@sun-asterisk/sungen 3.2.34-beta.1 → 3.3.0-alpha.1

package/src/harness/audit.ts

@@ -33,10 +33,10 @@ import { capabilityRegistry } from '../capabilities/registry';
 import { discoverAndRegisterCapabilities } from '../capabilities/discover';
 import { contextRouter } from '../capabilities/context-router';
 import { readTextFile, readViewpointText } from './read-text';
-import { isAiDraft } from './viewpoint-draft-header';
-import { viewpointReviewOpenFindings } from './viewpoint-review-open';
 import { uiInventoryCoverage, uiInventoryFindings, UiInventoryResult } from './ui-inventory';
 import { defectLinkFindings } from './defects';
+import { auditGateContext, auditGateVerdict, decisionFindings, relevelProvidedDepth } from './gates/audit-gate-verdict';
+import { GateOverrideReport } from './gates/gate-override-report';
 
 export interface AuditReport {
   screen: string;
@@ -100,6 +100,16 @@ export interface AuditReport {
   intent: IntentProfile;        // P3 — the intent profile that drove the thresholds
   provenance: Provenance;       // D1 — sungen version + catalog hash (diagnose cross-user score gaps)
   spec: SpecCoverageResult;     // G2 — spec-clause coverage (FR + validation-trigger matrix)
+  /** OES-11.6 — every gate-override touching this unit (applied / ineffective / rejected). */
+  gateOverrides: GateOverrideReport[];
+  /** OES-11.4 — the decisions this audit was judged under; readers re-audit when it moves. */
+  decisionsHash: string;
+  /** T3 — hash of .feature + spec + viewpoint + decisionsHash; a reader recomputes it. */
+  inputsHash: string;
+  /** T3 — what `sungen audit` exits with for this gateStatus (0 PASS · 2 FAIL). */
+  exitCode: number;
+  /** A capability gate sensor reported an error-level finding — recorded so a reader can re-evaluate the gate. */
+  gateSensorError: boolean;
 }
 
 /**
@@ -171,6 +181,8 @@ export function runAudit(screenDir: string, screenName: string): AuditReport {
   // `flows/<flow>`, `api/<area>`, `api/flows/<flow>`, else the bare screen. A bare screen matches
   // the old behaviour (so the audit-sample snapshot is unchanged); flows/api now resolve correctly.
   const catalogScreenName = catalogIdFromScreenDir(screenDir);
+  // OES-11.2 — the unit's gate rules: the catalog plus the gate-overrides in force for it.
+  const gates = auditGateContext(screenDir);
 
   const specPath = path.join(screenDir, 'requirements', 'spec.md');
   const featureText = readUnitFeatureText(screenDir, screenName);
@@ -204,9 +216,6 @@ export function runAudit(screenDir: string, screenName: string): AuditReport {
   const viewpointBaseline = checkViewpointBaseline(
     projectRootFromScreenDir(screenDir), catalogScreenName, viewpointPath);
   const viewpointText = readViewpointText(viewpointPath);
-  // `drafted-by: ai` — unreviewed until `--accept-viewpoint` removes it. Only a draft that declares ids
-  // has anything to accept; an id-less one is VIEWPOINT-MISSING's to report, not a dead-end finding.
-  const aiDraft = isAiDraft(viewpointText) && viewpointBaseline.status !== 'absent';
   // #569 — a flow WITH a contract is measured against ITS declaration, not a screen's.
   // Keyword page-type detection force-fit flows into screen checklists (a registration
   // flow judged as a `form` screen scored coverage 0%), so for contract-holding flows the
@@ -228,7 +237,9 @@ export function runAudit(screenDir: string, screenName: string): AuditReport {
   const isMobile = platform === 'mobile';
   const provided = capGate?.({ scenarios, viewpoints, catalog: gateCatalog, focus: intent.focus, cwd: projectRootFromScreenDir(screenDir), screenName: catalogScreenName, threshold: depthThresholdFor(intent.focus), businessCriticalMethods: intent.businessCriticalMethods, isMobile, viewpointText });
   const gate = provided?.gate ?? viewpointGate(scenarios, viewpoints, gateCatalog, isMobile, viewpointText);
-  const depth = provided?.depth ?? assertionDepth(scenarios, dataThemesFor(gateCatalog, gate.pageType), intent.focus);
+  const depth = provided?.depth
+    ? relevelProvidedDepth(provided.depth, gates.rules)
+    : assertionDepth(scenarios, dataThemesFor(gateCatalog, gate.pageType), intent.focus, gates.rules);
 
   // TQ-4 — deferral-aware coverage credit: an inherently cross-screen theme (cart / detail /
   // filter correctness) belongs in a FLOW, not on the screen. When a flow deeply covers a screen
@@ -251,7 +262,7 @@ export function runAudit(screenDir: string, screenName: string): AuditReport {
       gate.coverageRatio = gate.themesTotal ? gate.themesCovered / gate.themesTotal : 1;
     }
   }
-  const claim = claimProof(scenarios, intent.focus);
+  const claim = claimProof(scenarios, intent.focus, gates.rules);
   const taxonomy = taxonomyLint(scenarios);
   const balance = coverageBalance(scenarios);
   const duplicates = duplicateClusters(scenarios);
@@ -356,8 +367,7 @@ export function runAudit(screenDir: string, screenName: string): AuditReport {
   // so it can only measure what their author already thought of. One field run wrote the file
   // after generating the suite and said so in its own header — and both axes still read 100%,
   // because a first audit records the baseline silently (#622). Untrusted until a human accepts it.
-  // An AI draft rides the same hold: its line is only removed when a human accepts it.
-  const viewpointMoved = viewpointBaseline.status === 'changed' || adoptedPostHoc || aiDraft;
+  const viewpointMoved = viewpointBaseline.status === 'changed' || adoptedPostHoc;
   // #618 — is there a YARDSTICK at all? `atomicLedger` and `traceability` are both measured
   // against test-viewpoint.md, and four more sensors read it (the ledger, the declaration-integrity
   // checks, the browser-gesture check, the continuity check). When the file is absent every one of
@@ -671,12 +681,7 @@ export function runAudit(screenDir: string, screenName: string): AuditReport {
   } else if (balance.imbalanced && !flowScored) {
     findings.push(`BALANCE: ${balance.note} Stop expanding secondary viewpoints until business-core gaps are filled.`);
   }
-  if (aiDraft) {
-    findings.push(`VIEWPOINT-DRAFT-UNREVIEWED: \`requirements/test-viewpoint.md\` is marked \`drafted-by: ai\` and no QA has accepted it, so it is not recorded as the baseline and \`atomicLedger\` + \`traceability\` are held as unverified — a checklist the AI wrote would otherwise score the suite it generated at 100% by construction. QA: review the claims (add what you care about, delete what you do not), then \`sungen audit --screen ${screenName} --accept-viewpoint\` — it records the baseline and removes the \`drafted-by: ai\` line. A QA decision: an agent presents it and does not run it.`);
-  }
-  // create-test's checklist review of the viewpoint, still waiting for QA — advisory, never scored.
-  findings.push(...viewpointReviewOpenFindings(path.dirname(viewpointPath)));
-  if (adoptedPostHoc && !aiDraft) {   // a draft's own finding already says this and names the accept command
+  if (adoptedPostHoc) {
     findings.push(`VIEWPOINT-ADOPTED-POST-HOC: this is the first audit of \`requirements/test-viewpoint.md\` and the suite ALREADY has ${scenarios.length} scenarios, so the yardstick is being introduced against tests that already exist. A declaration written after the suite can only measure what its author already thought of — and if it was derived FROM the suite, \`atomicLedger\` and \`traceability\` read 100% by construction while anything dropped along the way stops being missing from anything. Both axes are held as unverified until you review the claims (add what you care about, delete what you do not) and confirm: \`sungen audit --screen ${screenName} --accept-viewpoint\` — a QA decision: an agent presents it and does not run it.`);
   }
   if (!hasYardstick) {
@@ -923,9 +928,10 @@ export function runAudit(screenDir: string, screenName: string): AuditReport {
   // taxonomy-match (scenarios must use the project's viewpoint IDs when defined),
   // AND scenario-id uniqueness — a reused id reaches the customer's deliverable as two rows
   // under one testcase id, and has already shipped a never-executed testcase as "Passed" (#694).
-  const gateStatus: 'PASS' | 'FAIL' =
-    gate.gaps.length === 0 && depth.verdict !== 'fail' && claim.verdict !== 'fail' && spec.verdict !== 'fail' && !taxonomyMismatch && !gateSensorError
-    && reusedIds.length === 0 ? 'PASS' : 'FAIL';
+  // The conjunction lives in gates/audit-gate-terms.ts, where the report reader re-evaluates it (OES-11.2).
+  findings.push(...decisionFindings(gates));
+  const verdict = auditGateVerdict(gates, { gate, depth, claim, spec, taxonomyMismatch, gateSensorError, reusedScenarioIds: reusedIds }, screenDir, screenName);
+  const gateStatus = verdict.gateStatus;
 
   return {
     screen: screenName,
@@ -957,5 +963,10 @@ export function runAudit(screenDir: string, screenName: string): AuditReport {
     findings,
     intent,
     provenance: getProvenance(),
+    gateOverrides: verdict.gateOverrides,
+    decisionsHash: verdict.decisionsHash,
+    inputsHash: verdict.inputsHash,
+    exitCode: verdict.exitCode,
+    gateSensorError,
   };
 }