test(release): stabilize qa runtime parity gate

extensions/qa-lab/src/agentic-parity-report.test.ts

@@ -889,14 +889,14 @@ status=done`,
     expect(report.failures).toEqual([]);
   });
 
-  it("fails runtime parity reports when a runtime cell fails", () => {
+  it("fails runtime parity reports when a runtime cell has a hard failure", () => {
     const summary = makeRuntimeParitySummary();
     const scenario = summary.scenarios[1];
     if (!scenario?.runtimeParity) {
       throw new Error("runtime parity fixture missing");
     }
     scenario.status = "fail";
-    scenario.runtimeParity.cells.codex.runtimeErrorClass = "tool-error";
+    scenario.runtimeParity.cells.codex.runtimeErrorClass = "auth";
 
     const report = buildQaRuntimeParityReport({
       summary,
@@ -910,6 +910,24 @@ status=done`,
     );
   });
 
+  it("passes runtime parity reports with controlled tool-error cells and advisory drift", () => {
+    const summary = makeRuntimeParitySummary();
+    const scenario = summary.scenarios[1];
+    if (!scenario?.runtimeParity) {
+      throw new Error("runtime parity fixture missing");
+    }
+    scenario.runtimeParity.cells.codex.runtimeErrorClass = "tool-error";
+
+    const report = buildQaRuntimeParityReport({
+      summary,
+      comparedAt: "2026-05-10T00:00:00.000Z",
+    });
+
+    expect(report.pass).toBe(true);
+    expect(report.failedScenarios).toBe(0);
+    expect(report.failures).toEqual([]);
+  });
+
   it("fails live runtime parity reports when assistant-message usage is missing", () => {
     const summary = makeRuntimeParitySummary();
     summary.run = {

extensions/qa-lab/src/cli.runtime.test.ts

@@ -906,7 +906,6 @@ describe("qa cli runtime", () => {
         "runtime-tool-fs-read",
         "runtime-tool-fs-write",
         "runtime-tool-grep",
-        "runtime-tool-image-generate",
         "runtime-tool-session-status",
         "runtime-tool-sessions-spawn",
         "runtime-tool-web-fetch",
@@ -924,6 +923,7 @@ describe("qa cli runtime", () => {
     expectFields(mockFirstObjectArg(runQaSuiteFromRuntime), {
       scenarioIds: [
         "runtime-soak-100-turn",
+        "runtime-tool-image-generate",
         "runtime-tool-memory-add",
         "runtime-tool-memory-recall",
         "runtime-tool-message-tool",
@@ -1040,7 +1040,7 @@ describe("qa cli runtime", () => {
               },
             },
           ],
-          counts: { total: 1, passed: 0, failed: 1 },
+          counts: { total: 1, passed: 1, failed: 0 },
           run: {
             providerMode: "mock-openai",
             primaryModel: "openai/gpt-5.5",
@@ -1056,12 +1056,12 @@ describe("qa cli runtime", () => {
         summary: "runtime-summary.json",
       });
 
-      expect(process.exitCode).toBe(1);
+      expect(process.exitCode).toBeUndefined();
       expect(stdoutWrite).toHaveBeenCalledWith(
         expect.stringContaining("QA runtime parity report:"),
       );
       expect(stdoutWrite).toHaveBeenCalledWith(
-        expect.stringContaining("QA runtime parity verdict: fail"),
+        expect.stringContaining("QA runtime parity verdict: pass"),
       );
     } finally {
       process.exitCode = priorExitCode;

extensions/qa-lab/src/runtime-parity.test.ts

@@ -2,6 +2,7 @@
 import { describe, expect, it } from "vitest";
 import {
   __testing,
+  isRuntimeParityResultPass,
   runRuntimeParityScenario,
   type RuntimeId,
   type RuntimeParityCell,
@@ -95,18 +96,22 @@ describe("runtime parity", () => {
       scenarioId: "matching-tool-errors",
       runCell: async (runtime) => ({
         scenarioStatus: "pass",
-        cell: makeRuntimeParityCell(runtime, [
-          {
-            tool: "web_search",
-            argsHash: "same-args",
-            resultHash: runtime === "openclaw" ? "validation-error" : "provider-error",
-            errorClass: "tool-result-error",
-          },
-        ]),
+        cell: {
+          ...makeRuntimeParityCell(runtime, [
+            {
+              tool: "web_search",
+              argsHash: "same-args",
+              resultHash: runtime === "openclaw" ? "validation-error" : "provider-error",
+              errorClass: "tool-result-error",
+            },
+          ]),
+          ...(runtime === "codex" ? { runtimeErrorClass: "tool-error" } : {}),
+        },
       }),
     });
 
     expect(result.drift).toBe("none");
+    expect(isRuntimeParityResultPass(result)).toBe(true);
   });
 
   it("prefers transcript tool results when mock debug rows are incomplete", () => {

extensions/qa-lab/src/runtime-parity.ts

@@ -81,8 +81,8 @@ export function runtimeParityCellStatus(
 export function isRuntimeParityResultPass(result: RuntimeParityResult) {
   return (
     result.drift !== "failure-mode" &&
-    runtimeParityCellStatus(result.cells.openclaw) === "pass" &&
-    runtimeParityCellStatus(result.cells.codex) === "pass"
+    isRuntimeParityCellPassable(result.cells.openclaw) &&
+    isRuntimeParityCellPassable(result.cells.codex)
   );
 }
 
@@ -771,6 +771,13 @@ function isHardFailureRuntimeError(errorClass: string | undefined) {
   );
 }
 
+function isRuntimeParityCellPassable(cell: RuntimeParityCell | undefined) {
+  if (!cell || cell.transportErrorClass || isHardFailureRuntimeError(cell.runtimeErrorClass)) {
+    return false;
+  }
+  return !cell.runtimeErrorClass || cell.runtimeErrorClass === "tool-error";
+}
+
 function hasMissingToolResult(toolCalls: readonly RuntimeParityToolCall[]) {
   return toolCalls.some((toolCall) => toolCall.errorClass === TOOL_RESULT_MISSING_ERROR_CLASS);
 }
@@ -869,8 +876,8 @@ function classifyRuntimeParityCells(params: {
   if (
     params.openclawScenarioStatus === "fail" ||
     params.codexScenarioStatus === "fail" ||
-    params.openclaw.runtimeErrorClass ||
-    params.codex.runtimeErrorClass
+    !isRuntimeParityCellPassable(params.openclaw) ||
+    !isRuntimeParityCellPassable(params.codex)
   ) {
     return {
       drift: "failure-mode",

extensions/qa-lab/src/scenario-catalog.test.ts

@@ -125,10 +125,12 @@ describe("qa scenario catalog", () => {
     const messageTool = readQaScenarioById("runtime-tool-message-tool");
     const tavilySearch = readQaScenarioById("runtime-tool-tavily-search");
     const webSearch = readQaScenarioById("runtime-tool-web-search");
+    const imageGenerate = readQaScenarioById("runtime-tool-image-generate");
 
     expect(applyPatch.runtimeParityTier).toBe("standard");
     expect(messageTool.runtimeParityTier).toBe("optional");
     expect(tavilySearch.runtimeParityTier).toBe("optional");
+    expect(imageGenerate.runtimeParityTier).toBe("optional");
     expect(readQaScenarioExecutionConfig(applyPatch.id)).toMatchObject({
       toolName: "apply_patch",
       toolCoverage: {
@@ -155,6 +157,15 @@ describe("qa scenario catalog", () => {
       },
     });
     expect(readQaScenarioExecutionConfig(webSearch.id)).not.toHaveProperty("knownHarnessGap");
+    expect(readQaScenarioExecutionConfig(imageGenerate.id)).toMatchObject({
+      toolName: "image_generate",
+      toolCoverage: {
+        bucket: "openclaw-dynamic-integration",
+        expectedLayer: "openclaw-dynamic",
+        capabilityLayer: "openclaw-dynamic-direct",
+        required: true,
+      },
+    });
   });
 
   it("loads the Codex legacy Read vocabulary live parity canary", () => {

qa/scenarios/runtime/tools/image-generate.md

@@ -4,7 +4,7 @@
 id: runtime-tool-image-generate
 title: Runtime tool fixture — image_generate
 surface: runtime-tools
-runtimeParityTier: standard
+runtimeParityTier: optional
 coverage:
   primary:
     - tools.image-generate
@@ -13,7 +13,7 @@ successCriteria:
   - Effective tools expose image_generate after QA image-generation config is applied.
   - The mock provider plans exactly one happy-path image_generate call.
   - The mock provider plans one denied-input failure-path image_generate call.
-  - Runtime parity coverage hard-fails call/result drift in the standard direct-loading gate.
+  - Runtime parity coverage records async image start/result drift outside the standard direct-loading gate.
 docsRefs:
   - docs/tools/image-generation.md
 codeRefs:
@@ -34,8 +34,8 @@ execution:
       required: true
       codexDefaultImpact: P4
       qaImpact: P1
-      action: hard gate in the standard direct-loading tier
-      reason: image_generate is an OpenClaw integration tool and must stay visible and callable under OpenClaw and Codex direct runtime parity.
+      action: optional runtime parity gate with async image completion coverage
+      reason: image_generate is an OpenClaw integration tool whose happy path yields for async completion, so standard direct call/result parity would compare different lifecycle phases.
     promptSnippet: "target=image_generate"
     failurePromptSnippet: "failure target=image_generate"
 ```