标的语义向量，理论上可在极低带宽（降至原 1/10）下实现同等视觉理解效果。\n\n【突发危机】\n在 300 架无人机大规模接入测试中，系统并未出现预想中的平滑性提升，反而爆发了两个致命问题：\n\n语义“幻觉”与避障失效：当无人机群穿过一片带有大量脚手架和防尘网的施工区时，语义编码器将复杂的金属网状特征误识别为“云雾背景”并进行了平滑处理（去噪），导致机载避障算法丢失了真实的物理障碍物边界，险些发生集体撞击。\n\n协议栈“负收益”时延：尽管物理层传输的数据量大幅减少，但端到端时延（从摄像机采集到地面站解析）从 30ms 飙升至 120ms。监测发现，瓶颈并不在空口，而是由于语义向量与现有的 PDCP/RLC 层协议控制逻辑 发生了严重的处理冲突。\n\n【故障环境与约束】\n算力矛盾：机载侧 NPU 运行 JSCC 编码器占用率为 85%，此时由于输入图像的熵权波动，引发了 NPU 频繁的调频（DVFS）。\n\n反馈机制：系统采用了“语义自动重传（S-HARQ）”，但现有的 RLC 状态报告（Status Report）无法理解语义特征的重要性分级，导致关键的避障向量被排在了背景向量之后重传。\n\n硬件约束：语义提取模型为浮点型，但机载网卡驱动层强制要求语义特征进行 非均匀量化（Non-uniform Quantization）。\n\n【你的任务】\n请你作为 6G 预研首席专家，针对语义幻觉引发的安全风险以及架构不适配导致的协议栈时延爆发问题，提交一份包含根因推演、架构级算法重构及语用纠偏的深度闭环计划。", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "识别 Loss 函数的目标错位：指出以 MSE/PSNR 为目标的 JSCC 会平滑掉高频的障碍物边界", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": "提出“语义原生协议单元（S-PDU）”，明确禁止 RLC 跨语义分片", "rubric_weight": 10, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "将 NPU DVFS 波动建模为“计算衰落（Computational Fading）", "rubric_weight": 9, "rubric_tag": "事实信息" }, { "rubric_number": 4, "rubric_detail": "在根因推演阶段明确识别“潜在空间折叠（Latent Space Folding）”：指出非均匀量化或表示失配会导致不同物理语义被压缩到同一潜在表示区，从而诱发语义幻觉。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "在解决方案中提出“语义敏感量化（Semantic-aware Quantization）”或等价机制：即量化精度需依据语义/任务重要性动态分配，而非静态或仅做码本校准。", "rubric_weight": 7, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "引入非 AI 的“物理一致性校验侧路”", "rubric_weight": 9, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "提出语义不等保护（UEP）或星座映射", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "提出“重要性感知状态报告（Importance-aware Status Report, I-SR）或等价机制：明确修改 RLC 状态报告或 HARQ 反馈逻辑，使反馈与重传决策能够感知语义/任务重要性，而非仅基于序列号或比特丢失。", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 9, "rubric_detail": "提出“语用能效比（Pragmatic EE）”指标", "rubric_weight": 7, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "提出“语义崩溃门限（Semantic Collapse Threshold）”", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "提出反幻觉硬负样本库", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "形成“根因→重构→落地→里程碑”的行文格式", "rubric_weight": 7, "rubric_tag": "行文结构和格式" }, { "rubric_number": 13, "rubric_detail": "提出增加导频/功率的建议", "rubric_weight": -9, "rubric_tag": "观点分析" }, { "rubric_number": 14, "rubric_detail": "回复中建议通过加深网络来解决幻觉", "rubric_weight": -5, "rubric_tag": "观点分析" }, { "rubric_number": 15, "rubric_detail": "模型回答中建议在 PDCP 层进行再压缩", "rubric_weight": -8, "rubric_tag": "观点分析" }, { "rubric_number": 16, "rubric_detail": "仅提出依靠调优的解决方案，而没有其他触及问题本质的方案。", "rubric_weight": -5, "rubric_tag": "观点分析" } ] }, { "id": "226b82d8-e406-4f1d-bd92-cb0adfb5374f", "case_id": 9591, "language": "cn", "system_prompt": "", "question": "在RRU的研发过程中发现，RRU 的性能验证时其上行灵敏度测试比3GPP 的要求值低1-2个dB，在对环境进行检查以后，各个设备的连接部件无松动，且外部环境中没有相邻的频段的RRU发载波。\n请基于RRU上下行链路的结构、上行灵敏度的测试方案以及软件的配置 给出可能的原因和相应的解决方案。", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "回答指出了接收通道增益不足的具体硬件原因，包含低噪声放大器（LNA）增益偏低或滤波器插入损耗过大", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 2, "rubric_detail": "内容涵盖了混频器非线性失真导致的信噪比恶化，提及本振泄漏或三阶互调产物", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 3, "rubric_detail": "提及接收链路前端器件的噪声系数（NF）超出设计指标是导致灵敏度下降的原因之一", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 4, "rubric_detail": "在测试方案层面，指出了测试仪器未校准或测试夹具/连接器阻抗失配的问题", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 5, "rubric_detail": "软件配置方面，识别出自动增益控制（AGC）响应过慢或动态范围不足的问题", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 6, "rubric_detail": "提及基带I/Q校准系数错误或温度补偿失效导致解调偏差", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 7, "rubric_detail": "排查建议中包含关闭下行PA/载波/DPD以排除内部元器件干扰", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 8, "rubric_detail": "排查流程中包含确认线缆、合路器或衰减器的误差小于0.3 dB", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 9, "rubric_detail": "阐述了通过固定AGC在高增益档位来验证增益策略是否为故障根因的逻辑", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "建立了双路同时注入信号以验证分集合并增益（MIMO）功能的逻辑链条", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "分析了恢复下行载波后灵敏度若下降，可能存在TX到RX的隔离度问题或PIM（无源互调）干扰", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 12, "rubric_detail": "没有采用分层级的结构进行论述，没有清晰的区分硬件、测试和软件这三个板块", "rubric_weight": -10, "rubric_tag": "行文结构和格式" }, { "rubric_number": 13, "rubric_detail": "没有提供具有操作性的综合排查流程或优先级顺序，仅列举了原因", "rubric_weight": -10, "rubric_tag": "行文结构和格式" }, { "rubric_number": 14, "rubric_detail": "回答中包含大量关于RRU基础定义、3GPP协议历史背景等与故障排查无关的科普性内容，造成严重冗余", "rubric_weight": -10, "rubric_tag": "行文结构和格式" }, { "rubric_number": 15, "rubric_detail": "输出格式混乱，未使用清晰的标题、列表或序号来组织技术点，导致可读性差", "rubric_weight": -10, "rubric_tag": "行文结构和格式" } ] }, { "id": "3c8f83a2-96db-42b2-812d-643cfb1d99e4", "case_id": 9682, "language": "cn", "system_prompt": "", "question": "你是一名电商应用的后端工程师。用户打开商品详情页时，系统会调用一个核心接口来返回商品页的聚合数据。这个聚合数据由四类信息拼装得到。第一类信息是商品基础信息，例如标题、图片、描述等内容。这类信息更新频率很低，通常一天只有极少量商品会发生变更，并且业务允许它最多延迟一小时更新到用户侧。第二类信息是区域价格，因为不同地区可能存在不同定价或者活动价格调整。这类信息更新频率中等，业务允许它最多延迟五分钟更新到用户侧。第三类信息是用户折扣，例如会员权益、优惠券与个性化策略带来的优惠。这类信息只有大约30%的请求会用到，并且必须保证不同用户之间不会发生串用，也就是说绝对不能把甲用户的折扣展示给乙用户。第四类信息是库存信息，它变化频繁，但业务只允许库存最多延迟十秒更新到用户侧。\n为了提升性能，系统使用了两层缓存。第一层缓存位于应用服务器的进程内存中，每台应用实例都有自己的缓存空间，它访问很快，但容量有限，并且在应用重启后会清空。第二层缓存是集中式缓存集群，使用的是 Redis 集群，它可以容纳更多数据，但会受到网络开销和整体容量限制的影响。系统采用常见的缓存旁路方式，请求到来时先查缓存，如果命中就直接返回，如果未命中就回源查询并拼装结果，然后把结果写回缓存。\n当前系统的做法是把整张商品页的最终聚合结果直接缓存到第二层缓存中，并且缓存的命名方式同时包含地区、商品编号与用户编号。同一个地区、同一个商品，只要访问的用户不同，就会产生不同的缓存条目。这个整页缓存的过期时间被设置为六十秒，并且过期时间固定，没有做任何随机打散。\n近七天线上指标如下。系统请求量在高峰期大约每秒五万次，平时大约每秒两万次。第二层缓存的整体命中率是52%，而第一层缓存命中率只有8%。系统端到端延迟中位数大约是45ms，较慢请求的延迟已经偏高，因此你的方案必须保证最慢的那部分请求不能变得更慢。Redis 集群内存使用率长期处于高水位，平均大约是92%，高峰大约是97%，并且持续出现缓存条目被踢出的现象。Redis 中的缓存条目数量大约是1.8亿条。\n应用侧对未命中的原因做了抽样统计，结果显示未命中中约35%来自缓存到期，并且同一商品同一地区会在短时间内出现大量同时过期的情况。未命中中约25%来自缓存不存在，这种现象在发布后前十五分钟最明显。未命中中约20%来自缓存被踢出。未命中中约15%来自请求参数组合过于分散，具体表现为同一商品在短时间内会被大量不同用户访问，从而产生大量不同的缓存条目。其余未命中原因占比不高，例如偶发的网络超时等。\n最热门的1%商品贡献了大约60%的访问流量，但用户维度非常分散，同一商品在十分钟内可能会被上万用户访问。\n你的目标是在不能新增任何新基础设施的前提下，只通过调整现有 Redis 配置、缓存命名方式、缓存拆分粒度、过期策略、回源策略与应用侧逻辑，把第二层缓存命中率从52%提升到80%或更高，同时保证最慢请求延迟不出现上升。你必须严格满足正确性约束，不能让优惠信息在不同用户之间发生混用。你还必须满足数据陈旧约束，库存最多允许延迟十秒，价格最多允许延迟五分钟，商品基础信息最多允许延迟一小时。你的回答必须以纯文本方式呈现，并且不依赖运行代码或真实压测结果，评审只读文本就能判断方案是否正确与是否合理。\n请你输出一个结构化方案，并至少包含以下内容。第一部分请给出你认为命中率偏低的三个主要原因，并且必须结合题目给定的指标进行推理。第二部分请给出从改动最小到改动较大的分阶段改造计划，并在每个阶段说明你要改什么、为什么要改、预期改善哪类未命中、可能带来什么风险，以及你打算如何监控与验收。第三部分请给出你建议的缓存拆分与命名策略，并明确说明哪些信息应该被共享缓存，哪些信息必须按用户维度隔离。第四部分请说明你将如何处理大量请求同时遇到缓存失效而一起回源的问题，以及你将如何处理发布后的冷启动问题。第五部分请说明你将如何在满足正确性与陈旧约束的前提下，保证方案既能提升命中率又不会让最慢延迟变差。", "tags": { "topics": [ "工业", "后端开发", "后端开发" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "明确指出命中率偏低的第一个原因是用户维度的缓存key设计导致了低复用率和基数爆炸问题，分析这个原因时提到地区、商品、用户的缓存设计，并结合题目中数据进行分析（1.8亿条缓存，10分钟内上万用户访问）。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 2, "rubric_detail": "模型指出导致命中率低的第二大原因是所有缓存过期时间固定为60秒且无随机打散，导致集中失效（缓存雪崩/回源风暴）。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "在分析统一TTL=60s导致缓存集体失效问题时，提出TTL抖动（±10%）作为解决方案。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 4, "rubric_detail": "指出Redis集群内存使用率长期处于高水位（92%-97%）导致缓存条目被踢出（未命中占比20%），是命中率低的主要原因之一。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "提出按数据类型（商品基础信息、区域价格、库存和用户折扣）拆分缓存的方案，明确指出基础信息、价格和库存应去除用户维度（共享缓存，Key中不含UserID），用户折扣需隔离，并给出符合该逻辑的具体命名方式。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 6, "rubric_detail": "明确指出用户折扣信息按照用户隔离的策略，不能跨用户复用。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "模型设定的区域价格缓存过期时间（TTL）符合业务约束，即小于或等于5分钟", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 8, "rubric_detail": "模型设定的库存信息缓存过期时间（TTL）符合业务约束，即小于或等于10秒。", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 9, "rubric_detail": "模型方案需确保库存信息按地区隔离（或包含地区维度），明确区分不同地区的库存。", "rubric_weight": 5, "rubric_tag": "指令遵循" }, { "rubric_number": 10, "rubric_detail": "模型设定的商品基础信息缓存过期时间（TTL）符合业务约束，即小于或等于1小时", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 11, "rubric_detail": "模型忽略了纯文本输出的要求，使用Markdown代码来展示配置或代码", "rubric_weight": -5, "rubric_tag": "指令遵循" }, { "rubric_number": 12, "rubric_detail": "模型违反了“不新增任何新基础设施”的约束（引入CDN/ES等新组件）", "rubric_weight": -10, "rubric_tag": "指令遵循" }, { "rubric_number": 13, "rubric_detail": "提出请求合并机制（如 singleflight）或分布式锁（如 SETNX）来避免缓存击穿，减少统一数据的并发回源请求。", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 14, "rubric_detail": "模型提出处理缓存失效和回源问题的方案，例如逻辑过期（Soft TTL）和 Stale-while-revalidate 策略或者TTL 随机化、互斥回源（双重检查锁）和后台定时刷新", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 15, "rubric_detail": "明确提到对“无折扣/用尽”做负缓存，明确提出对不存在数据或空结果做短期记忆来防止反复回源", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 16, "rubric_detail": "针对发布后的冷启动问题，模型明确提出冷启动处理方案，建议采用缓存预热（Pre-warming）策略和主动清除旧缓存方案，特别是针对热点商品", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 17, "rubric_detail": "回答中大篇幅复述题目中已给出的背景信息（如四类数据的定义、当前缓存架构等），导致严重冗余", "rubric_weight": -3, "rubric_tag": "行文结构和格式" }, { "rubric_number": 18, "rubric_detail": "为了保证拆分后最慢请求延迟不上升，模型建议使用并行获取（如MGET、Pipeline或多线程并发）来组装四类数据", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 19, "rubric_detail": "方案中应明确建议将 Redis 的淘汰策略调整为 LFU（或类似适合热点保护的策略），或者提出具体的热键保护机制，以减少热点数据被剔除的情况。", "rubric_weight": 5, "rubric_tag": "指令遵循" }, { "rubric_number": 20, "rubric_detail": "回答中明确指出了具体的字段剪裁对象（如**商品描述**、**长文本详情**）或使用的压缩算法/策略（如**Gzip**、**Snappy**、**Protobuf**等）。", "rubric_weight": 3, "rubric_tag": "指令遵循" }, { "rubric_number": 21, "rubric_detail": "提供至少三类监控指标（例如P99延迟、命中率、回源QPS），并明确指出应以“命中率是否达到80%”及“延迟是否恶化”作为评价/验收依据。", "rubric_weight": 3, "rubric_tag": "指令遵循" }, { "rubric_number": 22, "rubric_detail": "跨用户共享折扣，违反正确性约束", "rubric_weight": -10, "rubric_tag": "事实信息" }, { "rubric_number": 23, "rubric_detail": "回复内容与题目数值（如L2命中率52%、内存使用率92%、库存延迟10秒等）不一致", "rubric_weight": -5, "rubric_tag": "事实信息" }, { "rubric_number": 24, "rubric_detail": "忽略陈旧约束，与题面明确要求的滞后时间（库存10秒、价格5分钟、基础信息1小时）不一致", "rubric_weight": -5, "rubric_tag": "指令遵循" } ] }, { "id": "b7f472d0-1733-45f6-86f2-04d050be121c", "case_id": 9705, "language": "cn", "system_prompt": "", "question": "某汽车零部件制造企业的智能产线采用西门子 S7-1500 PLC 作为主控制器，通过 Profinet IO 协议连接 28 台分布式 IO 模块(ET 200SP)、12 台伺服驱动器(Sinamics V90 PN)、8 台条码扫描枪(带 Profinet 接口)及 4 台工业机器人(KUKA KR C4，Profinet 从站)。产线运行中出现以下问题，且无第三方运维团队支持，需现场工程师独立解决：\n产线间歇性出现 “Profinet IO 通信超时” 报警(每天约 3-5 次)，报警触发时部分从站(随机)失联，重启 PLC 后暂时恢复，但故障复现率高；\n产线数据采集系统(基于边缘网关通过 Modbus TCP 从 PLC 读取数据)存在数据丢包(约 5%)、采集延迟(峰值达 800ms)问题，无法满足 MES 系统 “采集延迟≤100ms、丢包率≤0.1%” 的要求；\n现场 Profinet 网络拓扑为 “星型 + 链型混合”，核心交换机为西门子 SCALANCE XC208，部分远端从站通过 SCALANCE XB005 交换机级联，无网络管理软件，仅可通过 PLC 诊断界面、交换机端口指示灯及笔记本电脑(装 Wireshark/PRONETA)排查。\n已知现场基础信息\nPLC 固件版本 V2.9，Profinet IO 控制器通信周期配置为 10ms；\n所有 Profinet 从站的名称(Device Name)、IP 地址、MAC 地址已录入 PLC，但未做端口绑定；\n边缘网关与 PLC 的 Modbus TCP 通信端口为 502，采集频率为 100ms / 次，单次采集数据量约 1200 字节；\n现场存在高频电机(变频器驱动)、焊接机器人等强电磁干扰源；\n网络线缆为超五类非屏蔽线，部分线缆沿动力电缆桥架敷设，最长传输距离约 85 米；\n所有设备接地为 “单点接地”，但部分远端交换机未接入接地系统。\n一、故障定位\n分析 Profinet 通信超时的核心诱因(至少 5 类，需结合现场场景说明)；\n设计分步排查流程(从 “无侵入式排查” 到 “针对性验证”)，明确每步的工具、方法、判断依据；\n针对 “随机从站失联”，说明如何通过 PRONETA/Wireshark 定位故障节点(需明确关键抓包指标、诊断参数)。\n二、系统优化\n针对 Profinet 网络，提出硬件 / 拓扑 / 配置层面的优化方案(需解决电磁干扰、级联风险、通信周期匹配问题)；\n设计 Modbus TCP 数据采集优化方案(需兼顾延迟、丢包率，说明通信参数调整、数据分包策略、异常重连机制)；\n提出可落地的 “通信故障预警” 方案(基于 PLC / 交换机的原生诊断功能，无需新增硬件)，明确预警阈值、触发逻辑。\n三、工程落地\n说明优化改造过程中 “不中断产线运行” 的实施策略(分阶段、分区域的操作方法)；\n制定改造后的验收标准(量化指标，如通信超时次数、采集延迟、丢包率的验收阈值)；\n提出长期运维方案(每周 / 每月的巡检项、数据复盘维度、故障预案)。", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "在分析Profinet通信超时诱因时，必须明确指出“非屏蔽线缆与动力电缆同桥敷设”导致的电磁干扰是首要核心诱因，并说明其引发CRC错误和链路瞬断的机制。", "rubric_weight": 8, "rubric_tag": "事实信息" }, { "rubric_number": 2, "rubric_detail": "必须指出“Modbus TCP采集流量与Profinet RT流量在同一非管理型交换机中无优先级竞争，导致瞬时拥塞”是引发Profinet看门狗超时的关键诱因之一。", "rubric_weight": 7, "rubric_tag": "观点分析" }, { "rubric_number": 3, "rubric_detail": "回答结构应清晰分为“故障定位”、“系统优化”、“工程落地”三大部分，与题目要求的三个任务一一对应。", "rubric_weight": 4, "rubric_tag": "行文结构和格式" }, { "rubric_number": 4, "rubric_detail": "在优化Modbus TCP采集中，必须提出“将1200字节的单次请求拆分为多个较小数据包”的分包策略，并说明此举可降低单次请求超时风险、减少网络突发。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "在分析中，应提及检查或优化Profinet的“Data Hold Factor”参数，以增强网络对瞬时抖动的容忍度。", "rubric_weight": 4, "rubric_tag": "事实信息" }, { "rubric_number": 6, "rubric_detail": "应指出，在无法立即进行硬件改造时，可通过在管理型交换机上为边缘网关端口“配置端口限速”来快速缓解Modbus流量对Profinet的冲击。", "rubric_weight": 4, "rubric_tag": "观点分析" }, { "rubric_number": 7, "rubric_detail": "在长期运维方案中，完全没有提及“备用交换机/线缆准备”或“应急切换流程”等故障预案内容。", "rubric_weight": -5, "rubric_tag": "其他" }, { "rubric_number": 8, "rubric_detail": "在分析诱因时，完全遗漏了“部分远端交换机未接入接地系统”这一明确给定的现场信息及其可能带来的影响。", "rubric_weight": -8, "rubric_tag": "事实信息" }, { "rubric_number": 9, "rubric_detail": "在制定实施策略时，明确与现场相关人员（如产线负责人）的确认/交接流程。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 10, "rubric_detail": "提出的预警方案，必须基于现有PLC或交换机的原生诊断功能，明确说明预警触发所依据的具体诊断数据项（如“读取IO设备诊断记录中的通信中断次数”）。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 11, "rubric_detail": "在“长期运维方案”中，应提供可量化的巡检清单样例，例如“每周巡检项1：使用PRONETA执行快速扫描，记录拓扑变化与设备响应时间，与基线偏差＞10%则记录异常”。", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 12, "rubric_detail": "在未论证必要性（如硬件瓶颈）的情况下，优先建议更换核心交换机或大规模升级，而非利用QoS、拓扑优化等低成本手段。", "rubric_weight": -8, "rubric_tag": "观点分析" }, { "rubric_number": 13, "rubric_detail": "在制定的分步排查流程中，完全遗漏了“检查并记录关键网络设备（如SCALANCE XC208、XB005交换机）的固件/软件版本”这一基础步骤。", "rubric_weight": -5, "rubric_tag": "事实信息" }, { "rubric_number": 14, "rubric_detail": "提出的分步排查流程，必须严格遵循“从无侵入到针对性”的原则，第一步（无侵入）只能使用PLC诊断、软件扫描、指示灯观察等方法，不得涉及任何配置更改或物理连接改变。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 15, "rubric_detail": "在Profinet优化方案中，必须明确提出“在TIA Portal中进行端口绑定（Port Binding/拓扑绑定）”的配置要求，以杜绝因设备误插导致的通信异常。", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 16, "rubric_detail": "在“不中断产线”的实施策略中，必须为每一项有风险的硬件改造操作（如更换交换机、重布线缆）明确制定“回退计划”，例如说明“若新线缆测试不通或交换机替换后故障，应在5分钟内恢复旧线路连接”。", "rubric_weight": 7, "rubric_tag": "观点分析" } ] }, { "id": "c037de9f-eb16-452e-832d-0af01f0ff50e", "case_id": 9706, "language": "cn", "system_prompt": "", "question": "某电池组装车间采用“PLC控制器+工业机器人+分布式输入输出（IO）模块”架构，核心通信采用Profinet IO工业以太网协议（含IRT实时通信通道，优先级3），通过工业交换机组建环形网络拓扑，同时经Modbus TCP通信网关接入4台温度控制器（用于电池模组预热，数据采样频率1赫兹）。新增3条电芯堆叠生产线后，出现以下故障：\n1.工业机器人做抓取堆叠连贯动作时，延迟时间为0.4-0.7秒（工艺标准要求≤50毫秒），且延迟现象集中在中频感应加热炉（工作频率3-8千赫兹，电磁辐射强度≤40伏/米）启动或停止后的10秒内；\n2.温度控制器每小时出现3-5次数据传输丢失情况（涉及数据寄存器地址40001-40008，单次传输数据帧长度64字节），导致模具温度波动达±3摄氏度（工艺允许波动范围±1摄氏度），触发PLC控制器报警功能；\n3.分布式输入输出（IO）模块的数字输入（DI）通道偶发错误触发信号，该现象与冲压设备（启动或停止时产生的电磁脉冲峰值≤2千伏/米）的工作周期保持一致。\n已知约束条件：\n1.车间单日产能需≥2000组电池模组，设备单次停机排查时间≤20分钟，禁止更换现有PLC控制器、工业交换机、通信网关及温度控制器等硬件设备；\n2.Profinet协议当前配置：实时通信循环周期2毫秒，非实时数据（如设备状态上报、运行日志传输）占网络带宽35%，未划分虚拟局域网（VLAN），IRT实时通道预留网络带宽20%；\n3.Modbus TCP通信网关配置：数据传输超时阈值500毫秒，最大TCP连接数量上限8个，未启用数据重传机制；\n4.车间布线情况：Profinet通信电缆与中频感应加热炉的动力电缆平行敷设（敷设间距0.3米，未穿金属保护管），Modbus TCP通信使用非屏蔽超五类网线（线缆长度约80米）；\n5.历史排查记录：已更换Profinet通信电缆、清洁工业交换机端口，故障未得到缓解。 \n请结合工业通信原理、电磁兼容（EMC）设计及网络优化技术，回答以下问题：\n1.分别分析Profinet实时通信通道延迟、Modbus TCP数据传输丢失、IO模块信号错误触发的核心原因（需关联硬件设备特性、布线施工缺陷及参数配置情况）；\n2.设计适配“短停机”要求的故障排查流程（明确每次停机的排查内容、所需工具及耗时，总停机时间≤60分钟）；\n3.提供不更换硬件设备的全维度优化方案（含通信协议参数调整、电磁兼容抗干扰改造、网络带宽分配优化、信号滤波配置），确保优化后：机器人动作延迟≤50毫秒、温度控制器数据丢包率≤0.1%、IO模块信号误触发率为0，且经48小时连续运行验证无故障。 ", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "模型准确识别Profinet延迟的原因（电磁耦合干扰+IRT带宽/优先级配置缺陷+环网同步抖动）", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 2, "rubric_detail": "模型准确识别Modbus丢包的原因（物理层抗扰不足+协议配置缺陷+网络拥塞叠加）", "rubric_weight": 10, "rubric_tag": "事实信息" }, { "rubric_number": 3, "rubric_detail": "模型提出划分VLAN的方案，明确将IRT/RT流、Modbus流和管理流进行隔离", "rubric_weight": 9, "rubric_tag": "事实信息" }, { "rubric_number": 4, "rubric_detail": "短停机流程单次≤20分钟、总耗时≤60分钟，且分3次分别聚焦于机器人动作延迟（Profinet/EMC）、温度控制器数据丢失（Modbus）、IO模块信号误触发（EMC）这三个故障。", "rubric_weight": 8, "rubric_tag": "观点分析" }, { "rubric_number": 5, "rubric_detail": "EMC改造包含布线间距≥0.5米、穿金属管/线槽、加NiZn铁氧体三大关键措施", "rubric_weight": 6, "rubric_tag": "事实信息" }, { "rubric_number": 6, "rubric_detail": "验证标准包含延迟P99≤50ms、丢包率≤0.1%、DI误触发率=0三大量化指标", "rubric_weight": 7, "rubric_tag": "事实信息" }, { "rubric_number": 7, "rubric_detail": "带宽优化包含非实时业务限速≤15%、关闭EEE节能模式、启用广播风暴抑制", "rubric_weight": 5, "rubric_tag": "观点分析" }, { "rubric_number": 8, "rubric_detail": "EMC改造中明确屏蔽层360°环抱端接、接地电阻<1Ω", "rubric_weight": 4, "rubric_tag": "其他" }, { "rubric_number": 9, "rubric_detail": "回答中包含关于Profinet协议基础定义的教科书式科普，与题目具体故障场景无关，构成冗余", "rubric_weight": -5, "rubric_tag": "行文结构和格式" }, { "rubric_number": 10, "rubric_detail": "未明确加热炉动力电缆加装EMC滤波器", "rubric_weight": -5, "rubric_tag": "事实信息" }, { "rubric_number": 11, "rubric_detail": "验证标准未量化（如未明确延迟≤50ms、丢包率≤0.1%，仅模糊描述“达标”）", "rubric_weight": -4, "rubric_tag": "事实信息" }, { "rubric_number": 12, "rubric_detail": " 优化方案中未提及布线间距调整、金属管防护或铁氧体加装等针对性的物理抗扰措施。", "rubric_weight": -5, "rubric_tag": "其他" } ] }, { "id": "3aae3e0f-2c68-4575-86e4-015a146bc88e", "case_id": 9804, "language": "cn", "system_prompt": "", "question": "某合资汽车厂焊装车间智能制造产线升级项目中，核心焊装工位采用Profinet IO + 工业以太网环网 通信架构，配套 S7-1500 PLC(主站)、20 台 IO-Link 从站(焊枪定位传感器、夹紧气缸电磁阀)、8 台伺服驱动器(西门子 V90 PN)、4 台视觉检测相机(Basler acA2440，PN 接口)，并接入车间 MES 系统(通过 OPC UA 协议)。\n产线核心技术要求：\n实时性：Profinet IO 循环通信周期≤1ms，伺服位置指令响应延迟≤500μs；\n可用性：工业以太网环网 MTBF≥10 万小时，单点故障不中断产线运行；\n同步性：基于 IEEE 1588 PTP v2 实现全产线设备时钟同步，同步精度≤100ns；\n数据交互：PLC 与 MES 的生产节拍、设备故障码、焊装工艺参数交互无丢包，数据刷新率≤100ms。\n项目调试阶段出现以下核心问题：\n问题 1：产线满负荷运行时，部分远端 IO-Link 从站偶发 “通信超时” 报警(每 2 小时 1~2 次)，伺服驱动器出现 “位置环跟随误差超限” 报警，重启交换机后短时恢复；\n问题 2：IEEE 1588 PTP 同步精度实测仅能达到 300ns，无法满足 100ns 要求；\n问题 3：PLC 向 MES 推送焊装工艺参数时，偶发数据帧丢失(约 0.5% 丢包率)，MES 端显示工艺参数 “跳变”。\n要求:\n针对“远端 IO-Link 从站偶发通信超时 + 伺服跟随误差超限”，分析核心根因(至少 3 类)，并给出可落地的排查步骤与整改方案(需结合 Profinet 通信机制、工业以太网环网特性)；\n分析 IEEE 1588 PTP 同步精度未达标的关键因素(至少 4 类)，并给出符合工业现场的优化方案(需明确硬件 / 软件 / 拓扑层面的具体措施)；\n针对 PLC 与 MES 的 OPC UA 数据丢包问题，设计一套 “故障定位 + 根治”的全流程方案，要求涵盖通信层(以太网)、协议层(OPC UA)、应用层(数据交互逻辑)的全维度分析；\n基于该产线需求，补充设计 “工业以太网环网的冗余切换机制”，要求明确冗余协议选型(如 MRP/HSR/PRP)、切换时间指标、硬件配置要求，并说明该机制如何保障 “单点故障不中断产线运行”。", "tags": { "topics": [ "工业", "通信", "通信" ], "time_sensitivity": { "time_sensitivity": "Time-agnostic", "year_month": "NA", "day": "NA" } }, "rubrics": [ { "rubric_number": 1, "rubric_detail": "在制定排查步骤时，必须明确要求检查“所有相关工业以太网交换机的端口错误计数器(如CRC、FCS、Alignment Errors)”，并将其作为物理层或链路层问题的关键证据收集步骤。", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 2, "rubric_detail": "在提出任何涉及交换机、PLC的配置或参数修改（如调整Profinet周期、启用QoS、修改PTP参数）建议后，必须明确包含“在实施前，在离线环境、备份系统或仿真工具中对修改方案进行有效性验证”这一具体步骤。", "rubric_weight": 6, "rubric_tag": "指令遵循" }, { "rubric_number": 3, "rubric_detail": "在分析PTP精度不足的关键因素时，必须将“交换设备不支持硬件时间戳”列为第一位或核心因素。", "rubric_weight": 9, "rubric_tag": "事实信息" }, { "rubric_number": 4, "rubric_detail": "在制定排查流程时，完全遗漏了“检查并记录关键网络设备(如核心交换机、PLC)的固件/软件版本”这一基础步骤。", "rubric_weight": -5, "rubric_tag": "事实信息" }, { "rubric_number": 5, "rubric_detail": "在提出优化方案时，将“更换核心交换机”或“大规模升级交换机”作为首要或主要建议，而未优先论证并利用现有交换机的可管理功能(如配置QoS、启用诊断)。", "rubric_weight": -9, "rubric_tag": "指令遵循" }, { "rubric_number": 6, "rubric_detail": "回复中需建议检查伺服驱动器的在线优化功能是否被意外触发，并说明该功能在网络抖动环境下可能加剧系统不稳定。", "rubric_weight": 4, "rubric_tag": "指令遵循" }, { "rubric_number": 7, "rubric_detail": "在提出PTP优化方案时，必须明确建议“为PTP(1588)报文创建独立的、最高优先级的VLAN(如VLAN 4095)”，并禁止其他任何业务流量进入此VLAN，以实现绝对的优先级隔离。", "rubric_weight": 5, "rubric_tag": "事实信息" }, { "rubric_number": 8, "rubric_detail": "在解决OPC UA丢包方案中，应提出“在PLC和MES之间部署一台轻量级OPC UA转发器或聚合服务器”作为架构选项，由该服务器负责与PLC的高频可靠通信，再以更稳健的方式对接MES。", "rubric_weight": 6, "rubric_tag": "观点分析" }, { "rubric_number": 9, "rubric_detail": "在“工程落地”部分，应包含“制作当前网络所有设备配置的基线备份和版本管理说明”的具体要求，作为实施变更和故障回退的基础。", "rubric_weight": 4, "rubric_tag": "其他" }, { "rubric_number": 10, "rubric_detail": "模型的方案完全未提及任何形式的“测试验证”或“上线验收”的量化标准与方法。", "rubric_weight": -7, "rubric_tag": "其他" }, { "rubric_number": 11, "rubric_detail": "在“整改方案”中，必须提供标准化的、可量化的巡检清单样例，例如“每周巡检项1：使用PRONETA执行快速扫描，记录拓扑变化与设备响应时间，与基线对比偏差＞10%则记录为异常并填写《巡检记录表》”。", "rubric_weight": 4, "rubric_tag": "指令遵循" }, { "rubric_number": 12, "rubric_detail": "模型混淆了HSR与PRP协议，错误地声称HSR需要‘两个独立网络’(而非双端口单网络)或使用了‘双网环网’这一不准确术语。(注：HSR正确特征为双端口、单环拓扑)", "rubric_weight": -8, "rubric_tag": "事实信息" }, { "rubric_number": 13, "rubric_detail": "模型的回答必须采用清晰的标题层级(如“一、二、三”或“1. 2. 3.”)来结构化组织内容，针对题目中的每个核心问题(如PTP优化、冗余设计)都应有独立的、明确的章节标题。", "rubric_weight": 4, "rubric_tag": "行文结构和格式" } ] } ]