session 2026-06-22: L1 N6 replay-wedge fix + PR #175 (clone-proven); Gwern/MemPalace/ARC studies

claude/memory/skill-harvest-register.md

@@ -220,3 +220,14 @@ The single place proposed skills live so they don't evaporate between sessions.
 | **`/wake-up` patch — surface the *why* when the thread touches the engine's purpose** | patch | When the active workstream is studium-engine / The Making / ARC-as-public-proof (the engine's reason-for-being), `/wake-up` should **actively weave the load-bearing telos + steward-formation "why"** into the briefing — not just leave the pointers in MEMORY.md. **Yardstick met, explicitly and painfully:** the steward had to re-disclose the chamber's childhood origin *"many times, across many threads … it clearly needs to be said again"* — the memory failing at its one job. Concrete: add to §2.a a conditional read of [[project-studium-engine-telos-chamber-of-voices]] + [[user-formation-flamenco-substrate]] when the thread is engine/Making/ARC-purpose, and one line in §3 holding the why. **Caveat (honest):** the prominent MEMORY.md pointers added this session may already largely close the gap, since wake reads MEMORY.md — so this is belt-and-suspenders, low-urgency, for steward judgment. Do NOT recite the why every wake (decorative); only when the thread touches it. | `/wake-up` §2.a + §3 | **PROPOSED** |
 
 *(One proposal, low-urgency. The pain it answers is real — losing the most important thing across threads is the exact failure the engine exists to end — but the file-prominence fix landed this session may suffice; surfaced for the steward to judge whether the wake-up patch adds enough over the pointers to be worth the weight.)*
+
+### Harvest 2026-06-22 (L1 N6 wedge fixed + clone-tested + PR #175)
+
+| Element | Kind | One-line | Where it lands | Status |
+|---|---|---|---|---|
+| **`/clone-test-runtime-fix` (CoW-clone + isolated-worktree harness)** | create skill OR ladder | When testing a runtime fix that needs **real live data** but must not touch the live instance: `/bin/cp -c -R` (APFS CoW) the live data dir to `/tmp`, `git worktree add --detach` the branch + symlink `node_modules`, `npm run build` in the worktree, run against the clone on an **alt port** with a separate `BM_DATA_DIR`. Live `dist/` + live instance untouched; the fix is proven against the worst-case real graph. Proven today (N6 fix proven on a clone of mindfabric-00's 688k-graph; PID 747 never touched). Recurs for any L1/BMF runtime fix. | new `/clone-test-runtime-fix` skill OR `reference-verification-ladder` | **PROPOSED** |
+| **verification-ladder: verify the RUNNING BINARY's provenance, not just source HEAD** | verification-ladder entry | Before reasoning about live behaviour, verify what the running process actually executes — compiled `dist/` build mtime + grep the compiled symbols — not just `git HEAD`. Today the whole "schema-drift" mechanism flipped on this: HEAD had the A1'' migration, but the running `dist/` was a **2026-05-24 build predating it** (zero A1'' symbols in dist) → "migration didn't fire" was wrong; "code never deployed" was right. Generalises [[live-state-discipline]] one layer down (source-as-committed ≠ code-as-running). | `reference-verification-ladder` | **PROPOSED** |
+| **Symmetria §3 flag: probe-confirms-hypothesis** | Symmetria §3 flag | A query/test **I constructed to match my hypothesis**, whose result I then read as *confirming* the hypothesis rather than testing whether the system actually does that. Caught today: I hand-wrote `EXPLAIN … WHERE coherence_evaluated=0`, saw it `SCAN`, and reported it as a live wedging code-path — but the build-map agent found **no code runs that query**. The probe matched my theory; the codebase didn't. Antidote: a constructed probe tests the *probe's* behaviour, not the *code's* — verify the code actually issues the query before citing the probe as evidence. Kin to `causal-story-before-reading-render`, specialised to self-authored SQL/test probes. | Symmetria §3 | **PROPOSED** |
+| **verification-ladder: quantify-the-removed-cost as the A/B control** | verification-ladder entry | When a fix *removes* a hot operation, the cleanest control isn't a flaky end-to-end before/after race — it's to **time the exact removed operation on real data**. Today: timing the wedging `json_each` scan on the real 676k graph = **2.56 s each** (vs 0.003 s indexed) × ≤20/event = the wedge, quantified decisively in seconds, no full-replay race needed. Bounded, reproducible, and the number goes straight into the PR. | `reference-verification-ladder` | **PROPOSED** |
+
+*(Four, all earned in use today on the N6 work. The clone-test harness and running-binary-provenance are the load-bearing two — the first is a reusable safe-test pattern for the steward's live substrate; the second flipped a wrong root-cause to the right one. The probe-confirms-hypothesis flag is a clean new §3 shape [self-authored probe ≠ code behaviour]. Minor/reinforcing: "assert-process-dead-from-a-pgrep-pattern-miss" recurred today [declared PID gone; it was alive, my pattern just didn't match the cmdline] — reinforces the existing `census-through-a-pattern` flag, no new entry needed. None created autonomously — surfaced for steward authorization.)*