File size: 17,418 Bytes
8c10cf2 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 | # Full reference trajectory configuration - RecruitmentAssistant-A2A project
#
# Version: ideal trajectory ported from the MCP version and adapted to the A2A architecture (applies to all models)
# Principle: only include steps explicitly required by the code design; do not rely on test statistics
#
# Migration notes:
# - This file is based on RecruitmentAssistant-MCP/reference_trajectory.yaml
# - It preserves the same business roles, Agent names, LLM/Tool names, and the dynamic 3x/4x reference trajectory design
# - Only the trajectory hierarchy is adapted to match the actual execution path under A2A:
# * Top-level SPAN: recruitment_orchestrator
# * Second-level SPAN: crew_execution
# * Inside each business stage (analyze_job / evaluate_candidates / prepare_interviews)
# add A2A-specific multi-layer SPANs: a2a_call_xxx → xxx_server_execution → Crew***.kickoff
#
# Code design + A2A architecture analysis (based on trace samples execution_path.md):
# 1. The top-level orchestrator runs 3 stages sequentially:
# - JobAnalysisCrew: Senior Job Requirements Analyst uses unified_web_search for market research
# - CandidateEvaluationCrew: Senior Candidate Evaluator uses candidate_evaluator_pro to evaluate candidates
# - InterviewCommunicationCrew: Senior Interview Materials Coordinator uses tools to generate interview materials and email templates
#
# 2. Tool usage (business design is consistent with the MCP version):
# - Job Analysis: unified_web_search (web search for market research; recommended 3-4 calls)
# - Candidate Evaluation: candidate_evaluator_pro (core scoring tool)
# - Interview Communication: comprehensive_interview_material_generator + email_template_generator
#
# 3. A2A execution flow (adds two extra server-related SPAN layers compared to MCP):
# - orchestrator creates the top-level SPAN: recruitment_orchestrator
# - orchestrator creates crew_execution SPAN, then runs 3 business stages in order:
# * analyze_job
# * evaluate_candidates
# * prepare_interviews
# - Within each stage, introduce the A2A-specific 3-level nesting:
# * Business method SPAN: analyze_job / evaluate_candidates / prepare_interviews
# * A2A call SPAN: a2a_call_job_analysis / a2a_call_candidate_evaluation / a2a_call_interview_communication
# * Server execution SPAN: job_analysis_server_execution / candidate_evaluation_server_execution / interview_communication_server_execution
# * Then enter the Crew chain inside the server: Chain: Crew***.kickoff → Agent → LLM/Tool
#
# 4. SPAN hierarchy under A2A (abstracted from execution_path.md):
# - [SPAN] recruitment_orchestrator
# └─ [SPAN] crew_execution
# ├─ [SPAN] analyze_job
# │ └─ [SPAN] a2a_call_job_analysis
# │ └─ [SPAN] job_analysis_server_execution
# │ └─ [Chain] Crew***.kickoff
# │ └─ [AGENT] Senior Job Requirements Analyst
# ├─ [SPAN] evaluate_candidates
# │ └─ [SPAN] a2a_call_candidate_evaluation
# │ └─ [SPAN] candidate_evaluation_server_execution
# │ └─ [Chain] Crew***.kickoff
# │ └─ [AGENT] Senior Candidate Evaluator
# └─ [SPAN] prepare_interviews
# └─ [SPAN] a2a_call_interview_communication
# └─ [SPAN] interview_communication_server_execution
# └─ [Chain] Crew***.kickoff
# └─ [AGENT] Senior Interview Materials Coordinator
#
# 5. The dynamic reference trajectory design remains unchanged (choose 3x/4x by unified_web_search count):
# - If unified_web_search under analyze_job SPAN ≤ 3, use reference_trajectory_3x
# - If unified_web_search under analyze_job SPAN ≥ 4, use reference_trajectory_4x
#
# 6. Important conventions:
# - Agent role names must match the role fields in each self-hosted server's agents.yaml
# - Tool names must match the actual invoked tool names: unified_web_search, candidate_evaluator_pro,
# comprehensive_interview_material_generator, email_template_generator
# - For [LLM], use "LLM: *" as a wildcard for any concrete model name
#
# Project name (used to match RESULTS/<model>/<project_name>/test_results)
project_name: "RecruitmentAssistant-A2A"
# Trajectory extraction type configuration
extract_types:
- "SPAN" # Multi-level SPAN structure (including A2A framework SPANs)
- "Chain" # Kickoff chain for each Crew
- "AGENT" # Agent for each Crew
- "LLM" # LLM calls
- "Tool" # Tool calls
# ============================================================================
# Reference trajectory (Ground Truth) - based on code design + A2A trace structure
# ============================================================================
#
# Diagram (ideal execution path; shows 3x version; 4x adds one extra search in Job Analysis):
#
# [SPAN] recruitment_orchestrator
# └─ [SPAN] crew_execution
# ├─ [SPAN] analyze_job
# │ └─ [SPAN] a2a_call_job_analysis
# │ └─ [SPAN] job_analysis_server_execution
# │ └─ [Chain] Crew***.kickoff
# │ └─ [AGENT] Senior Job Requirements Analyst
# │ ├─ [LLM] *
# │ ├─ [Tool] unified_web_search (1st search)
# │ ├─ [LLM] *
# │ ├─ [Tool] unified_web_search (2nd search)
# │ ├─ [LLM] *
# │ ├─ [Tool] unified_web_search (3rd/4th search)
# │ └─ [LLM] *
# ├─ [SPAN] evaluate_candidates
# │ └─ [SPAN] a2a_call_candidate_evaluation
# │ └─ [SPAN] candidate_evaluation_server_execution
# │ └─ [Chain] Crew***.kickoff
# │ └─ [AGENT] Senior Candidate Evaluator
# │ ├─ [LLM] *
# │ ├─ [Tool] candidate_profile_analyzer
# │ ├─ [LLM] *
# │ ├─ [Tool] candidate_evaluator_pro
# │ └─ [LLM] *
# └─ [SPAN] prepare_interviews
# └─ [SPAN] a2a_call_interview_communication
# └─ [SPAN] interview_communication_server_execution
# └─ [Chain] Crew***.kickoff
# └─ [AGENT] Senior Interview Materials Coordinator
# ├─ [LLM] *
# ├─ [Tool] comprehensive_interview_material_generator
# ├─ [LLM] *
# ├─ [Tool] email_template_generator
# └─ [LLM] *
#
# Notes:
# - Real traces may include extra LLM calls and auxiliary tools (e.g., candidate_profile_analyzer).
# They are not written into the reference trajectory, but are allowed in the In-order / Any-order / Precision / Recall metrics.
# - The reference trajectory captures only the key steps that should appear at minimum, and their required order.
# ============================================================================
# [Compatibility] Default reference trajectory (kept as the 3x version for older scripts and Markdown output)
reference_trajectory:
# ===== Top-level SPAN =====
- "SPAN: recruitment_orchestrator"
# ===== Second-level SPAN (contains 3 stages) =====
- "SPAN: crew_execution"
# ===== Crew 1: Job Analysis (3 searches; A2A multi-layer SPAN) =====
- "SPAN: analyze_job"
- "SPAN: a2a_call_job_analysis"
- "SPAN: job_analysis_server_execution"
- "Chain: Crew***.kickoff"
- "Agent: Senior Job Requirements Analyst"
- "LLM: *"
- "Tool: unified_web_search"
- "LLM: *"
- "Tool: unified_web_search"
- "LLM: *"
- "Tool: unified_web_search"
- "LLM: *"
# ===== Crew 2: Candidate Evaluation =====
- "SPAN: evaluate_candidates"
- "SPAN: a2a_call_candidate_evaluation"
- "SPAN: candidate_evaluation_server_execution"
- "Chain: Crew***.kickoff"
- "Agent: Senior Candidate Evaluator"
- "LLM: *"
- "Tool: candidate_profile_analyzer"
- "LLM: *"
- "Tool: candidate_evaluator_pro"
- "LLM: *"
# ===== Crew 3: Interview Communication =====
- "SPAN: prepare_interviews"
- "SPAN: a2a_call_interview_communication"
- "SPAN: interview_communication_server_execution"
- "Chain: Crew***.kickoff"
- "Agent: Senior Interview Materials Coordinator"
- "LLM: *"
- "Tool: comprehensive_interview_material_generator"
- "LLM: *"
- "Tool: email_template_generator"
- "LLM: *"
# ============================================================================
# Dynamic reference trajectory - Version 1: 3x unified_web_search
# ============================================================================
# Applies when unified_web_search count under analyze_job SPAN ≤ 3
reference_trajectory_3x:
# ===== Top-level SPAN =====
- "SPAN: recruitment_orchestrator"
# ===== Second-level SPAN (contains 3 stages) =====
- "SPAN: crew_execution"
# ===== Crew 1: Job Analysis (3 searches; A2A multi-layer SPAN) =====
- "SPAN: analyze_job"
- "SPAN: a2a_call_job_analysis"
- "SPAN: job_analysis_server_execution"
- "Chain: Crew***.kickoff"
- "Agent: Senior Job Requirements Analyst"
- "LLM: *" # Decide to call tools
- "Tool: unified_web_search" # Market research - search #1
- "LLM: *" # Analyze results and decide to continue searching
- "Tool: unified_web_search" # Market research - search #2
- "LLM: *" # Analyze results and decide to continue searching
- "Tool: unified_web_search" # Market research - search #3
- "LLM: *" # Produce the final job analysis report
# ===== Crew 2: Candidate Evaluation (A2A multi-layer SPAN) =====
- "SPAN: evaluate_candidates"
- "SPAN: a2a_call_candidate_evaluation"
- "SPAN: candidate_evaluation_server_execution"
- "Chain: Crew***.kickoff"
- "Agent: Senior Candidate Evaluator"
- "LLM: *" # Decide to call the auxiliary analysis tool
- "Tool: candidate_profile_analyzer" # Analyze candidate profiles and background
- "LLM: *" # Decide to call the scoring tool
- "Tool: candidate_evaluator_pro" # Evaluate all candidates (scoring, ranking, grading)
- "LLM: *" # Produce the final evaluation report
# ===== Crew 3: Interview Communication (A2A multi-layer SPAN) =====
- "SPAN: prepare_interviews"
- "SPAN: a2a_call_interview_communication"
- "SPAN: interview_communication_server_execution"
- "Chain: Crew***.kickoff"
- "Agent: Senior Interview Materials Coordinator"
- "LLM: *" # Decide to generate interview materials
- "Tool: comprehensive_interview_material_generator" # Generate interview questions for all stages
- "LLM: *" # Review materials and decide to generate email templates
- "Tool: email_template_generator" # Generate all types of email templates
- "LLM: *" # Final merge to produce a complete interview materials package
# ============================================================================
# Dynamic reference trajectory - Version 2: 4x unified_web_search
# ============================================================================
# Applies when unified_web_search count under analyze_job SPAN ≥ 4
reference_trajectory_4x:
# ===== Top-level SPAN =====
- "SPAN: recruitment_orchestrator"
# ===== Second-level SPAN (contains 3 stages) =====
- "SPAN: crew_execution"
# ===== Crew 1: Job Analysis (4 searches; A2A multi-layer SPAN) =====
- "SPAN: analyze_job"
- "SPAN: a2a_call_job_analysis"
- "SPAN: job_analysis_server_execution"
- "Chain: Crew***.kickoff"
- "Agent: Senior Job Requirements Analyst"
- "LLM: *" # Decide to call tools
- "Tool: unified_web_search" # Market research - search #1
- "LLM: *" # Analyze results and decide to continue searching
- "Tool: unified_web_search" # Market research - search #2
- "LLM: *" # Analyze results and decide to continue searching
- "Tool: unified_web_search" # Market research - search #3
- "LLM: *" # Analyze results and decide to continue searching
- "Tool: unified_web_search" # Market research - search #4
- "LLM: *" # Produce the final job analysis report
# ===== Crew 2: Candidate Evaluation (A2A multi-layer SPAN) =====
- "SPAN: evaluate_candidates"
- "SPAN: a2a_call_candidate_evaluation"
- "SPAN: candidate_evaluation_server_execution"
- "Chain: Crew***.kickoff"
- "Agent: Senior Candidate Evaluator"
- "LLM: *" # Decide to call the auxiliary analysis tool
- "Tool: candidate_profile_analyzer" # Analyze candidate profiles and background
- "LLM: *" # Decide to call the scoring tool
- "Tool: candidate_evaluator_pro" # Evaluate all candidates (scoring, ranking, grading)
- "LLM: *" # Produce the final evaluation report
# ===== Crew 3: Interview Communication (A2A multi-layer SPAN) =====
- "SPAN: prepare_interviews"
- "SPAN: a2a_call_interview_communication"
- "SPAN: interview_communication_server_execution"
- "Chain: Crew***.kickoff"
- "Agent: Senior Interview Materials Coordinator"
- "LLM: *" # Decide to generate interview materials
- "Tool: comprehensive_interview_material_generator" # Generate interview questions for all stages
- "LLM: *" # Review materials and decide to generate email templates
- "Tool: email_template_generator" # Generate all types of email templates
- "LLM: *" # Final merge to produce a complete interview materials package
# Target tool list (used by the single-tool use metric)
target_tools:
- "Tool: unified_web_search"
- "Tool: candidate_profile_analyzer"
- "Tool: candidate_evaluator_pro"
- "Tool: comprehensive_interview_material_generator"
- "Tool: email_template_generator"
# Models to evaluate (kept consistent with the MCP version)
models:
- "GPT-5"
- "GPT-4o-mini"
- "DeepSeek-V3-1"
- "DeepSeek-R1"
- "Gemini-2.5-flash"
- "Gemini-2.5-flash-nothinking"
- "Qwen3-235b"
# Permutable tool group configuration (for dynamic tool-order optimization)
permutable_tool_groups:
# Tools in the Candidate Evaluation stage can be in any order
candidate_evaluation_tools:
- "Tool: candidate_profile_analyzer"
- "Tool: candidate_evaluator_pro"
# Tools in the Interview Communication stage can be in any order
interview_communication_tools:
- "Tool: comprehensive_interview_material_generator"
- "Tool: email_template_generator"
# ============================================================================
# Usage - A2A architecture version
# ============================================================================
#
# 1. Trajectory hierarchy (tiered by type):
# - Level 1: SPAN (recruitment_orchestrator)
# - Level 2: SPAN (crew_execution)
# - Level 3: SPAN (analyze_job / evaluate_candidates / prepare_interviews)
# - Level 4: SPAN (a2a_call_job_analysis / a2a_call_candidate_evaluation /
# a2a_call_interview_communication)
# - Level 5: SPAN (job_analysis_server_execution /
# candidate_evaluation_server_execution /
# interview_communication_server_execution)
# - Level 6: Chain (Crew***.kickoff)
# - Level 7: AGENT (Senior roles for each stage)
# - Level 8: LLM / Tool (alternating)
#
# 2. Wildcards and exact matching:
# - SPAN names: must exactly match the names in this file
# - Chain name: "Crew***.kickoff"; evaluate_trajectory.py will automatically normalize
# the actual Crew_<UUID>.kickoff into this form
# - AGENT names: exact match the role fields in agents.yaml (e.g.,
# "Senior Job Requirements Analyst")
# - LLM: use "LLM: *" as a wildcard to match any model (gpt-5-chat-latest,
# deepseek-reasoner, gemini-2.5-flash, qwen3-235b-a22b, etc.)
# - Tool names: exact match tool names
#
# 3. Metric meanings (consistent with the MCP version):
# - Exact Match: Predicted trajectory exactly matches the reference trajectory in length and every step
# - In-order Match: Reference trajectory is a subsequence of the predicted trajectory, allowing extra steps
# - Any-order Match: Only requires the presence of all necessary steps, ignoring order
# - Precision: Proportion of steps in the predicted trajectory that are considered correct
# - Recall: Proportion of steps in the reference trajectory that are covered
# - Single-tool Use: Average usage of each tool in target_tools
# - unique_path_ratio: Number of unique complete trajectories / total samples
# - path_entropy: Shannon entropy based on trajectory frequencies, normalized to 0-1
#
# 4. How to use dynamic reference trajectories:
# - evaluate_trajectory.py automatically counts unified_web_search calls under analyze_job SPAN
# for each sample:
# * ≤ 3 → use reference_trajectory_3x
# * ≥ 4 → use reference_trajectory_4x
# - This treats models with fewer/more searches more fairly.
#
# 5. Run example:
# - Run in the RecruitmentAssistant-A2A directory:
# python3 evaluate_trajectory.py --config reference_trajectory.yaml
#
# 6. Possible differences vs. real A2A traces:
# - Real traces may include:
# * More LLM calls (thinking / planning / self-reflection)
# * Extra tools (e.g., multiple calls to candidate_profile_analyzer)
# * Orchestrator-level retries or business retries
# - These are not written into the reference trajectory, but will naturally be reflected in Precision/Recall;
# when analyzing results, focus on In-order Match, Recall, and path diversity metrics.
|