Workflow: Review Entire Work

Workflow: Review Entire Work

This workflow orchestrates a full "World-Class" review of the project by simulating five distinct expert personas.

Phase 1: Implementation Audit (Role: Auditor)

Reference Workflow: /​strategic_audit Focus: Contracts & Standards Tasks:

1. Contract Check: Compare docs/​contracts/ against src/. Ensure absolute compliance.
2. Sanity Check: Verify if random_state is fixed everywhere (Configs vs Code).
3. Artifact Check: Check results/ artifacts. Do they match the experiment_contract.md?

Phase 1.5: Pipeline Integrity & Alignment (Role: Architect)

Focus: Code Structure & Data Flow Tasks:

1. End-to-End Trace: Follow the data journey from dataset/ → loader.py → preprocessing.py → trainer.py → results/. Are there broken links or unused files?
2. Interface Consistency: Do the outputs of the Preprocessor (e.g., sparse matrices, array shapes) strictly match what the Models expect?
3. Logic Alignment: Does the actual python code in src/​pipelines match the theoretical flow described in the documentation?
4. Hidden Side-Effects: Look for "spaghetti code" or global state mutations that might cause invisible bugs (e.g., modifying dataframes in place unexpectedly).

Phase 2: Statistical Rigor Check (Role: Reviewer)

Reference Workflow: /​reviewer Focus: Numbers & Validity Tasks:

1. Analyze results/​tables/​aggregated_summary.csv and results/​experiment_log.csv.
2. Critical Check: Are Seed_count values > 1? Are there Confidence Intervals or Standard Deviations?
3. Significance Check: Do the "State-of-the-Art" claims hold up? If std is missing or high, flag it immediately.
4. Hard Truth: Are the improvements statistically significant or just random noise?

Phase 3: Paper Narrative Review (Role: Author)

Reference Workflow: /​author (or /​paper_review) Focus: Storytelling & Claims Tasks:

1. Read the latest PDF/​TeX draft in paper/ or the current task.md / README.md.
2. Alignment Check: Does the narrative align with the Phase 1, 1.5 & 2 findings? (e.g., claiming "Robust" when N=1).
3. Mitigation Strategy: If results are weak, propose specific pivots for the paper's contribution.

Phase 4: Meta-Review & Decision (Role: Area Chair)

Focus: Final Decision & Strategy Tasks:

1. Synthesis: Review the findings from Phases 1-3.
2. Critique the Reviewers: Was the Auditor too pedantic? Did the Architect miss a practical workaround?
3. Final Verdict: Provide a formal decision: Strong Reject, Reject, Weak Accept, Accept.
4. Action Plan: Generate a prioritized, bulleted list of immediate actions to fix Critical Flaws and move the paper to "Accept".