engineering/arborist
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
arborist/bench
History
russell@unturf.com b6bb31a836
docs+code: ground §12 judge pipeline in the actual judge_code.py 
The §12.1 pipeline I added was second-hand from benchmark-matrix.md
and got several things wrong against the code:
- listed 4 verdicts; actual is 5 (missing FABRICATED — the
  fabrication-vs-WRONG split that energy-cogs §5.5 leans on for the
  qwen-fabricates / hermes-abstains finding)
- "θ=0.85" was right by accident — but it's the code-judge-pinned
  _CODE_JUDGE_THETA_CONTRA constant, raised from the manifest 0.5
  default after measuring 114 FPs in the 0.5-0.75 band
- omitted the short-answer entity-grounding fast path (which runs
  BEFORE NLI per the 2026-05-19 Poland-Tusk smoke)
- omitted the HYBRID rescue ladder (NLI entail / entity rescue /
  2026-05-21 verbatim-quote-on-topic rescue)
- conflated WRONG and FABRICATED (the subject-in-gold split is what
  distinguishes "source has the topic but a different value" from
  "source silent on the topic")

Rewrote §12.1 grounded in `bench/judge_code.py:judge()` (its own
docstring at line 501-528 is the truth on rule order), with file:line
citations and the verdict-mapping in full.

Also fixed a real artifact-vs-doc drift INSIDE the judge: the
module-top docstring still claimed θ_contra default 0.5 and omitted
the short-path and the WRONG/FABRICATED split. Updated to match the
authoritative judge() docstring + current code.

No behavior change — docstring + benchmarks doc only.
2026-05-22 19:17:28 -04:00
..
batteries #000025 §10.11 + §10.13 + §10.14 — close the 5F battery 2026-05-11 07:41:37 -04:00
fixtures #000052 §3.2.2 step 3 part B: real-context MIS-CITE fixtures (n=20) — bge-large catches 100% of mis-cite at strictly 0% real-STRICT FP (the Zionist failure mode is fully covered with margin) 2026-05-13 15:27:58 -04:00
results feat(#000060): bench/jaggedness.py — deterministic retrieval jaggedness instrument 2026-05-21 08:38:07 -04:00
scripts feat(#000049 §7 #28): tinygrad NLI backend + deterministic engine-agreement A/B; ONNX-immunity rationale 2026-05-19 12:34:04 -04:00
analyze_judge_disagreement.py feat(#000057): reconcile code judge against Opus — 4 calibrated rules 2026-05-19 18:14:38 -04:00
control_ab.py fix(#000057): control_ab header prints actual model/answer_mode/judge 2026-05-21 12:37:24 -04:00
control_sweep.py feat: STOCK V.1 two-mode config family + wire treatment arms to the pin 2026-05-21 10:15:26 -04:00
emergent_log.jsonl #000006 — +30 emergent cycles (2026-05-12); verifier-ladder health re-confirmed 2026-05-12 11:28:57 -04:00
es_delta.py feat: cross-language Q&A (Operation Sandwich) + Windows quickstart — all default-OFF 2026-05-18 12:12:23 -04:00
es_join_patterns.py feat: cross-language Q&A (Operation Sandwich) + Windows quickstart — all default-OFF 2026-05-18 12:12:23 -04:00
es_roundtrip_analysis.py feat: cross-language Q&A (Operation Sandwich) + Windows quickstart — all default-OFF 2026-05-18 12:12:23 -04:00
jaggedness.py feat(#000060): bench/jaggedness.py — deterministic retrieval jaggedness instrument 2026-05-21 08:38:07 -04:00
judge.py feat(#000057): fail-closed Opus judge gate — ARBORIST_JUDGE_ENABLE=1 to run 2026-05-19 17:26:06 -04:00
judge_code.py docs+code: ground §12 judge pipeline in the actual judge_code.py 2026-05-22 19:17:28 -04:00
load_monitor.py feat: bench/load_monitor.py — stdlib request-load monitor for single-slot endpoints 2026-05-21 09:20:53 -04:00
make_es_questions.py feat: cross-language Q&A (Operation Sandwich) + Windows quickstart — all default-OFF 2026-05-18 12:12:23 -04:00
make_lang_questions.py feat: cross-language Q&A (Operation Sandwich) + Windows quickstart — all default-OFF 2026-05-18 12:12:23 -04:00
mine_questions.py fix(#000057): solo arm never ran — chat_completion(model=) was omitted 2026-05-19 11:49:30 -04:00
prometheus_sigma_trigger_probe.py #000012 Phase 1c follow-through: wire #000037 §12 Trigger 1 probe to fork_score_branches 2026-05-11 06:56:09 -04:00
qa_questions.txt aborist/arborist 2026-05-07 09:31:49 -04:00
qa_questions_accent.txt feat(retrieval): accent-fold (+30pp recall@1) + fold-search factory hardening 2026-05-18 19:23:22 -04:00
qa_questions_accent_map.json feat(retrieval): accent-fold (+30pp recall@1) + fold-search factory hardening 2026-05-18 19:23:22 -04:00
qa_questions_amp.txt feat(retrieval): accent-fold (+30pp recall@1) + fold-search factory hardening 2026-05-18 19:23:22 -04:00
qa_questions_amp_map.json feat(retrieval): accent-fold (+30pp recall@1) + fold-search factory hardening 2026-05-18 19:23:22 -04:00
qa_questions_brit.txt feat(retrieval): accent-fold (+30pp recall@1) + fold-search factory hardening 2026-05-18 19:23:22 -04:00
qa_questions_brit_map.json feat(retrieval): accent-fold (+30pp recall@1) + fold-search factory hardening 2026-05-18 19:23:22 -04:00
qa_questions_canonical_witness_npower.txt three-thread session output: stale TODOs, N-power probe, ForkScore Phase 1c 2026-05-10 07:46:35 -04:00
qa_questions_es.txt feat: cross-language Q&A (Operation Sandwich) + Windows quickstart — all default-OFF 2026-05-18 12:12:23 -04:00
qa_questions_es_map.json feat: cross-language Q&A (Operation Sandwich) + Windows quickstart — all default-OFF 2026-05-18 12:12:23 -04:00
qa_questions_fr.txt feat: cross-language Q&A (Operation Sandwich) + Windows quickstart — all default-OFF 2026-05-18 12:12:23 -04:00
qa_questions_fr_map.json feat: cross-language Q&A (Operation Sandwich) + Windows quickstart — all default-OFF 2026-05-18 12:12:23 -04:00
qa_questions_honorific.txt feat(retrieval): accent-fold (+30pp recall@1) + fold-search factory hardening 2026-05-18 19:23:22 -04:00
qa_questions_honorific_map.json feat(retrieval): accent-fold (+30pp recall@1) + fold-search factory hardening 2026-05-18 19:23:22 -04:00
qa_questions_hyphen.txt feat(retrieval): accent-fold (+30pp recall@1) + fold-search factory hardening 2026-05-18 19:23:22 -04:00
qa_questions_hyphen_map.json feat(retrieval): accent-fold (+30pp recall@1) + fold-search factory hardening 2026-05-18 19:23:22 -04:00
qa_questions_metacog_subset.txt qa(#000011 + 4 more): SOFT_PREFLIGHT_HINT impl + 5-task fan-out 2026-05-03 23:00:56 -04:00
qa_questions_numeral.txt fix(#000057): restore numeral fixture clobbered by b573c59 (40 entries, not 5) 2026-05-19 11:12:12 -04:00
qa_questions_numeral_map.json fix(#000057): restore numeral fixture clobbered by b573c59 (40 entries, not 5) 2026-05-19 11:12:12 -04:00
qa_questions_progressive_and.txt bench: progressive-AND fixture + 2026-05-09 A/B baseline report 2026-05-10 06:35:18 -04:00
qa_questions_quantifier_baseline.txt bench(#000008): harness extension — FC rate, violation kinds, raw brackets 2026-05-02 18:35:08 -04:00
qa_questions_quantifier_subset.txt ticket(#000008): §12 dry-run bench findings + --policy harness flag 2026-05-03 08:39:20 -04:00
qa_questions_smoke.txt speed: pytest-xdist, bench smoke, concurrency default; UTF surrogate fix 2026-05-02 09:29:40 -04:00
qa_questions_stale.txt feat(#000057): parallel + incremental + huge-N control sweep 2026-05-19 13:00:45 -04:00
qa_questions_stale_map.json feat(#000057): parallel + incremental + huge-N control sweep 2026-05-19 13:00:45 -04:00
qa_questions_warrant_chain_aggressive.txt bench: aggressive warrant fixture confirms Phase 3 is rescue-only, not default-path 2026-05-10 09:53:21 -04:00
qa_questions_warrant_chain_paraphrase.txt bench: Phase 3 paraphrase fixture investigation — empirically dormant on current corpus 2026-05-10 10:06:23 -04:00
qa_questions_warrant_chain_probe.txt bench: #000031 Phase 3 A/B finds mechanism dormant on warrant-targeted fixture 2026-05-10 09:37:27 -04:00
qa_sweep.py bench/qa_sweep: scrub lone surrogates from the NLI-shadow answer_text/context fields before json.dumps 2026-05-12 17:04:33 -04:00
recall_at_k.py feat(retrieval): accent-fold (+30pp recall@1) + fold-search factory hardening 2026-05-18 19:23:22 -04:00
run.sh aborist/arborist 2026-05-07 09:31:49 -04:00
score_with_code_judge.py feat(#000057): bench/score_with_code_judge.py — code-judge rescore tool 2026-05-19 17:46:52 -04:00
stock_v1.py feat: STOCK V.1 two-mode config family + wire treatment arms to the pin 2026-05-21 10:15:26 -04:00
watt_bench.py fix(#000057): real token usage + cost per input/output separately 2026-05-21 11:41:01 -04:00
watt_calibrate.py feat(#000057): bench/watt_calibrate.py — separate prefill vs decode energy 2026-05-21 11:48:36 -04:00
watt_probe.py fix(#000057): measure power STATES, not a duty-cycle blend; guarantee cache miss 2026-05-21 10:58:56 -04:00