Jyotisha

2,667 Commits 64 Branches 0 Tags

3 Commits

Include renames

Author	SHA1	Message	Date
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
Jesse_ChenandClaude Fable 5	b56147e524	docs(tasks): varga sensitivity weighting, and birth-time provenance labels 与上游流程对照补两处。①精度研究单加 M1b：上游有一张分盘分钟敏感度表（D60=2 / D30=4 / D24=5 / D9=13.3 / D1=120 分钟换一次上升），而本仓 11 张分盘等权再除以 22（且没有 D60）——20 分钟窗里 D30 和 D2 同分量，分辨率最高的盘被稀释。测 V1 反比配权 / V2 按窗宽选盘 / V3 加 D60，并报过拟合风险。 ②新任务书 BUG-690：录入已收 birth_time_source 三类，但校正链只用过一次，推算值与医院记录等同对待、输出直称「你的出生时间」。接进投影并按来源分档措辞；医院记录与校正结果冲突时怎么说挂给产品。 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0155nFCgCHtoA7jhSDGmZmMu	2026-09-14 14:44:15 +00:00
Jesse_ChenandClaude Fable 5	749c7541ad	docs(tasks): research the precision-adaptive probe gate, and ask for exact dates 出生时间每差 1 分钟，Vimshottari 边界只平移约 1.1 天；而 _boundary_windows 用固定 45 天（刷新 30 天）闸门，且全文不读证据精度。20 分钟窗内候选边界全挤在三周内被整批丢掉——这就是「六题之后引擎出不来题」的算术原因。研究单量闸门按年/月/日分档的收益，必测 ±3/±7/±14 天记错时真值是否被挤出（挤出即不推荐）。v4 实测有 59 件日精度事件可用。BUG-689 的邀请文案同步加 D1b：优先请用户说记得确切日期的事，但不得承诺能定到分钟。 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0155nFCgCHtoA7jhSDGmZmMu	2026-09-14 14:36:10 +00:00

Author

SHA1

Message

Date

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

Jesse_ChenandClaude Fable 5

b56147e524

docs(tasks): varga sensitivity weighting, and birth-time provenance labels

与上游流程对照补两处。①精度研究单加 M1b：上游有一张分盘分钟敏感度表
（D60=2 / D30=4 / D24=5 / D9=13.3 / D1=120 分钟换一次上升），而本仓 11 张分盘
等权再除以 22（且没有 D60）——20 分钟窗里 D30 和 D2 同分量，分辨率最高的盘被
稀释。测 V1 反比配权 / V2 按窗宽选盘 / V3 加 D60，并报过拟合风险。
②新任务书 BUG-690：录入已收 birth_time_source 三类，但校正链只用过一次，
推算值与医院记录等同对待、输出直称「你的出生时间」。接进投影并按来源分档措辞；
医院记录与校正结果冲突时怎么说挂给产品。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155nFCgCHtoA7jhSDGmZmMu

2026-09-14 14:44:15 +00:00

Jesse_ChenandClaude Fable 5

749c7541ad

docs(tasks): research the precision-adaptive probe gate, and ask for exact dates

出生时间每差 1 分钟，Vimshottari 边界只平移约 1.1 天；而 _boundary_windows 用固定
45 天（刷新 30 天）闸门，且全文不读证据精度。20 分钟窗内候选边界全挤在三周内被
整批丢掉——这就是「六题之后引擎出不来题」的算术原因。研究单量闸门按年/月/日分档
的收益，必测 ±3/±7/±14 天记错时真值是否被挤出（挤出即不推荐）。v4 实测有 59 件日
精度事件可用。BUG-689 的邀请文案同步加 D1b：优先请用户说记得确切日期的事，但不得
承诺能定到分钟。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155nFCgCHtoA7jhSDGmZmMu

2026-09-14 14:36:10 +00:00