- 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
Carry explicit local date intervals instead of inferring the day from clock
order. Cluster width, delivery, adoption, and reports keep the actual civil
date; adopted date is stored separately from the reported birth_date.
Algorithm identity is scoring-9 / spec-v5. Scoring weights, confirmation
thresholds, and Skill version are unchanged. Isolated Linux final-3 gates
passed; four pre-existing Python failures remain. This is not a production
release.
Add date-isolated caches and regression coverage, align scoring identity, and freeze full research reruns while preserving historical artifacts. Record unresolved cache/receipt identity and end-to-end acceptance gaps for branch review only.
Co-Authored-By: Claude Code <noreply@anthropic.com>
20 public AA cases, raman/mean. Lowering MIN_BOUNDARY_DAYS narrows ±10 from
15 to 11 minutes but drops top-1 0.80→0.75; wider radii get wider ranges.
Varga sensitivity weights (V1/V2) match production; D60 (V3) hurts ±10.
Day-precision events offset ±7 never squeeze the true minute out. Production
45/30 gate and equal varga weights stay unchanged.
Offline v4 holdout probe. Last round's width=window result was the
no-elimination metric; production still-valid ranges after six answers
are 15/33/56 minutes. Adjacent merge never fires under step-2 radii.
W1/W2 match baseline; W3 is uncertain after one coverage squeeze.
No production clustering or scoring defaults changed.
Build holdout v4 from the public AA set, correct the two v3 dates, and sweep
R1–R5 plus pairs offline. No production scoring defaults change. No
implementation brief: delivered width stays the full window on every radius.
Offline H1-H4 measurement on 20 public AA holdout cases.
Block unique top-1 did not rise; H4 made it worse; minute layer unchanged.
Leave production scoring untouched. Transits must not drive minute conclusions.
