Repository navigation
Description
Task
T-105b· Milestone M10 — Evaluation · Size M · Priority P0 · Epic #10Implements FR-EXP-003
Branch
spike/t-105b-run-exp-01-main-comparison— one task = one branch = one PR (doc 32 §6)
Summary
Execute EXP-01 and EXP-03 on k2d, produce reports, and evaluate SC-1 (MET-03 ≤ 50 % of random, p < 0.05), SC-2 (Â12 ≥ 0.64 on MET-02) and SC-3 (GT-A found in ≥ 90 % within 200) in docs/evidence/EXP-01.md, EXP-03.md.
Context / motivation
Answers the research question (doc 19 §1) with pre-registered statistics.
Scope
In scope
- Report negative or mixed results transparently.
Implementation guidance
- Plan compute: EXP-01 ≈ 48 000 k2d runs (doc 19 §4); size
ROBOPROBE_K2D_REPLICASandexperiments.max_concurrent_campaignsto the machine; checkpoints allow stop/resume across sessions. - After completion, generate reports and compute SC-1 (MET-03 median ratio PROBE/RANDOM ≤ 0.5, MWU p < 0.05), SC-2 (Â12(PROBE, RANDOM) on MET-02 ≥ 0.64) and SC-3 (fraction of PROBE seeds finding GT-A within 200 runs ≥ 0.9, with Wilson CI), using the reference set (T-103) for mode matching.
- EXP-03 ablation: each variant vs full PROBE with Holm correction across variants.
- State an explicit PASS/FAIL per criterion; negative results are reported, not hidden (doc 19 §5).
- Standards: respect the layering contracts (doc 16 §2) and conventions in doc 34; no hard-coded tunables — thresholds/defaults come from typed settings (doc 27); no magic numbers (ruff
PLR2004); public APIs typed (mypy strict) and documented.
Configuration keys
From the configuration reference (doc 27). Keys marked (new) must be added to the typed settings models and appear in the generated reference (T-003c).
experiments.reference_set.nexperiments.reference_set.seedexperiments.mode_match_radiusexperiments.bootstrap_resamples
Acceptance criteria
- Evidence docs with explicit pass/fail per SC criterion and links to raw data.
- SC table with numbers, CIs and verdicts.
- Raw data hashes recorded.
- Tests listed below added and green in CI; coverage of new code meets the gate in doc 18 §3.
- Docs, config reference and CHANGELOG
Unreleasedupdated where behaviour, keys or interfaces changed. - Definition of Done (doc 32 §3) satisfied.
Tests required
- N/A (evaluation).
Dependencies
Blocked by: #206 (T-104), #141 (T-079), #128 (T-103)
Blocks: #209 (T-105c), #222 (T-121)
Parallelization notes
Files / paths owned by this task — avoid editing other paths; shared files (pyproject.toml entry points, cli/main.py, api/routers/__init__.py, router.tsx, docker/compose.yaml, ci.yml) are append-only:
docs/evidence/EXP-01.mddocs/evidence/EXP-03.md
Long-running; other M10 work proceeds alongside.
Definition of Done
Per doc 32 §3 and AGENTS.md §5: lint, format, mypy strict and import-linter green; tests added and passing; docs updated; PR template completed with Closes #<this issue>; Conventional Commit messages referencing T-105b (doc 16 §6); commits authored only by the repository owner identity (AGENTS.md §4); CI green; reviewer checklist (doc 32 §5.1) satisfied.
References
Design docs: doc 19 §1 · doc 19 §4 · doc 19 §5 · doc 19 §6 · doc 01 §7
Requirements: FR-EXP-003 (traceability: doc 28)
Metadata
Metadata
Assignees
Labels
Projects
- StatusShow more project fieldsBacklog
Activity
configs/experiments/*.yaml) validated in CI #206fix(experiments): write the reference Parquet integrity manifest (#348)