thanatchon36 commited on
Commit
c8b99a6
·
verified ·
1 Parent(s): 004b0ba

Publish data for v0.1.0

Browse files
README.md ADDED
@@ -0,0 +1,117 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-4.0
3
+ pretty_name: "context-loop-engineering: corpus, taxonomy codes, GitHub diffusion and simulation results"
4
+ language:
5
+ - en
6
+ size_categories:
7
+ - 1K<n<10K
8
+ task_categories:
9
+ - text-classification
10
+ tags:
11
+ - context-engineering
12
+ - loop-engineering
13
+ - harness-engineering
14
+ - llm-agents
15
+ - systematic-review
16
+ - literature-review
17
+ - taxonomy
18
+ - simulation
19
+ configs:
20
+ - config_name: included_corpus
21
+ data_files: corpus/included_corpus.csv
22
+ default: true
23
+ - config_name: coded_corpus_all
24
+ data_files: corpus/coded_corpus.csv
25
+ - config_name: coupling_lift
26
+ data_files: taxonomy/table_coupling_lift.csv
27
+ - config_name: github_monthly
28
+ data_files: github/github_monthly.csv
29
+ - config_name: github_repos
30
+ data_files: github/github_repos.csv
31
+ - config_name: looparena
32
+ data_files: looparena/looparena_table.csv
33
+ - config_name: sim_e3_phase
34
+ data_files: simulation/model_e3_phase.csv
35
+ - config_name: sim_e6_global_sensitivity
36
+ data_files: simulation/model_e6_global_sensitivity.csv
37
+ ---
38
+
39
+ # context-loop-engineering: corpus, taxonomy codes, GitHub diffusion and simulation results
40
+
41
+ Data for the preprint *Context Engineering and Loop Engineering for LLM Agents: A Systematic Review, Unified
42
+ Taxonomy, and Use-Case Fit Framework* by Thanatchon Chongmankhong ([read online](https://thanatchon36.github.io/context-loop-engineering/paper/) ·
43
+ [PDF](https://github.com/thanatchon36/context-loop-engineering/blob/main/paper.pdf)). The study reviews 1,395 sources on the two engineered
44
+ layers around an LLM agent: **context engineering** (what the model sees at each step) and **loop engineering**
45
+ (whether, when and on what evidence the agent runs again when no person prompts it). It codes every source with a
46
+ taxonomy of 7 context operations × 8 loop levels, audits the codes against blind manual coding, measures how the
47
+ terms spread on GitHub, re-analyses the published LoopArena results and runs six simulation experiments.
48
+ Code and the full replication package: <https://github.com/thanatchon36/context-loop-engineering>; archived on Zenodo:
49
+ [10.5281/zenodo.24216599](https://doi.org/10.5281/zenodo.24216599).
50
+
51
+ Every number in the paper is regenerated from these files by `make all` in the GitHub repository.
52
+
53
+ ## Files
54
+
55
+ | path | rows | content |
56
+ |---|---:|---|
57
+ | `corpus/included_corpus.csv` | 1,395 | **the coded corpus** (544 academic, 274 grey, 577 repositories): title, URL, arXiv id, venue, date, source type, the reading lists and sections it came from, and its codes: `ctx_Compose` … `ctx_Guard` (7 context operations, 0/1), `loop_L1` … `loop_L8` (8 loop levels, 0/1), primary `layer` (context, loop, harness, evaluation), `contribution` (method, tool, practice guide, survey, benchmark, empirical), `domain` and `evaluative` |
58
+ | `corpus/coded_corpus.csv` | 1,492 | all 1,492 unique sources with `included` and `exclusion_reason` |
59
+ | `corpus/corpus.csv` | 1,492 | the deduplicated sources before coding |
60
+ | `corpus/items_all.csv` | 1,891 | every entry parsed from the 5 reading lists, with list, line and section path |
61
+ | `corpus/table_*.csv` | | academic sources by half-year and layer; contribution type by layer |
62
+ | `taxonomy/table_coupling_lift.csv` | 56 | co-occurrence of each loop level × context operation: observed, expected, lift, one-sided p, Holm-adjusted p, and whether the pair was predicted in the pre-analysis draft |
63
+ | `audit/` | | blind audit samples, manual codes and agreement results for both rounds; `coded_corpus_rules_v1.csv` and `audit_round1_rules_v1.json` are frozen records of codebook v1, kept because v1 was revised after its audit |
64
+ | `github/github_monthly.csv` | 23 | new repositories per month (Jan 2025 – Nov 2026) for "prompt", "context", "harness" and "loop engineering" |
65
+ | `github/github_counts.csv` | 100 | the same counts in long form with retrieval flags |
66
+ | `github/github_repos.csv` | 6,278 | metadata of every matching public repository: created, stars, forks, language, owner type, description, topics, matched terms, script flags |
67
+ | `looparena/looparena_table.csv` | 16 | LoopArena re-analysis: successes, runs, strict success rate with 95% Wilson interval, cost per run and per solved task, by setting (Type II condensed, Type III full) and loop policy |
68
+ | `simulation/model_e*.csv` | | results of the six Monte Carlo experiments: E1 verifier ceiling, E2 token cost, E3 phase diagram of context policies, E4 verification × context, E5 model strength, E6 global sensitivity (600 random draws) and retention grid |
69
+ | `taxonomy/codebook.json` | | the exact coding rules (codebook v2): regular expressions for scope, exclusions, each context operation, loop level, layer and contribution |
70
+ | `results.json` | | every summary statistic quoted in the paper |
71
+
72
+ ## Load
73
+
74
+ ```python
75
+ from datasets import load_dataset
76
+
77
+ corpus = load_dataset("thanatchon36/context-loop-engineering", "included_corpus", split="train")
78
+ lift = load_dataset("thanatchon36/context-loop-engineering", "coupling_lift", split="train")
79
+ ```
80
+
81
+ ## How the data were made
82
+
83
+ The sampling frame is five curated GitHub reading lists on context and harness engineering, retrieved on
84
+ 13 November 2026; GitHub counts and repository metadata come from the GitHub search API on the same day; the
85
+ LoopArena figures come from its published aggregate results. Sources were deduplicated, screened and coded by the
86
+ rule-based codebook in `taxonomy/codebook.json`. A held-out audit of 60 sources against blind manual coding gave
87
+ κ = 0.86 for the primary layer.
88
+
89
+ ## Limitations
90
+
91
+ - arXiv and bibliographic databases were not reachable from the analysis environment, so the literature was
92
+ sampled through curated lists; three of the five centre on harness engineering, and the frame favours recent,
93
+ English-language work.
94
+ - Codes rest on titles, list annotations and curators' section headings, not on full texts.
95
+ - The manual coder in the audit was the AI assistant that wrote the coding rules, so κ measures consistency
96
+ between the rules and careful reading, not agreement between independent human experts.
97
+ - LoopArena is one benchmark, one domain and one worker model, analysed from aggregate data.
98
+ - Simulation parameters are assumptions, not estimates; E6 varies them to show which conclusions are robust.
99
+
100
+ *Independent personal research. Does not represent the views of any employer or institution.*
101
+
102
+ ## Licence and citation
103
+
104
+ CC BY 4.0. Titles, descriptions and URLs of sources and repositories are those published by their authors and the
105
+ list curators. Raw third-party files are not included here; they are in the GitHub repository under their owners'
106
+ terms.
107
+
108
+ ```bibtex
109
+ @misc{chongmankhong2026context,
110
+ author = {Chongmankhong, Thanatchon},
111
+ title = {Context Engineering and Loop Engineering for LLM Agents: A Systematic Review, Unified Taxonomy, and Use-Case Fit Framework},
112
+ year = {2026},
113
+ howpublished = {Preprint},
114
+ doi = {10.5281/zenodo.24216599},
115
+ url = {https://thanatchon36.github.io/context-loop-engineering/paper/}
116
+ }
117
+ ```
audit/audit2_manual_codes.csv ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ source_id,layer,contribution
2
+ S1150,harness,tool
3
+ S1437,harness,tool
4
+ S0466,loop,method
5
+ S0789,harness,practice_guide
6
+ S0164,context,method
7
+ S0536,evaluation,benchmark
8
+ S1204,evaluation,benchmark
9
+ S1338,harness,tool
10
+ S1440,harness,tool
11
+ S0899,harness,practice_guide
12
+ S0249,loop,method
13
+ S0524,evaluation,benchmark
14
+ S0823,context,tool
15
+ S1051,harness,practice_guide
16
+ S0400,harness,survey
17
+ S1250,loop,tool
18
+ S0464,loop,empirical
19
+ S0603,harness,tool
20
+ S0252,harness,tool
21
+ S0308,context,survey
22
+ S0371,loop,method
23
+ S0788,harness,practice_guide
24
+ S0952,loop,tool
25
+ S0831,loop,practice_guide
26
+ S0709,context,tool
27
+ S0340,context,method
28
+ S0367,evaluation,benchmark
29
+ S1091,harness,empirical
30
+ S0988,harness,tool
31
+ S0514,evaluation,benchmark
32
+ S0350,context,method
33
+ S0794,harness,tool
34
+ S0824,context,method
35
+ S0163,context,method
36
+ S0951,loop,practice_guide
37
+ S0276,evaluation,empirical
38
+ S0202,context,method
39
+ S0972,harness,tool
40
+ S0351,context,method
41
+ S1457,harness,tool
42
+ S1097,harness,tool
43
+ S1182,harness,tool
44
+ S1241,harness,tool
45
+ S0497,loop,method
46
+ S1359,harness,tool
47
+ S0189,loop,method
48
+ S0412,harness,method
49
+ S1321,context,tool
50
+ S0882,harness,tool
51
+ S1456,loop,tool
52
+ S1229,harness,tool
53
+ S1067,harness,tool
54
+ S1366,harness,tool
55
+ S0017,loop,survey
56
+ S0661,loop,practice_guide
57
+ S0037,context,method
58
+ S0804,context,tool
59
+ S1008,harness,tool
60
+ S0137,context,method
61
+ S0544,evaluation,benchmark
audit/audit2_sample_blind.csv ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ source_id,source_type,title,description,sections
2
+ S1150,repository,HEAAL,Grammar-enforced safety constraints for AI agents via AIL (AI-Intent Language).,"Constraints, Guardrails & Safe Autonomy > Security, Authorization & Policy"
3
+ S1437,repository,ZeroClaw,"Single-binary agent runtime with providers, channels, tools, memory, SOPs, approval gates, sandboxing, ACP, and tool receipts.",Catalog > Reference Harness Implementations
4
+ S0466,academic,WorkTeam: Constructing Workflows from Natural Language with Multi-Agents,,Research Papers > System Intelligence > Agent Coordination || Applications > Enterprise Workflows and Digital Organizations
5
+ S0789,grey,Claude Agent SDK — Configure Permissions,"The most concrete reference for harness permission architecture: five-layer evaluation order (hooks → deny rules → permission mode → allow rules → canUseTool), ",Design Primitives > Permissions & Authorization
6
+ S0164,academic,REPLUG: Retrieval-Augmented Black-Box Language Models,,Implementation and Challenges > 1. Retrieval-Augmented Generation (RAG)
7
+ S0536,repository,Mem2ActBench,,"Benchmarks, Datasets, and Environments > Individual Intelligence"
8
+ S1204,repository,MCPMark,"A stress-testing benchmark for model and agent capabilities in real-world MCP tasks across tools like Notion, GitHub, and PostgreSQL, making harness MCP integra","Benchmarks > Tools, APIs & MCP"
9
+ S1338,repository,Capsule,Durable runtime that coordinates agent tasks inside isolated WebAssembly sandboxes with retries and lifecycle tracking.,Catalog > Execution Substrates & Sandboxing
10
+ S1440,repository,Crush,"Terminal coding agent that wires project tools, code, workflows, sessions, LSP context, model switching, and MCP extensions into an LLM runtime.",Catalog > Reference Harness Implementations
11
+ S0899,grey,Claude Code /doctor,"Anthropic's July 2026 bundled setup-checkup skill (alias /checkup) that audits harness hygiene: it deduplicates local and checked-in CLAUDE.md files, flags unus",Design Primitives > Debugging & Developer Experience
12
+ S0249,academic,CoMAS: Co-Evolving Multi-Agent Systems via Interaction Rewards,,Implementation and Challenges > 3. Agent Communication > Open Agent Protocols and Interoperability
13
+ S0524,academic,AgencyBench,,"Benchmarks, Datasets, and Environments > Individual Intelligence"
14
+ S0823,repository,Hindsight,"Agent memory system organized around three explicit operations—retain, recall, and reflect—with semantic, keyword, graph, and temporal retrieval plus an MCP ser",Design Primitives > Memory & State || Catalog > Context & Working-State Engineering
15
+ S1051,grey,Cloudflare Dynamic Workers,"V8 isolate-based sandboxing for AI-agent-generated code execution, now in open beta. Isolates start in milliseconds using megabytes of memory — 100x faster and ","Security, Sandbox & Permissions"
16
+ S0400,academic,"Internet of Agents: Fundamentals, Applications, and Challenges",,Research Papers > Individual Intelligence > Interaction Paradigm
17
+ S1250,repository,Cowork Forge,MIT-licensed multi-agent software-development workflow with specialized roles and a staged pipeline from requirements through delivery.,"Runtimes, Harnesses & Reference Implementations > Multi-Agent Orchestration"
18
+ S0464,academic,Multi-agent teams hold experts back,,Research Papers > System Intelligence > Agent Coordination
19
+ S0603,grey,Project ALICE,,Applications > Software Engineering and IT Operations
20
+ S0252,grey,Google Vertex AI Agent Engine,,Implementation and Challenges > 4. Tool Use and Function Calling > Hosted Agent Tools and Computer Use || Applications and Systems > Production Systems > Platform Stacks and Hosted Agent Runtimes
21
+ S0308,academic,"Graph-Based Agent Memory: Taxonomy, Techniques, and Applications",,Related Survey Papers
22
+ S0371,academic,Evolving Agents in the Dark: Retrospective Harness Optimization via Self-Preference,,Research Papers > Individual Intelligence > Runtime Orchestration
23
+ S0788,grey,Claude Code Auto Mode: A Safer Way to Skip Permissions,"Anthropic's engineering post on replacing approval fatigue (users approve 93% of prompts, making approvals meaningless) with a two-stage classifier: fast single",Design Primitives > Permissions & Authorization || Featured Harness Blogs || Catalog > Essential Readings & Ecosystem Maps
24
+ S0952,repository,AutoAgent,"Open-source library (April 2026) that automates the harness engineering loop itself: give it a task and a benchmark, and it iterates overnight on system prompts",Reference Implementations > Generators & Meta-Harnesses
25
+ S0831,grey,Building a C Compiler with a Team of Parallel Claudes,Anthropic's account of coordinating 16 Claude instances in parallel on a shared git repo without a central orchestrator: agents claim tasks via files in current,Design Primitives > Task Runners & Orchestration
26
+ S0709,repository,Context7,"MCP server and CLI that injects up-to-date, version-specific library documentation directly into agent context, eliminating hallucinated APIs and outdated code ",Design Primitives > Context Delivery & Compaction
27
+ S0340,academic,Escaping the Context Bottleneck: Active Context Curation for LLM Agents via Reinforcement Learning,,Research Papers > Model Intelligence > Context Engineering
28
+ S0367,academic,Harness-Bench: Measuring Harness Effects across Models in Realistic Agent Workflows,"Diagnostic benchmark that isolates the execution layer (context, tools, state, recovery) by evaluating model-harness pairings under shared tasks. The core findi","Research Papers > Individual Intelligence > Runtime Orchestration || Benchmarks, Datasets, and Environments > Individual Intelligence || Evals & Verification"
29
+ S1091,academic,AgentCgroup: Understanding and Controlling OS Resources of AI Agents,February 2026 empirical study of sandboxed coding-agent workloads finding that OS-level execution accounts for 56–74% of end-to-end latency and memory is the re,Production Infrastructure & Operations
30
+ S0988,repository,AIO Sandbox,"All-in-one agent sandbox combining browser, shell, filesystem, MCP servers, and VSCode Server in a single Docker container. Native MCP support exposes sandbox c",Reference Implementations > Demo Harnesses || Catalog > Execution Substrates & Sandboxing
31
+ S0514,academic,OlymMATH,,"Benchmarks, Datasets, and Environments > Model Intelligence"
32
+ S0350,academic,"Memori: A Persistent Memory Layer for Efficient, Context-Aware LLM Agents",,Research Papers > Individual Intelligence > Memory Management
33
+ S0794,repository,Agent Vault,"Infisical's open-source credential broker that sits between AI agents and the APIs they call, injecting real credentials onto outbound requests so agents never ","Design Primitives > Permissions & Authorization || Catalog > Guardrails, Security & Governance"
34
+ S0824,academic,ClawVM: Harness-Managed Virtual Memory for Stateful Tool-Using LLM Agents,"Applies virtual-memory semantics to agent context management, treating the context window as working memory with typed pages, minimum-fidelity invariants, and v",Design Primitives > Memory & State
35
+ S0163,academic,Prompt-Guided Retrieval Augmentation for Non-Knowledge-Intensive Tasks,,Implementation and Challenges > 1. Retrieval-Augmented Generation (RAG)
36
+ S0951,grey,HyperAgents: Self-Improving AI Systems,"Meta's framework integrating task-solving and meta-level improvement into a unified, editable program with metacognitive self-modification. Improved paper-revie",Reference Implementations > Generators & Meta-Harnesses
37
+ S0276,academic,Can Large Language Models Replace Humans in Systematic Reviews?,,Applications and Systems > Complex Research Systems
38
+ S0202,academic,Memory-Augmented Generative Adversarial Transformers,,Implementation and Challenges > 2. Memory Systems > Project Memory and Instruction Artifacts
39
+ S0972,repository,Anthropic Computer Use Demo,"Anthropic's reference harness for the screenshot-action loop: defines the screenshot, bash, and text_editor tool interface that makes desktop/browser control wo",Reference Implementations > Demo Harnesses
40
+ S0351,academic,LycheeMemory V2: Efficient Long-Term Memory for LLM Agents via Semantic Segment-Level Consolidation,,Research Papers > Individual Intelligence > Memory Management
41
+ S1457,repository,holaOS,"Desktop-first long-horizon agent environment with runtime, memory, tools, apps, and durable state.",Catalog > Reference Harness Implementations
42
+ S1097,repository,getagentseal/codeburn,"Free, local tool that tracks AI coding token usage and cost across 31 tools and agents by model, project, and task. Worth including because cross-tool cost visi",Production Infrastructure & Operations
43
+ S1182,repository,flameox,"Profiling and optimization toolkit for agents: bounded CLI and MCP workflows capture traces, preserve native evidence, and compare experiments behind runtime co","Evals & Observability > Telemetry, Tracing & Performance"
44
+ S1241,repository,Zaru,"Apache-2.0 terminal harness in Rust that writes every model reply, tool call, permission decision and tool result to a plain-text JSONL transcript, and runs pro","Runtimes, Harnesses & Reference Implementations > Coding-Agent Harnesses"
45
+ S0497,academic,Shepherd: Enabling Programmable Meta-Agents via Reversible Agentic Execution Traces,,Research Papers > System Intelligence > State Management
46
+ S1359,repository,MCP Registry,Official MCP registry service and publisher tooling that lets clients discover MCP servers through a shared server catalog.,"Catalog > Protocols, Tool Interfaces & Agent Contracts"
47
+ S0189,academic,Chain-of-verification reduces hallucination in large language models,,Implementation and Challenges > 1. Retrieval-Augmented Generation (RAG)
48
+ S0412,academic,DeltaBox: Scaling Stateful AI Agents with Millisecond-Level Sandbox Checkpoint/Rollback,,Research Papers > Individual Intelligence > Environment Feedback
49
+ S1321,repository,CCPM,Spec-driven project-manager skill that turns PRDs and GitHub issues into persistent context and parallel agent execution.,Catalog > Context & Working-State Engineering
50
+ S0882,repository,Arize Phoenix,"Self-hostable trace UI and eval runtime for agent workflows. Lets harness engineers audit and replay every reasoning step and tool call offline, without sending",Design Primitives > Observability & Tracing || Catalog > Observability & Reliability Operations
51
+ S1456,repository,Agent Orchestrator,Worktree-based orchestration layer for parallel coding agents with autonomous CI and review feedback handling.,Catalog > Reference Harness Implementations
52
+ S1229,repository,TeaQL Agent Kit,"Model-mediated coding-agent workflow for database-backed applications, combining deterministic domain-model evaluation and repair, typed code generation, and ev","Runtimes, Harnesses & Reference Implementations > Coding-Agent Harnesses || Catalog > Evaluation Harnesses & Benchmarks"
53
+ S1067,repository,claude-bulkhead,"PreToolUse hooks that keep parallel Claude Code sessions on one machine from interfering with each other: they block name-based process kills, port clashes, hea","Security, Sandbox & Permissions"
54
+ S1366,repository,LangChain MCP Adapters,Adapters connecting LangChain components with MCP servers.,"Catalog > Protocols, Tool Interfaces & Agent Contracts"
55
+ S0017,academic,The Landscape of Agentic Reinforcement Learning for LLMs: A Survey,,Related Survey
56
+ S0661,grey,Getting started with loops,"Anthropic's June 2026 practical taxonomy of agent loops: turn-based, goal-based (/goal), time-based (/loop, /schedule), and proactive loops. The framework for m",Design Primitives > Agent Loop
57
+ S0037,academic,A-MEM: Agentic Memory for LLM Agents,,Related Survey || Research Papers > Individual Intelligence > Memory Management
58
+ S0804,repository,TencentDB-Agent-Memory,Tencent's fully local agent memory system with a 4-tier progressive pipeline (Conversation → Atom → Scenario → Persona) and symbolic short-term memory via Merma,Design Primitives > Memory & State || Catalog > Context & Working-State Engineering
59
+ S1008,repository,ZCode,"Z.ai's official open-source coding agent harness (September 2026, Apache-2.0): a single TypeScript agent runtime — CLI, TUI, tools, and an RPC client SDK — serv",Reference Implementations > Demo Harnesses
60
+ S0137,academic,RankRAG: Unifying Context Ranking with Retrieval-Augmented Generation in LLMs,,Implementation and Challenges > 1. Retrieval-Augmented Generation (RAG)
61
+ S0544,academic,TaskBench,,"Benchmarks, Datasets, and Environments > System Intelligence"
audit/audit_manual_codes.csv ADDED
@@ -0,0 +1,101 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ source_id,layer,contribution
2
+ S1348,harness,tool
3
+ S1313,context,tool
4
+ S1098,loop,practice_guide
5
+ S0620,harness,tool
6
+ S0264,evaluation,method
7
+ S0981,harness,tool
8
+ S1221,harness,tool
9
+ S0566,evaluation,benchmark
10
+ S1038,harness,practice_guide
11
+ S0908,harness,practice_guide
12
+ S1206,evaluation,benchmark
13
+ S0896,harness,tool
14
+ S0561,evaluation,benchmark
15
+ S1171,evaluation,practice_guide
16
+ S1152,harness,tool
17
+ S1137,context,practice_guide
18
+ S0695,context,method
19
+ S0170,context,method
20
+ S0601,harness,tool
21
+ S0867,evaluation,practice_guide
22
+ S0358,context,method
23
+ S0366,harness,empirical
24
+ S0800,harness,practice_guide
25
+ S0953,loop,tool
26
+ S1299,loop,tool
27
+ S0418,loop,method
28
+ S1260,loop,tool
29
+ S0086,context,method
30
+ S0555,evaluation,benchmark
31
+ S0970,loop,tool
32
+ S0140,context,method
33
+ S0402,harness,empirical
34
+ S1397,harness,tool
35
+ S1350,harness,tool
36
+ S0220,harness,practice_guide
37
+ S0098,context,method
38
+ S0693,context,tool
39
+ S0048,context,method
40
+ S1379,evaluation,benchmark
41
+ S0002,context,survey
42
+ S0303,context,survey
43
+ S0987,harness,practice_guide
44
+ S0462,loop,method
45
+ S1332,harness,tool
46
+ S0468,loop,method
47
+ S0859,loop,tool
48
+ S0410,loop,method
49
+ S0674,loop,tool
50
+ S1226,harness,tool
51
+ S1417,harness,tool
52
+ S0110,loop,method
53
+ S1265,loop,tool
54
+ S0020,evaluation,benchmark
55
+ S1460,harness,tool
56
+ S0622,loop,method
57
+ S0480,loop,method
58
+ S0461,loop,method
59
+ S0167,context,method
60
+ S1390,harness,tool
61
+ S0401,harness,survey
62
+ S0047,context,method
63
+ S0664,harness,tool
64
+ S0711,context,tool
65
+ S1021,harness,practice_guide
66
+ S1050,harness,method
67
+ S0112,loop,method
68
+ S1023,harness,practice_guide
69
+ S1300,harness,tool
70
+ S0193,context,method
71
+ S0032,context,method
72
+ S0102,evaluation,benchmark
73
+ S1467,harness,tool
74
+ S0656,harness,empirical
75
+ S1020,harness,practice_guide
76
+ S0343,harness,method
77
+ S0296,context,tool
78
+ S0025,context,method
79
+ S0692,context,method
80
+ S1139,context,tool
81
+ S0812,context,practice_guide
82
+ S0171,context,method
83
+ S0736,harness,tool
84
+ S0301,loop,survey
85
+ S0106,context,method
86
+ S1249,loop,tool
87
+ S0977,harness,tool
88
+ S0356,context,method
89
+ S1084,evaluation,practice_guide
90
+ S1239,loop,tool
91
+ S0659,harness,practice_guide
92
+ S0388,evaluation,benchmark
93
+ S0457,harness,survey
94
+ S0121,harness,practice_guide
95
+ S0923,loop,empirical
96
+ S1111,harness,practice_guide
97
+ S1374,loop,tool
98
+ S0115,loop,method
99
+ S0368,harness,method
100
+ S0770,harness,tool
101
+ S0505,loop,method
audit/audit_results.json ADDED
@@ -0,0 +1,102 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "layer": {
3
+ "kappa": 0.663,
4
+ "agreement": 0.76,
5
+ "n": 100,
6
+ "labels": [
7
+ "context",
8
+ "evaluation",
9
+ "harness",
10
+ "loop"
11
+ ],
12
+ "confusion_rule_rows_manual_cols": [
13
+ [
14
+ 23,
15
+ 1,
16
+ 0,
17
+ 3
18
+ ],
19
+ [
20
+ 0,
21
+ 11,
22
+ 1,
23
+ 1
24
+ ],
25
+ [
26
+ 2,
27
+ 0,
28
+ 30,
29
+ 8
30
+ ],
31
+ [
32
+ 0,
33
+ 0,
34
+ 8,
35
+ 12
36
+ ]
37
+ ]
38
+ },
39
+ "contribution": {
40
+ "kappa": 0.781,
41
+ "agreement": 0.84,
42
+ "n": 100,
43
+ "labels": [
44
+ "benchmark",
45
+ "empirical",
46
+ "method",
47
+ "practice_guide",
48
+ "survey",
49
+ "tool"
50
+ ],
51
+ "confusion_rule_rows_manual_cols": [
52
+ [
53
+ 4,
54
+ 0,
55
+ 0,
56
+ 3,
57
+ 0,
58
+ 0
59
+ ],
60
+ [
61
+ 0,
62
+ 2,
63
+ 1,
64
+ 0,
65
+ 0,
66
+ 0
67
+ ],
68
+ [
69
+ 3,
70
+ 2,
71
+ 30,
72
+ 0,
73
+ 3,
74
+ 0
75
+ ],
76
+ [
77
+ 0,
78
+ 0,
79
+ 0,
80
+ 14,
81
+ 0,
82
+ 2
83
+ ],
84
+ [
85
+ 0,
86
+ 0,
87
+ 0,
88
+ 0,
89
+ 2,
90
+ 1
91
+ ],
92
+ [
93
+ 1,
94
+ 0,
95
+ 0,
96
+ 0,
97
+ 0,
98
+ 32
99
+ ]
100
+ ]
101
+ }
102
+ }
audit/audit_results_v2.json ADDED
@@ -0,0 +1,211 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "round1_v2_in_sample": {
3
+ "layer": {
4
+ "kappa": 0.86,
5
+ "agreement": 0.9,
6
+ "n": 100,
7
+ "labels": [
8
+ "context",
9
+ "evaluation",
10
+ "harness",
11
+ "loop"
12
+ ],
13
+ "confusion_rule_rows_manual_cols": [
14
+ [
15
+ 25,
16
+ 1,
17
+ 0,
18
+ 0
19
+ ],
20
+ [
21
+ 0,
22
+ 11,
23
+ 1,
24
+ 1
25
+ ],
26
+ [
27
+ 0,
28
+ 0,
29
+ 35,
30
+ 4
31
+ ],
32
+ [
33
+ 0,
34
+ 0,
35
+ 3,
36
+ 19
37
+ ]
38
+ ]
39
+ },
40
+ "contribution": {
41
+ "kappa": 0.781,
42
+ "agreement": 0.84,
43
+ "n": 100,
44
+ "labels": [
45
+ "benchmark",
46
+ "empirical",
47
+ "method",
48
+ "practice_guide",
49
+ "survey",
50
+ "tool"
51
+ ],
52
+ "confusion_rule_rows_manual_cols": [
53
+ [
54
+ 4,
55
+ 0,
56
+ 0,
57
+ 3,
58
+ 0,
59
+ 0
60
+ ],
61
+ [
62
+ 0,
63
+ 2,
64
+ 1,
65
+ 0,
66
+ 0,
67
+ 0
68
+ ],
69
+ [
70
+ 3,
71
+ 2,
72
+ 30,
73
+ 0,
74
+ 3,
75
+ 0
76
+ ],
77
+ [
78
+ 0,
79
+ 0,
80
+ 0,
81
+ 14,
82
+ 0,
83
+ 2
84
+ ],
85
+ [
86
+ 0,
87
+ 0,
88
+ 0,
89
+ 0,
90
+ 2,
91
+ 1
92
+ ],
93
+ [
94
+ 1,
95
+ 0,
96
+ 0,
97
+ 0,
98
+ 0,
99
+ 32
100
+ ]
101
+ ]
102
+ }
103
+ },
104
+ "round2_v2_held_out": {
105
+ "layer": {
106
+ "kappa": 0.856,
107
+ "agreement": 0.9,
108
+ "n": 60,
109
+ "labels": [
110
+ "context",
111
+ "evaluation",
112
+ "harness",
113
+ "loop"
114
+ ],
115
+ "confusion_rule_rows_manual_cols": [
116
+ [
117
+ 14,
118
+ 0,
119
+ 0,
120
+ 1
121
+ ],
122
+ [
123
+ 0,
124
+ 6,
125
+ 1,
126
+ 0
127
+ ],
128
+ [
129
+ 0,
130
+ 1,
131
+ 24,
132
+ 2
133
+ ],
134
+ [
135
+ 0,
136
+ 0,
137
+ 1,
138
+ 10
139
+ ]
140
+ ]
141
+ },
142
+ "contribution": {
143
+ "kappa": 0.764,
144
+ "agreement": 0.833,
145
+ "n": 60,
146
+ "labels": [
147
+ "benchmark",
148
+ "empirical",
149
+ "method",
150
+ "practice_guide",
151
+ "survey",
152
+ "tool"
153
+ ],
154
+ "confusion_rule_rows_manual_cols": [
155
+ [
156
+ 1,
157
+ 0,
158
+ 0,
159
+ 0,
160
+ 0,
161
+ 0
162
+ ],
163
+ [
164
+ 0,
165
+ 1,
166
+ 0,
167
+ 0,
168
+ 0,
169
+ 0
170
+ ],
171
+ [
172
+ 3,
173
+ 1,
174
+ 15,
175
+ 0,
176
+ 1,
177
+ 0
178
+ ],
179
+ [
180
+ 0,
181
+ 0,
182
+ 0,
183
+ 7,
184
+ 0,
185
+ 2
186
+ ],
187
+ [
188
+ 0,
189
+ 1,
190
+ 0,
191
+ 0,
192
+ 2,
193
+ 0
194
+ ],
195
+ [
196
+ 2,
197
+ 0,
198
+ 0,
199
+ 0,
200
+ 0,
201
+ 24
202
+ ]
203
+ ]
204
+ }
205
+ },
206
+ "pooled_v2_layer": {
207
+ "kappa": 0.859,
208
+ "agreement": 0.9,
209
+ "n": 160
210
+ }
211
+ }
audit/audit_round1_rules_v1.json ADDED
@@ -0,0 +1,102 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "layer": {
3
+ "kappa": 0.663,
4
+ "agreement": 0.76,
5
+ "n": 100,
6
+ "labels": [
7
+ "context",
8
+ "evaluation",
9
+ "harness",
10
+ "loop"
11
+ ],
12
+ "confusion_rule_rows_manual_cols": [
13
+ [
14
+ 23,
15
+ 1,
16
+ 0,
17
+ 3
18
+ ],
19
+ [
20
+ 0,
21
+ 11,
22
+ 1,
23
+ 1
24
+ ],
25
+ [
26
+ 2,
27
+ 0,
28
+ 30,
29
+ 8
30
+ ],
31
+ [
32
+ 0,
33
+ 0,
34
+ 8,
35
+ 12
36
+ ]
37
+ ]
38
+ },
39
+ "contribution": {
40
+ "kappa": 0.781,
41
+ "agreement": 0.84,
42
+ "n": 100,
43
+ "labels": [
44
+ "benchmark",
45
+ "empirical",
46
+ "method",
47
+ "practice_guide",
48
+ "survey",
49
+ "tool"
50
+ ],
51
+ "confusion_rule_rows_manual_cols": [
52
+ [
53
+ 4,
54
+ 0,
55
+ 0,
56
+ 3,
57
+ 0,
58
+ 0
59
+ ],
60
+ [
61
+ 0,
62
+ 2,
63
+ 1,
64
+ 0,
65
+ 0,
66
+ 0
67
+ ],
68
+ [
69
+ 3,
70
+ 2,
71
+ 30,
72
+ 0,
73
+ 3,
74
+ 0
75
+ ],
76
+ [
77
+ 0,
78
+ 0,
79
+ 0,
80
+ 14,
81
+ 0,
82
+ 2
83
+ ],
84
+ [
85
+ 0,
86
+ 0,
87
+ 0,
88
+ 0,
89
+ 2,
90
+ 1
91
+ ],
92
+ [
93
+ 1,
94
+ 0,
95
+ 0,
96
+ 0,
97
+ 0,
98
+ 32
99
+ ]
100
+ ]
101
+ }
102
+ }
audit/audit_sample_blind.csv ADDED
@@ -0,0 +1,101 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ source_id,source_type,title,description,sections
2
+ S1348,repository,skills,"CLI for installing, using, finding, updating, and initializing Agent Skills across OpenCode, Claude Code, Codex, Cursor, and other coding agents.","Catalog > Protocols, Tool Interfaces & Agent Contracts"
3
+ S1313,repository,code-review-graph,"Local-first code intelligence graph that incrementally maps dependencies and exposes targeted review context, blast-radius analysis, and risk-scored CI gates th",Catalog > Context & Working-State Engineering
4
+ S1098,grey,KernelEvolve: How Meta's Ranking Engineer Agent Optimizes AI Infrastructure,Meta's production-grade agentic kernel optimization system that autonomously generates optimized Triton kernels for hundreds of models serving billions of users,Production Infrastructure & Operations
5
+ S0620,repository,Hermes Agent,"Self-improving agent runtime with memory, skill creation, subagents, scheduled automations, and pluggable terminal backends.",Applications > General-Purpose Digital Agents and Personal Automation || Catalog > Referen
6
+ S0264,academic,BBScore: A Brownian Bridge Based Metric for Assessing Text Coherence,,Evaluation Paradigms for Context-Driven Systems > Context Quality Assessment
7
+ S0981,repository,Aider,"AI pair-programmer harness with an Architect mode that splits planning (one LLM) from coding (another), and git-aware tooling that uses version control as the u",Reference Implementations > Demo Harnesses || Catalog > Reference Harness Implementations
8
+ S1221,repository,DSH Studio,"Cross-platform desktop host for DeepSeek Harness with health probes, restart backoff, collision-free ports, and whole-process-tree cleanup.","Runtimes, Harnesses & Reference Implementations > Runtime Foundations & Control Layers"
9
+ S0566,academic,MAFBench,,"Benchmarks, Datasets, and Environments > System Intelligence"
10
+ S1038,grey,Under the Hood: Security Architecture of GitHub Agentic Workflows,"GitHub's March 9, 2026 architecture write-up is one of the clearest public descriptions of defense-in-depth for coding agents running inside CI: isolated agent ","Security, Sandbox & Permissions"
11
+ S0908,grey,More Visibility into Copilot Coding Agent Sessions,"GitHub's March 19, 2026 changelog is short but materially useful: setup-step logs, collapsed subagent traces, and clearer session-stage visibility are exactly t",Design Primitives > Debugging & Developer Experience
12
+ S1206,repository,TravelPlanner,"A benchmark for evaluating LLM agents on tool use and complex planning within multiple constraints, revealing how harness design handles multi-constraint satisf","Benchmarks > Tools, APIs & MCP"
13
+ S0896,repository,AgentOps,"Open-source agent engineering platform (YC W24) with session replay, cost tracking, and failure detection across 10+ frameworks including CrewAI, LangGraph, and","Design Primitives > Debugging & Developer Experience || Evals & Observability > Telemetry,"
14
+ S0561,academic,SyncBench,,"Benchmarks, Datasets, and Environments > System Intelligence"
15
+ S1171,grey,Evaluation best practices,OpenAI's general guide to building eval suites that match real-world distributions and catch regressions early.,Evals & Observability > Evaluation Design
16
+ S1152,repository,sofagent,"Commit-time audit harness for AI coding agents: 28 deterministic, zero-token audit rules over Git diffs (secrets, out-of-scope edits, blind modifications, promp","Constraints, Guardrails & Safe Autonomy > Security, Authorization & Policy"
17
+ S1137,grey,Writing a good CLAUDE.md,"A practical guide to creating durable, repo-local instructions that agents can repeatedly follow.","Context, Memory & Working State > Context Design & Delivery"
18
+ S0695,academic,A-RAG: Scaling Agentic Retrieval-Augmented Generation via Hierarchical Retrieval Interfaces,"Reframes RAG as a harness tool-design problem: instead of injecting retrieved documents into context at pipeline time, expose three retrieval tools (keyword sea",Design Primitives > Context Delivery & Compaction
19
+ S0170,academic,Iseeq: Information seeking question generation using dynamic meta-information retrieval and knowledge graphs,,Implementation and Challenges > 1. Retrieval-Augmented Generation (RAG)
20
+ S0601,grey,OpenCode,,Applications > Software Engineering and IT Operations
21
+ S0867,grey,Testing Agent Skills Systematically with Evals,"OpenAI's framework for skill regression testing: four eval dimensions (outcome, process, style, efficiency goals), JSONL trace capture for deterministic checks ",Design Primitives > Verification & CI Integration || Evals & Observability > Evaluation De
22
+ S0358,academic,SkillComposer: Learning to Evolve Agent Skills for Specification and Generalization,,Research Papers > Individual Intelligence > Skill Composition
23
+ S0366,academic,Configuring Agentic AI Coding Tools: An Exploratory Study,,Research Papers > Individual Intelligence > Runtime Orchestration
24
+ S0800,grey,Connections: Managed Credentials and Per-Caller Identity for Managed Deep Agents,LangChain's September 2026 implementation of on-behalf-of authorization as a harness primitive: credentials live in the workspace (never in .env or the build) a,Design Primitives > Permissions & Authorization
25
+ S0953,repository,metaharness,Open-source Python library (April 2026) that implements an outer optimization loop around executable harnesses for coding agents. Inspired by the Meta-Harness p,Reference Implementations > Generators & Meta-Harnesses
26
+ S1299,repository,Shannon,"Production-oriented multi-agent orchestration framework with Temporal workflows, WASI sandboxing, OPA approvals, token budgets, and OpenTelemetry tracing.",Catalog > Harness Architecture & Orchestration
27
+ S0418,academic,VFlow: Discovering Optimal Agentic Workflows for Verilog Generation,,Research Papers > System Intelligence > Task Organization
28
+ S1260,repository,stelow,"Agentic product-workflow harness with Shape Up boundaries, adversarial plan review, acceptance-based execution contracts, and audit loops.","Runtimes, Harnesses & Reference Implementations > Workflow, Profiles & Asset Management"
29
+ S0086,academic,CaMML: Context-Aware Multimodal Learner for Large Models,,"Components, Techniques and Architectures > Context Scaling"
30
+ S0555,academic,SILO-BENCH,,"Benchmarks, Datasets, and Environments > System Intelligence"
31
+ S0970,repository,Sage,"Agent harness with a recursive self-improvement (RSI) loop that separates task execution, evaluation, and package tuning. A reference for evidence-backed harnes",Reference Implementations > Generators & Meta-Harnesses
32
+ S0140,academic,Query Rewriting in Retrieval-Augmented Large Language Models,,Implementation and Challenges > 1. Retrieval-Augmented Generation (RAG)
33
+ S0402,academic,Beyond the Protocol: Unveiling Attack Vectors in the Model Context Protocol (MCP) Ecosystem,,Research Papers > Individual Intelligence > Interaction Paradigm
34
+ S1397,repository,OpenLIT,"OpenTelemetry-native AI engineering platform for LLM and coding-agent observability, evaluations, guardrails, prompt management, secrets, and dashboards.",Catalog > Observability & Reliability Operations
35
+ S1350,repository,Serena,"MCP toolkit that gives coding agents IDE-like semantic retrieval, editing, refactoring, debugging, and memory tools.","Catalog > Protocols, Tool Interfaces & Agent Contracts"
36
+ S0220,grey,Model Context Protocol Specification,,Implementation and Challenges > 3. Agent Communication > Open Agent Protocols and Interope
37
+ S0098,academic,Let's Ask GNN: Empowering Large Language Model for Graph In-Context Learning,,"Components, Techniques and Architectures > Structured Data Integration"
38
+ S0693,repository,context-mode,"MCP server that intercepts raw tool output before it enters the context window, sandboxing bulky data (Playwright snapshots, GitHub issues, logs) outside the LL",Design Primitives > Context Delivery & Compaction || Catalog > Context & Working-State Eng
39
+ S0048,academic,Large Language Models with Controllable Working Memory,,Related Survey || Implementation and Challenges > 2. Memory Systems > Project Memory and I
40
+ S1379,repository,WebArena,Self-hostable web environment and evaluation harness for autonomous web agents with reproducible end-to-end tasks.,Catalog > Evaluation Harnesses & Benchmarks
41
+ S0002,academic,The Prompt Report: A Systematic Survey of Prompt Engineering Techniques,,Related Survey
42
+ S0303,academic,"Integrating Graphs, Large Language Models, and Agents: Reasoning and Retrieval",,Related Survey Papers
43
+ S0987,grey,"Building NVIDIA Nemotron 3 Agents for Reasoning, Multimodal RAG, Voice, and Safety","NVIDIA's Nemotron 3 family (Super for long-context reasoning, Content Safety for multimodal moderation, VoiceChat for real-time speech) designed for scalable ag",Reference Implementations > Demo Harnesses
44
+ S0462,academic,Internet of agents: Weaving a web of heterogeneous agents for collaborative intelligence,,Research Papers > System Intelligence > Agent Coordination
45
+ S1332,repository,Sandbox Agent,"HTTP/SSE control server for running coding agents inside sandboxes with normalized sessions, permissions, event streaming, and replay.",Catalog > Execution Substrates & Sandboxing
46
+ S0468,academic,WebSwarm: Recursive Multi-Agent Orchestration for Deep-and-Wide Web Search,,Research Papers > System Intelligence > Agent Coordination
47
+ S0859,repository,LoopX,"Provider-neutral control plane that sits above existing harnesses (Codex, Claude Code, Cursor) and keeps long-horizon objectives, gates, todos, evidence, and qu",Design Primitives > Task Runners & Orchestration
48
+ S0410,academic,CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing,,Research Papers > Individual Intelligence > Environment Feedback
49
+ S0674,repository,AgentSPEX,"UIUC's open-source specification and execution language for LLM-agent workflows: declarative YAML with typed steps, branching, loops, and explicit state managem",Design Primitives > Agent Loop
50
+ S1226,repository,CUGA,"IBM Research's Apache-2.0 agent harness on LangGraph: a planner and code-act executor keep intermediate results in variables instead of the prompt, and policies","Runtimes, Harnesses & Reference Implementations > Runtime Foundations & Control Layers || "
51
+ S1417,repository,ClawKeeper,"Safety framework for autonomous agents combining skill-based policies, runtime plugins, and watcher-based governance.","Catalog > Guardrails, Security & Governance"
52
+ S0110,academic,Autonomous Tree-search Ability of Large Language Models,,"Components, Techniques and Architectures > Self-Generated Context"
53
+ S1265,repository,SAKO,"Repo-local task ledger and finish gate for Claude Code and Codex sessions: tasks, ownership and check evidence live in plain Markdown files, and SessionStart an","Runtimes, Harnesses & Reference Implementations > Workflow, Profiles & Asset Management"
54
+ S0020,academic,Episodic Memories Generation and Evaluation Benchmark for Large Language Models,,Related Survey
55
+ S1460,repository,Open SWE,Asynchronous open-source coding agent focused on software issue workflows.,Catalog > Reference Harness Implementations
56
+ S0622,academic,EconAgent,,Applications > Social and Economic Simulation
57
+ S0480,academic,AutoGRAMS: Autonomous Graphical Agent Modeling Software,,Research Papers > System Intelligence > State Management
58
+ S0461,academic,MoRSE: Task-Oriented Multi-Agent System with Mixture of Role-Subtask Experts,,Research Papers > System Intelligence > Agent Coordination
59
+ S0167,academic,Improving the Domain Adaptation of Retrieval Augmented Generation (RAG) Models for Open Domain Question Answering,,Implementation and Challenges > 1. Retrieval-Augmented Generation (RAG)
60
+ S1390,repository,RagaAI Catalyst,Agent observability and monitoring framework with timeline and graph analytics.,Catalog > Observability & Reliability Operations
61
+ S0401,academic,LLM Agent Communication Protocol (LACP) Requires Urgent Standardization: A Telecom-Inspired Protocol Is Necessary,,Research Papers > Individual Intelligence > Interaction Paradigm
62
+ S0047,academic,EM-LLM: Human-like Episodic Memory for Infinite Context LLMs,,Related Survey || Implementation and Challenges > 2. Memory Systems > Project Memory and I
63
+ S0664,repository,Life-Harness,Official implementation of a lifecycle-aware runtime harness that improves frozen LLM agents by adapting the model-environment interface across four layers: env,Design Primitives > Agent Loop
64
+ S0711,repository,OpenWiki,"LangChain's CLI that writes and maintains agent-readable wikis for codebases or purpose memory, turning documentation drift into a versioned, automatable harnes",Design Primitives > Context Delivery & Compaction
65
+ S1021,grey,"Building a safe, effective sandbox to enable Codex on Windows","OpenAI's July 2026 engineering deep-dive into sandboxing Codex on Windows, where no Seatbelt/seccomp-style capability isolation exists: why AppContainer, Window","Security, Sandbox & Permissions || Featured Harness Blogs || Catalog > Essential Readings "
66
+ S1050,academic,Grimlock: Guarding High-Agency Systems with eBPF and Attested Channels,"Roblox's agent guard for high-agency systems: eBPF-forced traffic interception channels every agent-to-agent and agent-to-service call through a guard, post-han","Security, Sandbox & Permissions"
67
+ S0112,academic,"Reflect, Retry, Reward: Self-Improving LLMs via Reinforcement Learning",,"Components, Techniques and Architectures > Self-Generated Context"
68
+ S1023,grey,Build zero-trust AI agents with Google's Agent Development Kit,"Google's August 2026 guide to hardening autonomous ADK agents against prompt injection and production-state mutation, using cryptographic write signatures, gVis","Security, Sandbox & Permissions"
69
+ S1300,repository,tRPC-Agent-Go,"Go framework for production agent systems with graph workflows, tools, memory, evaluation, protocols, and OpenTelemetry observability.",Catalog > Harness Architecture & Orchestration
70
+ S0193,academic,"Search-in-the-Chain: Towards Accurate, Credible, and Up-to-Date Large Language Models",,Implementation and Challenges > 1. Retrieval-Augmented Generation (RAG)
71
+ S0032,academic,TransformerFAM: Feedback Attention is Working Memory,,Related Survey
72
+ S0102,academic,StructuredRAG: JSON Response Formatting with Large Language Models,,"Components, Techniques and Architectures > Structured Data Integration"
73
+ S1467,repository,Google Agents CLI,"Google Cloud CLI and skill bundle that gives coding agents scaffold, evaluation, deployment, publishing, and observability workflows.",Catalog > Reference Harness Implementations
74
+ S0656,academic,"Harness Engineering: Anatomy, Architecture, and Evolution of Coding Agents","July 2026 source-code study of eleven production coding harnesses (Claude Code, Codex CLI, Gemini CLI, Mistral Vibe, OpenHands, Aider, Mini-SWE-Agent, Hermes, P",Foundations
75
+ S1020,grey,How we contain Claude across products,"Anthropic's May 2026 cross-product containment write-up: why environmental isolation must be the primary boundary, how model-layer defenses alone miss ~17% of o","Security, Sandbox & Permissions || Featured Harness Blogs || Catalog > Essential Readings "
76
+ S0343,academic,TALM: Tool Augmented Language Models,,Research Papers > Individual Intelligence > Tool Integration
77
+ S0296,grey,nv:context,,Applications and Systems > Production Systems > Coding Agents and Project Memory
78
+ S0025,academic,A Brain-inspired Memory Transformation based Differentiable Neural Computer,,Related Survey
79
+ S0692,academic,Active Context Compression: Autonomous Memory Management in LLM Agents,"Proposes a ""Focus Agent"" architecture where the agent autonomously decides when to consolidate interaction history into a persistent Knowledge block and prune r",Design Primitives > Context Delivery & Compaction
80
+ S1139,repository,DevProjex,"GUI, TUI, CLI, and read-only MCP tooling for selecting and exporting structured codebase context with live preview, secret redaction enforced in MCP mode, Git s","Context, Memory & Working State > Context Design & Delivery"
81
+ S0812,grey,Building Self-Correcting Memory in OpenWiki,"LangChain's August 2026 design account of memory that can forget: every claim the agent writes is persisted with versioned code evidence, and when that evidence",Design Primitives > Memory & State
82
+ S0171,academic,G-retriever: Retrieval-augmented generation for textual graph understanding and question answering,,Implementation and Challenges > 1. Retrieval-Augmented Generation (RAG)
83
+ S0736,repository,zerolang,"Experimental graph-first programming language where agents inspect and edit code through a compiler-derived ProgramGraph (node IDs, graph hashes, types, effects",Design Primitives > Tool Design
84
+ S0301,academic,"Graphs Meet AI Agents: Taxonomy, Progress, and Future Opportunities",,Related Survey Papers
85
+ S0106,academic,SelfCite: Self-Supervised Alignment for Context Attribution in Large Language Models,,"Components, Techniques and Architectures > Self-Generated Context"
86
+ S1249,repository,Squadron,"MIT-licensed declarative runtime for multi-agent workflows defined in HCL, including orchestration, state, dependency resolution, routing, persistence, and resu","Runtimes, Harnesses & Reference Implementations > Multi-Agent Orchestration"
87
+ S0977,repository,Goose,"Block's open-source, extensible AI agent donated to the Linux Foundation's Agentic AI Foundation in April 2026. Its MCP-native architecture treats every capabil",Reference Implementations > Demo Harnesses || Catalog > Reference Harness Implementations
88
+ S0356,academic,Harnessing LLM Agents with Skill Programs,,Research Papers > Individual Intelligence > Skill Composition
89
+ S1084,grey,Designing AI-Resistant Technical Evaluations,"Anthropic's January 21, 2026 write-up is the clearest account of a problem eval builders now have to treat as first-class: capable models can invalidate the tes",Evals & Verification
90
+ S1239,repository,Orbi,"Issue-driven harness on top of the Pi coding agent: each labelled GitHub issue runs in an isolated worktree, a separate review session must pass the exact commi","Runtimes, Harnesses & Reference Implementations > Coding-Agent Harnesses"
91
+ S0659,grey,Hooks – Codex,"OpenAI's lifecycle-hook framework for Codex: inject deterministic scripts at SessionStart, PreToolUse, PostToolUse, and other loop events to enforce guardrails,",Design Primitives > Agent Loop
92
+ S0388,academic,Harness-IF: Evaluating Instruction Following Across Instruction Surfaces in Coding Agents,,Research Papers > Individual Intelligence > Runtime Orchestration
93
+ S0457,academic,Agent Discovery in Internet of Agents: Challenges and Solutions,,Research Papers > System Intelligence > Agent Coordination
94
+ S0121,grey,Bring Your AI,,Implementation and Challenges > 0. Agent Harnesses and Runtime Systems
95
+ S0923,grey,Measuring AI Agent Autonomy in Practice,Anthropic's February 2026 empirical study of millions of real-world Claude Code interactions. Key finding: experienced users shift from per-action approval (20%,Design Primitives > Human-in-the-Loop
96
+ S1111,grey,"Building Governed Agents: A Framework for Cost, Control, and Compliance","LangChain's July 2026 framework for treating the agent gateway as a runtime control plane that enforces cost, control, and compliance policies across every mode",Production Infrastructure & Operations
97
+ S1374,repository,Agent Reinforcement Trainer,"Agent reinforcement-training harness for multi-step agents with GRPO, trajectory capture, reward assignment, training loops, and LangGraph/MCP examples.",Catalog > Evaluation Harnesses & Benchmarks
98
+ S0115,academic,Language Agent Tree Search Unifies Reasoning Acting and Planning in Language Models,"Unifies reasoning, acting, and planning via Monte Carlo Tree Search over agent trajectories. Directly informs harness design: external tool feedback as tree-sea","Components, Techniques and Architectures > Self-Generated Context || Research Papers > Sys"
99
+ S0368,academic,From Prompts to Contracts: Harness Engineering for Auditable Enterprise LLM Agents,,Research Papers > Individual Intelligence > Runtime Orchestration
100
+ S0770,repository,vurb.ts,"TypeScript framework for building production MCP servers with a ""Presenter"" perception layer that strips undeclared fields, redacts PII, and gates tool visibili",Design Primitives > Skills & MCP
101
+ S0505,academic,APEX: Autonomous Policy Exploration for Self-Evolving LLM Agents,,Research Papers > System Intelligence > State Management
audit/coded_corpus_rules_v1.csv ADDED
The diff for this file is too large to render. See raw diff
 
corpus/coded_corpus.csv ADDED
The diff for this file is too large to render. See raw diff
 
corpus/corpus.csv ADDED
The diff for this file is too large to render. See raw diff
 
corpus/included_corpus.csv ADDED
The diff for this file is too large to render. See raw diff
 
corpus/items_all.csv ADDED
The diff for this file is too large to render. See raw diff
 
corpus/table_contribution_by_layer.csv ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ layer,benchmark,empirical,method,practice_guide,survey,tool
2
+ context,0.0,0.031,0.548,0.124,0.09,0.207
3
+ evaluation,0.445,0.0,0.235,0.08,0.015,0.225
4
+ harness,0.0,0.017,0.143,0.236,0.032,0.573
5
+ loop,0.0,0.017,0.465,0.162,0.026,0.33
corpus/table_timeline_academic.csv ADDED
@@ -0,0 +1,15 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ half,context,evaluation,harness,loop
2
+ 2020H1,3,0,1,0
3
+ 2020H2,2,0,0,0
4
+ 2021H1,1,0,0,0
5
+ 2021H2,4,0,0,0
6
+ 2022H1,2,0,3,0
7
+ 2022H2,10,0,1,0
8
+ 2023H1,16,2,6,7
9
+ 2023H2,28,8,6,9
10
+ 2024H1,40,14,12,9
11
+ 2024H2,27,6,9,10
12
+ 2025H1,23,4,15,25
13
+ 2025H2,4,4,6,17
14
+ 2026H1,25,23,34,54
15
+ 2026H2,4,10,8,21
github/github_counts.csv ADDED
@@ -0,0 +1,101 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ term,year,month,n,incomplete,retrieved
2
+ context engineering,2023,,0,False,2026-11-13T09:16:18Z
3
+ context engineering,2024,,10,False,2026-11-13T09:16:25Z
4
+ context engineering,2025,1.0,3,False,2026-11-13T09:13:37Z
5
+ context engineering,2025,2.0,2,False,2026-11-13T09:13:44Z
6
+ context engineering,2025,3.0,2,False,2026-11-13T09:13:50Z
7
+ context engineering,2025,4.0,3,False,2026-11-13T09:13:57Z
8
+ context engineering,2025,5.0,4,False,2026-11-13T09:14:04Z
9
+ context engineering,2025,6.0,11,False,2026-11-13T09:14:11Z
10
+ context engineering,2025,7.0,184,False,2026-11-13T09:14:17Z
11
+ context engineering,2025,8.0,126,False,2026-11-13T09:14:30Z
12
+ context engineering,2025,9.0,76,False,2026-11-13T09:14:37Z
13
+ context engineering,2025,10.0,147,False,2026-11-13T09:14:44Z
14
+ context engineering,2025,11.0,111,False,2026-11-13T09:14:50Z
15
+ context engineering,2025,12.0,125,False,2026-11-13T09:14:57Z
16
+ context engineering,2026,1.0,132,False,2026-11-13T09:15:04Z
17
+ context engineering,2026,2.0,133,False,2026-11-13T09:15:11Z
18
+ context engineering,2026,3.0,207,False,2026-11-13T09:15:17Z
19
+ context engineering,2026,4.0,190,False,2026-11-13T09:15:24Z
20
+ context engineering,2026,5.0,191,False,2026-11-13T09:15:31Z
21
+ context engineering,2026,6.0,168,False,2026-11-13T09:15:38Z
22
+ context engineering,2026,7.0,181,False,2026-11-13T09:15:44Z
23
+ context engineering,2026,8.0,208,False,2026-11-13T09:15:51Z
24
+ context engineering,2026,9.0,199,False,2026-11-13T09:15:58Z
25
+ context engineering,2026,10.0,189,False,2026-11-13T09:16:04Z
26
+ context engineering,2026,11.0,63,False,2026-11-13T09:16:11Z
27
+ harness engineering,2023,,3,False,2026-11-13T09:19:06Z
28
+ harness engineering,2024,,3,False,2026-11-13T09:19:12Z
29
+ harness engineering,2025,1.0,0,False,2026-11-13T09:16:31Z
30
+ harness engineering,2025,2.0,1,False,2026-11-13T09:16:38Z
31
+ harness engineering,2025,3.0,0,False,2026-11-13T09:16:45Z
32
+ harness engineering,2025,4.0,2,False,2026-11-13T09:16:51Z
33
+ harness engineering,2025,5.0,0,False,2026-11-13T09:16:58Z
34
+ harness engineering,2025,6.0,1,False,2026-11-13T09:17:05Z
35
+ harness engineering,2025,7.0,2,False,2026-11-13T09:17:11Z
36
+ harness engineering,2025,8.0,0,False,2026-11-13T09:17:18Z
37
+ harness engineering,2025,9.0,1,False,2026-11-13T09:17:25Z
38
+ harness engineering,2025,10.0,0,False,2026-11-13T09:17:32Z
39
+ harness engineering,2025,11.0,0,False,2026-11-13T09:17:38Z
40
+ harness engineering,2025,12.0,2,False,2026-11-13T09:17:45Z
41
+ harness engineering,2026,1.0,4,False,2026-11-13T09:17:52Z
42
+ harness engineering,2026,2.0,25,False,2026-11-13T09:17:58Z
43
+ harness engineering,2026,3.0,242,False,2026-11-13T09:18:05Z
44
+ harness engineering,2026,4.0,408,False,2026-11-13T09:18:12Z
45
+ harness engineering,2026,5.0,321,False,2026-11-13T09:18:19Z
46
+ harness engineering,2026,6.0,255,False,2026-11-13T09:18:25Z
47
+ harness engineering,2026,7.0,261,False,2026-11-13T09:18:32Z
48
+ harness engineering,2026,8.0,204,False,2026-11-13T09:18:39Z
49
+ harness engineering,2026,9.0,360,False,2026-11-13T09:18:46Z
50
+ harness engineering,2026,10.0,270,False,2026-11-13T09:18:52Z
51
+ harness engineering,2026,11.0,110,False,2026-11-13T09:18:59Z
52
+ loop engineering,2023,,1,False,2026-11-13T09:21:53Z
53
+ loop engineering,2024,,0,False,2026-11-13T09:22:00Z
54
+ loop engineering,2025,1.0,0,False,2026-11-13T09:19:19Z
55
+ loop engineering,2025,2.0,0,False,2026-11-13T09:19:26Z
56
+ loop engineering,2025,3.0,1,False,2026-11-13T09:19:32Z
57
+ loop engineering,2025,4.0,0,False,2026-11-13T09:19:39Z
58
+ loop engineering,2025,5.0,1,False,2026-11-13T09:19:46Z
59
+ loop engineering,2025,6.0,0,False,2026-11-13T09:19:52Z
60
+ loop engineering,2025,7.0,1,False,2026-11-13T09:19:59Z
61
+ loop engineering,2025,8.0,0,False,2026-11-13T09:20:06Z
62
+ loop engineering,2025,9.0,0,False,2026-11-13T09:20:13Z
63
+ loop engineering,2025,10.0,0,False,2026-11-13T09:20:19Z
64
+ loop engineering,2025,11.0,1,False,2026-11-13T09:20:26Z
65
+ loop engineering,2025,12.0,1,False,2026-11-13T09:20:33Z
66
+ loop engineering,2026,1.0,2,False,2026-11-13T09:20:39Z
67
+ loop engineering,2026,2.0,5,False,2026-11-13T09:20:46Z
68
+ loop engineering,2026,3.0,3,False,2026-11-13T09:20:53Z
69
+ loop engineering,2026,4.0,8,False,2026-11-13T09:21:00Z
70
+ loop engineering,2026,5.0,14,False,2026-11-13T09:21:06Z
71
+ loop engineering,2026,6.0,375,False,2026-11-13T09:21:13Z
72
+ loop engineering,2026,7.0,426,False,2026-11-13T09:21:20Z
73
+ loop engineering,2026,8.0,304,False,2026-11-13T09:21:26Z
74
+ loop engineering,2026,9.0,143,False,2026-11-13T09:21:33Z
75
+ loop engineering,2026,10.0,62,False,2026-11-13T09:21:40Z
76
+ loop engineering,2026,11.0,42,False,2026-11-13T09:21:47Z
77
+ prompt engineering,2023,,1624,False,2026-11-13T09:24:41Z
78
+ prompt engineering,2024,,2742,False,2026-11-13T09:24:48Z
79
+ prompt engineering,2025,1.0,177,False,2026-11-13T09:22:07Z
80
+ prompt engineering,2025,2.0,232,False,2026-11-13T09:22:13Z
81
+ prompt engineering,2025,3.0,299,False,2026-11-13T09:22:20Z
82
+ prompt engineering,2025,4.0,437,False,2026-11-13T09:22:27Z
83
+ prompt engineering,2025,5.0,387,False,2026-11-13T09:22:34Z
84
+ prompt engineering,2025,6.0,497,False,2026-11-13T09:22:40Z
85
+ prompt engineering,2025,7.0,659,False,2026-11-13T09:22:47Z
86
+ prompt engineering,2025,8.0,610,False,2026-11-13T09:22:54Z
87
+ prompt engineering,2025,9.0,554,False,2026-11-13T09:23:00Z
88
+ prompt engineering,2025,10.0,624,False,2026-11-13T09:23:07Z
89
+ prompt engineering,2025,11.0,548,False,2026-11-13T09:23:14Z
90
+ prompt engineering,2025,12.0,735,False,2026-11-13T09:23:21Z
91
+ prompt engineering,2026,1.0,726,False,2026-11-13T09:23:27Z
92
+ prompt engineering,2026,2.0,826,False,2026-11-13T09:23:34Z
93
+ prompt engineering,2026,3.0,1330,False,2026-11-13T09:23:41Z
94
+ prompt engineering,2026,4.0,1267,False,2026-11-13T09:23:48Z
95
+ prompt engineering,2026,5.0,1431,False,2026-11-13T09:23:54Z
96
+ prompt engineering,2026,6.0,1834,False,2026-11-13T09:24:01Z
97
+ prompt engineering,2026,7.0,1898,False,2026-11-13T09:24:08Z
98
+ prompt engineering,2026,8.0,1355,False,2026-11-13T09:24:14Z
99
+ prompt engineering,2026,9.0,1470,False,2026-11-13T09:24:21Z
100
+ prompt engineering,2026,10.0,1385,False,2026-11-13T09:24:28Z
101
+ prompt engineering,2026,11.0,463,False,2026-11-13T09:24:35Z
github/github_monthly.csv ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ year,month,context engineering,harness engineering,loop engineering,prompt engineering
2
+ 2025,1,3.0,0.0,0.0,177.0
3
+ 2025,2,2.0,1.0,0.0,232.0
4
+ 2025,3,2.0,0.0,1.0,299.0
5
+ 2025,4,3.0,2.0,0.0,437.0
6
+ 2025,5,4.0,0.0,1.0,387.0
7
+ 2025,6,11.0,1.0,0.0,497.0
8
+ 2025,7,184.0,2.0,1.0,659.0
9
+ 2025,8,126.0,0.0,0.0,610.0
10
+ 2025,9,76.0,1.0,0.0,554.0
11
+ 2025,10,147.0,0.0,0.0,624.0
12
+ 2025,11,111.0,0.0,1.0,548.0
13
+ 2025,12,125.0,2.0,1.0,735.0
14
+ 2026,1,132.0,4.0,2.0,726.0
15
+ 2026,2,133.0,25.0,5.0,826.0
16
+ 2026,3,207.0,242.0,3.0,1330.0
17
+ 2026,4,190.0,408.0,8.0,1267.0
18
+ 2026,5,191.0,321.0,14.0,1431.0
19
+ 2026,6,168.0,255.0,375.0,1834.0
20
+ 2026,7,181.0,261.0,426.0,1898.0
21
+ 2026,8,208.0,204.0,304.0,1355.0
22
+ 2026,9,199.0,360.0,143.0,1470.0
23
+ 2026,10,189.0,270.0,62.0,1385.0
24
+ 2026,11,63.0,110.0,42.0,463.0
github/github_repos.csv ADDED
The diff for this file is too large to render. See raw diff
 
looparena/looparena_table.csv ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ setting,method,role,successes,runs,rate,lo,hi,cost_per_run,cost_per_success
2
+ type2,No control,reference,118,240,0.49166666666666664,0.42904808854732723,0.5545478196160049,1.6203,3.295525423728814
3
+ type2,Fixed control,reference,132,240,0.55,0.4867634425716991,0.6116611084483075,5.6226,10.222909090909091
4
+ type2,Qwen3.6-35B-A3B,controller,134,240,0.5583333333333333,0.49507767085201887,0.6197509720046555,4.2138,7.54710447761194
5
+ type2,Qwen3.7-Plus,controller,136,240,0.5666666666666667,0.5034094093437343,0.6278233253496076,6.036,10.651764705882352
6
+ type2,DeepSeek-V4-Flash-0731,controller,122,240,0.5083333333333333,0.44545218038399503,0.5709519114526727,9.8849,19.44570491803279
7
+ type2,GLM 5.2,controller,97,240,0.4041666666666667,0.3440630619540164,0.4672898819243045,3.539,8.756288659793814
8
+ type2,GPT-5.5,controller,136,240,0.5666666666666667,0.5034094093437343,0.6278233253496076,13.7986,24.350470588235297
9
+ type2,Claude Opus 4.8,controller,141,240,0.5875,0.5243161227556644,0.6479268415293468,10.8751,18.5108085106383
10
+ type3,No control,reference,66,240,0.275,0.22238768826588753,0.33470183214408356,2.6147,9.508000000000001
11
+ type3,Fixed control,reference,75,240,0.3125,0.2572003867927223,0.3737075468822535,9.474,30.316800000000004
12
+ type3,Qwen3.6-35B-A3B,controller,81,240,0.3375,0.28065337442469745,0.3994668347602816,11.5808,34.31348148148148
13
+ type3,Qwen3.7-Plus,controller,82,240,0.3416666666666667,0.28458018698925336,0.40374206811405955,12.6868,37.13209756097561
14
+ type3,DeepSeek-V4-Flash-0731,controller,68,240,0.2833333333333333,0.23008386871880204,0.3434097435278367,32.655,115.2529411764706
15
+ type3,GLM 5.2,controller,58,240,0.24166666666666667,0.1918497206713789,0.2996234323919212,11.9735,49.54551724137931
16
+ type3,GPT-5.5,controller,87,240,0.3625,0.30428862453168165,0.4250438601633006,34.6787,95.66537931034482
17
+ type3,Claude Opus 4.8,controller,81,240,0.3375,0.28065337442469745,0.3994668347602816,20.6288,61.12237037037036
results.json ADDED
@@ -0,0 +1,1330 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "flow": {
3
+ "lists": 5,
4
+ "entries": 1891,
5
+ "unique": 1492,
6
+ "excluded": 97,
7
+ "excluded_by_reason": {
8
+ "excluded section": 58,
9
+ "out of scope": 31,
10
+ "model architecture": 8
11
+ },
12
+ "included": 1395,
13
+ "entries_by_list": {
14
+ "aiboost_harness": 532,
15
+ "deepjlu_graph": 447,
16
+ "picrew_harness": 385,
17
+ "meirtz_context": 348,
18
+ "walkinglabs_harness": 179
19
+ },
20
+ "multi_list_sources": 180
21
+ },
22
+ "composition": {
23
+ "academic": {
24
+ "context": 190,
25
+ "evaluation": 92,
26
+ "harness": 110,
27
+ "loop": 152
28
+ },
29
+ "grey": {
30
+ "context": 39,
31
+ "evaluation": 38,
32
+ "harness": 146,
33
+ "loop": 51
34
+ },
35
+ "repository": {
36
+ "context": 61,
37
+ "evaluation": 70,
38
+ "harness": 346,
39
+ "loop": 100
40
+ }
41
+ },
42
+ "composition_layer_totals": {
43
+ "harness": 602,
44
+ "loop": 303,
45
+ "context": 290,
46
+ "evaluation": 200
47
+ },
48
+ "composition_type_totals": {
49
+ "repository": 577,
50
+ "academic": 544,
51
+ "grey": 274
52
+ },
53
+ "academic_dated_share": 0.945,
54
+ "timeline_academic": {
55
+ "context": {
56
+ "2020H1": 3,
57
+ "2020H2": 2,
58
+ "2021H1": 1,
59
+ "2021H2": 4,
60
+ "2022H1": 2,
61
+ "2022H2": 10,
62
+ "2023H1": 16,
63
+ "2023H2": 28,
64
+ "2024H1": 40,
65
+ "2024H2": 27,
66
+ "2025H1": 23,
67
+ "2025H2": 4,
68
+ "2026H1": 25,
69
+ "2026H2": 4
70
+ },
71
+ "evaluation": {
72
+ "2020H1": 0,
73
+ "2020H2": 0,
74
+ "2021H1": 0,
75
+ "2021H2": 0,
76
+ "2022H1": 0,
77
+ "2022H2": 0,
78
+ "2023H1": 2,
79
+ "2023H2": 8,
80
+ "2024H1": 14,
81
+ "2024H2": 6,
82
+ "2025H1": 4,
83
+ "2025H2": 4,
84
+ "2026H1": 23,
85
+ "2026H2": 10
86
+ },
87
+ "harness": {
88
+ "2020H1": 1,
89
+ "2020H2": 0,
90
+ "2021H1": 0,
91
+ "2021H2": 0,
92
+ "2022H1": 3,
93
+ "2022H2": 1,
94
+ "2023H1": 6,
95
+ "2023H2": 6,
96
+ "2024H1": 12,
97
+ "2024H2": 9,
98
+ "2025H1": 15,
99
+ "2025H2": 6,
100
+ "2026H1": 34,
101
+ "2026H2": 8
102
+ },
103
+ "loop": {
104
+ "2020H1": 0,
105
+ "2020H2": 0,
106
+ "2021H1": 0,
107
+ "2021H2": 0,
108
+ "2022H1": 0,
109
+ "2022H2": 0,
110
+ "2023H1": 7,
111
+ "2023H2": 9,
112
+ "2024H1": 9,
113
+ "2024H2": 10,
114
+ "2025H1": 25,
115
+ "2025H2": 17,
116
+ "2026H1": 54,
117
+ "2026H2": 21
118
+ }
119
+ },
120
+ "timeline_academic_yearly": {
121
+ "context": {
122
+ "2020": 5,
123
+ "2021": 5,
124
+ "2022": 12,
125
+ "2023": 44,
126
+ "2024": 67,
127
+ "2025": 27,
128
+ "2026": 29
129
+ },
130
+ "evaluation": {
131
+ "2020": 0,
132
+ "2021": 0,
133
+ "2022": 0,
134
+ "2023": 10,
135
+ "2024": 20,
136
+ "2025": 8,
137
+ "2026": 33
138
+ },
139
+ "harness": {
140
+ "2020": 1,
141
+ "2021": 0,
142
+ "2022": 4,
143
+ "2023": 12,
144
+ "2024": 21,
145
+ "2025": 21,
146
+ "2026": 42
147
+ },
148
+ "loop": {
149
+ "2020": 0,
150
+ "2021": 0,
151
+ "2022": 0,
152
+ "2023": 16,
153
+ "2024": 19,
154
+ "2025": 42,
155
+ "2026": 75
156
+ }
157
+ },
158
+ "academic_2026_layer_share": {
159
+ "loop": 0.419,
160
+ "harness": 0.235,
161
+ "evaluation": 0.184,
162
+ "context": 0.162
163
+ },
164
+ "academic_all_layer_share": {
165
+ "context": 0.37,
166
+ "loop": 0.296,
167
+ "harness": 0.196,
168
+ "evaluation": 0.138
169
+ },
170
+ "ops_counts": {
171
+ "Compose": 100,
172
+ "Select": 425,
173
+ "Write": 280,
174
+ "Compress": 96,
175
+ "Isolate": 109,
176
+ "Evolve": 33,
177
+ "Guard": 142
178
+ },
179
+ "levels_counts": {
180
+ "L1": 115,
181
+ "L2": 199,
182
+ "L3": 185,
183
+ "L4": 13,
184
+ "L5": 344,
185
+ "L6": 176,
186
+ "L7": 48,
187
+ "L8": 94
188
+ },
189
+ "ops_share_in_context_layer": {
190
+ "Compose": 0.186,
191
+ "Select": 0.534,
192
+ "Write": 0.441,
193
+ "Compress": 0.179,
194
+ "Isolate": 0.014,
195
+ "Evolve": 0.038,
196
+ "Guard": 0.021
197
+ },
198
+ "levels_share_in_loop_layer": {
199
+ "L1": 0.069,
200
+ "L2": 0.277,
201
+ "L3": 0.149,
202
+ "L4": 0.043,
203
+ "L5": 0.488,
204
+ "L6": 0.152,
205
+ "L7": 0.158,
206
+ "L8": 0.238
207
+ },
208
+ "contribution_by_layer": {
209
+ "benchmark": {
210
+ "context": 0.0,
211
+ "evaluation": 0.445,
212
+ "harness": 0.0,
213
+ "loop": 0.0
214
+ },
215
+ "empirical": {
216
+ "context": 0.031,
217
+ "evaluation": 0.0,
218
+ "harness": 0.017,
219
+ "loop": 0.017
220
+ },
221
+ "method": {
222
+ "context": 0.548,
223
+ "evaluation": 0.235,
224
+ "harness": 0.143,
225
+ "loop": 0.465
226
+ },
227
+ "practice_guide": {
228
+ "context": 0.124,
229
+ "evaluation": 0.08,
230
+ "harness": 0.236,
231
+ "loop": 0.162
232
+ },
233
+ "survey": {
234
+ "context": 0.09,
235
+ "evaluation": 0.015,
236
+ "harness": 0.032,
237
+ "loop": 0.026
238
+ },
239
+ "tool": {
240
+ "context": 0.207,
241
+ "evaluation": 0.225,
242
+ "harness": 0.573,
243
+ "loop": 0.33
244
+ }
245
+ },
246
+ "contribution_by_layer_counts": {
247
+ "benchmark": {
248
+ "context": 0,
249
+ "evaluation": 89,
250
+ "harness": 0,
251
+ "loop": 0
252
+ },
253
+ "empirical": {
254
+ "context": 9,
255
+ "evaluation": 0,
256
+ "harness": 10,
257
+ "loop": 5
258
+ },
259
+ "method": {
260
+ "context": 159,
261
+ "evaluation": 47,
262
+ "harness": 86,
263
+ "loop": 141
264
+ },
265
+ "practice_guide": {
266
+ "context": 36,
267
+ "evaluation": 16,
268
+ "harness": 142,
269
+ "loop": 49
270
+ },
271
+ "survey": {
272
+ "context": 26,
273
+ "evaluation": 3,
274
+ "harness": 19,
275
+ "loop": 8
276
+ },
277
+ "tool": {
278
+ "context": 60,
279
+ "evaluation": 45,
280
+ "harness": 345,
281
+ "loop": 100
282
+ }
283
+ },
284
+ "domain_by_layer": {
285
+ "dialogue": {
286
+ "context": 0.048,
287
+ "evaluation": 0.03,
288
+ "harness": 0.023,
289
+ "loop": 0.013
290
+ },
291
+ "general": {
292
+ "context": 0.738,
293
+ "evaluation": 0.615,
294
+ "harness": 0.488,
295
+ "loop": 0.601
296
+ },
297
+ "research": {
298
+ "context": 0.034,
299
+ "evaluation": 0.02,
300
+ "harness": 0.055,
301
+ "loop": 0.04
302
+ },
303
+ "software": {
304
+ "context": 0.166,
305
+ "evaluation": 0.265,
306
+ "harness": 0.387,
307
+ "loop": 0.314
308
+ },
309
+ "web_gui": {
310
+ "context": 0.014,
311
+ "evaluation": 0.07,
312
+ "harness": 0.047,
313
+ "loop": 0.033
314
+ }
315
+ },
316
+ "domain_totals": {
317
+ "general": 813,
318
+ "software": 429,
319
+ "research": 59,
320
+ "web_gui": 56,
321
+ "dialogue": 38
322
+ },
323
+ "evaluative_share_by_layer": {
324
+ "context": 0.031,
325
+ "evaluation": 0.445,
326
+ "harness": 0.017,
327
+ "loop": 0.017
328
+ },
329
+ "practice_share_by_layer": {
330
+ "context": 0.345,
331
+ "evaluation": 0.54,
332
+ "harness": 0.817,
333
+ "loop": 0.498
334
+ },
335
+ "first_academic_use": {
336
+ "context engineering": {
337
+ "first_arxiv_yymm": "2510",
338
+ "n_titles_or_notes": 2
339
+ },
340
+ "harness engineering": {
341
+ "first_arxiv_yymm": "2604",
342
+ "n_titles_or_notes": 5
343
+ },
344
+ "loop engineering": {
345
+ "first_arxiv_yymm": "2607",
346
+ "n_titles_or_notes": 1
347
+ }
348
+ },
349
+ "coupling": {
350
+ "n_sources": 1395,
351
+ "pairs_tested": 56,
352
+ "pairs_significant_holm": 6,
353
+ "significant_pairs": [
354
+ {
355
+ "level_name": "Improvement",
356
+ "op": "Evolve",
357
+ "both": 20,
358
+ "lift": 8.994,
359
+ "p_holm": 0.0
360
+ },
361
+ {
362
+ "level_name": "Scheduled",
363
+ "op": "Isolate",
364
+ "both": 11,
365
+ "lift": 2.933,
366
+ "p_holm": 0.0392
367
+ },
368
+ {
369
+ "level_name": "Session",
370
+ "op": "Isolate",
371
+ "both": 35,
372
+ "lift": 2.545,
373
+ "p_holm": 0.0
374
+ },
375
+ {
376
+ "level_name": "Action",
377
+ "op": "Compress",
378
+ "both": 20,
379
+ "lift": 2.527,
380
+ "p_holm": 0.0024
381
+ },
382
+ {
383
+ "level_name": "Session",
384
+ "op": "Write",
385
+ "both": 65,
386
+ "lift": 1.84,
387
+ "p_holm": 0.0
388
+ },
389
+ {
390
+ "level_name": "Orchestration",
391
+ "op": "Isolate",
392
+ "both": 48,
393
+ "lift": 1.786,
394
+ "p_holm": 0.0002
395
+ }
396
+ ],
397
+ "draft_pairs": 19,
398
+ "draft_pairs_lift_gt1": 12,
399
+ "draft_pairs_significant": 4,
400
+ "nondraft_significant": 2,
401
+ "mean_lift_draft": 1.812,
402
+ "mean_lift_nondraft": 0.973
403
+ },
404
+ "github_monthly": {
405
+ "2025-01": {
406
+ "context engineering": 3,
407
+ "harness engineering": 0,
408
+ "loop engineering": 0,
409
+ "prompt engineering": 177
410
+ },
411
+ "2025-02": {
412
+ "context engineering": 2,
413
+ "harness engineering": 1,
414
+ "loop engineering": 0,
415
+ "prompt engineering": 232
416
+ },
417
+ "2025-03": {
418
+ "context engineering": 2,
419
+ "harness engineering": 0,
420
+ "loop engineering": 1,
421
+ "prompt engineering": 299
422
+ },
423
+ "2025-04": {
424
+ "context engineering": 3,
425
+ "harness engineering": 2,
426
+ "loop engineering": 0,
427
+ "prompt engineering": 437
428
+ },
429
+ "2025-05": {
430
+ "context engineering": 4,
431
+ "harness engineering": 0,
432
+ "loop engineering": 1,
433
+ "prompt engineering": 387
434
+ },
435
+ "2025-06": {
436
+ "context engineering": 11,
437
+ "harness engineering": 1,
438
+ "loop engineering": 0,
439
+ "prompt engineering": 497
440
+ },
441
+ "2025-07": {
442
+ "context engineering": 184,
443
+ "harness engineering": 2,
444
+ "loop engineering": 1,
445
+ "prompt engineering": 659
446
+ },
447
+ "2025-08": {
448
+ "context engineering": 126,
449
+ "harness engineering": 0,
450
+ "loop engineering": 0,
451
+ "prompt engineering": 610
452
+ },
453
+ "2025-09": {
454
+ "context engineering": 76,
455
+ "harness engineering": 1,
456
+ "loop engineering": 0,
457
+ "prompt engineering": 554
458
+ },
459
+ "2025-10": {
460
+ "context engineering": 147,
461
+ "harness engineering": 0,
462
+ "loop engineering": 0,
463
+ "prompt engineering": 624
464
+ },
465
+ "2025-11": {
466
+ "context engineering": 111,
467
+ "harness engineering": 0,
468
+ "loop engineering": 1,
469
+ "prompt engineering": 548
470
+ },
471
+ "2025-12": {
472
+ "context engineering": 125,
473
+ "harness engineering": 2,
474
+ "loop engineering": 1,
475
+ "prompt engineering": 735
476
+ },
477
+ "2026-01": {
478
+ "context engineering": 132,
479
+ "harness engineering": 4,
480
+ "loop engineering": 2,
481
+ "prompt engineering": 726
482
+ },
483
+ "2026-02": {
484
+ "context engineering": 133,
485
+ "harness engineering": 25,
486
+ "loop engineering": 5,
487
+ "prompt engineering": 826
488
+ },
489
+ "2026-03": {
490
+ "context engineering": 207,
491
+ "harness engineering": 242,
492
+ "loop engineering": 3,
493
+ "prompt engineering": 1330
494
+ },
495
+ "2026-04": {
496
+ "context engineering": 190,
497
+ "harness engineering": 408,
498
+ "loop engineering": 8,
499
+ "prompt engineering": 1267
500
+ },
501
+ "2026-05": {
502
+ "context engineering": 191,
503
+ "harness engineering": 321,
504
+ "loop engineering": 14,
505
+ "prompt engineering": 1431
506
+ },
507
+ "2026-06": {
508
+ "context engineering": 168,
509
+ "harness engineering": 255,
510
+ "loop engineering": 375,
511
+ "prompt engineering": 1834
512
+ },
513
+ "2026-07": {
514
+ "context engineering": 181,
515
+ "harness engineering": 261,
516
+ "loop engineering": 426,
517
+ "prompt engineering": 1898
518
+ },
519
+ "2026-08": {
520
+ "context engineering": 208,
521
+ "harness engineering": 204,
522
+ "loop engineering": 304,
523
+ "prompt engineering": 1355
524
+ },
525
+ "2026-09": {
526
+ "context engineering": 199,
527
+ "harness engineering": 360,
528
+ "loop engineering": 143,
529
+ "prompt engineering": 1470
530
+ },
531
+ "2026-10": {
532
+ "context engineering": 189,
533
+ "harness engineering": 270,
534
+ "loop engineering": 62,
535
+ "prompt engineering": 1385
536
+ },
537
+ "2026-11": {
538
+ "context engineering": 63,
539
+ "harness engineering": 110,
540
+ "loop engineering": 42,
541
+ "prompt engineering": 463
542
+ }
543
+ },
544
+ "github_yearly_2023_2024": {
545
+ "context engineering": {
546
+ "2023": 0.0,
547
+ "2024": 10.0
548
+ },
549
+ "harness engineering": {
550
+ "2023": 3.0,
551
+ "2024": 3.0
552
+ },
553
+ "loop engineering": {
554
+ "2023": 1.0,
555
+ "2024": 0.0
556
+ },
557
+ "prompt engineering": {
558
+ "2023": 1624.0,
559
+ "2024": 2742.0
560
+ }
561
+ },
562
+ "github_term_summary": {
563
+ "context engineering": {
564
+ "sum_2025_01_to_2026_11": 2655,
565
+ "first_month_ge_10": "2025-06",
566
+ "peak_month": "2026-08",
567
+ "peak_n": 208
568
+ },
569
+ "harness engineering": {
570
+ "sum_2025_01_to_2026_11": 2469,
571
+ "first_month_ge_10": "2026-02",
572
+ "peak_month": "2026-04",
573
+ "peak_n": 408
574
+ },
575
+ "loop engineering": {
576
+ "sum_2025_01_to_2026_11": 1389,
577
+ "first_month_ge_10": "2026-05",
578
+ "peak_month": "2026-07",
579
+ "peak_n": 426
580
+ },
581
+ "prompt engineering": {
582
+ "sum_2025_01_to_2026_11": 19744,
583
+ "first_month_ge_10": "2025-01",
584
+ "peak_month": "2026-07",
585
+ "peak_n": 1898
586
+ }
587
+ },
588
+ "github_n_repos": 6278,
589
+ "github_communities": {
590
+ "context engineering": {
591
+ "repos": 2656,
592
+ "han_share": 0.035,
593
+ "org_share": 0.095,
594
+ "median_stars": 0.0,
595
+ "share_ge_100_stars": 0.023,
596
+ "top_languages": {
597
+ "Python": 940,
598
+ "TypeScript": 274,
599
+ "HTML": 161,
600
+ "JavaScript": 144,
601
+ "Jupyter Notebook": 143
602
+ },
603
+ "top_repos": [
604
+ {
605
+ "full_name": "dair-ai/Prompt-Engineering-Guide",
606
+ "stars": 80027,
607
+ "created": "2022-12-16"
608
+ },
609
+ {
610
+ "full_name": "gsd-build/get-shit-done",
611
+ "stars": 64144,
612
+ "created": "2025-12-14"
613
+ },
614
+ {
615
+ "full_name": "muratcankoylan/Agent-Skills-for-Context-Engineering",
616
+ "stars": 18345,
617
+ "created": "2025-12-21"
618
+ },
619
+ {
620
+ "full_name": "coleam00/context-engineering-intro",
621
+ "stars": 13914,
622
+ "created": "2025-07-02"
623
+ },
624
+ {
625
+ "full_name": "jasontang-ai/Context-Engineering",
626
+ "stars": 9281,
627
+ "created": "2025-06-29"
628
+ }
629
+ ]
630
+ },
631
+ "harness engineering": {
632
+ "repos": 2279,
633
+ "han_share": 0.114,
634
+ "org_share": 0.069,
635
+ "median_stars": 0.0,
636
+ "share_ge_100_stars": 0.024,
637
+ "top_languages": {
638
+ "Python": 725,
639
+ "TypeScript": 276,
640
+ "Shell": 174,
641
+ "HTML": 170,
642
+ "JavaScript": 153
643
+ },
644
+ "top_repos": [
645
+ {
646
+ "full_name": "walkinglabs/learn-harness-engineering",
647
+ "stars": 24433,
648
+ "created": "2026-03-29"
649
+ },
650
+ {
651
+ "full_name": "liyupi/ai-guide",
652
+ "stars": 22070,
653
+ "created": "2025-02-13"
654
+ },
655
+ {
656
+ "full_name": "deusyu/harness-engineering",
657
+ "stars": 6284,
658
+ "created": "2026-03-23"
659
+ },
660
+ {
661
+ "full_name": "ModelEngine-Group/nexent",
662
+ "stars": 6192,
663
+ "created": "2025-04-28"
664
+ },
665
+ {
666
+ "full_name": "ai-boost/awesome-harness-engineering",
667
+ "stars": 5140,
668
+ "created": "2026-03-29"
669
+ }
670
+ ]
671
+ },
672
+ "loop engineering": {
673
+ "repos": 1387,
674
+ "han_share": 0.087,
675
+ "org_share": 0.05,
676
+ "median_stars": 0.0,
677
+ "share_ge_100_stars": 0.012,
678
+ "top_languages": {
679
+ "Python": 493,
680
+ "JavaScript": 143,
681
+ "TypeScript": 135,
682
+ "HTML": 115,
683
+ "Shell": 111
684
+ },
685
+ "top_repos": [
686
+ {
687
+ "full_name": "HKUDS/DeepCode",
688
+ "stars": 16844,
689
+ "created": "2025-05-14"
690
+ },
691
+ {
692
+ "full_name": "cobusgreyling/loop-engineering",
693
+ "stars": 11763,
694
+ "created": "2026-06-09"
695
+ },
696
+ {
697
+ "full_name": "valkor-ai/loom",
698
+ "stars": 1488,
699
+ "created": "2026-06-09"
700
+ },
701
+ {
702
+ "full_name": "alchaincyf/loop-engineering-orange-book",
703
+ "stars": 1145,
704
+ "created": "2026-06-15"
705
+ },
706
+ {
707
+ "full_name": "dsta022/Loop-Engineering-for-VLA",
708
+ "stars": 845,
709
+ "created": "2026-05-30"
710
+ }
711
+ ]
712
+ }
713
+ },
714
+ "github_cooccurrence": {
715
+ "context engineering": 2625,
716
+ "harness engineering": 2242,
717
+ "loop engineering": 1368,
718
+ "context engineering;harness engineering": 24,
719
+ "harness engineering;loop engineering": 12,
720
+ "context engineering;loop engineering": 6,
721
+ "context engineering;harness engineering;loop engineering": 1
722
+ },
723
+ "looparena": {
724
+ "type2": {
725
+ "no_control": 0.4917,
726
+ "fixed_control": 0.55,
727
+ "controller_mean": 0.5319,
728
+ "controller_best": "Claude Opus 4.8",
729
+ "controller_best_rate": 0.5875,
730
+ "gains_vs_no_control": {
731
+ "Fixed control": {
732
+ "gain_pp": 5.83,
733
+ "ci_pp": [
734
+ -3.08,
735
+ 14.62
736
+ ],
737
+ "ci_pp_conservative": [
738
+ -9.45,
739
+ 20.74
740
+ ],
741
+ "cost_multiple": 3.47
742
+ },
743
+ "Qwen3.6-35B-A3B": {
744
+ "gain_pp": 6.67,
745
+ "ci_pp": [
746
+ -2.25,
747
+ 15.44
748
+ ],
749
+ "ci_pp_conservative": [
750
+ -8.64,
751
+ 21.54
752
+ ],
753
+ "cost_multiple": 2.6
754
+ },
755
+ "Qwen3.7-Plus": {
756
+ "gain_pp": 7.5,
757
+ "ci_pp": [
758
+ -1.42,
759
+ 16.25
760
+ ],
761
+ "ci_pp_conservative": [
762
+ -7.81,
763
+ 22.33
764
+ ],
765
+ "cost_multiple": 3.73
766
+ },
767
+ "DeepSeek-V4-Flash-0731": {
768
+ "gain_pp": 1.67,
769
+ "ci_pp": [
770
+ -7.23,
771
+ 10.52
772
+ ],
773
+ "ci_pp_conservative": [
774
+ -13.52,
775
+ 16.75
776
+ ],
777
+ "cost_multiple": 6.1
778
+ },
779
+ "GLM 5.2": {
780
+ "gain_pp": -8.75,
781
+ "ci_pp": [
782
+ -17.45,
783
+ 0.14
784
+ ],
785
+ "ci_pp_conservative": [
786
+ -23.48,
787
+ 6.54
788
+ ],
789
+ "cost_multiple": 2.18
790
+ },
791
+ "GPT-5.5": {
792
+ "gain_pp": 7.5,
793
+ "ci_pp": [
794
+ -1.42,
795
+ 16.25
796
+ ],
797
+ "ci_pp_conservative": [
798
+ -7.81,
799
+ 22.33
800
+ ],
801
+ "cost_multiple": 8.52
802
+ },
803
+ "Claude Opus 4.8": {
804
+ "gain_pp": 9.58,
805
+ "ci_pp": [
806
+ 0.67,
807
+ 18.29
808
+ ],
809
+ "ci_pp_conservative": [
810
+ -5.75,
811
+ 24.3
812
+ ],
813
+ "cost_multiple": 6.71
814
+ }
815
+ },
816
+ "no_control_cost_per_success": 3.3,
817
+ "cost_per_success": {
818
+ "No control": 3.3,
819
+ "Fixed control": 10.22,
820
+ "Qwen3.6-35B-A3B": 7.55,
821
+ "Qwen3.7-Plus": 10.65,
822
+ "DeepSeek-V4-Flash-0731": 19.45,
823
+ "GLM 5.2": 8.76,
824
+ "GPT-5.5": 24.35,
825
+ "Claude Opus 4.8": 18.51
826
+ }
827
+ },
828
+ "type3": {
829
+ "no_control": 0.275,
830
+ "fixed_control": 0.3125,
831
+ "controller_mean": 0.3174,
832
+ "controller_best": "GPT-5.5",
833
+ "controller_best_rate": 0.3625,
834
+ "gains_vs_no_control": {
835
+ "Fixed control": {
836
+ "gain_pp": 3.75,
837
+ "ci_pp": [
838
+ -4.39,
839
+ 11.82
840
+ ],
841
+ "ci_pp_conservative": [
842
+ -10.25,
843
+ 17.56
844
+ ],
845
+ "cost_multiple": 3.62
846
+ },
847
+ "Qwen3.6-35B-A3B": {
848
+ "gain_pp": 6.25,
849
+ "ci_pp": [
850
+ -1.99,
851
+ 14.38
852
+ ],
853
+ "ci_pp_conservative": [
854
+ -7.95,
855
+ 20.12
856
+ ],
857
+ "cost_multiple": 4.43
858
+ },
859
+ "Qwen3.7-Plus": {
860
+ "gain_pp": 6.67,
861
+ "ci_pp": [
862
+ -1.59,
863
+ 14.8
864
+ ],
865
+ "ci_pp_conservative": [
866
+ -7.57,
867
+ 20.54
868
+ ],
869
+ "cost_multiple": 4.85
870
+ },
871
+ "DeepSeek-V4-Flash-0731": {
872
+ "gain_pp": 0.83,
873
+ "ci_pp": [
874
+ -7.17,
875
+ 8.82
876
+ ],
877
+ "ci_pp_conservative": [
878
+ -12.92,
879
+ 14.54
880
+ ],
881
+ "cost_multiple": 12.49
882
+ },
883
+ "GLM 5.2": {
884
+ "gain_pp": -3.33,
885
+ "ci_pp": [
886
+ -11.11,
887
+ 4.49
888
+ ],
889
+ "ci_pp_conservative": [
890
+ -16.68,
891
+ 10.17
892
+ ],
893
+ "cost_multiple": 4.58
894
+ },
895
+ "GPT-5.5": {
896
+ "gain_pp": 8.75,
897
+ "ci_pp": [
898
+ 0.41,
899
+ 16.92
900
+ ],
901
+ "ci_pp_conservative": [
902
+ -5.64,
903
+ 22.65
904
+ ],
905
+ "cost_multiple": 13.26
906
+ },
907
+ "Claude Opus 4.8": {
908
+ "gain_pp": 6.25,
909
+ "ci_pp": [
910
+ -1.99,
911
+ 14.38
912
+ ],
913
+ "ci_pp_conservative": [
914
+ -7.95,
915
+ 20.12
916
+ ],
917
+ "cost_multiple": 7.89
918
+ }
919
+ },
920
+ "no_control_cost_per_success": 9.51,
921
+ "cost_per_success": {
922
+ "No control": 9.51,
923
+ "Fixed control": 30.32,
924
+ "Qwen3.6-35B-A3B": 34.31,
925
+ "Qwen3.7-Plus": 37.13,
926
+ "DeepSeek-V4-Flash-0731": 115.25,
927
+ "GLM 5.2": 49.55,
928
+ "GPT-5.5": 95.67,
929
+ "Claude Opus 4.8": 61.12
930
+ }
931
+ },
932
+ "horizon_drop": {
933
+ "No control": {
934
+ "type2": 0.4917,
935
+ "type3": 0.275,
936
+ "relative_drop": 0.441,
937
+ "cost_ratio": 1.61
938
+ },
939
+ "Fixed control": {
940
+ "type2": 0.55,
941
+ "type3": 0.3125,
942
+ "relative_drop": 0.432,
943
+ "cost_ratio": 1.68
944
+ },
945
+ "Qwen3.6-35B-A3B": {
946
+ "type2": 0.5583,
947
+ "type3": 0.3375,
948
+ "relative_drop": 0.396,
949
+ "cost_ratio": 2.75
950
+ },
951
+ "Qwen3.7-Plus": {
952
+ "type2": 0.5667,
953
+ "type3": 0.3417,
954
+ "relative_drop": 0.397,
955
+ "cost_ratio": 2.1
956
+ },
957
+ "DeepSeek-V4-Flash-0731": {
958
+ "type2": 0.5083,
959
+ "type3": 0.2833,
960
+ "relative_drop": 0.443,
961
+ "cost_ratio": 3.3
962
+ },
963
+ "GLM 5.2": {
964
+ "type2": 0.4042,
965
+ "type3": 0.2417,
966
+ "relative_drop": 0.402,
967
+ "cost_ratio": 3.38
968
+ },
969
+ "GPT-5.5": {
970
+ "type2": 0.5667,
971
+ "type3": 0.3625,
972
+ "relative_drop": 0.36,
973
+ "cost_ratio": 2.51
974
+ },
975
+ "Claude Opus 4.8": {
976
+ "type2": 0.5875,
977
+ "type3": 0.3375,
978
+ "relative_drop": 0.426,
979
+ "cost_ratio": 1.9
980
+ }
981
+ },
982
+ "typeI_vs_typeIII_spearman": {
983
+ "rho": 0.696,
984
+ "p": 0.125,
985
+ "n": 6
986
+ },
987
+ "relative_gain_best": {
988
+ "type2": 0.195,
989
+ "type3": 0.318
990
+ }
991
+ },
992
+ "model_e1": {
993
+ "moderate": {
994
+ "n1": 0.36,
995
+ "n20": 0.8,
996
+ "ceiling": 0.8,
997
+ "n20_heterogeneous": 0.682
998
+ },
999
+ "strong": {
1000
+ "n1": 0.38,
1001
+ "n20": 0.969,
1002
+ "ceiling": 0.969,
1003
+ "n20_heterogeneous": 0.861
1004
+ },
1005
+ "weak": {
1006
+ "n1": 0.32,
1007
+ "n20": 0.571,
1008
+ "ceiling": 0.571,
1009
+ "n20_heterogeneous": 0.512
1010
+ }
1011
+ },
1012
+ "model_e2_T300": {
1013
+ "append_kappa0.1": 5.95,
1014
+ "append_kappa1.0": 52.72,
1015
+ "mask_kappa0.1": 3.17,
1016
+ "mask_kappa1.0": 16.21,
1017
+ "compact_kappa0.1": 2.88,
1018
+ "compact_kappa1.0": 20.66,
1019
+ "reset_kappa0.1": 2.72,
1020
+ "reset_kappa1.0": 16.57
1021
+ },
1022
+ "model_e2_T100": {
1023
+ "append_kappa0.1": 1.5,
1024
+ "append_kappa1.0": 12.72,
1025
+ "mask_kappa0.1": 0.92,
1026
+ "mask_kappa1.0": 4.64,
1027
+ "compact_kappa0.1": 0.91,
1028
+ "compact_kappa1.0": 6.36,
1029
+ "reset_kappa0.1": 0.69,
1030
+ "reset_kappa1.0": 4.03
1031
+ },
1032
+ "model_e3_switch_T_by_gamma": {
1033
+ "0.0": 150,
1034
+ "0.5": 100,
1035
+ "1.0": 60,
1036
+ "2.0": 40,
1037
+ "3.0": 40,
1038
+ "4.0": 40
1039
+ },
1040
+ "model_e3_gamma2": {
1041
+ "append": {
1042
+ "10": 0.89,
1043
+ "20": 0.782,
1044
+ "40": 0.568,
1045
+ "60": 0.402,
1046
+ "80": 0.283,
1047
+ "100": 0.192,
1048
+ "150": 0.056,
1049
+ "200": 0.014,
1050
+ "300": 0.0
1051
+ },
1052
+ "compact": {
1053
+ "10": 0.89,
1054
+ "20": 0.782,
1055
+ "40": 0.568,
1056
+ "60": 0.406,
1057
+ "80": 0.286,
1058
+ "100": 0.2,
1059
+ "150": 0.076,
1060
+ "200": 0.024,
1061
+ "300": 0.002
1062
+ },
1063
+ "mask": {
1064
+ "10": 0.89,
1065
+ "20": 0.782,
1066
+ "40": 0.558,
1067
+ "60": 0.38,
1068
+ "80": 0.282,
1069
+ "100": 0.193,
1070
+ "150": 0.095,
1071
+ "200": 0.042,
1072
+ "300": 0.015
1073
+ },
1074
+ "reset": {
1075
+ "10": 0.89,
1076
+ "20": 0.782,
1077
+ "40": 0.594,
1078
+ "60": 0.464,
1079
+ "80": 0.35,
1080
+ "100": 0.28,
1081
+ "150": 0.168,
1082
+ "200": 0.108,
1083
+ "300": 0.045
1084
+ }
1085
+ },
1086
+ "model_e4": {
1087
+ "independent|0.0": {
1088
+ "append": 0.079,
1089
+ "compact": 0.072,
1090
+ "mask": 0.069,
1091
+ "reset": 0.124
1092
+ },
1093
+ "independent|0.25": {
1094
+ "append": 0.118,
1095
+ "compact": 0.104,
1096
+ "mask": 0.105,
1097
+ "reset": 0.176
1098
+ },
1099
+ "independent|0.5": {
1100
+ "append": 0.191,
1101
+ "compact": 0.201,
1102
+ "mask": 0.213,
1103
+ "reset": 0.279
1104
+ },
1105
+ "independent|0.75": {
1106
+ "append": 0.368,
1107
+ "compact": 0.374,
1108
+ "mask": 0.411,
1109
+ "reset": 0.483
1110
+ },
1111
+ "independent|0.9": {
1112
+ "append": 0.626,
1113
+ "compact": 0.655,
1114
+ "mask": 0.657,
1115
+ "reset": 0.725
1116
+ },
1117
+ "self-critique|0.0": {
1118
+ "append": 0.079,
1119
+ "compact": 0.072,
1120
+ "mask": 0.069,
1121
+ "reset": 0.124
1122
+ },
1123
+ "self-critique|0.25": {
1124
+ "append": 0.12,
1125
+ "compact": 0.09,
1126
+ "mask": 0.09,
1127
+ "reset": 0.166
1128
+ },
1129
+ "self-critique|0.5": {
1130
+ "append": 0.187,
1131
+ "compact": 0.153,
1132
+ "mask": 0.135,
1133
+ "reset": 0.252
1134
+ },
1135
+ "self-critique|0.75": {
1136
+ "append": 0.336,
1137
+ "compact": 0.242,
1138
+ "mask": 0.213,
1139
+ "reset": 0.396
1140
+ },
1141
+ "self-critique|0.9": {
1142
+ "append": 0.533,
1143
+ "compact": 0.351,
1144
+ "mask": 0.32,
1145
+ "reset": 0.557
1146
+ }
1147
+ },
1148
+ "model_e4_logodds_gap_reset_minus_append": {
1149
+ "0.0": 0.496,
1150
+ "0.25": 0.467,
1151
+ "0.5": 0.496,
1152
+ "0.75": 0.471,
1153
+ "0.9": 0.454
1154
+ },
1155
+ "model_e4_selfcritique_penalty_d09": {
1156
+ "append": 0.093,
1157
+ "compact": 0.304,
1158
+ "mask": 0.337,
1159
+ "reset": 0.167
1160
+ },
1161
+ "model_e5": {
1162
+ "eps0.005|gamma0.5|d0.5": {
1163
+ "reset_minus_append": -0.038,
1164
+ "best": "append"
1165
+ },
1166
+ "eps0.005|gamma0.5|d0.9": {
1167
+ "reset_minus_append": -0.022,
1168
+ "best": "append"
1169
+ },
1170
+ "eps0.005|gamma2.0|d0.5": {
1171
+ "reset_minus_append": 0.019,
1172
+ "best": "reset"
1173
+ },
1174
+ "eps0.005|gamma2.0|d0.9": {
1175
+ "reset_minus_append": 0.007,
1176
+ "best": "reset"
1177
+ },
1178
+ "eps0.01|gamma0.5|d0.5": {
1179
+ "reset_minus_append": -0.022,
1180
+ "best": "append"
1181
+ },
1182
+ "eps0.01|gamma0.5|d0.9": {
1183
+ "reset_minus_append": -0.008,
1184
+ "best": "append"
1185
+ },
1186
+ "eps0.01|gamma2.0|d0.5": {
1187
+ "reset_minus_append": 0.073,
1188
+ "best": "reset"
1189
+ },
1190
+ "eps0.01|gamma2.0|d0.9": {
1191
+ "reset_minus_append": 0.038,
1192
+ "best": "reset"
1193
+ },
1194
+ "eps0.02|gamma0.5|d0.5": {
1195
+ "reset_minus_append": 0.0,
1196
+ "best": "append"
1197
+ },
1198
+ "eps0.02|gamma0.5|d0.9": {
1199
+ "reset_minus_append": 0.011,
1200
+ "best": "reset"
1201
+ },
1202
+ "eps0.02|gamma2.0|d0.5": {
1203
+ "reset_minus_append": 0.083,
1204
+ "best": "reset"
1205
+ },
1206
+ "eps0.02|gamma2.0|d0.9": {
1207
+ "reset_minus_append": 0.095,
1208
+ "best": "reset"
1209
+ },
1210
+ "eps0.04|gamma0.5|d0.5": {
1211
+ "reset_minus_append": 0.027,
1212
+ "best": "reset"
1213
+ },
1214
+ "eps0.04|gamma0.5|d0.9": {
1215
+ "reset_minus_append": 0.033,
1216
+ "best": "reset"
1217
+ },
1218
+ "eps0.04|gamma2.0|d0.5": {
1219
+ "reset_minus_append": 0.06,
1220
+ "best": "reset"
1221
+ },
1222
+ "eps0.04|gamma2.0|d0.9": {
1223
+ "reset_minus_append": 0.149,
1224
+ "best": "reset"
1225
+ }
1226
+ },
1227
+ "model_e5_verifier_gain_append": {
1228
+ "eps0.005|gamma0.5": 0.215,
1229
+ "eps0.005|gamma2.0": 0.265,
1230
+ "eps0.01|gamma0.5": 0.308,
1231
+ "eps0.01|gamma2.0": 0.377,
1232
+ "eps0.02|gamma0.5": 0.423,
1233
+ "eps0.02|gamma2.0": 0.434,
1234
+ "eps0.04|gamma0.5": 0.456,
1235
+ "eps0.04|gamma2.0": 0.344
1236
+ },
1237
+ "model_e6_best_success_share": {
1238
+ "30": {
1239
+ "append": 0.707,
1240
+ "compact": 0.103,
1241
+ "mask": 0.05,
1242
+ "reset": 0.14
1243
+ },
1244
+ "100": {
1245
+ "append": 0.207,
1246
+ "compact": 0.15,
1247
+ "mask": 0.247,
1248
+ "reset": 0.397
1249
+ }
1250
+ },
1251
+ "model_e6_best_cost_share": {
1252
+ "30": {
1253
+ "append": 0.443,
1254
+ "compact": 0.143,
1255
+ "mask": 0.0,
1256
+ "reset": 0.413
1257
+ },
1258
+ "100": {
1259
+ "append": 0.02,
1260
+ "compact": 0.168,
1261
+ "mask": 0.077,
1262
+ "reset": 0.734
1263
+ }
1264
+ },
1265
+ "model_e6_retention_grid_best": {
1266
+ "0.02": {
1267
+ "0.02": "reset",
1268
+ "0.05": "reset",
1269
+ "0.1": "reset",
1270
+ "0.2": "reset",
1271
+ "0.4": "reset"
1272
+ },
1273
+ "0.05": {
1274
+ "0.02": "reset",
1275
+ "0.05": "reset",
1276
+ "0.1": "reset",
1277
+ "0.2": "reset",
1278
+ "0.4": "reset"
1279
+ },
1280
+ "0.1": {
1281
+ "0.02": "reset",
1282
+ "0.05": "reset",
1283
+ "0.1": "reset",
1284
+ "0.2": "reset",
1285
+ "0.4": "reset"
1286
+ },
1287
+ "0.2": {
1288
+ "0.02": "compact",
1289
+ "0.05": "compact",
1290
+ "0.1": "compact",
1291
+ "0.2": "reset",
1292
+ "0.4": "reset"
1293
+ },
1294
+ "0.4": {
1295
+ "0.02": "compact",
1296
+ "0.05": "compact",
1297
+ "0.1": "compact",
1298
+ "0.2": "mask",
1299
+ "0.4": "mask"
1300
+ }
1301
+ },
1302
+ "model_e6_tree_T100": {
1303
+ "text": "|--- lam_r <= 0.17\n| |--- s <= 1764.83\n| | |--- class: append\n| |--- s > 1764.83\n| | |--- delta_mask <= 0.13\n| | | |--- class: mask\n| | |--- delta_mask > 0.13\n| | | |--- class: reset\n|--- lam_r > 0.17\n| |--- delta_mask <= 0.27\n| | |--- s <= 2387.65\n| | | |--- class: append\n| | |--- s > 2387.65\n| | | |--- class: mask\n| |--- delta_mask > 0.27\n| | |--- lam_c <= 0.11\n| | | |--- class: compact\n| | |--- lam_c > 0.11\n| | | |--- class: reset\n",
1304
+ "train_accuracy": 0.64,
1305
+ "importances": {
1306
+ "gamma": 0.0,
1307
+ "eps_mean": 0.0,
1308
+ "rho": 0.0,
1309
+ "delta_mask": 0.287,
1310
+ "lam_c": 0.19,
1311
+ "lam_r": 0.297,
1312
+ "d": 0.0,
1313
+ "k": 0.0,
1314
+ "H": 0.0,
1315
+ "s": 0.226
1316
+ }
1317
+ },
1318
+ "model_e6_lowest_loss_rule_accuracy": 0.655,
1319
+ "model_e6_lowest_loss_rule_n": 238,
1320
+ "audit": {
1321
+ "round1_v1_layer_kappa": 0.663,
1322
+ "round1_v1_contribution_kappa": 0.781,
1323
+ "round2_v2_layer_kappa": 0.856,
1324
+ "round2_v2_layer_agreement": 0.9,
1325
+ "round2_v2_contribution_kappa": 0.764,
1326
+ "pooled_v2_layer_kappa": 0.859,
1327
+ "n_round1": 100,
1328
+ "n_round2": 60
1329
+ }
1330
+ }
simulation/model_e1_verifier.csv ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ verifier,f_n,f_p,n,p_correct,yield_,precision,p_correct_heterogeneous
2
+ strong,0.05,0.02,1,0.38,0.392,0.9693877551020408,0.37922914733606405
3
+ strong,0.05,0.02,2,0.6110399999999999,0.630336,0.9693877551020408,0.5396605684591286
4
+ strong,0.05,0.02,3,0.7515123199999999,0.775244288,0.9693877551020408,0.6272422043609486
5
+ strong,0.05,0.02,4,0.8369194905599999,0.863348527104,0.9693877551020408,0.6824322172017702
6
+ strong,0.05,0.02,5,0.8888470502604799,0.916915904479232,0.9693877551020408,0.7203940521715564
7
+ strong,0.05,0.02,6,0.9204190065583718,0.9494848699233731,0.9693877551020408,0.7480765640118906
8
+ strong,0.05,0.02,7,0.93961475598749,0.9692868009134108,0.9693877551020408,0.7691228997040812
9
+ strong,0.05,0.02,8,0.951285771640394,0.9813263749553538,0.9693877551020408,0.7856325610248954
10
+ strong,0.05,0.02,9,0.9583817491573595,0.9886464359728551,0.9693877551020408,0.798902968576435
11
+ strong,0.05,0.02,10,0.9626961034876745,0.9930970330714959,0.9693877551020408,0.8097800781190948
12
+ strong,0.05,0.02,11,0.9653192309205061,0.9958029961074695,0.9693877551020408,0.8188394518997456
13
+ strong,0.05,0.02,12,0.9669140923996676,0.9974482216333415,0.9693877551020408,0.8264864660779445
14
+ strong,0.05,0.02,13,0.967883768178998,0.9984485187530716,0.9693877551020408,0.8330149416842672
15
+ strong,0.05,0.02,14,0.9684733310528308,0.9990566994018676,0.9693877551020408,0.8386430577166983
16
+ strong,0.05,0.02,15,0.9688317852801211,0.9994264732363355,0.9693877551020408,0.843536210048346
17
+ strong,0.05,0.02,16,0.9690497254503136,0.999651295727692,0.9693877551020408,0.8478220443989654
18
+ strong,0.05,0.02,17,0.9691822330737907,0.9997879878024367,0.9693877551020408,0.8516006247796389
19
+ strong,0.05,0.02,18,0.9692627977088647,0.9998710965838815,0.9693877551020408,0.8549514820998355
20
+ strong,0.05,0.02,19,0.9693117810069898,0.999921626723,0.9693877551020408,0.8579386064601751
21
+ strong,0.05,0.02,20,0.9693415628522497,0.999952349047584,0.9693877551020408,0.8606140510859875
22
+ moderate,0.1,0.15,1,0.36000000000000004,0.45000000000000007,0.7999999999999999,0.3592697185289028
23
+ moderate,0.1,0.15,2,0.558,0.6975,0.7999999999999999,0.5032807034383595
24
+ moderate,0.1,0.15,3,0.6669,0.8336250000000001,0.7999999999999999,0.5731516783924651
25
+ moderate,0.1,0.15,4,0.7267950000000001,0.9084937500000001,0.7999999999999999,0.6116215152594321
26
+ moderate,0.1,0.15,5,0.75973725,0.9496715625000001,0.7999999999999999,0.6346412379542069
27
+ moderate,0.1,0.15,6,0.7778554875,0.972319359375,0.7999999999999999,0.6492253366665817
28
+ moderate,0.1,0.15,7,0.787820518125,0.98477564765625,0.7999999999999999,0.6588533752036455
29
+ moderate,0.1,0.15,8,0.79330128496875,0.9916266062109375,0.7999999999999999,0.6654097086481654
30
+ moderate,0.1,0.15,9,0.7963157067328124,0.9953946334160156,0.7999999999999999,0.669983625218776
31
+ moderate,0.1,0.15,10,0.7979736387030468,0.9974670483788086,0.7999999999999999,0.6732370573132703
32
+ moderate,0.1,0.15,11,0.7988855012866757,0.9986068766083447,0.7999999999999999,0.6755883577686487
33
+ moderate,0.1,0.15,12,0.7993870257076716,0.9992337821345896,0.7999999999999999,0.6773104266136725
34
+ moderate,0.1,0.15,13,0.7996628641392193,0.9995785801740242,0.7999999999999999,0.6785859628559515
35
+ moderate,0.1,0.15,14,0.7998145752765706,0.9997682190957133,0.7999999999999999,0.6795399560143333
36
+ moderate,0.1,0.15,15,0.7998980164021139,0.9998725205026423,0.7999999999999999,0.6802594958835628
37
+ moderate,0.1,0.15,16,0.7999439090211626,0.9999298862764533,0.7999999999999999,0.6808062237695293
38
+ moderate,0.1,0.15,17,0.7999691499616394,0.9999614374520494,0.7999999999999999,0.6812243645606199
39
+ moderate,0.1,0.15,18,0.7999830324789017,0.9999787905986272,0.7999999999999999,0.6815460253943291
40
+ moderate,0.1,0.15,19,0.7999906678633959,0.9999883348292449,0.7999999999999999,0.6817947598512817
41
+ moderate,0.1,0.15,20,0.7999948673248678,0.9999935841560847,0.7999999999999999,0.6819880067008564
42
+ weak,0.2,0.4,1,0.32000000000000006,0.56,0.5714285714285715,0.3193508609145803
43
+ weak,0.2,0.4,2,0.46080000000000004,0.8064,0.5714285714285715,0.43446086150949814
44
+ weak,0.2,0.4,3,0.5227520000000001,0.9148160000000001,0.5714285714285715,0.47902301263465574
45
+ weak,0.2,0.4,4,0.5500108800000001,0.96251904,0.5714285714285715,0.4973970376595722
46
+ weak,0.2,0.4,5,0.5620047872,0.9835083776,0.5714285714285715,0.5053854266378555
47
+ weak,0.2,0.4,6,0.567282106368,0.992743686144,0.5714285714285715,0.5090117889893732
48
+ weak,0.2,0.4,7,0.5696041268019201,0.99680722190336,0.5714285714285715,0.5107161400598187
49
+ weak,0.2,0.4,8,0.5706258157928449,0.9985951776374784,0.5714285714285715,0.511539810834581
50
+ weak,0.2,0.4,9,0.5710753589488518,0.9993818781604905,0.5714285714285715,0.5119469506606198
51
+ weak,0.2,0.4,10,0.5712731579374948,0.9997280263906159,0.5714285714285715,0.5121519504210691
52
+ weak,0.2,0.4,11,0.5713601894924978,0.999880331611871,0.5714285714285715,0.5122567642241219
53
+ weak,0.2,0.4,12,0.571398483376699,0.9999473459092232,0.5714285714285715,0.5123110494273051
54
+ weak,0.2,0.4,13,0.5714153326857475,0.9999768322000582,0.5714285714285715,0.5123394752141956
55
+ weak,0.2,0.4,14,0.571422746381729,0.9999898061680256,0.5714285714285715,0.5123545014875831
56
+ weak,0.2,0.4,15,0.5714260084079608,0.9999955147139312,0.5714285714285715,0.5123625102443372
57
+ weak,0.2,0.4,16,0.5714274436995027,0.9999980264741297,0.5714285714285715,0.5123668097488109
58
+ weak,0.2,0.4,17,0.5714280752277812,0.9999991316486171,0.5714285714285715,0.5123691327570494
59
+ weak,0.2,0.4,18,0.5714283531002238,0.9999996179253915,0.5714285714285715,0.5123703950510765
60
+ weak,0.2,0.4,19,0.5714284753640985,0.9999998318871722,0.5714285714285715,0.5123710844852685
61
+ weak,0.2,0.4,20,0.5714285291602034,0.9999999260303558,0.5714285714285715,0.5123714627799314
simulation/model_e2_cost.csv ADDED
@@ -0,0 +1,73 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ T,policy,kappa,priced_tokens,raw_tokens,peak_load
2
+ 10,append,0.1,49200.0,217500.0,33000.0
3
+ 10,append,1.0,217500.0,217500.0,33000.0
4
+ 10,mask,0.1,49200.0,217500.0,33000.0
5
+ 10,mask,1.0,217500.0,217500.0,33000.0
6
+ 10,compact,0.1,51450.0,217500.0,33000.0
7
+ 10,compact,1.0,217500.0,217500.0,33000.0
8
+ 10,reset,0.1,51450.0,217500.0,33000.0
9
+ 10,reset,1.0,217500.0,217500.0,33000.0
10
+ 20,append,0.1,118450.0,685000.0,58000.0
11
+ 20,append,1.0,685000.0,685000.0,58000.0
12
+ 20,mask,0.1,140950.0,685000.0,58000.0
13
+ 20,mask,1.0,685000.0,685000.0,58000.0
14
+ 20,compact,0.1,120700.0,685000.0,58000.0
15
+ 20,compact,1.0,685000.0,685000.0,58000.0
16
+ 20,reset,0.1,120700.0,685000.0,58000.0
17
+ 20,reset,1.0,685000.0,685000.0,58000.0
18
+ 40,append,0.1,331950.0,2370000.0,108000.0
19
+ 40,append,1.0,2370000.0,2370000.0,108000.0
20
+ 40,mask,0.1,327870.0,1638000.0,59200.0
21
+ 40,mask,1.0,1638000.0,1638000.0,59200.0
22
+ 40,compact,0.1,334200.0,2370000.0,108000.0
23
+ 40,compact,1.0,2370000.0,2370000.0,108000.0
24
+ 40,reset,0.1,250100.0,1430000.0,61000.0
25
+ 40,reset,1.0,1430000.0,1430000.0,61000.0
26
+ 60,append,0.1,645450.0,5055000.0,158000.0
27
+ 60,append,1.0,5055000.0,5055000.0,158000.0
28
+ 60,mask,0.1,519350.0,2615000.0,60400.0
29
+ 60,mask,1.0,2615000.0,2615000.0,60400.0
30
+ 60,compact,0.1,512650.0,3511000.0,118000.0
31
+ 60,compact,1.0,3511000.0,3511000.0,118000.0
32
+ 60,reset,0.1,388200.0,2235000.0,64000.0
33
+ 60,reset,1.0,2235000.0,2235000.0,64000.0
34
+ 80,append,0.1,1057250.0,8723000.0,200000.0
35
+ 80,append,1.0,8723000.0,8723000.0,200000.0
36
+ 80,mask,0.1,715390.0,3616000.0,61600.0
37
+ 80,mask,1.0,3616000.0,3616000.0,61600.0
38
+ 80,compact,0.1,733150.0,5266000.0,118000.0
39
+ 80,compact,1.0,5266000.0,5266000.0,118000.0
40
+ 80,reset,0.1,535000.0,3100000.0,67000.0
41
+ 80,reset,1.0,3100000.0,3100000.0,67000.0
42
+ 100,append,0.1,1502250.0,12723000.0,200000.0
43
+ 100,append,1.0,12723000.0,12723000.0,200000.0
44
+ 100,mask,0.1,915990.0,4641000.0,62800.0
45
+ 100,mask,1.0,4641000.0,4641000.0,62800.0
46
+ 100,compact,0.1,907250.0,6363500.0,119000.0
47
+ 100,compact,1.0,6363500.0,6363500.0,119000.0
48
+ 100,reset,0.1,690500.0,4025000.0,70000.0
49
+ 100,reset,1.0,4025000.0,4025000.0,70000.0
50
+ 150,append,0.1,2614750.0,22723000.0,200000.0
51
+ 150,append,1.0,22723000.0,22723000.0,200000.0
52
+ 150,mask,0.1,1437440.0,7308500.0,65800.0
53
+ 150,mask,1.0,7308500.0,7308500.0,65800.0
54
+ 150,compact,0.1,1404850.0,10021000.0,119000.0
55
+ 150,compact,1.0,10021000.0,10021000.0,119000.0
56
+ 150,reset,0.1,1118950.0,6482500.0,76000.0
57
+ 150,reset,1.0,6482500.0,6482500.0,76000.0
58
+ 200,append,0.1,3727250.0,32723000.0,200000.0
59
+ 200,append,1.0,32723000.0,32723000.0,200000.0
60
+ 200,mask,0.1,1987390.0,10126000.0,68800.0
61
+ 200,mask,1.0,10126000.0,10126000.0,68800.0
62
+ 200,compact,0.1,1932700.0,13981000.0,119000.0
63
+ 200,compact,1.0,13981000.0,13981000.0,119000.0
64
+ 200,reset,0.1,1598500.0,9550000.0,85000.0
65
+ 200,reset,1.0,9550000.0,9550000.0,85000.0
66
+ 300,append,0.1,5952250.0,52723000.0,200000.0
67
+ 300,append,1.0,52723000.0,52723000.0,200000.0
68
+ 300,mask,0.1,3172790.0,16211000.0,74800.0
69
+ 300,mask,1.0,16211000.0,16211000.0,74800.0
70
+ 300,compact,0.1,2884000.0,20663500.0,119000.0
71
+ 300,compact,1.0,20663500.0,20663500.0,119000.0
72
+ 300,reset,0.1,2724000.0,16575000.0,100000.0
73
+ 300,reset,1.0,16575000.0,16575000.0,100000.0
simulation/model_e3_phase.csv ADDED
@@ -0,0 +1,217 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ T,gamma,policy,success,se,cost,cost_per_success
2
+ 10,0.0,append,0.9133333333333333,0.005136651164444126,47670.65,52194.142335766424
3
+ 10,0.0,mask,0.9133333333333333,0.005136651164444126,47670.65,52194.142335766424
4
+ 10,0.0,compact,0.9133333333333333,0.005136651164444126,49920.65,54657.64598540146
5
+ 10,0.0,reset,0.9133333333333333,0.005136651164444126,49920.65,54657.64598540146
6
+ 10,0.5,append,0.9073333333333333,0.005294007793583092,47648.916666666664,52515.33798677443
7
+ 10,0.5,mask,0.9073333333333333,0.005294007793583092,47648.916666666664,52515.33798677443
8
+ 10,0.5,compact,0.9073333333333333,0.005294007793583092,49898.916666666664,54995.13225569434
9
+ 10,0.5,reset,0.9073333333333333,0.005294007793583092,49898.916666666664,54995.13225569434
10
+ 10,1.0,append,0.8983333333333333,0.005517564545931341,47378.433333333334,52740.37105751391
11
+ 10,1.0,mask,0.8983333333333333,0.005517564545931341,47378.433333333334,52740.37105751391
12
+ 10,1.0,compact,0.8983333333333333,0.005517564545931341,49628.433333333334,55245.00927643785
13
+ 10,1.0,reset,0.8983333333333333,0.005517564545931341,49628.433333333334,55245.00927643785
14
+ 10,2.0,append,0.8903333333333333,0.005704965348679134,47342.65,53174.07338075627
15
+ 10,2.0,mask,0.8903333333333333,0.005704965348679134,47342.65,53174.07338075627
16
+ 10,2.0,compact,0.8903333333333333,0.005704965348679134,49592.65,55701.216772744294
17
+ 10,2.0,reset,0.8903333333333333,0.005704965348679134,49592.65,55701.216772744294
18
+ 10,3.0,append,0.8796666666666667,0.005940067963197677,47054.01666666667,53490.735126942025
19
+ 10,3.0,mask,0.8796666666666667,0.005940067963197677,47054.01666666667,53490.735126942025
20
+ 10,3.0,compact,0.8796666666666667,0.005940067963197677,49304.01666666667,56048.52216748768
21
+ 10,3.0,reset,0.8796666666666667,0.005940067963197677,49304.01666666667,56048.52216748768
22
+ 10,4.0,append,0.871,0.006119885619846175,47023.46666666667,53987.90662074244
23
+ 10,4.0,mask,0.871,0.006119885619846175,47023.46666666667,53987.90662074244
24
+ 10,4.0,compact,0.871,0.006119885619846175,49273.46666666667,56571.14427860697
25
+ 10,4.0,reset,0.871,0.006119885619846175,49273.46666666667,56571.14427860697
26
+ 20,0.0,append,0.827,0.006905818802526848,108184.35,130815.41717049577
27
+ 20,0.0,mask,0.827,0.006905818802526848,128576.85,155473.82103990327
28
+ 20,0.0,compact,0.827,0.006905818802526848,110434.35,133536.09431680775
29
+ 20,0.0,reset,0.827,0.006905818802526848,110434.35,133536.09431680775
30
+ 20,0.5,append,0.8206666666666667,0.007004115192487174,107652.81666666667,131177.27457351747
31
+ 20,0.5,mask,0.8206666666666667,0.007004115192487174,127880.31666666667,155824.91876523153
32
+ 20,0.5,compact,0.8206666666666667,0.007004115192487174,109902.81666666667,133918.9480097482
33
+ 20,0.5,reset,0.8206666666666667,0.007004115192487174,109902.81666666667,133918.9480097482
34
+ 20,1.0,append,0.802,0.007275438131136845,107268.76666666666,133751.57938487115
35
+ 20,1.0,mask,0.802,0.007275438131136845,127496.26666666666,158972.90108063177
36
+ 20,1.0,compact,0.802,0.007275438131136845,109518.76666666666,136557.06566916042
37
+ 20,1.0,reset,0.802,0.007275438131136845,109518.76666666666,136557.06566916042
38
+ 20,2.0,append,0.7816666666666666,0.007542410951609769,105891.8,135469.25373134328
39
+ 20,2.0,mask,0.7816666666666666,0.007542410951609769,125819.3,160962.85714285713
40
+ 20,2.0,compact,0.7816666666666666,0.007542410951609769,108141.8,138347.7185501066
41
+ 20,2.0,reset,0.7816666666666666,0.007542410951609769,108141.8,138347.7185501066
42
+ 20,3.0,append,0.7566666666666667,0.007834160712806067,104036.56666666667,137493.25991189427
43
+ 20,3.0,mask,0.7566666666666667,0.007834160712806067,123506.56666666667,163224.53744493393
44
+ 20,3.0,compact,0.7566666666666667,0.007834160712806067,106286.56666666667,140466.8281938326
45
+ 20,3.0,reset,0.7566666666666667,0.007834160712806067,106286.56666666667,140466.8281938326
46
+ 20,4.0,append,0.7306666666666667,0.008099236360894133,103601.5,141790.37408759125
47
+ 20,4.0,mask,0.7306666666666667,0.008099236360894133,123101.5,168478.3302919708
48
+ 20,4.0,compact,0.7306666666666667,0.008099236360894133,105851.5,144869.75364963504
49
+ 20,4.0,reset,0.7306666666666667,0.008099236360894133,105851.5,144869.75364963504
50
+ 40,0.0,append,0.6926666666666667,0.008423767082003862,269569.95,389177.021174206
51
+ 40,0.0,mask,0.64,0.008763560920082658,268019.08,418779.8125
52
+ 40,0.0,compact,0.6926666666666667,0.008423767082003862,271819.95,392425.33686236764
53
+ 40,0.0,reset,0.6633333333333333,0.00862790477119102,205701.03333333333,310102.06030150753
54
+ 40,0.5,append,0.6656666666666666,0.008613062861250454,265707.31666666665,399159.71457185777
55
+ 40,0.5,mask,0.6213333333333333,0.008855849709320619,266327.0466666667,428637.9506437768
56
+ 40,0.5,compact,0.6656666666666666,0.008613062861250454,267957.31666666665,402539.7846770155
57
+ 40,0.5,reset,0.645,0.008736418030291362,203601.31666666668,315660.95607235143
58
+ 40,1.0,append,0.6226666666666667,0.00884972483355441,259891.18333333332,417384.1274089936
59
+ 40,1.0,mask,0.5873333333333334,0.008988379329053874,257472.09,438374.72758229286
60
+ 40,1.0,compact,0.6226666666666667,0.00884972483355441,262141.18333333332,420997.61777301924
61
+ 40,1.0,reset,0.622,0.008852796168443052,199578.61666666667,320865.94319399784
62
+ 40,2.0,append,0.568,0.009043892967080051,249046.98333333334,438462.99882629106
63
+ 40,2.0,mask,0.558,0.00906708332375963,253734.01666666666,454720.45997610514
64
+ 40,2.0,compact,0.568,0.009043892967080051,251296.98333333334,442424.2664319249
65
+ 40,2.0,reset,0.594,0.008965935534008707,196080.5,330101.85185185185
66
+ 40,3.0,append,0.5206666666666667,0.009120908011977917,240933.35,462740.10883482714
67
+ 40,3.0,mask,0.514,0.009125130136058335,244536.57,475752.08171206224
68
+ 40,3.0,compact,0.5206666666666667,0.009120908011977917,243183.35,467061.49167733674
69
+ 40,3.0,reset,0.561,0.00906051874894589,192722.35,343533.60071301245
70
+ 40,4.0,append,0.4736666666666667,0.009116039994711804,230693.83333333334,487038.3532723434
71
+ 40,4.0,mask,0.49,0.00912688336728371,242350.57333333333,494593.00680272107
72
+ 40,4.0,compact,0.4736666666666667,0.009116039994711804,232943.83333333334,491788.52920478536
73
+ 40,4.0,reset,0.526,0.009116358922289096,189077.65,359463.21292775666
74
+ 60,0.0,append,0.6006666666666667,0.008941778760817526,480357.6666666667,799707.5471698113
75
+ 60,0.0,mask,0.47633333333333333,0.009118477374519807,377071.3,791612.2463261022
76
+ 60,0.0,compact,0.5413333333333333,0.00909746403410598,389544.63333333336,719602.1551724138
77
+ 60,0.0,reset,0.543,0.009094888674414877,291737.15,537269.152854512
78
+ 60,0.5,append,0.529,0.009113341867833117,454286.7833333333,858765.1858853182
79
+ 60,0.5,mask,0.4563333333333333,0.009093829817010034,371834.75,814831.446311176
80
+ 60,0.5,compact,0.511,0.00912649987673259,378214.88333333336,740146.5427266797
81
+ 60,0.5,reset,0.5116666666666667,0.00912622391589002,285047.4666666667,557096.026058632
82
+ 60,1.0,append,0.484,0.009124034195464197,435092.56666666665,898951.5840220386
83
+ 60,1.0,mask,0.431,0.009041367890608883,359362.05333333334,833786.6666666666
84
+ 60,1.0,compact,0.47633333333333333,0.009118477374519807,366727.13333333336,769896.011196641
85
+ 60,1.0,reset,0.487,0.0091256232663857,278921.11666666664,572733.2991101985
86
+ 60,2.0,append,0.40166666666666667,0.008950429329657053,406440.01666666666,1011883.8589211619
87
+ 60,2.0,mask,0.37966666666666665,0.008860396697080195,343398.79,904474.4249341527
88
+ 60,2.0,compact,0.406,0.008965935534008707,343313.45,845599.6305418719
89
+ 60,2.0,reset,0.464,0.009105016932072853,270458.93333333335,582885.632183908
90
+ 60,3.0,append,0.3426666666666667,0.00866499129105587,377204.43333333335,1100791.1478599221
91
+ 60,3.0,mask,0.364,0.008784531860036708,332037.36666666664,912190.5677655678
92
+ 60,3.0,compact,0.3616666666666667,0.008772378789680119,323065.45,893268.5253456221
93
+ 60,3.0,reset,0.414,0.008992663676575478,259838.35,627628.8647342995
94
+ 60,4.0,append,0.2906666666666667,0.008290145868349877,349663.8333333333,1202971.9036697247
95
+ 60,4.0,mask,0.328,0.008571580951026479,320328.11,976610.0914634146
96
+ 60,4.0,compact,0.3406666666666667,0.00865280087387679,308070.4,904316.2426614482
97
+ 60,4.0,reset,0.3913333333333333,0.00891050981623303,251640.2,643032.8790459966
98
+ 80,0.0,append,0.5186666666666667,0.009122345377433656,709834.3666666667,1368575.2570694087
99
+ 80,0.0,mask,0.356,0.008741929611552207,461496.9733333333,1296339.8127340823
100
+ 80,0.0,compact,0.42,0.009011104260855048,498231.2,1186264.761904762
101
+ 80,0.0,reset,0.44,0.009062744249582096,361183.7166666667,820872.0833333334
102
+ 80,0.5,append,0.43833333333333335,0.009059014336048969,654803.35,1493847.9467680608
103
+ 80,0.5,mask,0.3433333333333333,0.008669016775381845,452203.8,1317098.4466019417
104
+ 80,0.5,compact,0.38333333333333336,0.008876727792170733,475336.95,1240009.4347826086
105
+ 80,0.5,reset,0.42433333333333334,0.009023572011784017,359357.2166666667,846874.8232521602
106
+ 80,1.0,append,0.37366666666666665,0.008832513588797716,616965.7666666667,1651112.6672613737
107
+ 80,1.0,mask,0.31666666666666665,0.008492916438399098,435422.23333333334,1375017.5789473683
108
+ 80,1.0,compact,0.3373333333333333,0.008632102014294386,454319.0333333333,1346795.5533596838
109
+ 80,1.0,reset,0.39966666666666667,0.008943027493010448,346484.31666666665,866933.2360300251
110
+ 80,2.0,append,0.2826666666666667,0.008221237178231798,550378.3,1947093.0424528301
111
+ 80,2.0,mask,0.282,0.00821535148365546,414654.07666666666,1470404.5271867611
112
+ 80,2.0,compact,0.286,0.008250333326599598,420316.7166666667,1469638.8694638694
113
+ 80,2.0,reset,0.35033333333333333,0.008710145212889944,330147.23333333334,942380.3044719315
114
+ 80,3.0,append,0.22433333333333333,0.007615945747550309,490276.38333333336,2185481.649331352
115
+ 80,3.0,mask,0.25366666666666665,0.007943968128689189,393819.6533333333,1552508.4888304863
116
+ 80,3.0,compact,0.25033333333333335,0.007909204670415258,395904.13333333336,1581507.8561917443
117
+ 80,3.0,reset,0.32466666666666666,0.0085490393655705,316786.7833333333,975729.3121149897
118
+ 80,4.0,append,0.17766666666666667,0.006978567718909619,454664.9,2559089.493433396
119
+ 80,4.0,mask,0.226,0.007635967522193897,375376.0933333333,1660956.1651917405
120
+ 80,4.0,compact,0.212,0.007462260604758677,366515.5833333333,1728847.0911949684
121
+ 80,4.0,reset,0.291,0.00829294881209332,303302.93333333335,1042278.1214203895
122
+ 100,0.0,append,0.398,0.008936740643732106,904989.3166666667,2273842.504187605
123
+ 100,0.0,mask,0.271,0.008114986136771894,527474.79,1946401.439114391
124
+ 100,0.0,compact,0.288,0.00826752683696884,553930.8833333333,1923371.1226851852
125
+ 100,0.0,reset,0.371,0.008819656077950735,429274.4166666667,1157073.8993710692
126
+ 100,0.5,append,0.3263333333333333,0.008560371660328945,830053.5666666667,2543575.791624106
127
+ 100,0.5,mask,0.25466666666666665,0.007954276743898123,514056.9666666667,2018548.2984293194
128
+ 100,0.5,compact,0.26066666666666666,0.008014976721853398,533474.3,2046576.5984654732
129
+ 100,0.5,reset,0.33866666666666667,0.008640438946325886,412524.5833333333,1218084.3996062991
130
+ 100,1.0,append,0.27066666666666667,0.008111847759547394,768399.2333333333,2838913.4236453203
131
+ 100,1.0,mask,0.23366666666666666,0.007725856059483107,493683.24333333335,2112767.089871612
132
+ 100,1.0,compact,0.23833333333333334,0.007778829293999528,515622.95,2163452.937062937
133
+ 100,1.0,reset,0.318,0.008502470229292191,406453.73333333334,1278156.394129979
134
+ 100,2.0,append,0.192,0.007191105617358155,659483.45,3434809.6354166665
135
+ 100,2.0,mask,0.19333333333333333,0.00721007525516795,461702.7366666667,2388117.603448276
136
+ 100,2.0,compact,0.2,0.007302967433402216,466298.86666666664,2331494.3333333335
137
+ 100,2.0,reset,0.2803333333333333,0.008200539729301471,376997.05,1344817.063020214
138
+ 100,3.0,append,0.14333333333333334,0.006397626874836452,582721.5,4065498.8372093025
139
+ 100,3.0,mask,0.169,0.006842002630809199,434311.32666666666,2569889.506903353
140
+ 100,3.0,compact,0.17066666666666666,0.006868759120237938,434573.4666666667,2546328.90625
141
+ 100,3.0,reset,0.24733333333333332,0.00787738441268326,360264.0833333333,1456593.3288409703
142
+ 100,4.0,append,0.11066666666666666,0.005727697488390821,512270.45,4628949.84939759
143
+ 100,4.0,mask,0.15,0.006519202405202649,409310.05,2728733.6666666665
144
+ 100,4.0,compact,0.145,0.0064284523798500675,398608.81666666665,2749026.3218390807
145
+ 100,4.0,reset,0.22233333333333333,0.007591688925446086,346251.61666666664,1557353.5982008995
146
+ 150,0.0,append,0.181,0.0070294380998768315,1233634.4666666666,6815660.036832413
147
+ 150,0.0,mask,0.14733333333333334,0.006471121031738427,635674.3733333333,4314531.945701358
148
+ 150,0.0,compact,0.11866666666666667,0.005904373771165714,652974.75,5502596.207865168
149
+ 150,0.0,reset,0.24533333333333332,0.007855887153145912,556905.1333333333,2269993.75
150
+ 150,0.5,append,0.135,0.006238990302925627,1068851.55,7917418.888888889
151
+ 150,0.5,mask,0.129,0.006119885619846175,609774.59,4726934.80620155
152
+ 150,0.5,compact,0.10033333333333333,0.005485330605519766,620170.1666666666,6181098.006644518
153
+ 150,0.5,reset,0.21966666666666668,0.00755895103441878,530142.1,2413393.4749620636
154
+ 150,1.0,append,0.102,0.00552557689295878,961513.4833333333,9426602.777777778
155
+ 150,1.0,mask,0.10766666666666666,0.005659050437295276,575638.6333333333,5346488.854489164
156
+ 150,1.0,compact,0.09333333333333334,0.005311064620902989,592405.5333333333,6347202.142857143
157
+ 150,1.0,reset,0.206,0.007383856264400962,519716.2833333333,2522894.579288026
158
+ 150,2.0,append,0.056,0.004197777189577042,786326.9166666666,14041552.083333334
159
+ 150,2.0,mask,0.09533333333333334,0.005361743773838535,543538.27,5701450.384615385
160
+ 150,2.0,compact,0.07633333333333334,0.004847905236819836,531389.2333333333,6961431.004366812
161
+ 150,2.0,reset,0.16833333333333333,0.006831232739465165,475451.5,2824464.3564356435
162
+ 150,3.0,append,0.044333333333333336,0.003758008731978896,665416.8666666667,15009403.007518796
163
+ 150,3.0,mask,0.08733333333333333,0.005154487437247349,509313.4266666667,5831833.129770992
164
+ 150,3.0,compact,0.065,0.004500925830685653,487679.55,7502762.307692308
165
+ 150,3.0,reset,0.148,0.006483209081928485,441763.51666666666,2984888.626126126
166
+ 150,4.0,append,0.035666666666666666,0.0033859787140281686,590755.45,16563236.915887851
167
+ 150,4.0,mask,0.07766666666666666,0.004886530997055599,478038.19333333336,6154998.197424892
168
+ 150,4.0,compact,0.05466666666666667,0.004150430588996047,451099.4166666667,8251818.597560976
169
+ 150,4.0,reset,0.12766666666666668,0.006092834285423298,419339.7833333333,3284645.822454308
170
+ 200,0.0,append,0.063,0.004435876463563881,1349005.2,21412780.95238095
171
+ 200,0.0,mask,0.08166666666666667,0.004999907406550053,704496.11,8626482.979591837
172
+ 200,0.0,compact,0.03966666666666667,0.0035633889404246546,693268.0833333334,17477346.63865546
173
+ 200,0.0,reset,0.157,0.006642062932553409,651995.9333333333,4152840.33970276
174
+ 200,0.5,append,0.04066666666666666,0.0036061470153470306,1144051.5166666666,28132414.344262294
175
+ 200,0.5,mask,0.064,0.004468556814006061,656413.1666666666,10256455.729166666
176
+ 200,0.5,compact,0.037,0.0034463023663050805,641942.9333333333,17349809.009009007
177
+ 200,0.5,reset,0.13633333333333333,0.006264890410202868,615423.45,4514108.435207824
178
+ 200,1.0,append,0.036,0.00340117626711701,1032435.2666666667,28678757.407407407
179
+ 200,1.0,mask,0.056,0.004197777189577042,626132.4,11180935.714285715
180
+ 200,1.0,compact,0.034,0.003308776208811953,611265.65,17978401.470588237
181
+ 200,1.0,reset,0.12933333333333333,0.006126614677569286,597346.9666666667,4618662.113402062
182
+ 200,2.0,append,0.014333333333333333,0.002170091310589556,813235.5333333333,56737362.79069767
183
+ 200,2.0,mask,0.041666666666666664,0.0036483126490454714,564469.0533333333,13547257.28
184
+ 200,2.0,compact,0.023666666666666666,0.0027752811002104247,561815.4166666666,23738679.57746479
185
+ 200,2.0,reset,0.108,0.005666745097496446,541076.1,5009963.888888889
186
+ 200,3.0,append,0.014333333333333333,0.002170091310589556,701534.7833333333,48944287.20930233
187
+ 200,3.0,mask,0.038,0.0034907496807037503,536843.78,14127467.894736841
188
+ 200,3.0,compact,0.017,0.002360155362117785,508902.05,29935414.70588235
189
+ 200,3.0,reset,0.07766666666666666,0.004886530997055599,488589.2,6290848.068669528
190
+ 200,4.0,append,0.005666666666666667,0.0013704689654221234,613870.4166666666,108330073.52941176
191
+ 200,4.0,mask,0.03266666666666666,0.0032454868949335146,494069.3466666667,15124571.836734693
192
+ 200,4.0,compact,0.012666666666666666,0.0020417494314290234,457246.43333333335,36098402.631578945
193
+ 200,4.0,reset,0.07433333333333333,0.004789150547118243,468757.35,6306152.69058296
194
+ 300,0.0,append,0.005666666666666667,0.0013704689654221234,1403777.3666666667,247725417.6470588
195
+ 300,0.0,mask,0.022333333333333334,0.0026978111841241196,760161.9,34037100.0
196
+ 300,0.0,compact,0.0026666666666666666,0.0009415511236882033,710481.4833333333,266430556.25
197
+ 300,0.0,reset,0.079,0.004924733495327438,754152.2333333333,9546230.801687764
198
+ 300,0.5,append,0.0033333333333333335,0.0010523342660045845,1176288.6666666667,352886600.0
199
+ 300,0.5,mask,0.019,0.0024925890154616345,715678.4333333333,37667285.96491228
200
+ 300,0.5,compact,0.0023333333333333335,0.0008808875995057276,661773.1833333333,283617078.5714286
201
+ 300,0.5,reset,0.07166666666666667,0.004709230494661719,725784.0833333334,10127219.767441861
202
+ 300,1.0,append,0.0036666666666666666,0.001103512909186268,1060411.4333333333,289203118.1818182
203
+ 300,1.0,mask,0.017666666666666667,0.0024051718965287807,677682.2966666666,38359375.283018865
204
+ 300,1.0,compact,0.0016666666666666668,0.0007447346034861208,623680.5333333333,374208320.0
205
+ 300,1.0,reset,0.066,0.004532990183091069,695706.8,10541012.121212121
206
+ 300,2.0,append,0.0003333333333333333,0.0003332777731473764,824375.8666666667,2473127600.0
207
+ 300,2.0,mask,0.014666666666666666,0.00219480868988283,629128.6,42895131.81818182
208
+ 300,2.0,compact,0.0023333333333333335,0.0008808875995057276,564232.95,241814121.42857143
209
+ 300,2.0,reset,0.045,0.0037848381735551126,619495.7833333333,13766572.962962963
210
+ 300,3.0,append,0.001,0.0005770615218501404,707209.0166666667,707209016.6666666
211
+ 300,3.0,mask,0.009666666666666667,0.0017863577676585601,567811.5833333334,58739129.31034483
212
+ 300,3.0,compact,0.0026666666666666666,0.0009415511236882033,510732.55,191524706.25
213
+ 300,3.0,reset,0.034333333333333334,0.003324382426634836,566764.6666666666,16507708.737864077
214
+ 300,4.0,append,0.001,0.0005770615218501404,620709.7,620709700.0
215
+ 300,4.0,mask,0.007666666666666666,0.0015924707102370713,528117.4733333334,68884887.82608695
216
+ 300,4.0,compact,0.0013333333333333333,0.0006662220739752263,464617.06666666665,348462800.0
217
+ 300,4.0,reset,0.028,0.003011976095522672,506097.0333333333,18074894.04761905
simulation/model_e4_verify_context.csv ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ d,verifier,policy,success,se,cost,cost_per_success
2
+ 0.0,independent,append,0.07933333333333334,0.004934219139693587,401446.26666666666,5060247.05882353
3
+ 0.0,independent,mask,0.06866666666666667,0.004617053734275267,301825.93333333335,4395523.300970874
4
+ 0.0,independent,compact,0.072,0.0047193219852008405,298336.4166666667,4143561.342592593
5
+ 0.0,independent,reset,0.124,0.006017308368365377,255563.68333333332,2060997.4462365592
6
+ 0.0,self-critique,append,0.07933333333333334,0.004934219139693587,401446.26666666666,5060247.05882353
7
+ 0.0,self-critique,mask,0.06866666666666667,0.004617053734275267,301825.93333333335,4395523.300970874
8
+ 0.0,self-critique,compact,0.072,0.0047193219852008405,298336.4166666667,4143561.342592593
9
+ 0.0,self-critique,reset,0.124,0.006017308368365377,255563.68333333332,2060997.4462365592
10
+ 0.25,independent,append,0.11833333333333333,0.005897190165820656,497987.45,4208344.647887324
11
+ 0.25,independent,mask,0.10466666666666667,0.005589023634337682,363620.55,3474081.687898089
12
+ 0.25,independent,compact,0.10433333333333333,0.005581155462473367,356551.31666666665,3417424.7603833866
13
+ 0.25,independent,reset,0.17633333333333334,0.006957966390857051,302098.88333333336,1713226.18147448
14
+ 0.25,self-critique,append,0.12,0.00593295878967653,498113.45,4150945.4166666665
15
+ 0.25,self-critique,mask,0.09033333333333333,0.005233648256625016,347236.05666666664,3843941.5867158673
16
+ 0.25,self-critique,compact,0.09,0.0052249401910452525,355457.81666666665,3949531.296296296
17
+ 0.25,self-critique,reset,0.16633333333333333,0.006798689960954623,298518.01666666666,1794697.49498998
18
+ 0.5,independent,append,0.19066666666666668,0.0071720031810945,642752.6833333333,3371080.506993007
19
+ 0.5,independent,mask,0.21333333333333335,0.007479354299720046,462458.36666666664,2167773.59375
20
+ 0.5,independent,compact,0.20133333333333334,0.007321161615987412,461136.88333333336,2290414.9834437086
21
+ 0.5,independent,reset,0.279,0.008188589622151058,375713.18333333335,1346642.2341696536
22
+ 0.5,self-critique,append,0.18733333333333332,0.00712365906809966,642586.05,3430174.644128114
23
+ 0.5,self-critique,mask,0.135,0.006238990302925627,422229.29333333333,3127624.3950617285
24
+ 0.5,self-critique,compact,0.153,0.006572442468367448,439580.2,2873073.202614379
25
+ 0.5,self-critique,reset,0.25233333333333335,0.007930137077886742,362926.93333333335,1438283.751651255
26
+ 0.75,independent,append,0.368,0.008804847149913128,927820.8,2521252.1739130435
27
+ 0.75,independent,mask,0.4106666666666667,0.008981825270243526,626751.07,1526179.5535714286
28
+ 0.75,independent,compact,0.37366666666666665,0.008832513588797716,608503.1333333333,1628465.1204281892
29
+ 0.75,independent,reset,0.4826666666666667,0.009123222302738501,498524.6666666667,1032854.9723756906
30
+ 0.75,self-critique,append,0.33566666666666667,0.008621572856417703,925395.5833333334,2756888.5302879843
31
+ 0.75,self-critique,mask,0.21266666666666667,0.007470822256297174,526616.02,2476250.8777429466
32
+ 0.75,self-critique,compact,0.24233333333333335,0.00782321095392612,569722.0166666667,2350984.938101788
33
+ 0.75,self-critique,reset,0.396,0.008929053701260846,462652.65,1168314.7727272727
34
+ 0.9,independent,append,0.6256666666666667,0.008835683125616054,1233313.3833333333,1971198.801278636
35
+ 0.9,independent,mask,0.657,0.008667006403597495,772038.3266666667,1175096.3876204973
36
+ 0.9,independent,compact,0.6553333333333333,0.00867701092073293,780176.55,1190503.3825025433
37
+ 0.9,independent,reset,0.7246666666666667,0.008155261877848945,610014.9166666666,841786.9135234591
38
+ 0.9,self-critique,append,0.533,0.009108805263772704,1205013.8,2260813.883677298
39
+ 0.9,self-critique,mask,0.31966666666666665,0.008514299518826918,621518.2366666667,1944269.7705943692
40
+ 0.9,self-critique,compact,0.35133333333333333,0.008715851884587879,687877.75,1957906.3092979128
41
+ 0.9,self-critique,reset,0.5573333333333333,0.009068496547368236,549639.2833333333,986194.8863636364
simulation/model_e5_model_strength.csv ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ eps_mean,gamma,d,policy,success,se,cost,cost_per_success
2
+ 0.005,0.5,0.5,append,0.7043333333333334,0.00833162426919043,1273358.85,1807892.3568386182
3
+ 0.005,0.5,0.5,mask,0.491,0.009127230321041173,692641.52,1410675.1934826884
4
+ 0.005,0.5,0.5,compact,0.5303333333333333,0.00911189495273237,738879.7166666667,1393236.4236329352
5
+ 0.005,0.5,0.5,reset,0.6663333333333333,0.008608778895643335,567846.2833333333,852195.5227613806
6
+ 0.005,0.5,0.9,append,0.919,0.004981264899601304,1454222.15,1582396.2459194777
7
+ 0.005,0.5,0.9,mask,0.8296666666666667,0.006863426959590204,855379.0433333333,1030991.2133386902
8
+ 0.005,0.5,0.9,compact,0.845,0.006607445295523326,869444.1333333333,1028927.9684418146
9
+ 0.005,0.5,0.9,reset,0.897,0.00554950448238399,661068.7333333333,736977.4061687106
10
+ 0.005,2.0,0.5,append,0.616,0.00887963963232743,1188592.7333333334,1929533.658008658
11
+ 0.005,2.0,0.5,mask,0.4746666666666667,0.009116984556716515,679718.5233333333,1431991.2710674158
12
+ 0.005,2.0,0.5,compact,0.48233333333333334,0.009123009168925366,706479.6666666666,1464712.5086385626
13
+ 0.005,2.0,0.5,reset,0.635,0.00878967196960918,557440.7666666667,877859.4750656168
14
+ 0.005,2.0,0.9,append,0.881,0.0059115423819282,1429830.3833333333,1622962.9776768824
15
+ 0.005,2.0,0.9,mask,0.815,0.007089311203024829,849743.09,1042629.5582822086
16
+ 0.005,2.0,0.9,compact,0.8326666666666667,0.006815005230834111,863757.7,1037339.1112890312
17
+ 0.005,2.0,0.9,reset,0.8876666666666667,0.0057652567319404924,660539.3,744129.8911002629
18
+ 0.01,0.5,0.5,append,0.5533333333333333,0.009076628514221852,1097495.5,1983425.6024096385
19
+ 0.01,0.5,0.5,mask,0.39566666666666667,0.008927757380879694,622223.3033333333,1572594.700926706
20
+ 0.01,0.5,0.5,compact,0.42533333333333334,0.009026348262889205,655804.3833333333,1541859.8354231976
21
+ 0.01,0.5,0.5,reset,0.5316666666666666,0.009110382762215541,509019.25,957402.9780564263
22
+ 0.01,0.5,0.9,append,0.8613333333333333,0.0063097338090240184,1414582.5666666667,1642317.2213622292
23
+ 0.01,0.5,0.9,mask,0.791,0.007423364017299254,837891.3133333334,1059281.0535187526
24
+ 0.01,0.5,0.9,compact,0.8003333333333333,0.0072983991141639835,847009.8666666667,1058321.3660974593
25
+ 0.01,0.5,0.9,reset,0.8536666666666667,0.0064529034521649984,648932.0833333334,760170.3436157751
26
+ 0.01,2.0,0.5,append,0.41633333333333333,0.008999997942386597,963354.3166666667,2313901.481184948
27
+ 0.01,2.0,0.5,mask,0.3586666666666667,0.00875642409679676,594590.7933333333,1657781.0223048327
28
+ 0.01,2.0,0.5,compact,0.359,0.00875821519108393,614567.45,1711887.0473537603
29
+ 0.01,2.0,0.5,reset,0.48966666666666664,0.009126759596962151,488265.5333333333,997138.5976855004
30
+ 0.01,2.0,0.9,append,0.7933333333333333,0.00739268908123775,1369755.9666666666,1726583.1512605043
31
+ 0.01,2.0,0.9,mask,0.7656666666666667,0.007733503350621507,830495.12,1084669.290378755
32
+ 0.01,2.0,0.9,compact,0.7613333333333333,0.007782563606954736,829889.1666666666,1090047.0665499123
33
+ 0.01,2.0,0.9,reset,0.8313333333333334,0.006836622514619876,639856.6333333333,769675.1804330393
34
+ 0.02,0.5,0.5,append,0.3323333333333333,0.008600152884084656,829982.8666666667,2497440.922768305
35
+ 0.02,0.5,0.5,mask,0.22766666666666666,0.007655815992989808,495880.31333333335,2178098.008784773
36
+ 0.02,0.5,0.5,compact,0.25133333333333335,0.007919698834528345,520399.45,2070554.8408488063
37
+ 0.02,0.5,0.5,reset,0.3323333333333333,0.008600152884084656,414821.7,1248209.7291875626
38
+ 0.02,0.5,0.9,append,0.7553333333333333,0.0078486705644733,1321079.1,1749001.4563106797
39
+ 0.02,0.5,0.9,mask,0.6966666666666667,0.008392898232478501,793949.4833333333,1139640.4066985645
40
+ 0.02,0.5,0.9,compact,0.7136666666666667,0.008253212618036599,807146.3333333334,1130985.053713218
41
+ 0.02,0.5,0.9,reset,0.7663333333333333,0.007725856059483108,617135.2833333333,805309.1996520226
42
+ 0.02,2.0,0.5,append,0.19266666666666668,0.007200606970300541,650209.5666666667,3374790.1384083047
43
+ 0.02,2.0,0.5,mask,0.19633333333333333,0.0072522767817092115,467132.47,2379282.529711375
44
+ 0.02,2.0,0.5,compact,0.19066666666666668,0.0071720031810945,454021.51666666666,2381231.730769231
45
+ 0.02,2.0,0.5,reset,0.276,0.008161372433604534,376207.6666666667,1363071.2560386474
46
+ 0.02,2.0,0.9,append,0.6266666666666667,0.00883092210276963,1232224.2666666666,1966315.319148936
47
+ 0.02,2.0,0.9,mask,0.6656666666666666,0.008613062861250454,778729.6966666667,1169849.3189784677
48
+ 0.02,2.0,0.9,compact,0.65,0.008708233651742087,780939.4833333333,1201445.358974359
49
+ 0.02,2.0,0.9,reset,0.7213333333333334,0.008185588057791075,603362.1666666666,836454.0203327172
50
+ 0.04,0.5,0.5,append,0.11966666666666667,0.0059258348372628914,505642.68333333335,4225426.323119777
51
+ 0.04,0.5,0.5,mask,0.09966666666666667,0.005469101761173774,355630.85333333333,3568202.54180602
52
+ 0.04,0.5,0.5,compact,0.111,0.005735241930380967,361726.23333333334,3258794.8948948947
53
+ 0.04,0.5,0.5,reset,0.14633333333333334,0.006452903452164999,288302.31666666665,1970175.284738041
54
+ 0.04,0.5,0.9,append,0.5756666666666667,0.009023572011784017,1164678.4833333334,2023182.0787492762
55
+ 0.04,0.5,0.9,mask,0.5616666666666666,0.009059014336048969,730536.0666666667,1300657.6854599407
56
+ 0.04,0.5,0.9,compact,0.5706666666666667,0.009037075895484156,738822.9833333333,1294666.4427570093
57
+ 0.04,0.5,0.9,reset,0.609,0.008909152597189027,560770.7166666667,920805.7744937055
58
+ 0.04,2.0,0.5,append,0.04066666666666666,0.0036061470153470306,349085.38333333336,8584066.803278688
59
+ 0.04,2.0,0.5,mask,0.069,0.004627418286690755,303526.76,4398938.550724638
60
+ 0.04,2.0,0.5,compact,0.056666666666666664,0.00422119870635327,285933.01666666666,5045876.764705882
61
+ 0.04,2.0,0.5,reset,0.101,0.005501484648104849,249096.46666666667,2466301.6501650163
62
+ 0.04,2.0,0.9,append,0.38466666666666666,0.008882533839361795,1000655.7,2601357.972270364
63
+ 0.04,2.0,0.9,mask,0.49333333333333335,0.009127897814859592,683102.9333333333,1384668.1081081082
64
+ 0.04,2.0,0.9,compact,0.467,0.009108805263772704,684276.05,1465259.2077087795
65
+ 0.04,2.0,0.9,reset,0.5333333333333333,0.009108400680852976,532317.35,998095.03125
simulation/model_e6_global_sensitivity.csv ADDED
The diff for this file is too large to render. See raw diff
 
simulation/model_e6_retention_grid.csv ADDED
@@ -0,0 +1,101 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ lam_c,lam_r,policy,success,se,cost,cost_per_success
2
+ 0.02,0.02,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
3
+ 0.02,0.02,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
4
+ 0.02,0.02,compact,0.29933333333333334,0.011824632168736832,498046.5,1663852.4498886415
5
+ 0.02,0.02,reset,0.3413333333333333,0.012242681865476177,400917.5333333333,1174563.0859375
6
+ 0.02,0.05,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
7
+ 0.02,0.05,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
8
+ 0.02,0.05,compact,0.29933333333333334,0.011824632168736832,498046.5,1663852.4498886415
9
+ 0.02,0.05,reset,0.338,0.012213544394100618,399465.73333333334,1181851.282051282
10
+ 0.02,0.1,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
11
+ 0.02,0.1,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
12
+ 0.02,0.1,compact,0.29933333333333334,0.011824632168736832,498046.5,1663852.4498886415
13
+ 0.02,0.1,reset,0.3,0.011832159566199232,389578.2,1298594.0
14
+ 0.02,0.2,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
15
+ 0.02,0.2,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
16
+ 0.02,0.2,compact,0.29933333333333334,0.011824632168736832,498046.5,1663852.4498886415
17
+ 0.02,0.2,reset,0.26066666666666666,0.011334888782149726,376418.6,1444061.125319693
18
+ 0.02,0.4,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
19
+ 0.02,0.4,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
20
+ 0.02,0.4,compact,0.29933333333333334,0.011824632168736832,498046.5,1663852.4498886415
21
+ 0.02,0.4,reset,0.17066666666666666,0.00971389230451438,336136.3,1969548.6328125
22
+ 0.05,0.02,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
23
+ 0.05,0.02,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
24
+ 0.05,0.02,compact,0.2853333333333333,0.011659566093190667,494069.0333333333,1731550.3504672898
25
+ 0.05,0.02,reset,0.3413333333333333,0.012242681865476177,400917.5333333333,1174563.0859375
26
+ 0.05,0.05,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
27
+ 0.05,0.05,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
28
+ 0.05,0.05,compact,0.2853333333333333,0.011659566093190667,494069.0333333333,1731550.3504672898
29
+ 0.05,0.05,reset,0.338,0.012213544394100618,399465.73333333334,1181851.282051282
30
+ 0.05,0.1,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
31
+ 0.05,0.1,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
32
+ 0.05,0.1,compact,0.2853333333333333,0.011659566093190667,494069.0333333333,1731550.3504672898
33
+ 0.05,0.1,reset,0.3,0.011832159566199232,389578.2,1298594.0
34
+ 0.05,0.2,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
35
+ 0.05,0.2,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
36
+ 0.05,0.2,compact,0.2853333333333333,0.011659566093190667,494069.0333333333,1731550.3504672898
37
+ 0.05,0.2,reset,0.26066666666666666,0.011334888782149726,376418.6,1444061.125319693
38
+ 0.05,0.4,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
39
+ 0.05,0.4,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
40
+ 0.05,0.4,compact,0.2853333333333333,0.011659566093190667,494069.0333333333,1731550.3504672898
41
+ 0.05,0.4,reset,0.17066666666666666,0.00971389230451438,336136.3,1969548.6328125
42
+ 0.1,0.02,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
43
+ 0.1,0.02,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
44
+ 0.1,0.02,compact,0.26466666666666666,0.01139058740721836,496438.13333333336,1875710.831234257
45
+ 0.1,0.02,reset,0.3413333333333333,0.012242681865476177,400917.5333333333,1174563.0859375
46
+ 0.1,0.05,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
47
+ 0.1,0.05,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
48
+ 0.1,0.05,compact,0.26466666666666666,0.01139058740721836,496438.13333333336,1875710.831234257
49
+ 0.1,0.05,reset,0.338,0.012213544394100618,399465.73333333334,1181851.282051282
50
+ 0.1,0.1,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
51
+ 0.1,0.1,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
52
+ 0.1,0.1,compact,0.26466666666666666,0.01139058740721836,496438.13333333336,1875710.831234257
53
+ 0.1,0.1,reset,0.3,0.011832159566199232,389578.2,1298594.0
54
+ 0.1,0.2,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
55
+ 0.1,0.2,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
56
+ 0.1,0.2,compact,0.26466666666666666,0.01139058740721836,496438.13333333336,1875710.831234257
57
+ 0.1,0.2,reset,0.26066666666666666,0.011334888782149726,376418.6,1444061.125319693
58
+ 0.1,0.4,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
59
+ 0.1,0.4,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
60
+ 0.1,0.4,compact,0.26466666666666666,0.01139058740721836,496438.13333333336,1875710.831234257
61
+ 0.1,0.4,reset,0.17066666666666666,0.00971389230451438,336136.3,1969548.6328125
62
+ 0.2,0.02,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
63
+ 0.2,0.02,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
64
+ 0.2,0.02,compact,0.22,0.01069579356569675,473442.0333333333,2152009.242424242
65
+ 0.2,0.02,reset,0.3413333333333333,0.012242681865476177,400917.5333333333,1174563.0859375
66
+ 0.2,0.05,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
67
+ 0.2,0.05,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
68
+ 0.2,0.05,compact,0.22,0.01069579356569675,473442.0333333333,2152009.242424242
69
+ 0.2,0.05,reset,0.338,0.012213544394100618,399465.73333333334,1181851.282051282
70
+ 0.2,0.1,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
71
+ 0.2,0.1,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
72
+ 0.2,0.1,compact,0.22,0.01069579356569675,473442.0333333333,2152009.242424242
73
+ 0.2,0.1,reset,0.3,0.011832159566199232,389578.2,1298594.0
74
+ 0.2,0.2,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
75
+ 0.2,0.2,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
76
+ 0.2,0.2,compact,0.22,0.01069579356569675,473442.0333333333,2152009.242424242
77
+ 0.2,0.2,reset,0.26066666666666666,0.011334888782149726,376418.6,1444061.125319693
78
+ 0.2,0.4,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
79
+ 0.2,0.4,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
80
+ 0.2,0.4,compact,0.22,0.01069579356569675,473442.0333333333,2152009.242424242
81
+ 0.2,0.4,reset,0.17066666666666666,0.00971389230451438,336136.3,1969548.6328125
82
+ 0.4,0.02,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
83
+ 0.4,0.02,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
84
+ 0.4,0.02,compact,0.166,0.009607080722050795,451183.43333333335,2717972.4899598393
85
+ 0.4,0.02,reset,0.3413333333333333,0.012242681865476177,400917.5333333333,1174563.0859375
86
+ 0.4,0.05,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
87
+ 0.4,0.05,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
88
+ 0.4,0.05,compact,0.166,0.009607080722050795,451183.43333333335,2717972.4899598393
89
+ 0.4,0.05,reset,0.338,0.012213544394100618,399465.73333333334,1181851.282051282
90
+ 0.4,0.1,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
91
+ 0.4,0.1,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
92
+ 0.4,0.1,compact,0.166,0.009607080722050795,451183.43333333335,2717972.4899598393
93
+ 0.4,0.1,reset,0.3,0.011832159566199232,389578.2,1298594.0
94
+ 0.4,0.2,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
95
+ 0.4,0.2,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
96
+ 0.4,0.2,compact,0.166,0.009607080722050795,451183.43333333335,2717972.4899598393
97
+ 0.4,0.2,reset,0.26066666666666666,0.011334888782149726,376418.6,1444061.125319693
98
+ 0.4,0.4,append,0.21266666666666667,0.01056533815693423,667952.6,3140842.946708464
99
+ 0.4,0.4,mask,0.23066666666666666,0.01087687318903601,490636.3466666667,2127036.1849710983
100
+ 0.4,0.4,compact,0.166,0.009607080722050795,451183.43333333335,2717972.4899598393
101
+ 0.4,0.4,reset,0.17066666666666666,0.00971389230451438,336136.3,1969548.6328125
taxonomy/codebook.json ADDED
@@ -0,0 +1,94 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "exclude_section": "pre-training|post-training|prompt engineering|ontology engineering|courses|tutorials|social media|related awesome|adjacent collections|^templates$|graph engineering$",
3
+ "architecture": "attention|positional|\\brope\\b|transformer architecture|kv[- ]?cache|state[- ]space|mamba|linear attention|sparse attention|position interpolation|length extrapolation|tokeni[sz]|speculative decoding|quantiz|pre-?train",
4
+ "in_scope": "agent|context|memory|retriev|\\brag\\b|tool|harness|loop|orchestrat|multi-agent|plan|reflect|verif|critic|refine|search|workflow|skill|\\bmcp\\b|sandbox|injection|compress|compaction|summar|benchmark|bench\\b|long-horizon|prompt|reason|llm|language model|coding|software|protocol|eval",
5
+ "context_operations": {
6
+ "Compose": "system prompt|instruction file|instructions? (file|artifact)|agents\\.md|claude\\.md|few-shot|in-context learning|prompt (template|design|cach)|prompt caching|kv reuse|\\bprefix\\b|spec-driven|specification|cursor rules|project memory|context file",
7
+ "Select": "retriev|\\brag\\b|graphrag|rerank|search engine|semantic search|code search|knowledge graph|tool (selection|retrieval)|skill|\\bmcp\\b|just-in-time|progressive disclosure|\\brecall\\b|grounding|knowledge (base|retrieval)|indexing|embedding",
8
+ "Write": "memory|memories|scratchpad|\\bnotes?\\b|notebook|state file|progress (file|log)|persist|working state|journal|ledger|wiki|knowledge store|episodic|long-term",
9
+ "Compress": "compress|compaction|\\bcompact|summari|prun|truncat|condens|observation mask|masking|lingua|folding|token (budget|reduction|efficien)|context (window )?management",
10
+ "Isolate": "sub-?agents?|isolat|separate context|context folding|recursive language model|delegat|fork(ed|ing)? (context|agent)|parallel agents",
11
+ "Evolve": "self-evolv|evolving context|evolv(e|ing) (prompt|playbook|context)|dspy|gepa|playbook|cheatsheet|prompt optimi|context optimi|learn(ed|s|ing)? to (retrieve|manage|compress|curate)|curation|experiential|experience (replay|library)|agentic context engineering|evolv\\w* (agent )?skills?",
12
+ "Guard": "injection|provenance|untrusted|guardrail|poison|privacy|jailbreak|exfiltrat|secret|data leak|trust boundar|context integrity"
13
+ },
14
+ "loop_levels": {
15
+ "L1 Action": "\\breact\\b|tool use|tool[- ]call|function[- ]call|agent loop|agentic loop|codeact|computer use|browser use|executable code action|act(ing)? in",
16
+ "L2 Plan": "\\bplan(s|ning|ner)?\\b|decompos|task graph|workflow|stateflow|replan|to-?do",
17
+ "L3 Verification": "verif|critic|critique|reflect|self-refine|\\brefine|evaluator|llm-as-a-judge|judge|quality gate|self-correct|proof|evidence-gated|\\btests?\\b|testing|code review|review loop|validator",
18
+ "L4 Search": "tree[- ]search|mcts|monte carlo|tree of thoughts|graph of thoughts|beam search|best-of-n|repeated sampling|\\blats\\b|search over|test-time (compute|scaling)|branch",
19
+ "L5 Orchestration": "multi-agent|orchestrat|coordinat|controller|supervisor|swarm|\\bteams?\\b|handoff|router|manager agent|worker",
20
+ "L6 Session": "long-running|long-horizon|multi-session|across sessions|context windows|\\bsession|resum|initializer|ralph|persistent agent|durable execution|checkpoint",
21
+ "L7 Scheduled": "schedul|\\bcron\\b|background agent|event-driven|trigger|daemon|unattended|outer loop|loop engineering|triage|always-on|ambient|autonomous (loop|agents? that run)",
22
+ "L8 Improvement": "self-improv|self-evolv|evolv|meta-harness|harness optimi|automated design|\\badas\\b|aflow|darwin|alphaevolve|funsearch|self-modif|(prompt|workflow|agent|harness) optimi[sz]ation|self-harness|automatic evolution"
23
+ },
24
+ "harness": "harness|runtime|sandbox|permission|authori[sz]|tool design|tool interface|protocol|\\bmcp\\b|execution|scaffold|observability|tracing|telemetry|infrastructure|container|\\bsdk\\b|framework|hooks?\\b|cli\\b|desktop|control plane",
25
+ "evaluation": "benchmark|\\bbench\\b|-bench|arena|leaderboard|\\bdataset|evaluat|\\beval",
26
+ "section_layer": [
27
+ [
28
+ "harness",
29
+ "runtime orchestration|interaction paradigm|agent loop|harness architecture|tool integration|reference harness|coding-agent harness|execution substrates|runtime foundations"
30
+ ],
31
+ [
32
+ "evaluation",
33
+ "benchmark|\\bevals?\\b|\\bevaluation|benchmarking|quality assessment"
34
+ ],
35
+ [
36
+ "context",
37
+ "context|memory|retrieval|\\brag\\b|knowledge|structured data|self-generated|compaction|working-state|instruction files|project memory|skill composition"
38
+ ],
39
+ [
40
+ "loop",
41
+ "agent loop|loop architecture|planning|orchestration|verification|human-in-the-loop|task runners|multi-agent|agent coordination|task organization|system evolution|environment feedback|interaction paradigm|state management|spec-driven|operating principles|operational autonomy|meta-harness|generators|workflow, profiles"
42
+ ],
43
+ [
44
+ "harness",
45
+ "harness|runtime|sandbox|security|permission|tool design|tool integration|skills|protocols|observability|debugging|production|execution|guardrails|reference|generators|tool use|function calling|agent communication|skill composition|foundations|platform|implementations|telemetry|browser, mcp|workflow, profiles|constraints"
46
+ ]
47
+ ],
48
+ "section_context_ops": {
49
+ "Select": "retrieval|\\brag\\b|knowledge|structured data|tool integration|skills",
50
+ "Write": "memory|state management|working-state|project memory",
51
+ "Compress": "compression|compaction|context management in production",
52
+ "Compose": "instruction files|context design|context delivery|prompt",
53
+ "Guard": "security|guardrails|permissions"
54
+ },
55
+ "section_loop_levels": {
56
+ "L1": "agent loop|tool use|function calling",
57
+ "L2": "planning|task organization|spec-driven|task decomposition",
58
+ "L3": "verification|quality gates|environment feedback",
59
+ "L5": "orchestration|multi-agent|agent coordination|task runners",
60
+ "L8": "system evolution|generators & meta-harnesses"
61
+ },
62
+ "contribution": [
63
+ [
64
+ "survey",
65
+ "\\bsurvey|systematic review|\\breview\\b|taxonomy|overview|position paper|perspective|\\bsok\\b|primer"
66
+ ],
67
+ [
68
+ "benchmark",
69
+ "benchmark|\\bbench\\b|-bench|arena|leaderboard|\\bdataset|evaluat|\\beval"
70
+ ],
71
+ [
72
+ "empirical",
73
+ "empirical|exploratory|\\bstudy\\b|analysis of|we find|measur|understanding|how (do|does|well)|lessons|investigat|uncovering|diagnos"
74
+ ]
75
+ ],
76
+ "domain": [
77
+ [
78
+ "software",
79
+ "\\bcode\\b|coding|software|\\bswe\\b|repositor|programming|developer|pull request|terminal|\\bci\\b|github|compiler|debug"
80
+ ],
81
+ [
82
+ "web_gui",
83
+ "\\bweb\\b|browser|\\bgui\\b|computer use|desktop|mobile|android|\\bos\\b|screen"
84
+ ],
85
+ [
86
+ "research",
87
+ "research|scien|discovery|literature|paper|chemistry|biolog|medical|clinical"
88
+ ],
89
+ [
90
+ "dialogue",
91
+ "conversation|dialog|personal|assistant|chatbot|user model|companion|customer"
92
+ ]
93
+ ]
94
+ }
taxonomy/table_coupling_lift.csv ADDED
@@ -0,0 +1,57 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ level,level_name,op,both,expected,lift,p_greater,in_draft,p_holm
2
+ L1,Action,Compose,10,8.24,1.213,0.30547763578496023,False,1.0
3
+ L1,Action,Select,24,35.04,0.685,0.9939949816511466,True,1.0
4
+ L1,Action,Write,18,23.08,0.78,0.9157430847822527,False,1.0
5
+ L1,Action,Compress,20,7.91,2.527,4.6120389332734616e-05,True,0.0023982602453022
6
+ L1,Action,Isolate,13,8.99,1.447,0.10474257247541333,False,1.0
7
+ L1,Action,Evolve,0,2.72,0.0,1.0,False,1.0
8
+ L1,Action,Guard,20,11.71,1.709,0.009016211479409732,False,0.4417943624910769
9
+ L2,Plan,Compose,12,14.27,0.841,0.7912346743969603,False,1.0
10
+ L2,Plan,Select,62,60.63,1.023,0.4393117871136316,True,1.0
11
+ L2,Plan,Write,37,39.94,0.926,0.7421806675984062,True,1.0
12
+ L2,Plan,Compress,8,13.69,0.584,0.9759549967712574,False,1.0
13
+ L2,Plan,Isolate,22,15.55,1.415,0.04914640756574053,False,1.0
14
+ L2,Plan,Evolve,2,4.71,0.425,0.961216840655643,False,1.0
15
+ L2,Plan,Guard,12,20.26,0.592,0.9903248718063031,False,1.0
16
+ L3,Verification,Compose,16,13.26,1.206,0.24132876618457783,True,1.0
17
+ L3,Verification,Select,48,56.36,0.852,0.9372464062635567,False,1.0
18
+ L3,Verification,Write,30,37.13,0.808,0.9363929630920199,False,1.0
19
+ L3,Verification,Compress,12,12.73,0.943,0.6386400736690434,False,1.0
20
+ L3,Verification,Isolate,20,14.46,1.384,0.0729720320547629,True,1.0
21
+ L3,Verification,Evolve,4,4.38,0.914,0.6564703519470709,False,1.0
22
+ L3,Verification,Guard,14,18.83,0.743,0.9223270126792508,False,1.0
23
+ L4,Search,Compose,1,0.93,1.073,0.6214219504923328,False,1.0
24
+ L4,Search,Select,3,3.96,0.757,0.8087595496609248,False,1.0
25
+ L4,Search,Write,6,2.61,2.299,0.029883958624855426,True,1.0
26
+ L4,Search,Compress,0,0.89,0.0,1.0,True,1.0
27
+ L4,Search,Isolate,5,1.02,4.922,0.00206049996779348,True,0.103024998389674
28
+ L4,Search,Evolve,0,0.31,0.0,1.0,False,1.0
29
+ L4,Search,Guard,1,1.32,0.756,0.7538897113909268,False,1.0
30
+ L5,Orchestration,Compose,11,24.66,0.446,0.9998841624994471,False,1.0
31
+ L5,Orchestration,Select,72,104.8,0.687,0.9999979313019455,False,1.0
32
+ L5,Orchestration,Write,62,69.05,0.898,0.8797105610011497,True,1.0
33
+ L5,Orchestration,Compress,18,23.67,0.76,0.9385103839288979,True,1.0
34
+ L5,Orchestration,Isolate,48,26.88,1.786,3.0664309517427362e-06,True,0.00016252084044236502
35
+ L5,Orchestration,Evolve,6,8.14,0.737,0.8612268443010418,False,1.0
36
+ L5,Orchestration,Guard,31,35.02,0.885,0.8227692318862216,False,1.0
37
+ L6,Session,Compose,19,12.62,1.506,0.037914147047543645,False,1.0
38
+ L6,Session,Select,51,53.62,0.951,0.7055536954834565,True,1.0
39
+ L6,Session,Write,65,35.33,1.84,1.7096928477017623e-08,True,9.403310662359692e-07
40
+ L6,Session,Compress,20,12.11,1.651,0.012766891141164513,False,0.6128107747758966
41
+ L6,Session,Isolate,35,13.75,2.545,1.877546158843905e-08,False,1.0138749257757087e-06
42
+ L6,Session,Evolve,2,4.16,0.48,0.9349309260331581,False,1.0
43
+ L6,Session,Guard,15,17.92,0.837,0.8177145542284223,False,1.0
44
+ L7,Scheduled,Compose,4,3.44,1.162,0.4555508094640406,False,1.0
45
+ L7,Scheduled,Select,15,14.62,1.026,0.5079151167685344,True,1.0
46
+ L7,Scheduled,Write,16,9.63,1.661,0.019804603103056834,True,0.9308163458436712
47
+ L7,Scheduled,Compress,6,3.3,1.816,0.10642793302661238,False,1.0
48
+ L7,Scheduled,Isolate,11,3.75,2.933,0.0007682306945926806,False,0.039179765424226715
49
+ L7,Scheduled,Evolve,0,1.14,0.0,1.0,False,1.0
50
+ L7,Scheduled,Guard,7,4.89,1.433,0.2093058895414234,True,1.0
51
+ L8,Improvement,Compose,9,6.74,1.336,0.22540967334306908,False,1.0
52
+ L8,Improvement,Select,22,28.64,0.768,0.9538803780966361,False,1.0
53
+ L8,Improvement,Write,20,18.87,1.06,0.42426540063160845,False,1.0
54
+ L8,Improvement,Compress,4,6.47,0.618,0.9024872572458577,False,1.0
55
+ L8,Improvement,Isolate,5,7.34,0.681,0.8754457846969024,False,1.0
56
+ L8,Improvement,Evolve,20,2.22,8.994,1.3962048796117363e-16,True,7.818747325825724e-15
57
+ L8,Improvement,Guard,1,9.57,0.105,0.9999714558063572,True,1.0