incremnt 0.8.7 → 0.8.8

package/src/remote.js

@@ -335,6 +335,7 @@ function endpointForCommand(baseUrl, normalizedCommand, options) {
       if (options.from) url.searchParams.set('from', options.from);
       if (options.to) url.searchParams.set('to', options.to);
       if (options.limit) url.searchParams.set('limit', options.limit);
+      if (options['include-signals']) url.searchParams.set('includeSignals', 'true');
       return url;
     }
     case 'coach-observations-current': {
@@ -486,6 +487,20 @@ export async function buildWriteRequest(commandId, options, sessionState) {
   }
 
   switch (commandId) {
+    case 'programs-create': {
+      if (!options.file) {
+        const error = new Error('--file is required for programs create.');
+        error.code = 'MISSING_OPTION';
+        throw error;
+      }
+      const raw = await fs.readFile(options.file, 'utf8');
+      const draft = JSON.parse(raw);
+      return {
+        method: 'POST',
+        url: resolveServiceUrl(baseUrl, '/cli/programs').toString(),
+        body: draft
+      };
+    }
     case 'programs-propose': {
       if (!options.file) {
         const error = new Error('--file is required for programs propose.');
@@ -632,6 +647,42 @@ export async function buildWriteRequest(commandId, options, sessionState) {
 }
 
 const remoteWriteCommandHandlers = {
+  'programs-create': async (options, sessionState) => {
+    const baseUrl = sessionState.session?.transport?.baseUrl;
+    if (!baseUrl) throw notImplementedError();
+
+    const filePath = options.file;
+    if (!filePath) {
+      const error = new Error('--file is required for programs create.');
+      error.code = 'MISSING_OPTION';
+      throw error;
+    }
+
+    const raw = await fs.readFile(filePath, 'utf8');
+    const draft = JSON.parse(raw);
+
+    const endpoint = resolveServiceUrl(baseUrl, '/cli/programs');
+    const response = await fetch(endpoint, {
+      method: 'POST',
+      headers: {
+        'Content-Type': 'application/json',
+        Authorization: `Bearer ${sessionState.session?.auth?.accessToken ?? ''}`
+      },
+      body: JSON.stringify(draft)
+    });
+
+    if (response.status === 401) throw authenticationFailedError();
+    if (response.status === 403) await throwForbiddenResponse(response, sessionState, 'programs-create');
+    if (!response.ok) {
+      const payload = await response.json().catch(() => null);
+      const error = new Error(payload?.error ?? `Unexpected error (HTTP ${response.status}).`);
+      error.code = 'REMOTE_HTTP_ERROR';
+      throw error;
+    }
+
+    return response.json();
+  },
+
   'programs-propose': async (options, sessionState) => {
     const baseUrl = sessionState.session?.transport?.baseUrl;
     if (!baseUrl) throw notImplementedError();

package/src/score-context.js

@@ -4,6 +4,11 @@
 // They are NOT persisted — pure projections of (current snapshot, previous
 // snapshot) into agent-friendly explanatory shape. See GitHub issue #498.
 
+import {
+  publicTrainingLanguageViolations,
+  replacePublicTrainingLanguageTerms
+} from './training-language.js';
+
 // Score bands. Inclusive lower bound, exclusive upper bound (except 'peak').
 //   0..40   weak
 //   40..60  developing
@@ -29,7 +34,7 @@ export function computeScoreBand(score) {
 // Component-keyed action templates surfaced as `recommendedNextActions` for
 // each top-2 negative driver. Keep these short, imperative, single-line.
 const COMPONENT_ACTIONS = {
-  coverage: 'Add the missing muscle work before adding more of what is already covered.',
+  coverage: 'Train the missing muscles before adding more of what is already trained.',
   recovery: 'Go easier next session unless sleep and soreness feel clearly better.',
   stimulus: 'Put the short muscle groups earlier next time.',
   execution: 'Follow the planned work more closely next session.',
@@ -56,8 +61,12 @@ function actionForDriver(driver) {
 
 function driverDisplayMessage(driver) {
   if (!driver || typeof driver !== 'object') return null;
-  if (typeof driver.message === 'string' && driver.message.trim()) return driver.message;
-  if (typeof driver.label === 'string' && driver.label.trim()) return driver.label;
+  if (typeof driver.message === 'string' && driver.message.trim()) {
+    return replacePublicTrainingLanguageTerms(driver.message);
+  }
+  if (typeof driver.label === 'string' && driver.label.trim()) {
+    return replacePublicTrainingLanguageTerms(driver.label);
+  }
   return null;
 }
 
@@ -152,6 +161,22 @@ export function enrichScoreSnapshots(snapshots) {
   });
 }
 
+function includeSignalsOption(options = {}) {
+  const value = options.includeSignals ?? options['include-signals'];
+  return value === true || value === 'true' || value === '1';
+}
+
+export function compactScoreHistorySnapshots(snapshots, options = {}) {
+  if (!Array.isArray(snapshots)) return [];
+  if (includeSignalsOption(options)) return snapshots;
+  return snapshots.map((snapshot) => {
+    if (!snapshot || typeof snapshot !== 'object') return snapshot;
+    const compact = { ...snapshot };
+    delete compact.signals;
+    return compact;
+  });
+}
+
 export function enrichScoreSnapshot(current, previous) {
   if (!current || typeof current !== 'object') return current;
 
@@ -187,3 +212,33 @@ export function enrichScoreSnapshot(current, previous) {
   enriched.summaryText = computeSummaryText(enriched);
   return enriched;
 }
+
+export function scoreContextPublicLanguageViolations(enriched) {
+  if (!enriched || typeof enriched !== 'object') return [];
+  const publicTexts = [];
+  if (typeof enriched.summaryText === 'string') {
+    publicTexts.push({ field: 'summaryText', text: enriched.summaryText });
+  }
+  if (Array.isArray(enriched.recommendedNextActions)) {
+    enriched.recommendedNextActions.forEach((action, index) => {
+      if (typeof action?.driverMessage === 'string') {
+        publicTexts.push({
+          field: `recommendedNextActions[${index}].driverMessage`,
+          text: action.driverMessage
+        });
+      }
+      if (typeof action?.action === 'string') {
+        publicTexts.push({
+          field: `recommendedNextActions[${index}].action`,
+          text: action.action
+        });
+      }
+    });
+  }
+
+  return publicTexts
+    .filter(({ text }) => text.trim())
+    .flatMap(({ field, text }) => (
+      publicTrainingLanguageViolations(text).map((violation) => ({ ...violation, field }))
+    ));
+}

package/src/summary-evals.js

@@ -17,6 +17,7 @@ import { formatIncrementScorePrelude, isScoreQuestion } from './score-prelude.js
 import {
   AI_PROMPT_VERSIONS,
   generateAskAnswer,
+  generateAskAnswerStarterGraph,
   generateCheckpointSummary,
   generateCoachingSummary,
   generateVitalsSummary,
@@ -25,7 +26,14 @@ import {
 import { computeScoreBand } from './score-context.js';
 import { stripXMLTagBlocks } from './prompt-security.js';
 import { extractAskProgramDraft, hasProgramDraftBlock } from './program-draft.js';
+import { extractCoachAdvice } from './coach-advice.js';
 import { findAskAnswerExerciseMentions, verifyAskAnswer } from './ask-answer-verifier.js';
+import {
+  STARTER_GRAPH_VERSION,
+  applyStarterPromptStructuredIntentHint,
+  starterPromptToolPlan
+} from './ask-starter-prompts.js';
+import { publicTrainingLanguageViolations } from './training-language.js';
 
 const __filename = fileURLToPath(import.meta.url);
 const __dirname = path.dirname(__filename);
@@ -120,8 +128,9 @@ export function buildSummaryEvalContext(snapshot, testCase) {
       const question = testCase.context?.question ?? testCase.question ?? '';
       const today = testCase.context?.today ?? testCase.today ?? null;
       const history = Array.isArray(testCase.context?.history) ? testCase.context.history : [];
+      const routeToday = today ? new Date(today) : new Date();
       const routed = question
-        ? askRoutedContext(snapshot, question, { exclude: new Set(testCase.exclude ?? []), history, today: today ?? new Date() })
+        ? askRoutedContext(snapshot, question, { exclude: new Set(testCase.exclude ?? []), history, today: routeToday })
         : null;
       // Mirror production: the live /cli/ask path prepends the Increment Score
       // prelude to the routed context. Including it here means a live eval feeds
@@ -132,10 +141,11 @@ export function buildSummaryEvalContext(snapshot, testCase) {
       const routedContext = routed?.context ?? null;
       const trainingData = testCase.context?.trainingData
         ?? (prelude && routedContext ? `${prelude}\n\n${routedContext}` : (routedContext ?? prelude));
+      const routedMetadata = applyStarterPromptEvalMetadata(routed?.metadata ?? null, testCase);
       return {
         ...(testCase.context ?? {}),
         trainingData,
-        routedMetadata: routed?.metadata ?? null
+        routedMetadata
       };
     }
     case 'scoreCommentary':
@@ -152,25 +162,101 @@ function summaryEvalGenerationMetadata(result) {
       model: result.model,
       durationMs: result.durationMs,
       fallback: result.fallback,
+      promptSurface: result.promptSurface,
+      promptVersion: result.promptVersion,
+      graphVersion: result.graph?.version,
+      starterPromptFamily: result.graph?.family,
+      graphNodes: result.graph?.nodes,
+      graphToolNames: result.graph?.toolNames,
       langfuseTraceId: result.langfuseTraceId,
       langfuseObservationId: result.langfuseObservationId
     }).filter(([, value]) => value !== undefined && value !== null)
   );
 }
 
+function evalEvidenceLabel(section, toolName) {
+  const cleaned = String(section || toolName || 'evidence')
+    .replace(/^observation_/, '')
+    .replace(/_/g, ' ')
+    .trim();
+  return cleaned ? cleaned.replace(/\b\w/g, (char) => char.toUpperCase()) : 'Evidence';
+}
+
+function applyStarterPromptEvalMetadata(metadata, testCase) {
+  const starterPrompt = testCase.starterPrompt;
+  if (!starterPrompt?.family || !metadata || typeof metadata !== 'object') return metadata;
+
+  const hintedMetadata = applyStarterPromptStructuredIntentHint(metadata, starterPrompt);
+  const plannedCalls = starterPromptToolPlan(starterPrompt);
+  if (plannedCalls.length === 0) return metadata;
+
+  const names = plannedCalls.map((call) => call.name).filter(Boolean);
+  const toolParams = Object.fromEntries(plannedCalls.map((call) => [call.name, call.params ?? {}]));
+  const provenanceEntries = plannedCalls.map((call) => ({
+    section: call.name,
+    toolName: call.name,
+    sourceIds: []
+  }));
+  const contextBundle = hintedMetadata.contextBundle && typeof hintedMetadata.contextBundle === 'object'
+    ? hintedMetadata.contextBundle
+    : null;
+
+  return {
+    ...hintedMetadata,
+    starterPromptFamily: starterPrompt.family,
+    graphVersion: STARTER_GRAPH_VERSION,
+    graphNodes: ['prepare_evidence', 'synthesize'],
+    graphToolNames: names,
+    toolsUsed: [...new Set([...(hintedMetadata.toolsUsed ?? []), ...names])],
+    toolParams: {
+      ...(hintedMetadata.toolParams ?? {}),
+      ...toolParams
+    },
+    evidencePlan: hintedMetadata.evidencePlan && typeof hintedMetadata.evidencePlan === 'object'
+      ? {
+          ...hintedMetadata.evidencePlan,
+          executedTools: [...new Set([...(hintedMetadata.evidencePlan.executedTools ?? []), ...names])]
+        }
+      : hintedMetadata.evidencePlan,
+    provenance: [...(hintedMetadata.provenance ?? []), ...provenanceEntries],
+    ...(contextBundle
+      ? {
+          contextBundle: {
+            ...contextBundle,
+            executedTools: [...new Set([...(contextBundle.executedTools ?? []), ...names])],
+            evidenceUsed: [
+              ...(contextBundle.evidenceUsed ?? []),
+              ...provenanceEntries.map((entry) => ({
+                label: evalEvidenceLabel(entry.section, entry.toolName),
+                section: entry.section,
+                toolName: entry.toolName,
+                sourceTimestamp: null,
+                sourceIds: [],
+                noteSourceIds: [],
+                missingDataFlags: []
+              }))
+            ]
+          }
+        }
+      : {})
+  };
+}
+
 function buildAskEvalStructuredMetadata(testCase, context, output) {
   if (testCase.surface !== 'ask') return {};
   const parsedAsk = extractAskProgramDraft(output, {
     canonicalizeExerciseName: canonicalExerciseName
   });
-  const answer = stripXMLTagBlocks(parsedAsk.answerText);
+  const adviceParsed = extractCoachAdvice(parsedAsk.answerText);
+  const answer = stripXMLTagBlocks(adviceParsed.answerText);
   const question = context?.question ?? testCase.context?.question ?? testCase.question ?? '';
   const routingMetadata = context?.routedMetadata ?? null;
   return {
     routingMetadata,
     structured: buildAskStructuredResponse(answer, routingMetadata ?? {}, {
       programDraft: askStructuredProgramDraft(parsedAsk, routingMetadata),
-      question
+      question,
+      coachAdvice: adviceParsed.coachAdvice ?? testCase.coachAdviceCandidate ?? null
     })
   };
 }
@@ -215,13 +301,24 @@ export async function generateSummaryEvalOutputWithMetadata(testCase, context, s
       const trainingContext = context.trainingData
         ?? askRoutedContext(snapshot, question, { exclude: new Set(testCase.exclude ?? []) }).context
         ?? askContext(snapshot, { exclude: new Set(testCase.exclude ?? []) });
-      result = await generateAskAnswer(trainingContext, question, {
-        apiKey,
-        history: context.history ?? [],
-        tone: context.tone,
-        model: context.model,
-        routingMetadata: context.routedMetadata ?? undefined
-      });
+      result = testCase.starterPrompt?.family
+        ? await generateAskAnswerStarterGraph(trainingContext, question, {
+          apiKey,
+          history: context.history ?? [],
+          tone: context.tone,
+          model: context.model,
+          snapshot,
+          starterPrompt: testCase.starterPrompt,
+          routingMetadata: context.routedMetadata ?? undefined,
+          exclude: testCase.exclude ?? []
+        })
+        : await generateAskAnswer(trainingContext, question, {
+          apiKey,
+          history: context.history ?? [],
+          tone: context.tone,
+          model: context.model,
+          routingMetadata: context.routedMetadata ?? undefined
+        });
       break;
     }
     default:
@@ -1185,7 +1282,10 @@ function hasUnqualifiedDeclineLanguage(window) {
   const text = normalizeText(window);
   const decline = /\b(drop(?:ped|ping|s)?(?: off)?|drop-off|declin(?:e|ed|ing)|regress(?:ed|ion|ing)?|fell|fall(?:ing)?|decreas(?:e|ed|ing)|lower|worse|slid|slipped)\b/i;
   if (!decline.test(text)) return false;
-  if (/\b(?:no|not|isn'?t|wasn'?t|without|rather than)\b.{0,45}\b(drop(?:ped|ping|s)?(?: off)?|drop-off|declin(?:e|ed|ing)?|decreas(?:e|ed|ing)?|regress(?:ed|ion|ing)?|fall(?:ing)?|fell|lower|worse|slid|slipped)\b/i.test(text)) return false;
+  if (/\b(?:volume|tonnage|work rate|work done|weekly work)\b.{0,80}\b(drop(?:ped|ping|s)?(?: off)?|drop-off|declin(?:e|ed|ing)?|decreas(?:e|ed|ing)?|fell|fall(?:ing)?|lower|down|dip(?:ped)?|pullback)\b/i.test(text)) return false;
+  if (/\b(drop(?:ped|ping|s)?(?: off)?|drop-off|declin(?:e|ed|ing)?|decreas(?:e|ed|ing)?|fell|fall(?:ing)?|lower|down|dip(?:ped)?|pullback)\b.{0,80}\b(?:volume|tonnage|work rate|work done|weekly work)\b/i.test(text)) return false;
+  if (/\b(?:no|not|isn'?t|wasn'?t|weren'?t|doesn'?t|don'?t|didn'?t|without|rather than)\b.{0,60}\b(drop(?:ped|ping|s)?(?: off)?|drop-off|declin(?:e|ed|ing)?|decreas(?:e|ed|ing)?|regress(?:ed|ion|ing)?|fall(?:ing)?|fell|lower|worse|slid|slipped)\b/i.test(text)) return false;
+  if (/\b(drop(?:ped|ping|s)?(?: off)?|drop-off|declin(?:e|ed|ing)?|decreas(?:e|ed|ing)?|regress(?:ed|ion|ing)?|fall(?:ing)?|fell|lower|worse|slid|slipped)\b.{0,60}\b(?:no|not|isn'?t|wasn'?t|weren'?t|doesn'?t|don'?t|didn'?t)\b/i.test(text)) return false;
   if (/\b(?:rep|reps)\b.{0,20}\b(drop(?:ped|ping|s)?(?: off)?|drop-off|slip(?:ped|ping)?|fell|fall(?:ing)?|lower|declin(?:e|ed|ing)?|decreas(?:e|ed|ing)?|worse)\b/i.test(text)) return false;
   if (/\b(drop(?:ped|ping|s)?(?: off)?|drop-off|slip(?:ped|ping)?|fell|fall(?:ing)?|lower|declin(?:e|ed|ing)?|decreas(?:e|ed|ing)?|worse)\b.{0,20}\b(?:rep|reps)\b/i.test(text)) return false;
   return true;
@@ -1338,13 +1438,26 @@ export function evaluateAskScoreVoice(output, testCase) {
     hits.add('day-over-day delta number');
   }
 
-  return {
+  const rawScoreResult = {
     key: 'ask_score_voice',
     passed: hits.size === 0,
     reason: hits.size === 0
       ? 'Ask answer does not recite raw Increment Score component sub-scores.'
       : `Ask answer recites raw score internals: ${[...hits].join(', ')}. Speak in training reality, not raw sub-scores.`
   };
+
+  if (!rawScoreResult.passed) return rawScoreResult;
+
+  const publicLanguageHits = publicTrainingLanguageViolations(text).map((violation) => violation.term);
+  if (publicLanguageHits.length > 0) {
+    return {
+      key: 'ask_score_voice',
+      passed: false,
+      reason: `Ask answer uses internal training language: ${publicLanguageHits.join(', ')}. Speak in normal training language.`
+    };
+  }
+
+  return rawScoreResult;
 }
 
 function relevantSessionsForStaleness(snapshot, testCase) {
@@ -1591,8 +1704,10 @@ function outputCallsStaleEvidenceRecent(outputText, row) {
   const normalized = normalizeText(outputText);
   const claimsRecent = /\brecent(?:ly)?\b/i.test(normalized);
   if (!claimsRecent) return false;
-  const explicitlyNotRecent = /\b(?:not|isn'?t|wasn'?t|no longer)\s+(?:a\s+)?recent\b/i.test(normalized)
-    || /\brecent\b.{0,20}\b(?:not|isn'?t|wasn'?t)\b/i.test(normalized);
+  const explicitlyNotRecent = /\b(?:not|isn'?t|wasn'?t|no longer)\s+(?:\w+\s+){0,4}?recent\b/i.test(normalized)
+    || /\b(?:no|not enough|without|missing|lack(?:ing)?|insufficient|don'?t have enough|do not have enough)\s+(?:\w+\s+){0,4}?recent\s+(?:work|sessions?|data|logs?|evidence|history)\b/i.test(normalized)
+    || /\brecent\b.{0,45}\b(?:not|isn'?t|wasn'?t|stale|old|older|latest logged|last logged)\b/i.test(normalized)
+    || /\b(?:only|latest|last)\s+logged\b.{0,80}\b(?:stale|old|older|days?\s+ago)\b/i.test(normalized);
   if (explicitlyNotRecent) return false;
   const daysAgo = Number(row?.daysAgo);
   return !Number.isFinite(daysAgo) || !new RegExp(`\\b${daysAgo}\\s+days?\\s+ago\\b`, 'i').test(normalized);
@@ -1746,12 +1861,19 @@ function evaluateAskEvidencePlan(_output, context, testCase) {
       }
     }
 
-    for (const key of ['requiredTools', 'optionalTools', 'executedTools', 'evidenceGaps']) {
+    for (const key of ['requiredTools', 'optionalTools', 'executedTools', 'executedDynamicTools', 'evidenceGaps']) {
       if (Array.isArray(expected[key]) && !arrayEquals(plan[key], expected[key])) {
         failures.push(`Expected evidencePlan.${key} to equal ${expected[key].join(', ')}; got ${(plan[key] ?? []).join(', ')}.`);
       }
     }
 
+    if (Array.isArray(expected.dynamicTools)) {
+      const actualDynamicTools = (plan.dynamicTools ?? []).map((item) => item.toolName);
+      if (!arrayEquals(actualDynamicTools, expected.dynamicTools)) {
+        failures.push(`Expected evidencePlan.dynamicTools to equal ${expected.dynamicTools.join(', ')}; got ${actualDynamicTools.join(', ')}.`);
+      }
+    }
+
     if (Array.isArray(expected.excludedExecutedTools)) {
       const executed = new Set(plan.executedTools ?? []);
       const hits = expected.excludedExecutedTools.filter((toolName) => executed.has(toolName));
@@ -1955,6 +2077,35 @@ function evaluateAskStructuredResponse(_output, context, testCase, structured) {
         failures.push(`Expected programDraft present=${expected.programDraftPresent}, got ${hasProgramDraft}.`);
       }
     }
+
+    if (typeof expected.coachAdvicePresent === 'boolean') {
+      const hasCoachAdvice = structured.coachAdvice != null;
+      if (hasCoachAdvice !== expected.coachAdvicePresent) {
+        failures.push(`Expected coachAdvice present=${expected.coachAdvicePresent}, got ${hasCoachAdvice}.`);
+      }
+    }
+
+    if (expected.coachAdviceKind && structured.coachAdvice?.kind !== expected.coachAdviceKind) {
+      failures.push(`Expected coachAdvice kind ${expected.coachAdviceKind}, got ${structured.coachAdvice?.kind ?? 'null'}.`);
+    }
+
+    if (expected.coachAdviceTarget && structured.coachAdvice?.target !== expected.coachAdviceTarget) {
+      failures.push(`Expected coachAdvice target ${expected.coachAdviceTarget}, got ${structured.coachAdvice?.target ?? 'null'}.`);
+    }
+
+    requireStructuredStrings(
+      structured.coachAdvice?.evidenceRefs,
+      expected.requiredCoachAdviceEvidenceRefs,
+      'coachAdvice evidence ref(s)',
+      failures
+    );
+
+    requireStructuredStrings(
+      structured.coachAdvice?.outcomeCriteria,
+      expected.requiredCoachAdviceOutcomeCriteria,
+      'coachAdvice outcome criteria',
+      failures
+    );
   }
 
   return {
@@ -2278,42 +2429,47 @@ export function evaluateSummaryOutputFromSnapshot(testCase, snapshot, output) {
     ? extractAskProgramDraft(output, { canonicalizeExerciseName: canonicalExerciseName })
     : null;
   const visibleOutput = parsedAsk
-    ? stripXMLTagBlocks(parsedAsk.answerText)
+    ? parsedAsk.answerText
     : output;
+  const adviceParsed = testCase.surface === 'ask'
+    ? extractCoachAdvice(visibleOutput)
+    : { answerText: visibleOutput, coachAdvice: null };
+  const finalVisibleOutput = stripXMLTagBlocks(adviceParsed.answerText);
   const structuredAsk = testCase.surface === 'ask'
-    ? buildAskStructuredResponse(visibleOutput, context.routedMetadata ?? {}, {
+    ? buildAskStructuredResponse(finalVisibleOutput, context.routedMetadata ?? {}, {
         programDraft: askStructuredProgramDraft(structuredParsedAsk, context.routedMetadata),
-        question: context.question ?? testCase.context?.question ?? testCase.question ?? ''
+        question: context.question ?? testCase.context?.question ?? testCase.question ?? '',
+        coachAdvice: adviceParsed.coachAdvice ?? testCase.coachAdviceCandidate ?? null
       })
     : null;
 
   const checks = [
-    evaluateNoInsight(visibleOutput, testCase),
-    evaluateShape(visibleOutput, testCase, context),
-    evaluateRequiredMentions(visibleOutput, testCase),
-    evaluateAnyOfMentions(visibleOutput, testCase),
-    evaluateForbiddenPhrases(visibleOutput, testCase),
-    evaluateForbiddenMentions(visibleOutput, testCase),
-    evaluateExerciseMentions(visibleOutput, snapshot, context, testCase.surface, testCase),
-    evaluateWorkoutClaims(visibleOutput, context, testCase),
-    evaluateAskClaims(visibleOutput, snapshot, testCase, context),
-    evaluateAskDirectionalConsistency(visibleOutput, snapshot, testCase),
-    evaluateAskScoreVoice(visibleOutput, testCase),
-    evaluateAskSelfReference(visibleOutput, testCase),
-    evaluateAskVolunteeredScore(visibleOutput, testCase, context),
-    evaluateAskStaleness(visibleOutput, snapshot, testCase),
-    evaluateAskToolProvenance(visibleOutput, context, testCase, snapshot),
-    evaluateFormulaVersion(visibleOutput, snapshot, testCase),
-    evaluateAskEvidencePlan(visibleOutput, context, testCase),
-    evaluateAskMetadata(visibleOutput, context, testCase),
-    evaluateAskStructuredResponse(visibleOutput, context, testCase, structuredAsk),
-    evaluateScoreCommentaryAction(visibleOutput, context, testCase),
-    evaluateScoreCommentarySynthesis(visibleOutput, context, testCase),
-    evaluateScoreCommentaryExerciseInvention(visibleOutput, snapshot, context, testCase),
-    evaluateScoreCommentaryBand(visibleOutput, context, testCase),
-    evaluateScoreCommentaryTone(visibleOutput, testCase),
-    evaluateScoreCommentaryLength(visibleOutput, testCase),
-    evaluatePersonaMotivation(visibleOutput, context, testCase),
+    evaluateNoInsight(finalVisibleOutput, testCase),
+    evaluateShape(finalVisibleOutput, testCase, context),
+    evaluateRequiredMentions(finalVisibleOutput, testCase),
+    evaluateAnyOfMentions(finalVisibleOutput, testCase),
+    evaluateForbiddenPhrases(finalVisibleOutput, testCase),
+    evaluateForbiddenMentions(finalVisibleOutput, testCase),
+    evaluateExerciseMentions(finalVisibleOutput, snapshot, context, testCase.surface, testCase),
+    evaluateWorkoutClaims(finalVisibleOutput, context, testCase),
+    evaluateAskClaims(finalVisibleOutput, snapshot, testCase, context),
+    evaluateAskDirectionalConsistency(finalVisibleOutput, snapshot, testCase),
+    evaluateAskScoreVoice(finalVisibleOutput, testCase),
+    evaluateAskSelfReference(finalVisibleOutput, testCase),
+    evaluateAskVolunteeredScore(finalVisibleOutput, testCase, context),
+    evaluateAskStaleness(finalVisibleOutput, snapshot, testCase),
+    evaluateAskToolProvenance(finalVisibleOutput, context, testCase, snapshot),
+    evaluateFormulaVersion(finalVisibleOutput, snapshot, testCase),
+    evaluateAskEvidencePlan(finalVisibleOutput, context, testCase),
+    evaluateAskMetadata(finalVisibleOutput, context, testCase),
+    evaluateAskStructuredResponse(finalVisibleOutput, context, testCase, structuredAsk),
+    evaluateScoreCommentaryAction(finalVisibleOutput, context, testCase),
+    evaluateScoreCommentarySynthesis(finalVisibleOutput, context, testCase),
+    evaluateScoreCommentaryExerciseInvention(finalVisibleOutput, snapshot, context, testCase),
+    evaluateScoreCommentaryBand(finalVisibleOutput, context, testCase),
+    evaluateScoreCommentaryTone(finalVisibleOutput, testCase),
+    evaluateScoreCommentaryLength(finalVisibleOutput, testCase),
+    evaluatePersonaMotivation(finalVisibleOutput, context, testCase),
     evaluateProgramDraft(output, testCase, parsedAsk)
   ];