diff --git a/tests/wp0/v21-learning-policy-p1-decision-record.test.js b/tests/wp0/v21-learning-policy-p1-decision-record.test.js
new file mode 100644
index 000000000..c22e79db6
--- /dev/null
+++ b/tests/wp0/v21-learning-policy-p1-decision-record.test.js
@@ -0,0 +1,94 @@
+'use strict';
+
+const test = require('node:test');
+const assert = require('node:assert/strict');
+
+const { createLearningPolicyDecisionContract } = require('../../backend/services/learningPolicyDecisionContract');
+
+test('Learned Policy DecisionRecord binds canonical identity, persona, features, action and generation provenance', () => {
+  const authorityCalls = [];
+  const contract = createLearningPolicyDecisionContract({
+    personContextAuthority: {
+      resolve(input) {
+        authorityCalls.push(input);
+        return {
+          authority: 'PersonContextAuthority',
+          found: true,
+          personId: 'person-1',
+          contactIds: ['contact-1'],
+          conversationIds: ['conversation-1']
+        };
+      }
+    }
+  });
+
+  const record = contract.createDecisionRecord({
+    contactId: 'contact-1',
+    conversationId: 'conversation-1',
+    personaProfileId: 'persona-owner-v7',
+    featureBundle: {
+      relationshipStage: 'warming',
+      interactionBand: 'balanced',
+      targetLanguage: 'en'
+    },
+    candidateStrategyBranch: 'screen_and_advance',
+    policyVersion: 'vw-p1-v1',
+    policyArtifactId: 'artifact-sha256:abc123',
+    generation: {
+      modelBrainExecutionId: 'exec-1',
+      candidatePlanId: 'plan-1'
+    }
+  });
+
+  assert.equal(authorityCalls.length, 1);
+  assert.equal(record.personId, 'person-1');
+  assert.deepEqual(record.contactIds, ['contact-1']);
+  assert.deepEqual(record.conversationIds, ['conversation-1']);
+  assert.equal(record.personaProfileId, 'persona-owner-v7');
+  assert.equal(record.candidateStrategyBranch, 'screen_and_advance');
+  assert.equal(record.actionProbability, 1);
+  assert.equal(record.exploration, false);
+  assert.equal(record.policyVersion, 'vw-p1-v1');
+  assert.equal(record.policyArtifactId, 'artifact-sha256:abc123');
+  assert.match(record.decisionId, /^decision:/);
+  assert.equal(Object.isFrozen(record), true);
+});
+
+test('DecisionRecord fails closed on identity/persona ambiguity and rejects raw chat bodies in features', () => {
+  const ambiguous = createLearningPolicyDecisionContract({
+    personContextAuthority: {
+      resolve() {
+        return { authority: 'PersonContextAuthority', found: true, personId: 'person-1', contactIds: ['contact-1'], conversationIds: ['conversation-2'] };
+      }
+    }
+  });
+
+  assert.throws(() => ambiguous.createDecisionRecord({
+    contactId: 'contact-1',
+    conversationId: 'conversation-1',
+    personaProfileId: 'persona-owner-v7',
+    featureBundle: { relationshipStage: 'warming' },
+    candidateStrategyBranch: 'natural_hook',
+    policyVersion: 'vw-p1-v1',
+    policyArtifactId: 'artifact-sha256:abc123',
+    generation: { modelBrainExecutionId: 'exec-1' }
+  }), error => error?.reasonCode === 'LEARNING_POLICY_IDENTITY_BINDING_MISMATCH');
+
+  const valid = createLearningPolicyDecisionContract({
+    personContextAuthority: {
+      resolve() {
+        return { authority: 'PersonContextAuthority', found: true, personId: 'person-1', contactIds: ['contact-1'], conversationIds: ['conversation-1'] };
+      }
+    }
+  });
+  assert.throws(() => valid.createDecisionRecord({
+    contactId: 'contact-1',
+    conversationId: 'conversation-1',
+    personaProfileId: 'persona-owner-v7',
+    featureBundle: { rawChatBody: 'private text' },
+    candidateStrategyBranch: 'natural_hook',
+    policyVersion: 'vw-p1-v1',
+    policyArtifactId: 'artifact-sha256:abc123',
+    generation: { modelBrainExecutionId: 'exec-1' }
+  }), error => error?.reasonCode === 'LEARNING_POLICY_FEATURE_BUNDLE_PRIVATE_BODY_FORBIDDEN');
+});
diff --git a/tests/wp0/v21-learning-policy-p1-outcome-binding.test.js b/tests/wp0/v21-learning-policy-p1-outcome-binding.test.js
new file mode 100644
index 000000000..3db7b6960
--- /dev/null
+++ b/tests/wp0/v21-learning-policy-p1-outcome-binding.test.js
@@ -0,0 +1,62 @@
+'use strict';
+
+const test = require('node:test');
+const assert = require('node:assert/strict');
+
+const { createLearningOutcomeAttributionService } = require('../../backend/services/learningOutcomeAttributionService');
+
+test('raw OutcomeVector is immutable, decision-bound and permanently non-trainable', () => {
+  const service = createLearningOutcomeAttributionService();
+  const vector = service.createOutcomeVector({
+    decisionId: 'decision:abc',
+    outcomes: [
+      { outcomeId: 'outcome-1', type: 'reply_received', value: 1, evidenceRef: 'message-99' },
+      { outcomeId: 'outcome-2', type: 'conversation_continued', value: 1, evidenceRef: 'conversation-1' }
+    ],
+    observedAt: '2026-08-14T11:00:00.000Z'
+  });
+
+  assert.equal(vector.signalType, 'policy_outcome_observed');
+  assert.equal(vector.learningEligible, false);
+  assert.equal(vector.decisionId, 'decision:abc');
+  assert.equal(Object.isFrozen(vector), true);
+  assert.equal(Object.isFrozen(vector.outcomes), true);
+});
+
+test('trainable binding requires the immutable eligible candidate_sent anchor and Learning-approved Langfuse Score', () => {
+  const service = createLearningOutcomeAttributionService();
+  const vector = service.createOutcomeVector({
+    decisionId: 'decision:abc',
+    outcomes: [{ outcomeId: 'outcome-1', type: 'reply_received', value: 1, evidenceRef: 'message-99' }],
+    observedAt: '2026-08-14T11:00:00.000Z'
+  });
+  const source = {
+    signal_id: 'signal-1',
+    signal_type: 'candidate_sent',
+    learning_eligible: true,
+    signal: { decisionRecord: { decisionId: 'decision:abc' } }
+  };
+  const score = {
+    authority: 'Langfuse',
+    approvedByLearning: true,
+    scoreId: 'score-1',
+    eligibleSourceSignalId: 'signal-1',
+    decisionId: 'decision:abc',
+    outcomeIds: ['outcome-1'],
+    outcomeEvidenceSetRef: 'evidence-set-1',
+    rewardPolicyVersion: 'reward-v1',
+    value: 1
+  };
+
+  const bound = service.bindTrainableOutcome({ eligibleSourceSignal: source, outcomeVector: vector, score });
+  assert.equal(bound.eligibleSourceSignalId, 'signal-1');
+  assert.equal(bound.decisionId, 'decision:abc');
+  assert.equal(bound.reward.authority, 'Langfuse');
+  assert.equal(vector.learningEligible, false, 'binding may never upgrade the raw outcome row');
+
+  assert.throws(() => service.bindTrainableOutcome({
+    eligibleSourceSignal: { ...source, learning_eligible: false },
+    outcomeVector: vector,
+    score
+  }), error => error?.reasonCode === 'LEARNING_POLICY_ELIGIBLE_SOURCE_SIGNAL_REQUIRED');
+});
diff --git a/tests/wp0/v21-learning-policy-p1-production-consumption.test.js b/tests/wp0/v21-learning-policy-p1-production-consumption.test.js
new file mode 100644
index 000000000..a629ababf
--- /dev/null
+++ b/tests/wp0/v21-learning-policy-p1-production-consumption.test.js
@@ -0,0 +1,24 @@
+'use strict';
+
+const test = require('node:test');
+const assert = require('node:assert/strict');
+const fs = require('node:fs');
+const path = require('node:path');
+
+const ROOT = path.resolve(__dirname, '..', '..');
+const source = () => fs.readFileSync(path.join(ROOT, 'backend/services/contextAwareReplyBrain.js'), 'utf8');
+
+test('production reply brain consumes Learned Policy before existing frontier generation', () => {
+  const text = source();
+  assert.match(text, /require\(['"]\.\/learningPolicyRuntimeAdapter['"]\)/);
+  assert.match(text, /selectLearnedPolicyAction\s*\(/);
+  assert.match(text, /candidateStrategyBranch/);
+  assert.match(text, /aiGateway\.execute\s*\(/, 'existing Model Brain gateway must remain the final generator');
+});
+
+test('Learned Policy production consumption cannot take provider/model credential or final-text authority', () => {
+  const text = source();
+  assert.doesNotMatch(text, /learnedPolicy[^\n]{0,120}(apiKey|credential|providerCredential)/i);
+  assert.doesNotMatch(text, /learnedPolicy[^\n]{0,120}modelId\s*:/i);
+  assert.doesNotMatch(text, /learnedPolicy[^\n]{0,120}(finalReply|finalText)\s*:/i);
+});
diff --git a/tests/wp0/v21-learning-policy-p1-projection.test.js b/tests/wp0/v21-learning-policy-p1-projection.test.js
new file mode 100644
index 000000000..97050511f
--- /dev/null
+++ b/tests/wp0/v21-learning-policy-p1-projection.test.js
@@ -0,0 +1,72 @@
+'use strict';
+
+const test = require('node:test');
+const assert = require('node:assert/strict');
+
+const { createLearningDeepTrainingContract } = require('../../backend/services/learningDeepTrainingContract');
+
+test('projectPolicy reads only immutable eligible source signals and joins raw outcomes by decisionId', async () => {
+  const repository = {
+    async listLearningSignals(query) {
+      assert.equal(query.learningEligible, true);
+      return [{
+        signal_id: 'signal-1',
+        signal_type: 'candidate_sent',
+        learning_eligible: true,
+        scope_type: 'relationship',
+        scope_id: 'person-1',
+        signal: {
+          decisionRecord: {
+            decisionId: 'decision:abc',
+            candidateStrategyBranch: 'natural_hook',
+            featureBundle: { relationshipStage: 'warming' },
+            policyVersion: 'vw-p1-v1',
+            policyArtifactId: 'artifact-sha256:abc123'
+          }
+        }
+      }];
+    },
+    async listPolicyOutcomeSignals({ decisionIds }) {
+      assert.deepEqual(decisionIds, ['decision:abc']);
+      return [{
+        signal_id: 'raw-outcome-1',
+        signal_type: 'policy_outcome_observed',
+        learning_eligible: false,
+        signal: {
+          decisionId: 'decision:abc',
+          outcomes: [{ outcomeId: 'outcome-1', type: 'reply_received', value: 1, evidenceRef: 'message-99' }]
+        }
+      }];
+    }
+  };
+  const contract = createLearningDeepTrainingContract({ repository, dataPolicy: { minimize: async value => ({ allowed: true, text: value.text || '' }) } });
+  const projection = await contract.projectPolicy({
+    scopeType: 'relationship',
+    scopeId: 'person-1',
+    approvedScoresBySignalId: {
+      'signal-1': {
+        authority: 'Langfuse', approvedByLearning: true, scoreId: 'score-1', name: 'policy_reward', value: 1, traceId: 'trace-1',
+        eligibleSourceSignalId: 'signal-1', decisionId: 'decision:abc', outcomeIds: ['outcome-1'], outcomeEvidenceSetRef: 'evidence-set-1', rewardPolicyVersion: 'reward-v1'
+      }
+    }
+  });
+
+  assert.equal(projection.authority, 'Learning');
+  assert.equal(projection.readOnly, true);
+  assert.equal(projection.trajectory.length, 1);
+  assert.equal(projection.trajectory[0].decisionId, 'decision:abc');
+  assert.equal(projection.trajectory[0].outcomes[0].outcomeId, 'outcome-1');
+});
+
+test('projectPolicy never upgrades raw outcome eligibility', async () => {
+  const raw = { signal_id: 'raw-outcome-1', signal_type: 'policy_outcome_observed', learning_eligible: false, signal: { decisionId: 'decision:abc', outcomes: [] } };
+  const repository = {
+    async listLearningSignals() {
+      return [{ signal_id: 'signal-1', signal_type: 'candidate_sent', learning_eligible: true, scope_type: 'relationship', scope_id: 'person-1', signal: { decisionRecord: { decisionId: 'decision:abc', candidateStrategyBranch: 'natural_hook', featureBundle: {}, policyVersion: 'vw-p1-v1', policyArtifactId: 'artifact-sha256:abc123' } } }];
+    },
+    async listPolicyOutcomeSignals() { return [raw]; }
+  };
+  const contract = createLearningDeepTrainingContract({ repository, dataPolicy: { minimize: async () => ({ allowed: true, text: '' }) } });
+  await contract.projectPolicy({ scopeType: 'relationship', scopeId: 'person-1', approvedScoresBySignalId: { 'signal-1': { authority: 'Langfuse', approvedByLearning: true, scoreId: 'score-1', name: 'policy_reward', value: 1, traceId: 'trace-1', eligibleSourceSignalId: 'signal-1', decisionId: 'decision:abc', outcomeIds: [], outcomeEvidenceSetRef: 'evidence-set-1', rewardPolicyVersion: 'reward-v1' } } });
+  assert.equal(raw.learning_eligible, false);
+});
diff --git a/tests/wp0/v21-learning-policy-p1-supply-chain.test.js b/tests/wp0/v21-learning-policy-p1-supply-chain.test.js
new file mode 100644
index 000000000..3360b8dad
--- /dev/null
+++ b/tests/wp0/v21-learning-policy-p1-supply-chain.test.js
@@ -0,0 +1,25 @@
+'use strict';
+
+const test = require('node:test');
+const assert = require('node:assert/strict');
+const fs = require('node:fs');
+const path = require('node:path');
+
+const ROOT = path.resolve(__dirname, '..', '..');
+const read = p => fs.readFileSync(path.join(ROOT, p), 'utf8');
+
+test('Vowpal Wabbit supply chain is pinned to 9.11.2 with exact upstream provenance and BSD-3-Clause notice', () => {
+  const pyproject = read('runtime/learning-growth/python/pyproject.toml');
+  const lock = read('runtime/learning-growth/python/uv.lock');
+  const upstream = JSON.parse(read('config/upstreams/v21-learning-growth-brain-p0.json'));
+  const notices = read('THIRD_PARTY_NOTICES.md');
+  const license = read('third_party/licenses/vowpal-wabbit-BSD-3-Clause.txt');
+
+  assert.match(pyproject, /vowpalwabbit\s*==\s*9\.11\.2/i);
+  assert.match(lock, /name\s*=\s*["']vowpalwabbit["']/i);
+  assert.match(lock, /version\s*=\s*["']9\.11\.2["']/i);
+  assert.match(JSON.stringify(upstream), /122bae254a5b8bc2b774d13b33d53e6dbc2cfba7/);
+  assert.match(notices, /Vowpal Wabbit/i);
+  assert.match(notices, /BSD-3-Clause/i);
+  assert.match(license, /Redistribution and use in source and binary forms/i);
+});
diff --git a/tests/wp0/v21-learning-policy-p1-vw-runtime.test.js b/tests/wp0/v21-learning-policy-p1-vw-runtime.test.js
new file mode 100644
index 000000000..d0cfb13fd
--- /dev/null
+++ b/tests/wp0/v21-learning-policy-p1-vw-runtime.test.js
@@ -0,0 +1,40 @@
+'use strict';
+
+const test = require('node:test');
+const assert = require('node:assert/strict');
+const fs = require('node:fs');
+const path = require('node:path');
+
+const ROOT = path.resolve(__dirname, '..', '..');
+const read = p => fs.readFileSync(path.join(ROOT, p), 'utf8');
+
+const { createLearningPolicyRuntimeAdapter } = require('../../backend/services/learningPolicyRuntimeAdapter');
+
+test('Learning runtime adapter delegates the action head to sealed Vowpal Wabbit and keeps P1 deterministic', async () => {
+  const calls = [];
+  const adapter = createLearningPolicyRuntimeAdapter({
+    invokeVowpalWabbit: async input => {
+      calls.push(input);
+      return { action: 'playful_attraction', policyVersion: 'vw-p1-v1', policyArtifactId: 'artifact-sha256:abc123' };
+    }
+  });
+  const decision = await adapter.selectLearnedPolicyAction({
+    featureBundle: { relationshipStage: 'warming', interactionBand: 'balanced' },
+    allowedActions: ['natural_hook', 'playful_attraction', 'direct_advance', 'screen_and_advance', 'leave_aftertaste']
+  });
+
+  assert.equal(calls.length, 1);
+  assert.equal(decision.candidateStrategyBranch, 'playful_attraction');
+  assert.equal(decision.actionProbability, 1);
+  assert.equal(decision.exploration, false);
+  assert.equal(decision.providerRoutingAuthority, undefined);
+  assert.equal(decision.finalReply, undefined);
+});
+
+test('sealed Learning Python entrypoint exposes the VW policy action mode without provider credentials or local text generation', () => {
+  const source = read('runtime/learning-growth/python/learning_entrypoint.py');
+  assert.match(source, /vowpalwabbit/i);
+  assert.match(source, /learned[_-]policy/i);
+  assert.doesNotMatch(source, /(OPENAI_API_KEY|ANTHROPIC_API_KEY|provider[_-]?credential)/i);
+  assert.doesNotMatch(source, /(generate[_-]?reply|final[_-]?reply)/i);
+});