Add marriage benchmark summary bridge

2026-06-28 17:19:30 +08:00
parent 838f5d97b7
commit 42c0d5e068
4 changed files with 267 additions and 0 deletions
@@ -12,9 +12,11 @@ This file is the small index for the current engineering fronts that still drive
 ## Relationship Adjudication

 - `/Users/wuyongnaren/Documents/印度占星/docs/research/marriage_adjudicator_first_pass_audit_2026_06_27.md`
+- `/Users/wuyongnaren/Documents/印度占星/docs/research/marriage_benchmark_summary_bridge_audit_2026_06_28.md`
 - `/Users/wuyongnaren/Documents/印度占星/docs/research/isolated_asset_bridge_audit_2026_06_28.md`
 - `/Users/wuyongnaren/Documents/印度占星/references/event_judgment_marriage.md`
 - `/Users/wuyongnaren/Documents/印度占星/docs/superpowers/specs/2026-06-28-jaimini-marriage-bridge-v1-design.md`
+- `/Users/wuyongnaren/Documents/印度占星/scripts/marriage_benchmark_summary.py`

 ## Wealth Adjudication

@@ -0,0 +1,59 @@
+# Marriage Benchmark Summary Bridge Audit - 2026-06-28
+
+## Scope
+
+This pass turns the v6.1 marriage verification dataset from a large historical JSON asset into a small adjudicator-ready summary.
+
+## Entrypoint
+
+- `/Users/wuyongnaren/Documents/印度占星/scripts/marriage_benchmark_summary.py`
+- Source dataset: `/Users/wuyongnaren/Documents/印度占星/tests/test-data/verify-results-v6.1.json`
+
+## Why This Was Needed
+
+The v6.1 benchmark already contains:
+
+- 18 public cases
+- 26 marriage events
+- Rao P1-P8 hit data
+- UL / Argala / D7 / D60 / Karakamsha context
+- divorce markers
+
+But the useful event statistics were buried inside a large JSON file and a generated report. The new summary helper makes the benchmark reusable by strict adjudicators without requiring every agent to manually parse the full file.
+
+## Current Summary
+
+- `case_count = 18`
+- `ascendant_match_count = 18`
+- `marriage_event_count = 26`
+- `divorce_event_count = 15`
+- Rao hit distribution:
+  - `2/8`: 2 events
+  - `3/8`: 2 events
+  - `4/8`: 9 events
+  - `5/8`: 8 events
+  - `6/8`: 2 events
+  - `7/8`: 3 events
+
+## Label-Lift Seed Cases
+
+The helper exposes the strongest `label_lift_failure_seed` candidates:
+
+- `Britney Spears|Kevin Federline|2004-10-06`
+- `Elon Musk|Justine Wilson|2000-01-01`
+- `Tom Cruise|Katie Holmes|2006-11-18`
+- `Albert Einstein|Mileva Maric|1903-01-06`
+- `Nelson Mandela|Winnie Madikizela|1958-06-14`
+
+These are high-signal calibration targets for future relationship adjudicator work.
+
+## Boundary
+
+- The helper does not recompute astrology.
+- It does not alter the source dataset.
+- It does not promote `legal_marriage` or any other label by itself.
+- It only exposes benchmark evidence in a stable shape.
+
+## Regression Coverage
+
+- `/Users/wuyongnaren/Documents/印度占星/tests/test_marriage_benchmark_summary.py`