feat(continual-learning): activate negative-feedback loop (RSI steps 1-2)

Step 1 — policy.ts shouldIntervene: suppress on a prior dismissal alone.
The former `&& !priorOutcome.wasRealCollision` term was dead code (outcomes
record wasRealCollision hardcoded true), so the 85 real dismissals in Atlas
were never used. Now a prior accepted===false suppresses the next identical
nudge. Spec: continual-learning/policy.md:41, spec.md:163.

Step 2 — podman.ts handle: only escalate severity to 'critical' (the spoken
alert trigger) when the recalled prior was an accepted *real* collision,
instead of blanket-escalating every recall. Surfaces learned routing in
preferredAction; stops dismissed/false priors over-escalating to voice.
Spec: continual-learning/policy.md:62-63, plan.md:66.

Documentation-first: adds PLAN.md section 8 "P0.5 - RSI negative-feedback
activation" with both rungs + follow-ups. No schema change. backend typecheck
passes. Independent of the MongoDB-cleanup handoff (Codex).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

backend/src/agent/podman.ts

@@ -100,7 +100,14 @@ export class PodMan {
     if (this.activeConflicts.has(key)) return; // single-shot: already voiced, still unresolved
 
     const prior = await recallSimilar(collision); // Loop A: exact/vector recall raises confidence
-    if (prior) collision.severity = 'critical';
+    // Only escalate to critical (which triggers the spoken alert) when the
+    // recalled prior was an *accepted real* collision. Blanket-escalating every
+    // recall — including dismissed/false-positive priors — masked the learned
+    // routing in preferredAction and made recalled noise scream "CRITICAL".
+    // (RSI Step 2 — continual-learning/policy.md:62-63, plan.md:66)
+    if (prior?.priorOutcome?.accepted && prior?.priorOutcome?.wasRealCollision) {
+      collision.severity = 'critical';
+    }
     if (!shouldIntervene(collision, prior)) return; // Loop B: policy gate
 
     this.activeConflicts.add(key); // claim now we're alerting; re-armed in onScreenFrame on resolution

backend/src/memory/policy.ts

@@ -12,7 +12,12 @@ export function shouldIntervene(collision: Collision, prior: RecalledCollision |
   if (collision.severity === 'info') return false;
 
   const priorOutcome = prior?.priorOutcome;
-  if (priorOutcome && !priorOutcome.accepted && !priorOutcome.wasRealCollision) return false;
+  // Suppress when the identical prior was dismissed (accepted === false). The
+  // former `&& !priorOutcome.wasRealCollision` term was dead code: outcomes are
+  // recorded with wasRealCollision hardcoded true, so the gate never fired and
+  // the 85 real dismissals in Atlas were ignored. Dismissals are the negative
+  // signal per continual-learning/policy.md:41 + spec.md:163. (RSI Step 1)
+  if (priorOutcome && !priorOutcome.accepted) return false;
 
   const cooldown = cooldownMs();
   const last = lastNudgeByPod.get(collision.podId) ?? 0;

docs/PLAN.md

@@ -484,6 +484,37 @@ artifact.
       recorded backup.
     - Keep backup video on a separate device.
 
+### P0.5 - RSI negative-feedback activation (continual-learning)
+
+The continual-learning loop records outcomes but never feeds the negative
+signal back. Live Atlas (2026-06-28): `outcomes` = 22 accepted / 85 dismissed,
+yet `wasRealCollision` is `true` in 107/107 (hardcoded), so the suppression
+gate is dead and dismissals are unused. These two rungs activate the loop with
+no schema change. Owner: RSI track. Independent of the MongoDB-cleanup handoff.
+
+1. **Step 1 - suppress on prior dismissal alone** ✅
+   - `backend/src/memory/policy.ts` `shouldIntervene`: remove the dead
+     `&& !priorOutcome.wasRealCollision` term so a prior `accepted === false`
+     suppresses the next identical-signature nudge.
+   - Spec: `docs/continual-learning/policy.md:41` (dismissed = negative signal),
+     `spec.md:163` (dismissals adapt suppression).
+   - Caveat: recall is single-shot most-recent (`memory/vectors.ts`), so this is
+     "last-outcome-wins" until Step 3 (derive `wasRealCollision`) lands.
+
+2. **Step 2 - gate the recall severity escalation** ✅
+   - `backend/src/agent/podman.ts` `handle`: only force `severity = 'critical'`
+     when the recalled prior was an accepted *real* collision, instead of
+     blanket-escalating every recall. Surfaces the learned routing in
+     `preferredAction`; stops dismissed/false priors over-escalating to voice.
+   - Spec: `docs/continual-learning/policy.md:62-63` (prefer prior accepted
+     kind), `plan.md:66` (second similar event behaves differently).
+
+Follow-ups (separate rungs, not in this change): Step 3 derive
+`wasRealCollision` from git overlap; Step 4-5 `strategy_versions` +
+Gemini-proposed `LearningProposal` slice; seed a clean demo pod with a repeated
+dismissed signature (the historic 85 dismissals are orphaned — `collisionId`
+resolves to no collision — so they cannot drive the demo verifier).
+
 ### P1 - polish the money moment
 
 - Add visible live inference captions in the PWA.