diff --git a/specs/restamp-a.md b/specs/restamp-a.md
new file mode 100644
index 0000000..f37c27e
--- /dev/null
+++ b/specs/restamp-a.md
@@ -0,0 +1,119 @@
+# RE-STAMP A — 862 ROWS RECOVERED BY HONEST JOIN
+
+**Date:** 2026-08-02 · served path, champion and reference ruler **byte-identical**
+(no application code changed) · database only.
+
+---
+
+## RESULT
+
+| | rows | books | settled | pending |
+|---|---:|---|---:|---:|
+| **RE-STAMPED** (`price_source='archive_restamp'`) | **862** | **862/862 takeable, 0 non-takeable** | 47 | **812** |
+| still quarantined | **144** | 0 takeable | 0 | 143 |
+
+**Holdout verification, exactly as the committed scripts select:**
+
+```
+holdout rows total ................................. 2,792
+  re-stamped rows INCLUDED ......................... 862
+  re-stamped but non-takeable (must be 0) .......... 0
+still excluded ..................................... 144
+quarantined rows leaked into holdout (must be 0) ... 0
+NON-TAKEABLE rows in the holdout since 2026-08-01 .. 0
+```
+
+**812 pending clean rows recovered** into the accruing verdicts, and the
+contamination is fully out of the measurement population.
+
+---
+
+## 🔴 862, NOT 936 — a deviation worth stating plainly
+
+The order authorised **936**, from the earlier finding that a takeable book posted
+the **same LINE**. Requiring what a re-stamp actually needs — **that takeable
+book's price for the GRADED SIDE, at LOCK TIME** — resolves **862**.
+
+**Why the other 74 were left out:**
+
+| reason | rows |
+|---|---:|
+| takeable side-price exists **only OUTSIDE the lock window** | **73** |
+| genuinely **one-sided** takeable market | 5 |
+
+*(overlapping; 74 distinct)*
+
+**I did not widen the window to reach 936.** A takeable price captured hours after
+the grade is **a later market moment, not a lock price**. Substituting it would be
+exactly the reconstruct-vs-join line this order was fenced against — and it would
+have been invisible in the totals, showing only as a cleaner-looking 936.
+
+Those 74 were also **re-tagged**, because their old label had become a lie:
+`recoverable_same_line` → **`no_takeable_lock_price_for_side`**. A future attempt
+reading the old tag would have been invited to widen the window and call it
+recovery.
+
+---
+
+## THE JOIN
+
+Same `player_key + stat + game_date + LINE`, **takeable books only**, from
+**`lock_lines`** (lock time — *not* `closing_captures`, which is the wrong moment
+for a lock price). Where several takeable books locked the same prop, the **latest
+lock at or before the grade**, deterministically.
+
+**`takeable` was RECOMPUTED, not carried over** — the old flag was computed *from*
+the contaminated price and was wrong on its own terms. **101 rows had their flag
+change**, which is the direct measure of how wrong it was.
+
+Four distinct takeable books supplied the recovered prices.
+
+## PROVENANCE
+
+`price_source` — `native_takeable` vs **`archive_restamp`** — travels with the
+data, on the same principle as `is_proxy` on the opportunity axis.
+
+**A value recovered by a later join is not identical in kind to one captured
+natively at grade time, even when it is the same number.** Any future bias check
+can segment on it; `NULL` predates the distinction and claims nothing.
+
+## WHAT STAYS OUT — 144 rows
+
+| tag | rows |
+|---|---:|
+| `no_takeable_lock_price_for_side` | 74 |
+| `no_takeable_quote` | 49 |
+| `takeable_line_differs` | 21 |
+
+**Re-stamping any of these would invent a lock price, or a line, we never
+captured.** None has settled, so nothing is lost from the record — only 143
+pending rows stay out of the holdouts, honestly.
+
+---
+
+## NEXT
+
+**(b) Combined structural-hardening order** — shared, tested guards for:
+- **`Number(null) === 0` → unknown ≠ zero** (6× recurring, most recently in my own
+  new module).
+- **Takeable by BOOK IDENTITY, never by price extremity or one-sidedness.** This
+  audit produced direct evidence for that invariant: **5 of the excluded rows are
+  one-sided takeable markets, and `betrivers`/`hardrockbet` legitimately quote
+  one side only.** A thin, juiced or one-sided takeable market is **REAL**, not
+  contamination — and a guard that inferred takeability from price shape would
+  throw away real markets while still admitting a DFS book at −119.
+
+**(c) `hits`** — inherits both guards, and must model thin/one-sided takeable
+markets as a **normal state**, not a clean two-sided assumption.
+
+**Carry-forward:** tb-v1 verdict · third pre-registered branch · **524 ≠ failure**
+(100s Cloudflare origin timeout vs a ~115s snapshot; the work completes).
+
+## TAGS
+
+**VERIFIED:** 862 re-stamped, 862/862 takeable · 0 non-takeable in the holdout
+since 2026-08-01 · 0 quarantined rows leaked · 144 still excluded · 101 takeable
+flags corrected · 812 pending rows recovered.
+
+**DEVIATION:** 862 not 936 — the 74 lack a takeable LOCK-TIME price for the graded
+side; the window was not widened to reach the authorised number.