[Stack 14/17] Fix D7: match Clojure repness metric formula (product of 4 signed values)

[jucor]

Summary

Changes the representativeness metric from a weighted sum of absolutes to the
Clojure product formula (repness.clj:188-190).

Before (Python):

• agree_metric = pa * (|pat| + |rat|) — weighted sum, tolerant of weak factors
• disagree_metric = (1 - pd) * (|pdt| + |rdt|) — doubly wrong: uses (1-pd) and sum

After (Clojure formula):

• agree_metric = ra * rat * pa * pat — product of 4 signed values
• disagree_metric = rd * rdt * pd * pdt — product of 4 signed values

The product formula is more conservative: any factor near zero kills the entire
metric, requiring ALL dimensions (probability, significance, relative
representativeness) to be strong simultaneously.

The old disagree formula was doubly wrong:

1. Used (1 - pd) instead of pd — high metric when disagree probability is LOW
2. Used a weighted sum of absolutes instead of a signed product

No feature flag for the old formula — it has no defensible behavior.

Test plan

• 5 new D7 formula tests (agree product, disagree product, zero-kills-metric,
  sign preservation, multiple known values)
• Updated unit tests in test_repness_unit.py and test_old_format_repness.py
• Full test suite passes (excluding DynamoDB/MinIO tests)
• Private dataset tests pass (--include-local)
• Golden snapshots re-recorded for all 7 datasets

🤖 Generated with Claude Code

Squashed commits

• Add PR description and update plan/journal for D7 fix

commit-id:80eaa87c

---

Stack:

• [Stack 17/17] Fix K-means k divergence: preserve vote-encounter row order #2524
• [Stack 16/17] Fix D15: match Clojure moderation handling (zero out columns, don't remove) #2523
• [Stack 15/17] Fix D8: match Clojure repful classification (rat > rdt) #2522
• [Stack 14/17] Fix D7: match Clojure repness metric formula (product of 4 signed values) #2521 ⬅
• [Stack 13/17] Fix D6: match Clojure two-proportion test formula (+1 pseudocount) #2520
• [Stack 12/17] Fix D5: match Clojure prop_test formula (Wilson-score-like with +1 pseudocount) #2519
• [Stack 11/17] Fix D9: z-score thresholds from two-tailed to one-tailed #2518
• [Stack 10/17] Fix test DB connection: use DATABASE_URL with dotenv #2517
• [Stack 9/17] Vectorize participant info computation (3-15x speedup) #2516
• [Stack 8/17] Speed up regression tests #2515
• [Stack 7/17] Fix D4: pseudocount formula #2514
• [Stack 6/17] Fix D2: in-conv participant threshold + D2c vote count source #2513
• [Stack 5/17] Per-discrepancy test infrastructure #2512
• [Stack 4/17] Deep analysis of Python-Clojure discrepancies and fix plan #2511
• [Stack 3/17] Speed up CI: replace pip with uv pip in Dockerfile (~2x faster installs) #2510
• [Stack 2/17] Add SKIP_GOLDEN env var to disable golden snapshot tests #2509
• [Stack 1/17] CI: add concurrency group to E2E workflow; run CI on spr/ branches #2508

---

⚠️ Part of a stack created by spr. Do not merge manually using the UI - doing so may have unexpected results.