Tyler-R-Kendrick · Tyler-R-Kendrick · Jul 27, 2026 · Jul 25, 2026 · Jul 25, 2026
diff --git a/docs/brains/repo/MOC.md b/docs/brains/repo/MOC.md
@@ -47,6 +47,16 @@ reads this section to seed the autotrain hypothesis loop and prunes it as
   space + complete bridge supervision? (see valid-edit flow project)
 - [[recursive-depth-supervision-objective]] — arithmetic/compatibility contract
   is supported; real-suite mode selection remains open.
+- [[recursive-recurrence-health]] — **updated 2026-07-25 (SLM-421)**: the n=2
+  SLM-282 negative was underpowered noise, not a robust contraction
+  violation. A powered n=20 Wilson-interval rerun (fresh disjoint seeds,
+  preregistered `min_pass_rate=0.5` locked before observing them) found
+  `recursive_core_positive` (18/20, Wilson 95% CI [0.699, 0.972]). This
+  satisfies only one of PR #853-#856's two LAR3 reopening conditions; LAR3
+  stays closed pending a powered rerun of SLM-317's value gate (blocked —
+  that harness is not yet merged to `main`). Next open question: land
+  SLM-317's harness on `main` (via its own PR #852) so its value gate can
+  get the same powered-rerun treatment.
 
 ## Dead ends (do not re-propose)
 
@@ -55,5 +65,3 @@ hypothesis loop's novelty audit can exclude them.
 
 - _(seed)_ E249 / E252 local-preference chain — negative; see
   [`research-lineage.md`](../../design/research-lineage.md) "Exact-state local decision preference".
-- [[recursive-recurrence-health]] — negative: seed 1 / R=4 / example `b`
-  regressed at the final depth, so the LAR3 activation prerequisite failed.
diff --git a/docs/brains/repo/recursive-recurrence-health.md b/docs/brains/repo/recursive-recurrence-health.md
@@ -1,10 +1,11 @@
 ---
 type: concept
-status: dead-end
+status: active
 tags: [recurrence, diagnostics, fixture]
 created: 2026-07-23
-linear: SLM-282
-design: docs/design/iter-slm282-recurrence-health-20260723.md
+updated: 2026-07-25
+linear: SLM-282, SLM-421
+design: docs/design/iter-slm282-recurrence-health-20260723.md, docs/design/iter-slm282-recurrence-health-powered-rerun-20260725.md
 sources: "[[deeploop-source]], [[training-free-looped-transformers-source]], https://arxiv.org/abs/2106.14342"
 ---
 
@@ -32,9 +33,31 @@ non-finite, or if matched initialization/data/optimizer controls diverge.
 
 ## Status & next step
 
-Negative. Only seed 0 passed. For seed 1 / R=4 / example `b`, CE regressed from
-`17.487688` at depth 3 to `17.855324` at depth 4, so the two-seed prerequisite
-failed even though the token-weighted aggregate improved. All required
-telemetry was finite, but several local directional gains also exceed one.
-Do not activate LAR3 from this result; no ship, checkpoint, or
-production-default claim follows.
+**SLM-282 (n=2, 2026-07-23): negative.** Only seed 0 passed. For seed 1 / R=4 /
+example `b`, CE regressed from `17.487688` at depth 3 to `17.855324` at depth
+4, so the fixed two-of-two-seed prerequisite failed even though the
+token-weighted aggregate improved. All required telemetry was finite, but
+several local directional gains also exceed one. This n=2 record is immutable
+and remains the historical evidence — PR #853/#854/#855/#856 correctly cited
+it (alongside SLM-317's inconclusive value gate) to gate-close LAR3-01..04 as
+`not_authorized`.
+
+**SLM-421 (n=20, powered rerun, 2026-07-25): `recursive_core_positive`.** The
+fixed 2-seed rule could not distinguish a real contraction violation from
+fixture-scale noise, so SLM-421 added an additive Wilson-interval `power_rule`
+to the same unchanged harness/model/data recipe and reran on 20 fresh seeds
+(2..21, disjoint from SLM-282's 0/1) with a preregistered `min_pass_rate=0.5`
+locked before any of those seeds were observed. Observed: 18/20 passed
+(rate 0.90, Wilson 95% CI [0.699, 0.972]) — the lower bound clears 0.5, so the
+`recursive_core_positive` LAR3-reopening condition is now satisfied for this
+`as_is` shared_recursive-tower fixture configuration. The seed-1 regression
+that drove the SLM-282 negative reads as fixture-scale noise at n=2, not a
+systematic property of the core.
+
+**Net effect on LAR3**: only one of the two PR #853-#856 reopening conditions
+is now met. LAR3 stays closed — the second condition (a passing SLM-317-style
+valid-state repair advancement screen) is untouched by this issue; SLM-317's
+own harness is not yet merged to `main`, so its powered rerun is a separate,
+not-yet-landable follow-up. No ship, checkpoint, or production-default claim
+follows from either record; residual_delta remains a fixture-only
+counterfactual, never a production default.