@validate.qa/runner 1.0.89 → 1.0.91

package/dist/cli.js

@@ -80738,7 +80738,7 @@ Return ONLY one valid JSON object matching the schema below. Do NOT include <thi
 2. **One purpose per scenario** - each scenario tests ONE thing (happy path, validation, journey, or bug)
 3. **Happy paths first** - then pair edge cases to the same form/flow
 4. **Multi-screen journeys** - only when screen transitions were actually observed. When an observed target element sits below the fold, include an explicit swipe step to scroll it into view before the tap (mobile elements off-screen are not hittable until revealed)
-5. **Bug reports** - for observed crashes, ANRs, or broken features
+5. **Bug reports** - for observed *product* defects (wrong row deleted, validation bypassed, count wrong). Do NOT plan scenarios whose expected outcome is system ANR / "isn't responding" freezes \u2014 those are environment noise
 6. **Steps are human-readable** - "Type into the email field", not "driver.findElement(AppiumBy.accessibilityId('email')).sendKeys()"
 7. **Expected outcomes are plain-English success criteria** - describe the user-visible product result, not the controls touched along the way. Steps already cover buttons, inputs, tabs, sheets, and menus. Expected outcomes should say what the user can now do or rely on (record saved, message sent, search results filtered, validation explained, home screen reached). Name the target screen or exact copy only when it is the real product outcome; do NOT list "heading visible", "button visible", or similar element inventory unless that exact control is the behavior under test.
 8. **Auth-aware** - mark scenarios that need a signed-in session. Credentials are filled ON-DEVICE from the stored TEST_USER_EMAIL / TEST_USER_PASSWORD keys (key-based injection at run time) \u2014 say "type the test account email", never a literal value and never a code idiom like process.env
@@ -81150,7 +81150,7 @@ ${context.userNotes.trim()}`);
         const requiredBlock = (0, scenario_taxonomy_js_1.formatRequiredScenarioKindsPrompt)(signals);
         if (requiredBlock) {
           sections.push(requiredBlock);
-          sections.push('## Scenario kind coverage rules\n1. Prefer scenarios whose expectedOutcomes prove durable product results (state deltas, validation errors, persistence), not only "control X is visible".\n2. For negative_validation: empty required field and/or malformed email-like input when an email field was observed.\n3. For crud_identity: assert the specific entity identity is gone/changed \u2014 never only that the list screen is still visible.\n4. For count_delta: mention the before/after numeric expectation in expectedOutcomes.\n5. If a required kind is impossible given evidence, say so in "reasoning" \u2014 do not invent unobserved screens.');
+          sections.push('## Scenario kind coverage rules\n1. Prefer scenarios whose expectedOutcomes prove durable product results (state deltas, validation errors, persistence), not only "control X is visible".\n2. For negative_validation: empty required field AND/OR malformed email-like input; expected outcome must be blocked submit or visible validation error \u2014 never success Home after invalid submit.\n3. For crud_identity: capture entity identity before delete/edit, then assert that identity is gone/changed (state delta) \u2014 never only that the list screen is still visible.\n4. For count_delta: capture before count, mutate, assert count changed as expected.\n5. For double_submit: assert count +1 not +2 (or single confirmation), not merely that a button remains visible.\n6. Never plan expectedOutcomes that are ANR / System UI freezes.\n7. If a required kind is impossible given evidence, say so in "reasoning" \u2014 do not invent unobserved screens.');
         }
       }
       const orderedScreens = plannableTargetAppScreens(graph, context.appId).sort((a, b) => a.firstSeenIteration - b.firstSeenIteration || a.screenId.localeCompare(b.screenId));
@@ -82026,6 +82026,123 @@ var require_discovery_lint_flags = __commonJS({
   }
 });
 
+// ../runner-core/dist/services/mobile/anr-env.js
+var require_anr_env = __commonJS({
+  "../runner-core/dist/services/mobile/anr-env.js"(exports2) {
+    "use strict";
+    Object.defineProperty(exports2, "__esModule", { value: true });
+    exports2.textLooksLikeAnrOrSystemUi = textLooksLikeAnrOrSystemUi;
+    exports2.isAnrOrSystemUiEnvironmentFailure = isAnrOrSystemUiEnvironmentFailure;
+    exports2.stepLooksLikeAnrChromeAssert = stepLooksLikeAnrChromeAssert;
+    var ANR_TEXT_RE = /system\s*ui\s*isn['’]?t\s*responding|app\s*isn['’]?t\s*responding|application\s*not\s*responding|\banr\b|close\s*app|wait\s*and\s*see|isn['’]?t\s*responding/i;
+    function textLooksLikeAnrOrSystemUi(text) {
+      if (!text || typeof text !== "string")
+        return false;
+      return ANR_TEXT_RE.test(text);
+    }
+    function isAnrOrSystemUiEnvironmentFailure(error2, logs2) {
+      const blob = `${error2 ?? ""}
+${logs2 ?? ""}`;
+      if (textLooksLikeAnrOrSystemUi(blob))
+        return true;
+      if (/\bcold-boot\b/i.test(blob) && /\b(froze|hang|anr|unresponsive)\b/i.test(blob))
+        return true;
+      if (/\bno actionable elements\b/i.test(blob) && /\b(system|anr|responding)\b/i.test(blob)) {
+        return true;
+      }
+      return false;
+    }
+    function stepLooksLikeAnrChromeAssert(step) {
+      if (step.action !== "assertVisible" && step.action !== "assertText")
+        return false;
+      return textLooksLikeAnrOrSystemUi([step.description, step.assertion, step.value].filter(Boolean).join(" "));
+    }
+  }
+});
+
+// ../runner-core/dist/services/mobile/oracle-quality.js
+var require_oracle_quality = __commonJS({
+  "../runner-core/dist/services/mobile/oracle-quality.js"(exports2) {
+    "use strict";
+    Object.defineProperty(exports2, "__esModule", { value: true });
+    exports2.assessStrongOracleQuality = assessStrongOracleQuality;
+    var anr_env_js_1 = require_anr_env();
+    var CHROME_ONLY_RE = /\b(tab|home tab|catalog tab|forms tab|account tab|back|menu|navigation|nav bar|bottom nav|status bar|query|search field)\b/i;
+    var PRODUCT_OUTCOME_TEXT_RE = /\b(error|invalid|required|must|cannot|failed|denied|blocked|empty|missing|deleted|removed|saved|success|created|updated|signed?\s*out|logged?\s*out|wrong|incorrect)\b/i;
+    function stepBlob(step) {
+      return [step.description, step.assertion, step.value, step.target?.primary].filter(Boolean).join(" ");
+    }
+    function isChromeOnlyVisible(step) {
+      if (step.action !== "assertVisible")
+        return false;
+      const blob = stepBlob(step);
+      if (PRODUCT_OUTCOME_TEXT_RE.test(blob))
+        return false;
+      return CHROME_ONLY_RE.test(blob) || blob.length < 40;
+    }
+    function assessStrongOracleQuality(steps) {
+      const productSteps = steps.filter((s) => s.action !== "launchApp");
+      const anrAsserts = productSteps.filter(anr_env_js_1.stepLooksLikeAnrChromeAssert);
+      const nonAnrAsserts = productSteps.filter((s) => (s.action === "assertVisible" || s.action === "assertText" || s.action === "assertStateDelta") && !(0, anr_env_js_1.stepLooksLikeAnrChromeAssert)(s));
+      if (productSteps.length > 0 && anrAsserts.length > 0 && nonAnrAsserts.length === 0) {
+        return {
+          strong: false,
+          unpairedCaptures: 0,
+          hasStateDelta: false,
+          hasStrongAssertText: false,
+          hasStrongValidationVisible: false,
+          anrOnly: true,
+          reason: "only asserts system ANR / freeze chrome \u2014 not a product test"
+        };
+      }
+      const captures = steps.filter((s) => s.action === "captureState").length;
+      const deltas = steps.filter((s) => s.action === "assertStateDelta").length;
+      const unpairedCaptures = Math.max(0, captures - deltas);
+      const hasStateDelta = deltas > 0;
+      const hasStrongAssertText = steps.some((s) => {
+        if (s.action !== "assertText")
+          return false;
+        if ((0, anr_env_js_1.stepLooksLikeAnrChromeAssert)(s))
+          return false;
+        const blob = stepBlob(s);
+        if (/^(email|password|username|name|query|search)$/i.test((s.assertion ?? "").trim())) {
+          return false;
+        }
+        return blob.length >= 3;
+      });
+      const hasStrongValidationVisible = steps.some((s) => {
+        if (s.action !== "assertVisible")
+          return false;
+        if ((0, anr_env_js_1.stepLooksLikeAnrChromeAssert)(s) || isChromeOnlyVisible(s))
+          return false;
+        return PRODUCT_OUTCOME_TEXT_RE.test(stepBlob(s));
+      });
+      const strong = hasStateDelta || hasStrongAssertText || hasStrongValidationVisible;
+      let reason;
+      if (strong && hasStateDelta) {
+        reason = unpairedCaptures > 0 ? `has assertStateDelta but ${unpairedCaptures} unpaired captureState (prefer 1:1 pairing)` : "has assertStateDelta durable state proof";
+      } else if (strong && hasStrongAssertText) {
+        reason = "has product assertText outcome";
+      } else if (strong) {
+        reason = "has validation/error assertVisible outcome";
+      } else if (unpairedCaptures > 0) {
+        reason = `captureState without assertStateDelta (${unpairedCaptures}) \u2014 not strong`;
+      } else {
+        reason = "no durable state/text/validation oracle \u2014 smoke/nav only";
+      }
+      return {
+        strong,
+        unpairedCaptures,
+        hasStateDelta,
+        hasStrongAssertText,
+        hasStrongValidationVisible,
+        anrOnly: false,
+        reason
+      };
+    }
+  }
+});
+
 // ../runner-core/dist/services/mobile/mobile-generator.js
 var require_mobile_generator = __commonJS({
   "../runner-core/dist/services/mobile/mobile-generator.js"(exports2) {
@@ -82052,6 +82169,8 @@ var require_mobile_generator = __commonJS({
     var mobile_session_keepalive_js_1 = require_mobile_session_keepalive();
     var discovery_lint_flags_js_1 = require_discovery_lint_flags();
     var scenario_taxonomy_js_1 = require_scenario_taxonomy();
+    var anr_env_js_1 = require_anr_env();
+    var oracle_quality_js_1 = require_oracle_quality();
     var mobile_source_parser_js_1 = require_mobile_source_parser();
     var mobile_source_parser_js_2 = require_mobile_source_parser();
     var mobile_observation_js_1 = require_mobile_observation();
@@ -82197,7 +82316,7 @@ var require_mobile_generator = __commonJS({
       type: "function",
       function: {
         name: "mobile_capture_state",
-        description: 'Capture list labels or numeric badge counts into a named state key BEFORE a mutating action (delete/create/toggle). Pair with mobile_assert_state_delta afterward to prove durable identity/count outcomes \u2014 stronger than "list still visible".',
+        description: "Capture list labels or numeric badge counts into a named state key BEFORE a mutating action (delete/create/toggle). MUST pair with mobile_assert_state_delta afterward \u2014 capture alone is not a strong oracle and will not tag the test @strong-oracle.",
         parameters: {
           type: "object",
           properties: {
@@ -82559,9 +82678,11 @@ The scenario name + goal + steps describe ONE product behavior from survey/disco
    - Navigation scenarios: you MUST land on the destination and assert something unique to it.
 4. PROVE the outcome with an assert tool that matches the scenario \u2014 not ambient chrome:
    - mobile_assert_visible({ref}) / mobile_assert_text({ref, expected})
-   - For list delete/create or count changes: mobile_capture_state({stateKey, kind}) BEFORE the mutation, then mobile_assert_state_delta({stateKey, expect}) AFTER \u2014 proves identity/count change, not only "list screen still visible".
+   - For list delete/create or count changes: mobile_capture_state({stateKey, kind}) BEFORE the mutation, then mobile_assert_state_delta({stateKey, expect}) AFTER \u2014 BOTH are required. captureState alone does NOT prove the outcome.
+   - Negative validation: leave a required field EMPTY (or type clearly invalid email-like input), submit, assert the app BLOCKS success \u2014 error message / still on form / no navigation to Home. Do NOT fill a valid password-only path and call it "empty email".
    - After sign-in \u2192 signed-in product signal; after like \u2192 liked/unlike control or durable like state; after open Settings \u2192 a settings-specific control.
    - Do NOT assert bottom-nav tab labels alone, notification badge counts ("N unread"), or one-time empty-state CTAs as the scenario outcome.
+   - NEVER assert system freeze chrome ("System UI isn't responding", ANR dialogs, Wait/Close app). If the app freezes, finish_generate with verified:false (environment) \u2014 do NOT write a test whose only proof is the ANR dialog.
 5. Call finish_generate with verified:true ONLY if you completed the goal AND asserted a matching durable outcome. If the app cannot complete the goal, finish with verified:false and explain \u2014 never invent a green path from unrelated screens.
 
 ## Rules
@@ -82570,7 +82691,7 @@ The scenario name + goal + steps describe ONE product behavior from survey/disco
 - Assert only DURABLE elements \u2014 screen chrome that exists on every visit (tab labels, headers, nav bars, the entity you just created). NEVER assert one-time / data-dependent UI: empty-state CTAs, first-run or onboarding prompts, or counters \u2014 they vanish as soon as the account has data, and the test then fails forever on a healthy app.
 - Do NOT trigger irreversible or destructive side effects \u2014 real payments; sending messages to others; and ESPECIALLY account/data deletion, deactivation, or any "type DELETE to confirm" gate (the runner hard-blocks destructive taps and confirmation keystrokes). Note them and assert what you safely can.
 - STATE OWNERSHIP for mutation scenarios (edit / rename / archive / remove): CREATE the record first in this same recording, then act on the row YOU created \u2014 never mutate pre-existing data (it may not exist on a clean install, so the replay fails forever, and another test may depend on it). If the app offers no way to create one, do NOT act on someone else's data: assert what you can safely prove and finish with verified:false, explaining why in the description.
-- If the APP itself misbehaves (crash, error screen, a flow that cannot proceed as specified), do not force a green path: assert the failure you observed (e.g. the error text), then call finish_generate with suspectedBug:true and a concrete bugDescription of what is broken in the app.
+- If the APP itself misbehaves (product error screen, broken validation, wrong entity deleted), assert the product failure and call finish_generate with suspectedBug:true and a concrete bugDescription. If the device shows ANR / "System UI isn't responding" / freeze dialogs, that is ENVIRONMENT \u2014 finish with verified:false and do NOT assert the freeze dialog as the test outcome.
 - Stay inside this app.${targetLine} Do not open or use app stores, browsers, email apps, settings, or any other app. If an action leaves the target app, the runner will relaunch the target app and reject that action.
 - NEVER tap the phone's home screen / launcher (app icons, "Predicted app" chips, search bars, widgets). Those are NOT part of the app under test \u2014 tapping them is junk that poisons the test. If the app ever backgrounds or you find yourself on the launcher, use mobile_relaunch to cold-start the app; do NOT tap a launcher icon or app-prediction to get back in.
 - Do NOT re-do a sequence you already completed. If you filled a form then navigated away, do NOT come back and fill it again \u2014 the scenario needs ONE clean pass. Filling the login form, going back, and re-filling it is exactly the kind of wandering that produces a bloated, broken test.
@@ -84162,6 +84283,9 @@ ${statePacket}`, redactionCredentials) },
     function weakAssertionTargetReason(element) {
       if (isAlwaysPresentContainer(element))
         return "always-present app/system container";
+      if ((0, anr_env_js_1.textLooksLikeAnrOrSystemUi)([element.text, element.label, element.accessibilityId, element.resourceId].filter(Boolean).join(" "))) {
+        return "system ANR / freeze dialog chrome (environment, not product)";
+      }
       const hasIdentitySignal = Boolean(nonEmpty3(element.accessibilityId) || nonEmpty3(element.resourceId) || nonEmpty3(element.text) || nonEmpty3(element.label));
       if (!hasIdentitySignal)
         return "no identity locator (accessibility id / resource id / text) for the assert gate";
@@ -85023,6 +85147,30 @@ ${statePacket}`, redactionCredentials) },
       return [action.selector, action.value, action.innerText, action.metadata?.targetLabel].filter((v) => typeof v === "string" && v.trim().length > 0).join(" ");
     }
     function evaluateMobileAssertionStrength(actions) {
+      const anrAsserts = actions.filter((a) => {
+        if (a.type !== "ASSERT_VISIBLE" && a.type !== "ASSERT_TEXT")
+          return false;
+        return (0, anr_env_js_1.textLooksLikeAnrOrSystemUi)([a.value, a.innerText, a.metadata?.targetLabel, a.selector].filter(Boolean).join(" "));
+      });
+      if (anrAsserts.length > 0) {
+        const otherAsserts = actions.filter((a) => (a.type === "ASSERT_VISIBLE" || a.type === "ASSERT_TEXT" || a.type === "ASSERT_STATE_DELTA") && !anrAsserts.includes(a));
+        if (otherAsserts.length === 0) {
+          return {
+            ok: false,
+            severity: "block",
+            reason: `only asserts system ANR / "isn't responding" chrome \u2014 not a product test (discard or re-record on a healthy session)`,
+            signals: {
+              apiTest: false,
+              interacted: actions.some((a) => a.type === "TAP" || a.type === "TYPE"),
+              mutated: false,
+              navOnly: false,
+              hasOutcomeProof: false,
+              errorBanner: false,
+              terminal: null
+            }
+          };
+        }
+      }
       if (actions.some((a) => a.type === "ASSERT_STATE_DELTA")) {
         return {
           ok: true,
@@ -85169,9 +85317,10 @@ ${statePacket}`, redactionCredentials) },
       }
       const baseStrengthVerdict = evaluateMobileAssertionStrength(actions);
       const bugDescription = recording.finish?.bugDescription?.trim() ?? "";
-      const bugWitnessHonored = process.env.DISCOVERY_BUG_WITNESS_ENABLED === "true" && recording.finish?.suspectedBug === true && bugDescription.length >= 12 && recording.assertCount > 0;
+      const bugLooksLikeAnr = (0, anr_env_js_1.textLooksLikeAnrOrSystemUi)(bugDescription);
+      const bugWitnessHonored = process.env.DISCOVERY_BUG_WITNESS_ENABLED === "true" && recording.finish?.suspectedBug === true && bugDescription.length >= 12 && recording.assertCount > 0 && !bugLooksLikeAnr;
       if (recording.finish?.suspectedBug === true && !bugWitnessHonored) {
-        log2("  suspectedBug declared but NOT honored: " + (process.env.DISCOVERY_BUG_WITNESS_ENABLED !== "true" ? "bug-witness channel disabled (DISCOVERY_BUG_WITNESS_ENABLED)." : bugDescription.length < 12 ? "bugDescription too short." : "no honored assert backs the observation."));
+        log2("  suspectedBug declared but NOT honored: " + (bugLooksLikeAnr ? "ANR/System UI freeze is environment noise, not a product bug-witness." : process.env.DISCOVERY_BUG_WITNESS_ENABLED !== "true" ? "bug-witness channel disabled (DISCOVERY_BUG_WITNESS_ENABLED)." : bugDescription.length < 12 ? "bugDescription too short." : "no honored assert backs the observation."));
       }
       const strengthVerdict = bugWitnessHonored ? {
         ...baseStrengthVerdict,
@@ -85210,12 +85359,22 @@ ${statePacket}`, redactionCredentials) },
       const strengthBlocked = !strengthVerdict.ok || strengthVerdict.severity === "block" || strengthVerdict.severity === "route-bug" || !verified || coldResetUnavailable;
       {
         const kind = (0, scenario_taxonomy_js_1.inferScenarioKind)(scenario);
-        const hasStateOracle = compiled.steps.some((s) => s.action === "assertStateDelta" || s.action === "assertText" || s.action === "captureState");
-        const strong = verified && !strengthBlocked && strengthVerdict.severity === "none" && ((0, scenario_taxonomy_js_1.isStrongOracleKind)(kind) || hasStateOracle);
+        const oracleQuality = (0, oracle_quality_js_1.assessStrongOracleQuality)(compiled.steps);
+        if (oracleQuality.anrOnly) {
+          log2(`  discarding ANR-chrome-only recording: ${oracleQuality.reason}`);
+          return null;
+        }
+        if (oracleQuality.unpairedCaptures > 0) {
+          log2(`  oracle quality: ${oracleQuality.reason}`);
+        }
+        const strong = verified && !strengthBlocked && strengthVerdict.severity === "none" && oracleQuality.strong && ((0, scenario_taxonomy_js_1.isStrongOracleKind)(kind) || oracleQuality.hasStateDelta || oracleQuality.hasStrongAssertText);
         if (strong)
           tags.push("@strong-oracle");
-        else if (!tags.includes("@smoke") && (kind === "smoke_nav" || kind === "navigation_integrity")) {
+        else if (!tags.includes("@smoke")) {
           tags.push("@smoke");
+          if (!oracleQuality.strong) {
+            log2(`  tagged @smoke (not strong-oracle): ${oracleQuality.reason}`);
+          }
         }
       }
       const strengthReason = coldResetUnavailable ? `Cold app-data reset unavailable${coldResetReason ? ` (${coldResetReason})` : ""} \u2014 generation may not match cold-boot replay. ${strengthVerdict.reason}` : !verified && recording.assertCount === 0 ? "Mobile scenario finished unverified \u2014 the agent never proved the scenario outcome live, so the recording may not demonstrate the goal (e.g. a happy path that never completed). Review before promotion." : strengthVerdict.reason;
@@ -85475,6 +85634,7 @@ var require_failure_triage = __commonJS({
     exports2.listStrongOracleKinds = listStrongOracleKinds;
     exports2.triageMobileFailure = triageMobileFailure;
     exports2.triageToHealTerminal = triageToHealTerminal;
+    var anr_env_js_1 = require_anr_env();
     var STRONG_ACTIONS = /* @__PURE__ */ new Set(["assertStateDelta", "assertText", "captureState"]);
     var CHROME_VISIBLE_RE = /\b(tab|home|settings|back|menu|navigation|nav bar|bottom nav|status bar)\b/i;
     function hasStrongMobileOracle(steps) {
@@ -85564,6 +85724,15 @@ ${logs2 ?? ""}`.trim();
         failureCount,
         errorPrefix: prefix
       };
+      if ((0, anr_env_js_1.isAnrOrSystemUiEnvironmentFailure)(input.error, input.logs)) {
+        return {
+          verdict: "config_issue",
+          reason: "ANR / System UI freeze (device or app hung) \u2014 treat as environment, not product bug",
+          shouldHeal: false,
+          healVerdict: "config_issue",
+          evidence
+        };
+      }
       if (prefix === "CONFIG_ISSUE" || prefix === "RATE_LIMIT_BLOCKED") {
         return {
           verdict: "config_issue",
@@ -85812,6 +85981,7 @@ var require_mobile_heal_postprocess = __commonJS({
     exports2.detectMobileStepSecretLeak = detectMobileStepSecretLeak;
     exports2.computeMobileHealOutcome = computeMobileHealOutcome2;
     var oracle_protection_js_1 = require_oracle_protection();
+    var anr_env_js_1 = require_anr_env();
     var REDACT_KEY_RE = /email|username|user|phone|password|secret|token|api[_-]?key|access[_-]?key|\bpass\b|credential|bearer|\bpat\b|session|cookie|jwt|otp|pin/i;
     var DETECT_KEY_RE = /password|secret|token|api[_]?key|\bpass\b|credential|bearer|\bpat\b/i;
     var MIN_REDACT_LEN = 4;
@@ -85882,21 +86052,26 @@ var require_mobile_heal_postprocess = __commonJS({
       const oracleWeakening = isVerifiedFix && args.originalSteps ? (0, oracle_protection_js_1.rejectOracleWeakeningPatch)(args.originalSteps, heal.healedSteps ?? []) : null;
       if (oracleWeakening)
         isVerifiedFix = false;
-      const isConfigIssue = heal.verdict === "config_issue";
-      const isAppBug = heal.verdict === "app_bug" || Boolean(oracleWeakening && heal.passed);
+      const anrEnv = (0, anr_env_js_1.isAnrOrSystemUiEnvironmentFailure)(heal.healReason ?? args.finalError ?? args.reproduceError, typeof args.finalError === "string" ? args.finalError : void 0);
+      const isConfigIssue = heal.verdict === "config_issue" || heal.verdict === "app_bug" && anrEnv;
+      const isAppBug = !anrEnv && (heal.verdict === "app_bug" || Boolean(oracleWeakening && heal.passed));
       const extraTags = new Set((heal.extraTags ?? []).filter((t) => typeof t === "string" && t.length > 0));
       if (secretLeak)
         extraTags.add("@needs-review");
-      if (oracleWeakening) {
+      if (oracleWeakening && !anrEnv) {
         extraTags.add("@needs-review");
         extraTags.add("@bug-found");
       }
+      if (anrEnv) {
+        extraTags.delete("@bug-found");
+        extraTags.add("@needs-review");
+      }
       if (!heal.passed && NEEDS_REVIEW_VERDICTS.has(heal.verdict))
         extraTags.add("@needs-review");
       let status;
       if (secretLeak)
         status = "FAILED";
-      else if (oracleWeakening)
+      else if (oracleWeakening && !anrEnv)
         status = "FAILED";
       else if (isVerifiedFix)
         status = "PASSED";
@@ -85909,10 +86084,11 @@ var require_mobile_heal_postprocess = __commonJS({
         error2 = void 0;
       } else if (secretLeak) {
         error2 = `HEAL_SECRET_LEAK: healed steps contained a literal secret (${secretLeak}); patch dropped, tagged @needs-review`;
-      } else if (oracleWeakening) {
+      } else if (oracleWeakening && !anrEnv) {
         error2 = `APP_BUG: heal attempted to drop strong oracle (${oracleWeakening}); patch refused`;
       } else if (isConfigIssue) {
-        error2 = `CONFIG_ISSUE: ${heal.healReason ?? args.finalError ?? args.reproduceError ?? "environment/rate-limit blocker"}`;
+        const reason = heal.healReason ?? args.finalError ?? args.reproduceError ?? "environment/rate-limit blocker";
+        error2 = anrEnv ? `CONFIG_ISSUE: ANR / System UI freeze (not a product assertion failure) \u2014 ${reason}` : `CONFIG_ISSUE: ${reason}`;
       } else if (isAppBug) {
         error2 = `APP_BUG: ${heal.healReason ?? args.finalError ?? args.reproduceError ?? "application defect found during heal"}`;
       } else if (heal.passed && heal.verdict === "fix" && !hasFix) {
@@ -86355,6 +86531,7 @@ var require_mobile_heal_loop = __commonJS({
     var failure_triage_js_1 = require_failure_triage();
     var oracle_protection_js_1 = require_oracle_protection();
     var network_oracle_js_1 = require_network_oracle();
+    var anr_env_js_1 = require_anr_env();
     var mobile_observation_js_1 = require_mobile_observation();
     var selector_registry_context_js_1 = require_selector_registry_context();
     var mobile_session_keepalive_js_1 = require_mobile_session_keepalive();
@@ -86366,12 +86543,13 @@ Reason over steps + error + tree + registry and return working healed steps. Do
 2. STEP RESULTS: which steps passed before the failure.
 3. TREE: does the current screen match what the failed step expects?
 3b. STALE EMPTY-STATE ASSERT: a failing assert on one-time / data-dependent UI (empty-state CTAs, first-run prompts, onboarding banners) is NOT a locator problem \u2014 the element is gone because the account now has data. REPLACE the assertion with a durable landing-screen anchor (tab label, header, nav bar) that proves the same outcome (verdict fix).
+3c. ANR / "System UI isn't responding" / freeze dialogs: this is ENVIRONMENT (config_issue), NOT app_bug. Do not quarantine product bugs for device freezes.
 4. EXTERNAL AUTH HANDOFF: if the failure says the foreground app changed outside the target app after a Login / Sign In tap, and the steps never enter email/password credentials, do NOT heal by choosing Google/Apple/Facebook/SSO or another external app. If the test later expects Logout/session state, return test_invalid or prereq_unmet: the recorded test lacks a real credential setup.
 5. ENTRY-SCREEN DRIFT ("unexpected screen"/"screen drift"): the app started on a DIFFERENT screen than recorded; the fix is usually a MISSING NAVIGATION step (e.g. tap the onboarding/welcome CTA present in the tree before the recorded first step). Author it from the tree if you can (verdict fix); if a single static tree is insufficient, return needs_device.
 ## Response \u2014 STRICT JSON, exactly one verdict:
 {"verdict":"fix","steps":[<FULL healed MobileTestStep array, all steps, your changes applied, order/metadata preserved>],"reason":"what changed and why","confidence":0.0-1.0}
-{"verdict":"app_bug","reason":"app crashed/blank/5xx (NOT rate-limit)"}
-{"verdict":"config_issue","reason":"429/rate limit/lockout/quota"}
+{"verdict":"app_bug","reason":"product defect with durable oracle evidence (wrong entity, validation bypass, wrong count) \u2014 NOT ANR/System UI"}
+{"verdict":"config_issue","reason":"429/rate limit/lockout/quota OR ANR/System UI freeze/device unresponsive"}
 {"verdict":"needs_device","reason":"cannot fix from a static tree; need live re-observation"}
 {"verdict":"prereq_unmet","reason":"required entity/state absent for the account"}
 {"verdict":"state_stale","reason":"leftover data blocks a baseline assertion"}
@@ -86382,7 +86560,7 @@ For a fix, return the COMPLETE steps array (not a diff).
 - NEVER remove or downgrade assertStateDelta, captureState, or assertText outcome proof to go green.
 - NEVER replace a durable outcome assert with mere assertVisible of a tab/header.
 - You MAY fix locators, waits, and navigation around those oracles.
-- If the app truly violates the oracle, return verdict "app_bug" instead of stripping the assert.`;
+- If the app truly violates the oracle on a healthy UI, return verdict "app_bug" instead of stripping the assert.`;
     var DEFAULT_SCRIPT_MAX_ATTEMPTS = 6;
     var RECOVERABLE_MAX = 3;
     var MAX_CONSECUTIVE_INVALID_VERDICTS = 3;
@@ -87350,6 +87528,18 @@ ${args.directive}`);
           providerUnavailable: true
         };
       }
+      if (script.verdict === "app_bug" && (0, anr_env_js_1.isAnrOrSystemUiEnvironmentFailure)(script.healReason ?? ctx.error, ctx.logs)) {
+        return {
+          passed: false,
+          verdict: "config_issue",
+          healedSteps: script.healedSteps,
+          healedAppiumCode: script.healedAppiumCode,
+          healReason: `ANR / System UI freeze (environment) \u2014 ${script.healReason ?? ctx.error ?? "device unresponsive"}`,
+          confidence: script.confidence,
+          extraTags: script.extraTags,
+          removeTags: [...script.removeTags ?? [], "@bug-found"].filter((t, i, a) => a.indexOf(t) === i)
+        };
+      }
       const shouldEscalate = script.verdict === "needs_device" || script.verdict === "fix" && !script.passed;
       if (shouldEscalate && options.runLiveHeal) {
         try {
@@ -355106,7 +355296,7 @@ var require_package4 = __commonJS({
   "package.json"(exports2, module2) {
     module2.exports = {
       name: "@validate.qa/runner",
-      version: "1.0.89",
+      version: "1.0.91",
       description: "validate local test runner - execute Playwright tests on your machine, connected to validate.qa cloud.",
       bin: {
         "validate-runner": "dist/cli.js"

package/dist/cli.mjs

@@ -80743,7 +80743,7 @@ Return ONLY one valid JSON object matching the schema below. Do NOT include <thi
 2. **One purpose per scenario** - each scenario tests ONE thing (happy path, validation, journey, or bug)
 3. **Happy paths first** - then pair edge cases to the same form/flow
 4. **Multi-screen journeys** - only when screen transitions were actually observed. When an observed target element sits below the fold, include an explicit swipe step to scroll it into view before the tap (mobile elements off-screen are not hittable until revealed)
-5. **Bug reports** - for observed crashes, ANRs, or broken features
+5. **Bug reports** - for observed *product* defects (wrong row deleted, validation bypassed, count wrong). Do NOT plan scenarios whose expected outcome is system ANR / "isn't responding" freezes \u2014 those are environment noise
 6. **Steps are human-readable** - "Type into the email field", not "driver.findElement(AppiumBy.accessibilityId('email')).sendKeys()"
 7. **Expected outcomes are plain-English success criteria** - describe the user-visible product result, not the controls touched along the way. Steps already cover buttons, inputs, tabs, sheets, and menus. Expected outcomes should say what the user can now do or rely on (record saved, message sent, search results filtered, validation explained, home screen reached). Name the target screen or exact copy only when it is the real product outcome; do NOT list "heading visible", "button visible", or similar element inventory unless that exact control is the behavior under test.
 8. **Auth-aware** - mark scenarios that need a signed-in session. Credentials are filled ON-DEVICE from the stored TEST_USER_EMAIL / TEST_USER_PASSWORD keys (key-based injection at run time) \u2014 say "type the test account email", never a literal value and never a code idiom like process.env
@@ -81155,7 +81155,7 @@ ${context.userNotes.trim()}`);
         const requiredBlock = (0, scenario_taxonomy_js_1.formatRequiredScenarioKindsPrompt)(signals);
         if (requiredBlock) {
           sections.push(requiredBlock);
-          sections.push('## Scenario kind coverage rules\n1. Prefer scenarios whose expectedOutcomes prove durable product results (state deltas, validation errors, persistence), not only "control X is visible".\n2. For negative_validation: empty required field and/or malformed email-like input when an email field was observed.\n3. For crud_identity: assert the specific entity identity is gone/changed \u2014 never only that the list screen is still visible.\n4. For count_delta: mention the before/after numeric expectation in expectedOutcomes.\n5. If a required kind is impossible given evidence, say so in "reasoning" \u2014 do not invent unobserved screens.');
+          sections.push('## Scenario kind coverage rules\n1. Prefer scenarios whose expectedOutcomes prove durable product results (state deltas, validation errors, persistence), not only "control X is visible".\n2. For negative_validation: empty required field AND/OR malformed email-like input; expected outcome must be blocked submit or visible validation error \u2014 never success Home after invalid submit.\n3. For crud_identity: capture entity identity before delete/edit, then assert that identity is gone/changed (state delta) \u2014 never only that the list screen is still visible.\n4. For count_delta: capture before count, mutate, assert count changed as expected.\n5. For double_submit: assert count +1 not +2 (or single confirmation), not merely that a button remains visible.\n6. Never plan expectedOutcomes that are ANR / System UI freezes.\n7. If a required kind is impossible given evidence, say so in "reasoning" \u2014 do not invent unobserved screens.');
         }
       }
       const orderedScreens = plannableTargetAppScreens(graph, context.appId).sort((a, b) => a.firstSeenIteration - b.firstSeenIteration || a.screenId.localeCompare(b.screenId));
@@ -82031,6 +82031,123 @@ var require_discovery_lint_flags = __commonJS({
   }
 });
 
+// ../runner-core/dist/services/mobile/anr-env.js
+var require_anr_env = __commonJS({
+  "../runner-core/dist/services/mobile/anr-env.js"(exports) {
+    "use strict";
+    Object.defineProperty(exports, "__esModule", { value: true });
+    exports.textLooksLikeAnrOrSystemUi = textLooksLikeAnrOrSystemUi;
+    exports.isAnrOrSystemUiEnvironmentFailure = isAnrOrSystemUiEnvironmentFailure;
+    exports.stepLooksLikeAnrChromeAssert = stepLooksLikeAnrChromeAssert;
+    var ANR_TEXT_RE = /system\s*ui\s*isn['’]?t\s*responding|app\s*isn['’]?t\s*responding|application\s*not\s*responding|\banr\b|close\s*app|wait\s*and\s*see|isn['’]?t\s*responding/i;
+    function textLooksLikeAnrOrSystemUi(text) {
+      if (!text || typeof text !== "string")
+        return false;
+      return ANR_TEXT_RE.test(text);
+    }
+    function isAnrOrSystemUiEnvironmentFailure(error2, logs2) {
+      const blob = `${error2 ?? ""}
+${logs2 ?? ""}`;
+      if (textLooksLikeAnrOrSystemUi(blob))
+        return true;
+      if (/\bcold-boot\b/i.test(blob) && /\b(froze|hang|anr|unresponsive)\b/i.test(blob))
+        return true;
+      if (/\bno actionable elements\b/i.test(blob) && /\b(system|anr|responding)\b/i.test(blob)) {
+        return true;
+      }
+      return false;
+    }
+    function stepLooksLikeAnrChromeAssert(step) {
+      if (step.action !== "assertVisible" && step.action !== "assertText")
+        return false;
+      return textLooksLikeAnrOrSystemUi([step.description, step.assertion, step.value].filter(Boolean).join(" "));
+    }
+  }
+});
+
+// ../runner-core/dist/services/mobile/oracle-quality.js
+var require_oracle_quality = __commonJS({
+  "../runner-core/dist/services/mobile/oracle-quality.js"(exports) {
+    "use strict";
+    Object.defineProperty(exports, "__esModule", { value: true });
+    exports.assessStrongOracleQuality = assessStrongOracleQuality;
+    var anr_env_js_1 = require_anr_env();
+    var CHROME_ONLY_RE = /\b(tab|home tab|catalog tab|forms tab|account tab|back|menu|navigation|nav bar|bottom nav|status bar|query|search field)\b/i;
+    var PRODUCT_OUTCOME_TEXT_RE = /\b(error|invalid|required|must|cannot|failed|denied|blocked|empty|missing|deleted|removed|saved|success|created|updated|signed?\s*out|logged?\s*out|wrong|incorrect)\b/i;
+    function stepBlob(step) {
+      return [step.description, step.assertion, step.value, step.target?.primary].filter(Boolean).join(" ");
+    }
+    function isChromeOnlyVisible(step) {
+      if (step.action !== "assertVisible")
+        return false;
+      const blob = stepBlob(step);
+      if (PRODUCT_OUTCOME_TEXT_RE.test(blob))
+        return false;
+      return CHROME_ONLY_RE.test(blob) || blob.length < 40;
+    }
+    function assessStrongOracleQuality(steps) {
+      const productSteps = steps.filter((s) => s.action !== "launchApp");
+      const anrAsserts = productSteps.filter(anr_env_js_1.stepLooksLikeAnrChromeAssert);
+      const nonAnrAsserts = productSteps.filter((s) => (s.action === "assertVisible" || s.action === "assertText" || s.action === "assertStateDelta") && !(0, anr_env_js_1.stepLooksLikeAnrChromeAssert)(s));
+      if (productSteps.length > 0 && anrAsserts.length > 0 && nonAnrAsserts.length === 0) {
+        return {
+          strong: false,
+          unpairedCaptures: 0,
+          hasStateDelta: false,
+          hasStrongAssertText: false,
+          hasStrongValidationVisible: false,
+          anrOnly: true,
+          reason: "only asserts system ANR / freeze chrome \u2014 not a product test"
+        };
+      }
+      const captures = steps.filter((s) => s.action === "captureState").length;
+      const deltas = steps.filter((s) => s.action === "assertStateDelta").length;
+      const unpairedCaptures = Math.max(0, captures - deltas);
+      const hasStateDelta = deltas > 0;
+      const hasStrongAssertText = steps.some((s) => {
+        if (s.action !== "assertText")
+          return false;
+        if ((0, anr_env_js_1.stepLooksLikeAnrChromeAssert)(s))
+          return false;
+        const blob = stepBlob(s);
+        if (/^(email|password|username|name|query|search)$/i.test((s.assertion ?? "").trim())) {
+          return false;
+        }
+        return blob.length >= 3;
+      });
+      const hasStrongValidationVisible = steps.some((s) => {
+        if (s.action !== "assertVisible")
+          return false;
+        if ((0, anr_env_js_1.stepLooksLikeAnrChromeAssert)(s) || isChromeOnlyVisible(s))
+          return false;
+        return PRODUCT_OUTCOME_TEXT_RE.test(stepBlob(s));
+      });
+      const strong = hasStateDelta || hasStrongAssertText || hasStrongValidationVisible;
+      let reason;
+      if (strong && hasStateDelta) {
+        reason = unpairedCaptures > 0 ? `has assertStateDelta but ${unpairedCaptures} unpaired captureState (prefer 1:1 pairing)` : "has assertStateDelta durable state proof";
+      } else if (strong && hasStrongAssertText) {
+        reason = "has product assertText outcome";
+      } else if (strong) {
+        reason = "has validation/error assertVisible outcome";
+      } else if (unpairedCaptures > 0) {
+        reason = `captureState without assertStateDelta (${unpairedCaptures}) \u2014 not strong`;
+      } else {
+        reason = "no durable state/text/validation oracle \u2014 smoke/nav only";
+      }
+      return {
+        strong,
+        unpairedCaptures,
+        hasStateDelta,
+        hasStrongAssertText,
+        hasStrongValidationVisible,
+        anrOnly: false,
+        reason
+      };
+    }
+  }
+});
+
 // ../runner-core/dist/services/mobile/mobile-generator.js
 var require_mobile_generator = __commonJS({
   "../runner-core/dist/services/mobile/mobile-generator.js"(exports) {
@@ -82057,6 +82174,8 @@ var require_mobile_generator = __commonJS({
     var mobile_session_keepalive_js_1 = require_mobile_session_keepalive();
     var discovery_lint_flags_js_1 = require_discovery_lint_flags();
     var scenario_taxonomy_js_1 = require_scenario_taxonomy();
+    var anr_env_js_1 = require_anr_env();
+    var oracle_quality_js_1 = require_oracle_quality();
     var mobile_source_parser_js_1 = require_mobile_source_parser();
     var mobile_source_parser_js_2 = require_mobile_source_parser();
     var mobile_observation_js_1 = require_mobile_observation();
@@ -82202,7 +82321,7 @@ var require_mobile_generator = __commonJS({
       type: "function",
       function: {
         name: "mobile_capture_state",
-        description: 'Capture list labels or numeric badge counts into a named state key BEFORE a mutating action (delete/create/toggle). Pair with mobile_assert_state_delta afterward to prove durable identity/count outcomes \u2014 stronger than "list still visible".',
+        description: "Capture list labels or numeric badge counts into a named state key BEFORE a mutating action (delete/create/toggle). MUST pair with mobile_assert_state_delta afterward \u2014 capture alone is not a strong oracle and will not tag the test @strong-oracle.",
         parameters: {
           type: "object",
           properties: {
@@ -82564,9 +82683,11 @@ The scenario name + goal + steps describe ONE product behavior from survey/disco
    - Navigation scenarios: you MUST land on the destination and assert something unique to it.
 4. PROVE the outcome with an assert tool that matches the scenario \u2014 not ambient chrome:
    - mobile_assert_visible({ref}) / mobile_assert_text({ref, expected})
-   - For list delete/create or count changes: mobile_capture_state({stateKey, kind}) BEFORE the mutation, then mobile_assert_state_delta({stateKey, expect}) AFTER \u2014 proves identity/count change, not only "list screen still visible".
+   - For list delete/create or count changes: mobile_capture_state({stateKey, kind}) BEFORE the mutation, then mobile_assert_state_delta({stateKey, expect}) AFTER \u2014 BOTH are required. captureState alone does NOT prove the outcome.
+   - Negative validation: leave a required field EMPTY (or type clearly invalid email-like input), submit, assert the app BLOCKS success \u2014 error message / still on form / no navigation to Home. Do NOT fill a valid password-only path and call it "empty email".
    - After sign-in \u2192 signed-in product signal; after like \u2192 liked/unlike control or durable like state; after open Settings \u2192 a settings-specific control.
    - Do NOT assert bottom-nav tab labels alone, notification badge counts ("N unread"), or one-time empty-state CTAs as the scenario outcome.
+   - NEVER assert system freeze chrome ("System UI isn't responding", ANR dialogs, Wait/Close app). If the app freezes, finish_generate with verified:false (environment) \u2014 do NOT write a test whose only proof is the ANR dialog.
 5. Call finish_generate with verified:true ONLY if you completed the goal AND asserted a matching durable outcome. If the app cannot complete the goal, finish with verified:false and explain \u2014 never invent a green path from unrelated screens.
 
 ## Rules
@@ -82575,7 +82696,7 @@ The scenario name + goal + steps describe ONE product behavior from survey/disco
 - Assert only DURABLE elements \u2014 screen chrome that exists on every visit (tab labels, headers, nav bars, the entity you just created). NEVER assert one-time / data-dependent UI: empty-state CTAs, first-run or onboarding prompts, or counters \u2014 they vanish as soon as the account has data, and the test then fails forever on a healthy app.
 - Do NOT trigger irreversible or destructive side effects \u2014 real payments; sending messages to others; and ESPECIALLY account/data deletion, deactivation, or any "type DELETE to confirm" gate (the runner hard-blocks destructive taps and confirmation keystrokes). Note them and assert what you safely can.
 - STATE OWNERSHIP for mutation scenarios (edit / rename / archive / remove): CREATE the record first in this same recording, then act on the row YOU created \u2014 never mutate pre-existing data (it may not exist on a clean install, so the replay fails forever, and another test may depend on it). If the app offers no way to create one, do NOT act on someone else's data: assert what you can safely prove and finish with verified:false, explaining why in the description.
-- If the APP itself misbehaves (crash, error screen, a flow that cannot proceed as specified), do not force a green path: assert the failure you observed (e.g. the error text), then call finish_generate with suspectedBug:true and a concrete bugDescription of what is broken in the app.
+- If the APP itself misbehaves (product error screen, broken validation, wrong entity deleted), assert the product failure and call finish_generate with suspectedBug:true and a concrete bugDescription. If the device shows ANR / "System UI isn't responding" / freeze dialogs, that is ENVIRONMENT \u2014 finish with verified:false and do NOT assert the freeze dialog as the test outcome.
 - Stay inside this app.${targetLine} Do not open or use app stores, browsers, email apps, settings, or any other app. If an action leaves the target app, the runner will relaunch the target app and reject that action.
 - NEVER tap the phone's home screen / launcher (app icons, "Predicted app" chips, search bars, widgets). Those are NOT part of the app under test \u2014 tapping them is junk that poisons the test. If the app ever backgrounds or you find yourself on the launcher, use mobile_relaunch to cold-start the app; do NOT tap a launcher icon or app-prediction to get back in.
 - Do NOT re-do a sequence you already completed. If you filled a form then navigated away, do NOT come back and fill it again \u2014 the scenario needs ONE clean pass. Filling the login form, going back, and re-filling it is exactly the kind of wandering that produces a bloated, broken test.
@@ -84167,6 +84288,9 @@ ${statePacket}`, redactionCredentials) },
     function weakAssertionTargetReason(element) {
       if (isAlwaysPresentContainer(element))
         return "always-present app/system container";
+      if ((0, anr_env_js_1.textLooksLikeAnrOrSystemUi)([element.text, element.label, element.accessibilityId, element.resourceId].filter(Boolean).join(" "))) {
+        return "system ANR / freeze dialog chrome (environment, not product)";
+      }
       const hasIdentitySignal = Boolean(nonEmpty3(element.accessibilityId) || nonEmpty3(element.resourceId) || nonEmpty3(element.text) || nonEmpty3(element.label));
       if (!hasIdentitySignal)
         return "no identity locator (accessibility id / resource id / text) for the assert gate";
@@ -85028,6 +85152,30 @@ ${statePacket}`, redactionCredentials) },
       return [action.selector, action.value, action.innerText, action.metadata?.targetLabel].filter((v) => typeof v === "string" && v.trim().length > 0).join(" ");
     }
     function evaluateMobileAssertionStrength(actions) {
+      const anrAsserts = actions.filter((a) => {
+        if (a.type !== "ASSERT_VISIBLE" && a.type !== "ASSERT_TEXT")
+          return false;
+        return (0, anr_env_js_1.textLooksLikeAnrOrSystemUi)([a.value, a.innerText, a.metadata?.targetLabel, a.selector].filter(Boolean).join(" "));
+      });
+      if (anrAsserts.length > 0) {
+        const otherAsserts = actions.filter((a) => (a.type === "ASSERT_VISIBLE" || a.type === "ASSERT_TEXT" || a.type === "ASSERT_STATE_DELTA") && !anrAsserts.includes(a));
+        if (otherAsserts.length === 0) {
+          return {
+            ok: false,
+            severity: "block",
+            reason: `only asserts system ANR / "isn't responding" chrome \u2014 not a product test (discard or re-record on a healthy session)`,
+            signals: {
+              apiTest: false,
+              interacted: actions.some((a) => a.type === "TAP" || a.type === "TYPE"),
+              mutated: false,
+              navOnly: false,
+              hasOutcomeProof: false,
+              errorBanner: false,
+              terminal: null
+            }
+          };
+        }
+      }
       if (actions.some((a) => a.type === "ASSERT_STATE_DELTA")) {
         return {
           ok: true,
@@ -85174,9 +85322,10 @@ ${statePacket}`, redactionCredentials) },
       }
       const baseStrengthVerdict = evaluateMobileAssertionStrength(actions);
       const bugDescription = recording.finish?.bugDescription?.trim() ?? "";
-      const bugWitnessHonored = process.env.DISCOVERY_BUG_WITNESS_ENABLED === "true" && recording.finish?.suspectedBug === true && bugDescription.length >= 12 && recording.assertCount > 0;
+      const bugLooksLikeAnr = (0, anr_env_js_1.textLooksLikeAnrOrSystemUi)(bugDescription);
+      const bugWitnessHonored = process.env.DISCOVERY_BUG_WITNESS_ENABLED === "true" && recording.finish?.suspectedBug === true && bugDescription.length >= 12 && recording.assertCount > 0 && !bugLooksLikeAnr;
       if (recording.finish?.suspectedBug === true && !bugWitnessHonored) {
-        log2("  suspectedBug declared but NOT honored: " + (process.env.DISCOVERY_BUG_WITNESS_ENABLED !== "true" ? "bug-witness channel disabled (DISCOVERY_BUG_WITNESS_ENABLED)." : bugDescription.length < 12 ? "bugDescription too short." : "no honored assert backs the observation."));
+        log2("  suspectedBug declared but NOT honored: " + (bugLooksLikeAnr ? "ANR/System UI freeze is environment noise, not a product bug-witness." : process.env.DISCOVERY_BUG_WITNESS_ENABLED !== "true" ? "bug-witness channel disabled (DISCOVERY_BUG_WITNESS_ENABLED)." : bugDescription.length < 12 ? "bugDescription too short." : "no honored assert backs the observation."));
       }
       const strengthVerdict = bugWitnessHonored ? {
         ...baseStrengthVerdict,
@@ -85215,12 +85364,22 @@ ${statePacket}`, redactionCredentials) },
       const strengthBlocked = !strengthVerdict.ok || strengthVerdict.severity === "block" || strengthVerdict.severity === "route-bug" || !verified || coldResetUnavailable;
       {
         const kind = (0, scenario_taxonomy_js_1.inferScenarioKind)(scenario);
-        const hasStateOracle = compiled.steps.some((s) => s.action === "assertStateDelta" || s.action === "assertText" || s.action === "captureState");
-        const strong = verified && !strengthBlocked && strengthVerdict.severity === "none" && ((0, scenario_taxonomy_js_1.isStrongOracleKind)(kind) || hasStateOracle);
+        const oracleQuality = (0, oracle_quality_js_1.assessStrongOracleQuality)(compiled.steps);
+        if (oracleQuality.anrOnly) {
+          log2(`  discarding ANR-chrome-only recording: ${oracleQuality.reason}`);
+          return null;
+        }
+        if (oracleQuality.unpairedCaptures > 0) {
+          log2(`  oracle quality: ${oracleQuality.reason}`);
+        }
+        const strong = verified && !strengthBlocked && strengthVerdict.severity === "none" && oracleQuality.strong && ((0, scenario_taxonomy_js_1.isStrongOracleKind)(kind) || oracleQuality.hasStateDelta || oracleQuality.hasStrongAssertText);
         if (strong)
           tags.push("@strong-oracle");
-        else if (!tags.includes("@smoke") && (kind === "smoke_nav" || kind === "navigation_integrity")) {
+        else if (!tags.includes("@smoke")) {
           tags.push("@smoke");
+          if (!oracleQuality.strong) {
+            log2(`  tagged @smoke (not strong-oracle): ${oracleQuality.reason}`);
+          }
         }
       }
       const strengthReason = coldResetUnavailable ? `Cold app-data reset unavailable${coldResetReason ? ` (${coldResetReason})` : ""} \u2014 generation may not match cold-boot replay. ${strengthVerdict.reason}` : !verified && recording.assertCount === 0 ? "Mobile scenario finished unverified \u2014 the agent never proved the scenario outcome live, so the recording may not demonstrate the goal (e.g. a happy path that never completed). Review before promotion." : strengthVerdict.reason;
@@ -85480,6 +85639,7 @@ var require_failure_triage = __commonJS({
     exports.listStrongOracleKinds = listStrongOracleKinds;
     exports.triageMobileFailure = triageMobileFailure;
     exports.triageToHealTerminal = triageToHealTerminal;
+    var anr_env_js_1 = require_anr_env();
     var STRONG_ACTIONS = /* @__PURE__ */ new Set(["assertStateDelta", "assertText", "captureState"]);
     var CHROME_VISIBLE_RE = /\b(tab|home|settings|back|menu|navigation|nav bar|bottom nav|status bar)\b/i;
     function hasStrongMobileOracle(steps) {
@@ -85569,6 +85729,15 @@ ${logs2 ?? ""}`.trim();
         failureCount,
         errorPrefix: prefix
       };
+      if ((0, anr_env_js_1.isAnrOrSystemUiEnvironmentFailure)(input.error, input.logs)) {
+        return {
+          verdict: "config_issue",
+          reason: "ANR / System UI freeze (device or app hung) \u2014 treat as environment, not product bug",
+          shouldHeal: false,
+          healVerdict: "config_issue",
+          evidence
+        };
+      }
       if (prefix === "CONFIG_ISSUE" || prefix === "RATE_LIMIT_BLOCKED") {
         return {
           verdict: "config_issue",
@@ -85817,6 +85986,7 @@ var require_mobile_heal_postprocess = __commonJS({
     exports.detectMobileStepSecretLeak = detectMobileStepSecretLeak;
     exports.computeMobileHealOutcome = computeMobileHealOutcome2;
     var oracle_protection_js_1 = require_oracle_protection();
+    var anr_env_js_1 = require_anr_env();
     var REDACT_KEY_RE = /email|username|user|phone|password|secret|token|api[_-]?key|access[_-]?key|\bpass\b|credential|bearer|\bpat\b|session|cookie|jwt|otp|pin/i;
     var DETECT_KEY_RE = /password|secret|token|api[_]?key|\bpass\b|credential|bearer|\bpat\b/i;
     var MIN_REDACT_LEN = 4;
@@ -85887,21 +86057,26 @@ var require_mobile_heal_postprocess = __commonJS({
       const oracleWeakening = isVerifiedFix && args.originalSteps ? (0, oracle_protection_js_1.rejectOracleWeakeningPatch)(args.originalSteps, heal.healedSteps ?? []) : null;
       if (oracleWeakening)
         isVerifiedFix = false;
-      const isConfigIssue = heal.verdict === "config_issue";
-      const isAppBug = heal.verdict === "app_bug" || Boolean(oracleWeakening && heal.passed);
+      const anrEnv = (0, anr_env_js_1.isAnrOrSystemUiEnvironmentFailure)(heal.healReason ?? args.finalError ?? args.reproduceError, typeof args.finalError === "string" ? args.finalError : void 0);
+      const isConfigIssue = heal.verdict === "config_issue" || heal.verdict === "app_bug" && anrEnv;
+      const isAppBug = !anrEnv && (heal.verdict === "app_bug" || Boolean(oracleWeakening && heal.passed));
       const extraTags = new Set((heal.extraTags ?? []).filter((t) => typeof t === "string" && t.length > 0));
       if (secretLeak)
         extraTags.add("@needs-review");
-      if (oracleWeakening) {
+      if (oracleWeakening && !anrEnv) {
         extraTags.add("@needs-review");
         extraTags.add("@bug-found");
       }
+      if (anrEnv) {
+        extraTags.delete("@bug-found");
+        extraTags.add("@needs-review");
+      }
       if (!heal.passed && NEEDS_REVIEW_VERDICTS.has(heal.verdict))
         extraTags.add("@needs-review");
       let status;
       if (secretLeak)
         status = "FAILED";
-      else if (oracleWeakening)
+      else if (oracleWeakening && !anrEnv)
         status = "FAILED";
       else if (isVerifiedFix)
         status = "PASSED";
@@ -85914,10 +86089,11 @@ var require_mobile_heal_postprocess = __commonJS({
         error2 = void 0;
       } else if (secretLeak) {
         error2 = `HEAL_SECRET_LEAK: healed steps contained a literal secret (${secretLeak}); patch dropped, tagged @needs-review`;
-      } else if (oracleWeakening) {
+      } else if (oracleWeakening && !anrEnv) {
         error2 = `APP_BUG: heal attempted to drop strong oracle (${oracleWeakening}); patch refused`;
       } else if (isConfigIssue) {
-        error2 = `CONFIG_ISSUE: ${heal.healReason ?? args.finalError ?? args.reproduceError ?? "environment/rate-limit blocker"}`;
+        const reason = heal.healReason ?? args.finalError ?? args.reproduceError ?? "environment/rate-limit blocker";
+        error2 = anrEnv ? `CONFIG_ISSUE: ANR / System UI freeze (not a product assertion failure) \u2014 ${reason}` : `CONFIG_ISSUE: ${reason}`;
       } else if (isAppBug) {
         error2 = `APP_BUG: ${heal.healReason ?? args.finalError ?? args.reproduceError ?? "application defect found during heal"}`;
       } else if (heal.passed && heal.verdict === "fix" && !hasFix) {
@@ -86360,6 +86536,7 @@ var require_mobile_heal_loop = __commonJS({
     var failure_triage_js_1 = require_failure_triage();
     var oracle_protection_js_1 = require_oracle_protection();
     var network_oracle_js_1 = require_network_oracle();
+    var anr_env_js_1 = require_anr_env();
     var mobile_observation_js_1 = require_mobile_observation();
     var selector_registry_context_js_1 = require_selector_registry_context();
     var mobile_session_keepalive_js_1 = require_mobile_session_keepalive();
@@ -86371,12 +86548,13 @@ Reason over steps + error + tree + registry and return working healed steps. Do
 2. STEP RESULTS: which steps passed before the failure.
 3. TREE: does the current screen match what the failed step expects?
 3b. STALE EMPTY-STATE ASSERT: a failing assert on one-time / data-dependent UI (empty-state CTAs, first-run prompts, onboarding banners) is NOT a locator problem \u2014 the element is gone because the account now has data. REPLACE the assertion with a durable landing-screen anchor (tab label, header, nav bar) that proves the same outcome (verdict fix).
+3c. ANR / "System UI isn't responding" / freeze dialogs: this is ENVIRONMENT (config_issue), NOT app_bug. Do not quarantine product bugs for device freezes.
 4. EXTERNAL AUTH HANDOFF: if the failure says the foreground app changed outside the target app after a Login / Sign In tap, and the steps never enter email/password credentials, do NOT heal by choosing Google/Apple/Facebook/SSO or another external app. If the test later expects Logout/session state, return test_invalid or prereq_unmet: the recorded test lacks a real credential setup.
 5. ENTRY-SCREEN DRIFT ("unexpected screen"/"screen drift"): the app started on a DIFFERENT screen than recorded; the fix is usually a MISSING NAVIGATION step (e.g. tap the onboarding/welcome CTA present in the tree before the recorded first step). Author it from the tree if you can (verdict fix); if a single static tree is insufficient, return needs_device.
 ## Response \u2014 STRICT JSON, exactly one verdict:
 {"verdict":"fix","steps":[<FULL healed MobileTestStep array, all steps, your changes applied, order/metadata preserved>],"reason":"what changed and why","confidence":0.0-1.0}
-{"verdict":"app_bug","reason":"app crashed/blank/5xx (NOT rate-limit)"}
-{"verdict":"config_issue","reason":"429/rate limit/lockout/quota"}
+{"verdict":"app_bug","reason":"product defect with durable oracle evidence (wrong entity, validation bypass, wrong count) \u2014 NOT ANR/System UI"}
+{"verdict":"config_issue","reason":"429/rate limit/lockout/quota OR ANR/System UI freeze/device unresponsive"}
 {"verdict":"needs_device","reason":"cannot fix from a static tree; need live re-observation"}
 {"verdict":"prereq_unmet","reason":"required entity/state absent for the account"}
 {"verdict":"state_stale","reason":"leftover data blocks a baseline assertion"}
@@ -86387,7 +86565,7 @@ For a fix, return the COMPLETE steps array (not a diff).
 - NEVER remove or downgrade assertStateDelta, captureState, or assertText outcome proof to go green.
 - NEVER replace a durable outcome assert with mere assertVisible of a tab/header.
 - You MAY fix locators, waits, and navigation around those oracles.
-- If the app truly violates the oracle, return verdict "app_bug" instead of stripping the assert.`;
+- If the app truly violates the oracle on a healthy UI, return verdict "app_bug" instead of stripping the assert.`;
     var DEFAULT_SCRIPT_MAX_ATTEMPTS = 6;
     var RECOVERABLE_MAX = 3;
     var MAX_CONSECUTIVE_INVALID_VERDICTS = 3;
@@ -87355,6 +87533,18 @@ ${args.directive}`);
           providerUnavailable: true
         };
       }
+      if (script.verdict === "app_bug" && (0, anr_env_js_1.isAnrOrSystemUiEnvironmentFailure)(script.healReason ?? ctx.error, ctx.logs)) {
+        return {
+          passed: false,
+          verdict: "config_issue",
+          healedSteps: script.healedSteps,
+          healedAppiumCode: script.healedAppiumCode,
+          healReason: `ANR / System UI freeze (environment) \u2014 ${script.healReason ?? ctx.error ?? "device unresponsive"}`,
+          confidence: script.confidence,
+          extraTags: script.extraTags,
+          removeTags: [...script.removeTags ?? [], "@bug-found"].filter((t, i, a) => a.indexOf(t) === i)
+        };
+      }
       const shouldEscalate = script.verdict === "needs_device" || script.verdict === "fix" && !script.passed;
       if (shouldEscalate && options.runLiveHeal) {
         try {
@@ -355111,7 +355301,7 @@ var require_package4 = __commonJS({
   "package.json"(exports, module) {
     module.exports = {
       name: "@validate.qa/runner",
-      version: "1.0.89",
+      version: "1.0.91",
       description: "validate local test runner - execute Playwright tests on your machine, connected to validate.qa cloud.",
       bin: {
         "validate-runner": "dist/cli.js"

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@validate.qa/runner",
-  "version": "1.0.89",
+  "version": "1.0.91",
   "description": "validate local test runner - execute Playwright tests on your machine, connected to validate.qa cloud.",
   "bin": {
     "validate-runner": "dist/cli.js"