feat: add rectification input evidence contract

This commit is contained in:
732642856
2026-07-21 17:49:43 +08:00
parent 8df4a74fec
commit 31dc3a9b97
11 changed files with 261 additions and 16 deletions
@@ -0,0 +1,9 @@
# Minute Rectification P0 Plan
1. Add failing tests for canonical input hashes, candidate fingerprints and
pending stability probes.
2. Add a small shared input-contract helper and wire it into the sensitivity
scanner and private three-engine receipt.
3. Correct public-case coordinate defaults and test them.
4. Add a semantic evidence-hash helper without replacing raw artifact hashes.
5. Run focused tests and inspect the resulting JSON contracts.
@@ -0,0 +1,32 @@
# Minute Rectification P0: Input Evidence Contract
## Goal
Make a candidate-minute calculation reproducible across the local scanner and
the three-engine receipt without changing scoring weights or confirmation rules.
## Scope
- Canonicalize birth input with Lahiri and mean-node defaults.
- Hash the canonical input for every scanned candidate minute.
- Publish the required `-5`, `-2`, `-1`, `+1`, `+2`, and `+5` minute
stability probes as pending evidence, not as a confidence result.
- Correct the public Steve Jobs comparison defaults to the canonical San
Francisco coordinates.
- Preserve raw evidence hashes and add a separate semantic hash for known
order-insensitive aspect lists.
## Non-goals
- No new questionnaire format, scoring weights, or candidate confirmation path.
- No claim that a candidate is accurate to the minute.
- No raw birth input or external raw response exposed to the browser.
## Acceptance
- Equivalent input mappings have the same canonical hash.
- Every candidate row carries a distinct input fingerprint when the minute
differs.
- Stability probes are explicit and always state that minute confirmation is
blocked pending public blind holdout evidence.
- The public comparison defaults match the canonical Steve Jobs source.