docs: free proof surface recorded (/record, hollow-preserving, CLV honest-absent)

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QJs13VsyiSKYQP6rj3NNmc
2026-07-31 07:56:08 -04:00
parent 5930f18d81
commit e8b15c705a
2 changed files with 51 additions and 0 deletions
@@ -764,3 +764,21 @@ gated rows because a result is a fact, not a judgment.
 a leak; `max-age=30` on this endpoint means post-deploy checks must bust the cache.

 319 suites / 3,970 tests green, web build exit 0.
+
+---
+
+# FREE PROOF SURFACE — `/record` (shipped 2026-07-31)
+
+Tier-record-forward, presentation-only (`src/` untouched). Reads `/api/accuracy` + `/api/ledger/model`
+and prints them as-is: **B 60% n512, C 57% n413 with C honestly below B; A/D/F hollow with real
+samples**. Sport slicing is client-side (the endpoints ignore `?sport=`).
+
+**The rule that matters:** a withheld percentage stays null — A is 1/2 and a test asserts we do **not**
+derive 50%. **CLV renders an honest absence**: `beat_close_pct` is null behind the capture-reliability
+guard, so the panel says NOT PUBLISHED YET and explains why, and tests forbid the computable-but-wrong
+34/937 = 3.6% fallback. Calibration and accuracy-over-time are **named as held**, not faked — no honest
+source exists yet.
+
+Cache-busted fingerprint: page served, CLV honest-absent, guard named, held pieces named, **no "3.6"**,
+**no hard-coded percentage in the markup**. The hollow-explanation line is client-conditional → Chrome
+audit. 320 suites / 3,986 tests green, web build exit 0.