Jesse_ChenandClaude Opus 5.5 0f5442cea2 research(rectification): offline R1/R2/R3 — answer-flip tolerance, V1/V2 rerun, dasha shift arithmetic ...

- R1: flipping 1 answer keeps truth in range 98-100% but cuts head hit by
  a third or more; 2 flips squeeze truth out in 7-10% of ±30/±60 replays
  (two flips = 8 points = SEPARATION_LEAD).
- R2: weights do apply (research scorer == production at V0); V1/V2 are
  identity at ±30/±60 by construction and leave six-question metrics
  unchanged at ±10 -> no_benefit (measured). Supplementary V1n does not
  pass the gate.
- R3: boundary shift is ~3.8 days/minute (1.3-5.9), not 1.1; the 45-day
  gate is ~8-34 minutes. The _representative_pairs hypothesis is refuted
  (all-pairs adds no dated probes); the bottleneck is monthly evaluation.
  New finding recorded as BUG-1048 (investigating): _boundary_windows
  year-straddle exemption and positional zip misalignment bypass the gate.
- Dated errata appended (no deletions) to the 09-14/09-16 briefs and
  research docs; README board row -> 待验收. No production code, scoring,
  thresholds, gates or Skill changed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8

2026-09-26 14:47:48 +08:00

..

jev_intent_samples

research(jev-intent): 修复轮重造语料并全量对照

2026-09-19 11:30:31 +08:00

answer_flip_tolerance.py

research(rectification): offline R1/R2/R3 — answer-flip tolerance, V1/V2 rerun, dasha shift arithmetic

2026-09-26 14:47:48 +08:00

cluster_width_lib.py

docs(research): measure why delivered range width equals the search window

2026-09-14 22:11:47 +08:00

cluster_width_probe.py

docs(research): measure why delivered range width equals the search window

2026-09-14 22:11:47 +08:00

dasha_shift_per_minute.py

research(rectification): offline R1/R2/R3 — answer-flip tolerance, V1/V2 rerun, dasha shift arithmetic

2026-09-26 14:47:48 +08:00

fewer_probes_card_replay.py

feat(rectification): 少问几道就出卡，交付卡以范围为主（D1–D4）

2026-09-26 14:27:12 +08:00

guided_collect_holdout_replay.py

fix(rectification): 有题就接着问，题问完才出卡；引导窗口不再硬贴领域

2026-09-16 12:16:55 +00:00

holdout_v4_build.py

docs(research): measure minute-resolution scoring; no variant clears all radii

2026-09-14 20:43:27 +08:00

house_lord_gochara_supply.py

docs(research): measure house-lord gochara rules with no benefit

2026-09-14 01:07:11 +08:00

jev_intent_corpus_build.py

research(jev-intent): 修复轮重造语料并全量对照

2026-09-19 11:30:31 +08:00

jev_intent_current.py

research(rectification): DeepSeek Flash 抽 33% 对照现行分类器提示词

2026-09-19 09:45:29 +08:00

jev_intent_probe.py

research(jev-intent): fix2 把来源 B 现行与高置信错误补进报告

2026-09-19 12:28:43 +08:00

jev_intent_questions.py

research(rectification): Jev 意图分类离线对照，结论不可接

2026-09-19 09:27:10 +08:00

midnight_date_anchor_regression.py

fix(rectification): anchor candidate windows to civil dates across midnight

2026-09-21 02:55:00 +08:00

minute_resolution_lib.py

docs(research): measure minute-resolution scoring; no variant clears all radii

2026-09-14 20:43:27 +08:00

minute_resolution_sweep.py

docs(research): measure minute-resolution scoring; no variant clears all radii

2026-09-14 20:43:27 +08:00

offline_research_20260926_lib.py

research(rectification): offline R1/R2/R3 — answer-flip tolerance, V1/V2 rerun, dasha shift arithmetic

2026-09-26 14:47:48 +08:00

precision_gate_lib.py

docs(research): measure precision-adaptive probe gates; no variant clears the bar

2026-09-15 09:04:03 +08:00

precision_gate_sweep.py

docs(research): measure precision-adaptive probe gates; no variant clears the bar

2026-09-15 09:04:03 +08:00

probe_supply_after_six.py

docs(research): measure dated probe supply after six answers

2026-09-13 12:59:46 +08:00

reported_offset_sweep.py

fix(rectification): anchor candidate windows to civil dates across midnight

2026-09-21 02:55:00 +08:00

representative_pairs_probe.py

research(rectification): offline R1/R2/R3 — answer-flip tolerance, V1/V2 rerun, dasha shift arithmetic

2026-09-26 14:47:48 +08:00

sealed_holdout_rerun.py

fix(rectification): anchor candidate windows to civil dates across midnight

2026-09-21 02:55:00 +08:00

varga_sensitivity_rerun.py

research(rectification): offline R1/R2/R3 — answer-flip tolerance, V1/V2 rerun, dasha shift arithmetic

2026-09-26 14:47:48 +08:00