open-multi-agent-kit 1.0.0 → 1.2.2

package/dist/metacognition/verification.d.ts

@@ -0,0 +1,93 @@
+/**
+ * Time-aware independent verification (Jev audit algorithm A7, finding F07).
+ *
+ * F07 is the confusion between "each condition was satisfied at some point"
+ * and "they hold together now". Keeping `CompleteCurrent` and `CompleteEver`
+ * apart resolves it without throwing away valid history.
+ *
+ * `unknown` is a third value, not a synonym for `false`. A missing field or an
+ * incomplete collection scope means the evidence is absent; any binary rule
+ * that negates it into `true` manufactures a verification that never happened.
+ *
+ * Evidence strength is ordered but deliberately not a universal trust score:
+ * a backend response still has to be matched to the right account and request.
+ * The point is to hold a verification signal that does not share the executing
+ * model's uncertainty.
+ */
+export type ConditionValue = "true" | "false" | "unknown";
+export interface ConditionObservation {
+    readonly snapshotId: string;
+    /** Target generation this snapshot belongs to. */
+    readonly generation: number;
+    readonly values: Readonly<Record<string, ConditionValue>>;
+}
+export interface UnsatisfiedCondition {
+    readonly condition: string;
+    readonly value: ConditionValue;
+}
+export type CompletionResult = {
+    readonly complete: true;
+} | {
+    readonly complete: false;
+    readonly unsatisfied: readonly UnsatisfiedCondition[];
+};
+/**
+ * Every current condition must hold in the *same* snapshot.
+ *
+ * Evaluating them across snapshots would let a condition that has since
+ * flipped keep satisfying the gate.
+ */
+export declare function completeCurrent(conditions: readonly string[], snapshot: ConditionObservation): CompletionResult;
+/**
+ * Each historical condition must have been observed true at least once in the
+ * supplied evidence set, which the caller scopes to the current task.
+ */
+export declare function completeEver(conditions: readonly string[], history: readonly ConditionObservation[]): CompletionResult;
+export type EvidenceKind = "authoritative-query" | "structured-remote-linked" | "independent-ui" | "page-success-text" | "model-completion-claim";
+export declare function evidenceStrength(kind: EvidenceKind): number;
+export type VerificationLevel = "REMOTE_EFFECT_VERIFIED" | "UI_VERIFIED" | "CLAIMED_ONLY";
+/**
+ * Label a result by the strongest evidence actually held.
+ *
+ * A UI observation is reported as `UI_VERIFIED`, never as
+ * `REMOTE_EFFECT_VERIFIED`: if the target system offers no authoritative
+ * query, the honest move is to lower the claimed level, not to borrow one.
+ */
+export declare function strongestEvidence(kinds: readonly EvidenceKind[]): VerificationLevel;
+export interface EvidenceItem {
+    readonly id: string;
+    /** Common-cause group, e.g. one DOM snapshot that many readings derive from. */
+    readonly sourceGroup: string;
+}
+export interface EvidenceGrouping {
+    readonly independentCount: number;
+    readonly byGroup: Readonly<Record<string, number>>;
+}
+/**
+ * Collapse correlated readings into their source groups.
+ *
+ * Fifty screenshots of one screen are one witness. Text, accessibility names
+ * and rendered copy from the same DOM share a common cause, so counting them
+ * separately inflates apparent independent confirmation.
+ */
+export declare function independentEvidenceGroups(items: readonly EvidenceItem[]): EvidenceGrouping;
+export interface VerifierSample {
+    readonly verdict: ConditionValue;
+    /** Externally established truth, not another model's opinion. */
+    readonly truth: "true" | "false";
+}
+export interface VerifierQuality {
+    /** Undefined when no negative case was sampled; that is absent evidence. */
+    readonly falsePositiveRate: number | undefined;
+    readonly falseNegativeRate: number | undefined;
+    readonly unknownRate: number;
+    readonly samples: number;
+}
+/**
+ * Measure the verifier before reporting any success rate.
+ *
+ * A mislabelling verifier contaminates every downstream selector and bandit
+ * trained on its labels, so its error rates are the prior question.
+ */
+export declare function verifierQuality(samples: readonly VerifierSample[]): VerifierQuality;
+//# sourceMappingURL=verification.d.ts.map

package/dist/metacognition/verification.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"verification.d.ts","sourceRoot":"","sources":["../../src/metacognition/verification.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAIH,MAAM,MAAM,cAAc,GAAG,MAAM,GAAG,OAAO,GAAG,SAAS,CAAC;AAE1D,MAAM,WAAW,oBAAoB;IACpC,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,kDAAkD;IAClD,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,MAAM,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC,CAAC;CAC1D;AAED,MAAM,WAAW,oBAAoB;IACpC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,KAAK,EAAE,cAAc,CAAC;CAC/B;AAED,MAAM,MAAM,gBAAgB,GACzB;IAAE,QAAQ,CAAC,QAAQ,EAAE,IAAI,CAAA;CAAE,GAC3B;IAAE,QAAQ,CAAC,QAAQ,EAAE,KAAK,CAAC;IAAC,QAAQ,CAAC,WAAW,EAAE,SAAS,oBAAoB,EAAE,CAAA;CAAE,CAAC;AAEvF;;;;;GAKG;AACH,wBAAgB,eAAe,CAAC,UAAU,EAAE,SAAS,MAAM,EAAE,EAAE,QAAQ,EAAE,oBAAoB,GAAG,gBAAgB,CAS/G;AAED;;;GAGG;AACH,wBAAgB,YAAY,CAC3B,UAAU,EAAE,SAAS,MAAM,EAAE,EAC7B,OAAO,EAAE,SAAS,oBAAoB,EAAE,GACtC,gBAAgB,CAWlB;AAED,MAAM,MAAM,YAAY,GACrB,qBAAqB,GACrB,0BAA0B,GAC1B,gBAAgB,GAChB,mBAAmB,GACnB,wBAAwB,CAAC;AAU5B,wBAAgB,gBAAgB,CAAC,IAAI,EAAE,YAAY,GAAG,MAAM,CAI3D;AAED,MAAM,MAAM,iBAAiB,GAAG,wBAAwB,GAAG,aAAa,GAAG,cAAc,CAAC;AAE1F;;;;;;GAMG;AACH,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,SAAS,YAAY,EAAE,GAAG,iBAAiB,CAOnF;AAED,MAAM,WAAW,YAAY;IAC5B,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,gFAAgF;IAChF,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;CAC7B;AAED,MAAM,WAAW,gBAAgB;IAChC,QAAQ,CAAC,gBAAgB,EAAE,MAAM,CAAC;IAClC,QAAQ,CAAC,OAAO,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC;CACnD;AAED;;;;;;GAMG;AACH,wBAAgB,yBAAyB,CAAC,KAAK,EAAE,SAAS,YAAY,EAAE,GAAG,gBAAgB,CAW1F;AAED,MAAM,WAAW,cAAc;IAC9B,QAAQ,CAAC,OAAO,EAAE,cAAc,CAAC;IACjC,iEAAiE;IACjE,QAAQ,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC;CACjC;AAED,MAAM,WAAW,eAAe;IAC/B,4EAA4E;IAC5E,QAAQ,CAAC,iBAAiB,EAAE,MAAM,GAAG,SAAS,CAAC;IAC/C,QAAQ,CAAC,iBAAiB,EAAE,MAAM,GAAG,SAAS,CAAC;IAC/C,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;CACzB;AAED;;;;;GAKG;AACH,wBAAgB,eAAe,CAAC,OAAO,EAAE,SAAS,cAAc,EAAE,GAAG,eAAe,CAuBnF","sourcesContent":["/**\n * Time-aware independent verification (Jev audit algorithm A7, finding F07).\n *\n * F07 is the confusion between \"each condition was satisfied at some point\"\n * and \"they hold together now\". Keeping `CompleteCurrent` and `CompleteEver`\n * apart resolves it without throwing away valid history.\n *\n * `unknown` is a third value, not a synonym for `false`. A missing field or an\n * incomplete collection scope means the evidence is absent; any binary rule\n * that negates it into `true` manufactures a verification that never happened.\n *\n * Evidence strength is ordered but deliberately not a universal trust score:\n * a backend response still has to be matched to the right account and request.\n * The point is to hold a verification signal that does not share the executing\n * model's uncertainty.\n */\n\nimport { ensure } from \"./validation.ts\";\n\nexport type ConditionValue = \"true\" | \"false\" | \"unknown\";\n\nexport interface ConditionObservation {\n\treadonly snapshotId: string;\n\t/** Target generation this snapshot belongs to. */\n\treadonly generation: number;\n\treadonly values: Readonly<Record<string, ConditionValue>>;\n}\n\nexport interface UnsatisfiedCondition {\n\treadonly condition: string;\n\treadonly value: ConditionValue;\n}\n\nexport type CompletionResult =\n\t| { readonly complete: true }\n\t| { readonly complete: false; readonly unsatisfied: readonly UnsatisfiedCondition[] };\n\n/**\n * Every current condition must hold in the *same* snapshot.\n *\n * Evaluating them across snapshots would let a condition that has since\n * flipped keep satisfying the gate.\n */\nexport function completeCurrent(conditions: readonly string[], snapshot: ConditionObservation): CompletionResult {\n\tensure(Array.isArray(conditions) && conditions.length > 0, \"current conditions must be non-empty\");\n\tconst unsatisfied: UnsatisfiedCondition[] = [];\n\tfor (const condition of conditions) {\n\t\t// An absent field is unknown, never false.\n\t\tconst value = snapshot.values[condition] ?? \"unknown\";\n\t\tif (value !== \"true\") unsatisfied.push({ condition, value });\n\t}\n\treturn unsatisfied.length === 0 ? { complete: true } : { complete: false, unsatisfied };\n}\n\n/**\n * Each historical condition must have been observed true at least once in the\n * supplied evidence set, which the caller scopes to the current task.\n */\nexport function completeEver(\n\tconditions: readonly string[],\n\thistory: readonly ConditionObservation[],\n): CompletionResult {\n\tensure(Array.isArray(conditions) && conditions.length > 0, \"ever conditions must be non-empty\");\n\tensure(Array.isArray(history) && history.length > 0, \"history must be non-empty to judge an ever condition\");\n\tconst unsatisfied: UnsatisfiedCondition[] = [];\n\tfor (const condition of conditions) {\n\t\tconst seenTrue = history.some((snapshot) => snapshot.values[condition] === \"true\");\n\t\tif (seenTrue) continue;\n\t\tconst anyKnown = history.some((snapshot) => snapshot.values[condition] === \"false\");\n\t\tunsatisfied.push({ condition, value: anyKnown ? \"false\" : \"unknown\" });\n\t}\n\treturn unsatisfied.length === 0 ? { complete: true } : { complete: false, unsatisfied };\n}\n\nexport type EvidenceKind =\n\t| \"authoritative-query\"\n\t| \"structured-remote-linked\"\n\t| \"independent-ui\"\n\t| \"page-success-text\"\n\t| \"model-completion-claim\";\n\nconst EVIDENCE_RANK: Readonly<Record<EvidenceKind, number>> = {\n\t\"authoritative-query\": 4,\n\t\"structured-remote-linked\": 3,\n\t\"independent-ui\": 2,\n\t\"page-success-text\": 1,\n\t\"model-completion-claim\": 0,\n};\n\nexport function evidenceStrength(kind: EvidenceKind): number {\n\tconst rank = EVIDENCE_RANK[kind];\n\tensure(rank !== undefined, `unknown evidence kind: ${String(kind)}`);\n\treturn rank;\n}\n\nexport type VerificationLevel = \"REMOTE_EFFECT_VERIFIED\" | \"UI_VERIFIED\" | \"CLAIMED_ONLY\";\n\n/**\n * Label a result by the strongest evidence actually held.\n *\n * A UI observation is reported as `UI_VERIFIED`, never as\n * `REMOTE_EFFECT_VERIFIED`: if the target system offers no authoritative\n * query, the honest move is to lower the claimed level, not to borrow one.\n */\nexport function strongestEvidence(kinds: readonly EvidenceKind[]): VerificationLevel {\n\tensure(Array.isArray(kinds) && kinds.length > 0, \"evidence set must be non-empty\");\n\tlet best = -1;\n\tfor (const kind of kinds) best = Math.max(best, evidenceStrength(kind));\n\tif (best >= EVIDENCE_RANK[\"structured-remote-linked\"]) return \"REMOTE_EFFECT_VERIFIED\";\n\tif (best >= EVIDENCE_RANK[\"independent-ui\"]) return \"UI_VERIFIED\";\n\treturn \"CLAIMED_ONLY\";\n}\n\nexport interface EvidenceItem {\n\treadonly id: string;\n\t/** Common-cause group, e.g. one DOM snapshot that many readings derive from. */\n\treadonly sourceGroup: string;\n}\n\nexport interface EvidenceGrouping {\n\treadonly independentCount: number;\n\treadonly byGroup: Readonly<Record<string, number>>;\n}\n\n/**\n * Collapse correlated readings into their source groups.\n *\n * Fifty screenshots of one screen are one witness. Text, accessibility names\n * and rendered copy from the same DOM share a common cause, so counting them\n * separately inflates apparent independent confirmation.\n */\nexport function independentEvidenceGroups(items: readonly EvidenceItem[]): EvidenceGrouping {\n\tensure(Array.isArray(items) && items.length > 0, \"evidence items must be non-empty\");\n\tconst byGroup: Record<string, number> = {};\n\tfor (const item of items) {\n\t\tensure(\n\t\t\ttypeof item.sourceGroup === \"string\" && item.sourceGroup.length > 0,\n\t\t\t\"each evidence item needs a declared sourceGroup\",\n\t\t);\n\t\tbyGroup[item.sourceGroup] = (byGroup[item.sourceGroup] ?? 0) + 1;\n\t}\n\treturn { independentCount: Object.keys(byGroup).length, byGroup };\n}\n\nexport interface VerifierSample {\n\treadonly verdict: ConditionValue;\n\t/** Externally established truth, not another model's opinion. */\n\treadonly truth: \"true\" | \"false\";\n}\n\nexport interface VerifierQuality {\n\t/** Undefined when no negative case was sampled; that is absent evidence. */\n\treadonly falsePositiveRate: number | undefined;\n\treadonly falseNegativeRate: number | undefined;\n\treadonly unknownRate: number;\n\treadonly samples: number;\n}\n\n/**\n * Measure the verifier before reporting any success rate.\n *\n * A mislabelling verifier contaminates every downstream selector and bandit\n * trained on its labels, so its error rates are the prior question.\n */\nexport function verifierQuality(samples: readonly VerifierSample[]): VerifierQuality {\n\tensure(Array.isArray(samples) && samples.length > 0, \"verifier samples must be non-empty\");\n\tlet positives = 0;\n\tlet negatives = 0;\n\tlet falsePositives = 0;\n\tlet falseNegatives = 0;\n\tlet unknowns = 0;\n\tfor (const sample of samples) {\n\t\tif (sample.verdict === \"unknown\") unknowns += 1;\n\t\tif (sample.truth === \"true\") {\n\t\t\tpositives += 1;\n\t\t\tif (sample.verdict === \"false\") falseNegatives += 1;\n\t\t} else {\n\t\t\tnegatives += 1;\n\t\t\tif (sample.verdict === \"true\") falsePositives += 1;\n\t\t}\n\t}\n\treturn {\n\t\tfalsePositiveRate: negatives > 0 ? falsePositives / negatives : undefined,\n\t\tfalseNegativeRate: positives > 0 ? falseNegatives / positives : undefined,\n\t\tunknownRate: unknowns / samples.length,\n\t\tsamples: samples.length,\n\t};\n}\n"]}

package/dist/metacognition/verification.js

@@ -0,0 +1,132 @@
+/**
+ * Time-aware independent verification (Jev audit algorithm A7, finding F07).
+ *
+ * F07 is the confusion between "each condition was satisfied at some point"
+ * and "they hold together now". Keeping `CompleteCurrent` and `CompleteEver`
+ * apart resolves it without throwing away valid history.
+ *
+ * `unknown` is a third value, not a synonym for `false`. A missing field or an
+ * incomplete collection scope means the evidence is absent; any binary rule
+ * that negates it into `true` manufactures a verification that never happened.
+ *
+ * Evidence strength is ordered but deliberately not a universal trust score:
+ * a backend response still has to be matched to the right account and request.
+ * The point is to hold a verification signal that does not share the executing
+ * model's uncertainty.
+ */
+import { ensure } from "./validation.js";
+/**
+ * Every current condition must hold in the *same* snapshot.
+ *
+ * Evaluating them across snapshots would let a condition that has since
+ * flipped keep satisfying the gate.
+ */
+export function completeCurrent(conditions, snapshot) {
+    ensure(Array.isArray(conditions) && conditions.length > 0, "current conditions must be non-empty");
+    const unsatisfied = [];
+    for (const condition of conditions) {
+        // An absent field is unknown, never false.
+        const value = snapshot.values[condition] ?? "unknown";
+        if (value !== "true")
+            unsatisfied.push({ condition, value });
+    }
+    return unsatisfied.length === 0 ? { complete: true } : { complete: false, unsatisfied };
+}
+/**
+ * Each historical condition must have been observed true at least once in the
+ * supplied evidence set, which the caller scopes to the current task.
+ */
+export function completeEver(conditions, history) {
+    ensure(Array.isArray(conditions) && conditions.length > 0, "ever conditions must be non-empty");
+    ensure(Array.isArray(history) && history.length > 0, "history must be non-empty to judge an ever condition");
+    const unsatisfied = [];
+    for (const condition of conditions) {
+        const seenTrue = history.some((snapshot) => snapshot.values[condition] === "true");
+        if (seenTrue)
+            continue;
+        const anyKnown = history.some((snapshot) => snapshot.values[condition] === "false");
+        unsatisfied.push({ condition, value: anyKnown ? "false" : "unknown" });
+    }
+    return unsatisfied.length === 0 ? { complete: true } : { complete: false, unsatisfied };
+}
+const EVIDENCE_RANK = {
+    "authoritative-query": 4,
+    "structured-remote-linked": 3,
+    "independent-ui": 2,
+    "page-success-text": 1,
+    "model-completion-claim": 0,
+};
+export function evidenceStrength(kind) {
+    const rank = EVIDENCE_RANK[kind];
+    ensure(rank !== undefined, `unknown evidence kind: ${String(kind)}`);
+    return rank;
+}
+/**
+ * Label a result by the strongest evidence actually held.
+ *
+ * A UI observation is reported as `UI_VERIFIED`, never as
+ * `REMOTE_EFFECT_VERIFIED`: if the target system offers no authoritative
+ * query, the honest move is to lower the claimed level, not to borrow one.
+ */
+export function strongestEvidence(kinds) {
+    ensure(Array.isArray(kinds) && kinds.length > 0, "evidence set must be non-empty");
+    let best = -1;
+    for (const kind of kinds)
+        best = Math.max(best, evidenceStrength(kind));
+    if (best >= EVIDENCE_RANK["structured-remote-linked"])
+        return "REMOTE_EFFECT_VERIFIED";
+    if (best >= EVIDENCE_RANK["independent-ui"])
+        return "UI_VERIFIED";
+    return "CLAIMED_ONLY";
+}
+/**
+ * Collapse correlated readings into their source groups.
+ *
+ * Fifty screenshots of one screen are one witness. Text, accessibility names
+ * and rendered copy from the same DOM share a common cause, so counting them
+ * separately inflates apparent independent confirmation.
+ */
+export function independentEvidenceGroups(items) {
+    ensure(Array.isArray(items) && items.length > 0, "evidence items must be non-empty");
+    const byGroup = {};
+    for (const item of items) {
+        ensure(typeof item.sourceGroup === "string" && item.sourceGroup.length > 0, "each evidence item needs a declared sourceGroup");
+        byGroup[item.sourceGroup] = (byGroup[item.sourceGroup] ?? 0) + 1;
+    }
+    return { independentCount: Object.keys(byGroup).length, byGroup };
+}
+/**
+ * Measure the verifier before reporting any success rate.
+ *
+ * A mislabelling verifier contaminates every downstream selector and bandit
+ * trained on its labels, so its error rates are the prior question.
+ */
+export function verifierQuality(samples) {
+    ensure(Array.isArray(samples) && samples.length > 0, "verifier samples must be non-empty");
+    let positives = 0;
+    let negatives = 0;
+    let falsePositives = 0;
+    let falseNegatives = 0;
+    let unknowns = 0;
+    for (const sample of samples) {
+        if (sample.verdict === "unknown")
+            unknowns += 1;
+        if (sample.truth === "true") {
+            positives += 1;
+            if (sample.verdict === "false")
+                falseNegatives += 1;
+        }
+        else {
+            negatives += 1;
+            if (sample.verdict === "true")
+                falsePositives += 1;
+        }
+    }
+    return {
+        falsePositiveRate: negatives > 0 ? falsePositives / negatives : undefined,
+        falseNegativeRate: positives > 0 ? falseNegatives / positives : undefined,
+        unknownRate: unknowns / samples.length,
+        samples: samples.length,
+    };
+}
+//# sourceMappingURL=verification.js.map

package/dist/metacognition/verification.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"verification.js","sourceRoot":"","sources":["../../src/metacognition/verification.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAEH,OAAO,EAAE,MAAM,EAAE,MAAM,iBAAiB,CAAC;AAoBzC;;;;;GAKG;AACH,MAAM,UAAU,eAAe,CAAC,UAA6B,EAAE,QAA8B,EAAoB;IAChH,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,UAAU,CAAC,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,sCAAsC,CAAC,CAAC;IACnG,MAAM,WAAW,GAA2B,EAAE,CAAC;IAC/C,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;QACpC,2CAA2C;QAC3C,MAAM,KAAK,GAAG,QAAQ,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,SAAS,CAAC;QACtD,IAAI,KAAK,KAAK,MAAM;YAAE,WAAW,CAAC,IAAI,CAAC,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC,CAAC;IAC9D,CAAC;IACD,OAAO,WAAW,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,KAAK,EAAE,WAAW,EAAE,CAAC;AAAA,CACxF;AAED;;;GAGG;AACH,MAAM,UAAU,YAAY,CAC3B,UAA6B,EAC7B,OAAwC,EACrB;IACnB,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,UAAU,CAAC,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,mCAAmC,CAAC,CAAC;IAChG,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,sDAAsD,CAAC,CAAC;IAC7G,MAAM,WAAW,GAA2B,EAAE,CAAC;IAC/C,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;QACpC,MAAM,QAAQ,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,MAAM,CAAC,SAAS,CAAC,KAAK,MAAM,CAAC,CAAC;QACnF,IAAI,QAAQ;YAAE,SAAS;QACvB,MAAM,QAAQ,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,MAAM,CAAC,SAAS,CAAC,KAAK,OAAO,CAAC,CAAC;QACpF,WAAW,CAAC,IAAI,CAAC,EAAE,SAAS,EAAE,KAAK,EAAE,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC;IACxE,CAAC;IACD,OAAO,WAAW,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,KAAK,EAAE,WAAW,EAAE,CAAC;AAAA,CACxF;AASD,MAAM,aAAa,GAA2C;IAC7D,qBAAqB,EAAE,CAAC;IACxB,0BAA0B,EAAE,CAAC;IAC7B,gBAAgB,EAAE,CAAC;IACnB,mBAAmB,EAAE,CAAC;IACtB,wBAAwB,EAAE,CAAC;CAC3B,CAAC;AAEF,MAAM,UAAU,gBAAgB,CAAC,IAAkB,EAAU;IAC5D,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,CAAC,CAAC;IACjC,MAAM,CAAC,IAAI,KAAK,SAAS,EAAE,0BAA0B,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACrE,OAAO,IAAI,CAAC;AAAA,CACZ;AAID;;;;;;GAMG;AACH,MAAM,UAAU,iBAAiB,CAAC,KAA8B,EAAqB;IACpF,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,gCAAgC,CAAC,CAAC;IACnF,IAAI,IAAI,GAAG,CAAC,CAAC,CAAC;IACd,KAAK,MAAM,IAAI,IAAI,KAAK;QAAE,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,gBAAgB,CAAC,IAAI,CAAC,CAAC,CAAC;IACxE,IAAI,IAAI,IAAI,aAAa,CAAC,0BAA0B,CAAC;QAAE,OAAO,wBAAwB,CAAC;IACvF,IAAI,IAAI,IAAI,aAAa,CAAC,gBAAgB,CAAC;QAAE,OAAO,aAAa,CAAC;IAClE,OAAO,cAAc,CAAC;AAAA,CACtB;AAaD;;;;;;GAMG;AACH,MAAM,UAAU,yBAAyB,CAAC,KAA8B,EAAoB;IAC3F,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,kCAAkC,CAAC,CAAC;IACrF,MAAM,OAAO,GAA2B,EAAE,CAAC;IAC3C,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QAC1B,MAAM,CACL,OAAO,IAAI,CAAC,WAAW,KAAK,QAAQ,IAAI,IAAI,CAAC,WAAW,CAAC,MAAM,GAAG,CAAC,EACnE,iDAAiD,CACjD,CAAC;QACF,OAAO,CAAC,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;IAClE,CAAC;IACD,OAAO,EAAE,gBAAgB,EAAE,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,OAAO,EAAE,CAAC;AAAA,CAClE;AAgBD;;;;;GAKG;AACH,MAAM,UAAU,eAAe,CAAC,OAAkC,EAAmB;IACpF,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,oCAAoC,CAAC,CAAC;IAC3F,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,cAAc,GAAG,CAAC,CAAC;IACvB,IAAI,cAAc,GAAG,CAAC,CAAC;IACvB,IAAI,QAAQ,GAAG,CAAC,CAAC;IACjB,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC9B,IAAI,MAAM,CAAC,OAAO,KAAK,SAAS;YAAE,QAAQ,IAAI,CAAC,CAAC;QAChD,IAAI,MAAM,CAAC,KAAK,KAAK,MAAM,EAAE,CAAC;YAC7B,SAAS,IAAI,CAAC,CAAC;YACf,IAAI,MAAM,CAAC,OAAO,KAAK,OAAO;gBAAE,cAAc,IAAI,CAAC,CAAC;QACrD,CAAC;aAAM,CAAC;YACP,SAAS,IAAI,CAAC,CAAC;YACf,IAAI,MAAM,CAAC,OAAO,KAAK,MAAM;gBAAE,cAAc,IAAI,CAAC,CAAC;QACpD,CAAC;IACF,CAAC;IACD,OAAO;QACN,iBAAiB,EAAE,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,cAAc,GAAG,SAAS,CAAC,CAAC,CAAC,SAAS;QACzE,iBAAiB,EAAE,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,cAAc,GAAG,SAAS,CAAC,CAAC,CAAC,SAAS;QACzE,WAAW,EAAE,QAAQ,GAAG,OAAO,CAAC,MAAM;QACtC,OAAO,EAAE,OAAO,CAAC,MAAM;KACvB,CAAC;AAAA,CACF","sourcesContent":["/**\n * Time-aware independent verification (Jev audit algorithm A7, finding F07).\n *\n * F07 is the confusion between \"each condition was satisfied at some point\"\n * and \"they hold together now\". Keeping `CompleteCurrent` and `CompleteEver`\n * apart resolves it without throwing away valid history.\n *\n * `unknown` is a third value, not a synonym for `false`. A missing field or an\n * incomplete collection scope means the evidence is absent; any binary rule\n * that negates it into `true` manufactures a verification that never happened.\n *\n * Evidence strength is ordered but deliberately not a universal trust score:\n * a backend response still has to be matched to the right account and request.\n * The point is to hold a verification signal that does not share the executing\n * model's uncertainty.\n */\n\nimport { ensure } from \"./validation.ts\";\n\nexport type ConditionValue = \"true\" | \"false\" | \"unknown\";\n\nexport interface ConditionObservation {\n\treadonly snapshotId: string;\n\t/** Target generation this snapshot belongs to. */\n\treadonly generation: number;\n\treadonly values: Readonly<Record<string, ConditionValue>>;\n}\n\nexport interface UnsatisfiedCondition {\n\treadonly condition: string;\n\treadonly value: ConditionValue;\n}\n\nexport type CompletionResult =\n\t| { readonly complete: true }\n\t| { readonly complete: false; readonly unsatisfied: readonly UnsatisfiedCondition[] };\n\n/**\n * Every current condition must hold in the *same* snapshot.\n *\n * Evaluating them across snapshots would let a condition that has since\n * flipped keep satisfying the gate.\n */\nexport function completeCurrent(conditions: readonly string[], snapshot: ConditionObservation): CompletionResult {\n\tensure(Array.isArray(conditions) && conditions.length > 0, \"current conditions must be non-empty\");\n\tconst unsatisfied: UnsatisfiedCondition[] = [];\n\tfor (const condition of conditions) {\n\t\t// An absent field is unknown, never false.\n\t\tconst value = snapshot.values[condition] ?? \"unknown\";\n\t\tif (value !== \"true\") unsatisfied.push({ condition, value });\n\t}\n\treturn unsatisfied.length === 0 ? { complete: true } : { complete: false, unsatisfied };\n}\n\n/**\n * Each historical condition must have been observed true at least once in the\n * supplied evidence set, which the caller scopes to the current task.\n */\nexport function completeEver(\n\tconditions: readonly string[],\n\thistory: readonly ConditionObservation[],\n): CompletionResult {\n\tensure(Array.isArray(conditions) && conditions.length > 0, \"ever conditions must be non-empty\");\n\tensure(Array.isArray(history) && history.length > 0, \"history must be non-empty to judge an ever condition\");\n\tconst unsatisfied: UnsatisfiedCondition[] = [];\n\tfor (const condition of conditions) {\n\t\tconst seenTrue = history.some((snapshot) => snapshot.values[condition] === \"true\");\n\t\tif (seenTrue) continue;\n\t\tconst anyKnown = history.some((snapshot) => snapshot.values[condition] === \"false\");\n\t\tunsatisfied.push({ condition, value: anyKnown ? \"false\" : \"unknown\" });\n\t}\n\treturn unsatisfied.length === 0 ? { complete: true } : { complete: false, unsatisfied };\n}\n\nexport type EvidenceKind =\n\t| \"authoritative-query\"\n\t| \"structured-remote-linked\"\n\t| \"independent-ui\"\n\t| \"page-success-text\"\n\t| \"model-completion-claim\";\n\nconst EVIDENCE_RANK: Readonly<Record<EvidenceKind, number>> = {\n\t\"authoritative-query\": 4,\n\t\"structured-remote-linked\": 3,\n\t\"independent-ui\": 2,\n\t\"page-success-text\": 1,\n\t\"model-completion-claim\": 0,\n};\n\nexport function evidenceStrength(kind: EvidenceKind): number {\n\tconst rank = EVIDENCE_RANK[kind];\n\tensure(rank !== undefined, `unknown evidence kind: ${String(kind)}`);\n\treturn rank;\n}\n\nexport type VerificationLevel = \"REMOTE_EFFECT_VERIFIED\" | \"UI_VERIFIED\" | \"CLAIMED_ONLY\";\n\n/**\n * Label a result by the strongest evidence actually held.\n *\n * A UI observation is reported as `UI_VERIFIED`, never as\n * `REMOTE_EFFECT_VERIFIED`: if the target system offers no authoritative\n * query, the honest move is to lower the claimed level, not to borrow one.\n */\nexport function strongestEvidence(kinds: readonly EvidenceKind[]): VerificationLevel {\n\tensure(Array.isArray(kinds) && kinds.length > 0, \"evidence set must be non-empty\");\n\tlet best = -1;\n\tfor (const kind of kinds) best = Math.max(best, evidenceStrength(kind));\n\tif (best >= EVIDENCE_RANK[\"structured-remote-linked\"]) return \"REMOTE_EFFECT_VERIFIED\";\n\tif (best >= EVIDENCE_RANK[\"independent-ui\"]) return \"UI_VERIFIED\";\n\treturn \"CLAIMED_ONLY\";\n}\n\nexport interface EvidenceItem {\n\treadonly id: string;\n\t/** Common-cause group, e.g. one DOM snapshot that many readings derive from. */\n\treadonly sourceGroup: string;\n}\n\nexport interface EvidenceGrouping {\n\treadonly independentCount: number;\n\treadonly byGroup: Readonly<Record<string, number>>;\n}\n\n/**\n * Collapse correlated readings into their source groups.\n *\n * Fifty screenshots of one screen are one witness. Text, accessibility names\n * and rendered copy from the same DOM share a common cause, so counting them\n * separately inflates apparent independent confirmation.\n */\nexport function independentEvidenceGroups(items: readonly EvidenceItem[]): EvidenceGrouping {\n\tensure(Array.isArray(items) && items.length > 0, \"evidence items must be non-empty\");\n\tconst byGroup: Record<string, number> = {};\n\tfor (const item of items) {\n\t\tensure(\n\t\t\ttypeof item.sourceGroup === \"string\" && item.sourceGroup.length > 0,\n\t\t\t\"each evidence item needs a declared sourceGroup\",\n\t\t);\n\t\tbyGroup[item.sourceGroup] = (byGroup[item.sourceGroup] ?? 0) + 1;\n\t}\n\treturn { independentCount: Object.keys(byGroup).length, byGroup };\n}\n\nexport interface VerifierSample {\n\treadonly verdict: ConditionValue;\n\t/** Externally established truth, not another model's opinion. */\n\treadonly truth: \"true\" | \"false\";\n}\n\nexport interface VerifierQuality {\n\t/** Undefined when no negative case was sampled; that is absent evidence. */\n\treadonly falsePositiveRate: number | undefined;\n\treadonly falseNegativeRate: number | undefined;\n\treadonly unknownRate: number;\n\treadonly samples: number;\n}\n\n/**\n * Measure the verifier before reporting any success rate.\n *\n * A mislabelling verifier contaminates every downstream selector and bandit\n * trained on its labels, so its error rates are the prior question.\n */\nexport function verifierQuality(samples: readonly VerifierSample[]): VerifierQuality {\n\tensure(Array.isArray(samples) && samples.length > 0, \"verifier samples must be non-empty\");\n\tlet positives = 0;\n\tlet negatives = 0;\n\tlet falsePositives = 0;\n\tlet falseNegatives = 0;\n\tlet unknowns = 0;\n\tfor (const sample of samples) {\n\t\tif (sample.verdict === \"unknown\") unknowns += 1;\n\t\tif (sample.truth === \"true\") {\n\t\t\tpositives += 1;\n\t\t\tif (sample.verdict === \"false\") falseNegatives += 1;\n\t\t} else {\n\t\t\tnegatives += 1;\n\t\t\tif (sample.verdict === \"true\") falsePositives += 1;\n\t\t}\n\t}\n\treturn {\n\t\tfalsePositiveRate: negatives > 0 ? falsePositives / negatives : undefined,\n\t\tfalseNegativeRate: positives > 0 ? falseNegatives / positives : undefined,\n\t\tunknownRate: unknowns / samples.length,\n\t\tsamples: samples.length,\n\t};\n}\n"]}

package/dist/metacognition/verifier.d.ts

@@ -0,0 +1,49 @@
+export type MutantClass = "killed-intended" | "survived" | "equivalent" | "invalid-compile" | "environment-failure" | "timeout" | "killed-unintended";
+export interface MutantRecord {
+    readonly mutantId: string;
+    readonly obligationId: string;
+    /** The host classifies the outcome; this module never guesses. */
+    readonly classification: MutantClass;
+    /** Whether the intended obligation violation was what the check caught. */
+    readonly intendedViolation: boolean;
+}
+export interface VerifierEvaluation {
+    readonly obligationId: string;
+    /** D_v over valid non-equivalent mutants; "unknown" when empty. */
+    readonly detectionRate: number | "unknown";
+    readonly evaluated: number;
+    readonly excluded: Readonly<Record<Exclude<MutantClass, "killed-intended" | "survived">, number>>;
+    readonly survivingMutantIds: readonly string[];
+    readonly blindSpots: readonly string[];
+}
+export declare function evaluateVerifier(obligationId: string, mutants: readonly MutantRecord[], checksValidOnHealthy: boolean): VerifierEvaluation;
+/**
+ * §7.4 metamorphic relation: Pre(x,T) ⟹ R(f(x), f(T(x))). The relation must
+ * be a declared required property of this program — never a universal rule.
+ */
+export interface MetamorphicRelation {
+    readonly relationId: string;
+    readonly description: string;
+    readonly precondition: string;
+    readonly requiredProperty: boolean;
+}
+export declare function relationApplicable(rel: MetamorphicRelation): boolean;
+/** §10.1 progress key: same key + no new adopted evidence = repetition. */
+export interface ProgressKey {
+    readonly goalScope: string;
+    readonly candidateFamily: string;
+    readonly errorClass: string;
+    readonly approach: string;
+    readonly checkMethod: string;
+    readonly sourceFamily: string;
+}
+export declare function progressKey(k: ProgressKey): string;
+export declare function isStagnant(current: ProgressKey, history: readonly {
+    key: string;
+    newEvidence: boolean;
+    resolvedObligations: number;
+    newValidChecks: number;
+}[], maxRepetitions: number): boolean;
+/** §10.2 net strategy-switch score with hysteresis margin. */
+export declare function netSwitchValue(estimatedLossDelta: number, cost: number, switchCost: number): number;
+//# sourceMappingURL=verifier.d.ts.map

package/dist/metacognition/verifier.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"verifier.d.ts","sourceRoot":"","sources":["../../src/metacognition/verifier.ts"],"names":[],"mappings":"AAYA,MAAM,MAAM,WAAW,GACpB,iBAAiB,GACjB,UAAU,GACV,YAAY,GACZ,iBAAiB,GACjB,qBAAqB,GACrB,SAAS,GACT,mBAAmB,CAAC;AACvB,MAAM,WAAW,YAAY;IAC5B,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,kEAAkE;IAClE,QAAQ,CAAC,cAAc,EAAE,WAAW,CAAC;IACrC,2EAA2E;IAC3E,QAAQ,CAAC,iBAAiB,EAAE,OAAO,CAAC;CACpC;AACD,MAAM,WAAW,kBAAkB;IAClC,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,mEAAmE;IACnE,QAAQ,CAAC,aAAa,EAAE,MAAM,GAAG,SAAS,CAAC;IAC3C,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,QAAQ,EAAE,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,WAAW,EAAE,iBAAiB,GAAG,UAAU,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC;IAClG,QAAQ,CAAC,kBAAkB,EAAE,SAAS,MAAM,EAAE,CAAC;IAC/C,QAAQ,CAAC,UAAU,EAAE,SAAS,MAAM,EAAE,CAAC;CACvC;AAID,wBAAgB,gBAAgB,CAC/B,YAAY,EAAE,MAAM,EACpB,OAAO,EAAE,SAAS,YAAY,EAAE,EAChC,oBAAoB,EAAE,OAAO,GAC3B,kBAAkB,CAsDpB;AAED;;;GAGG;AACH,MAAM,WAAW,mBAAmB;IACnC,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,gBAAgB,EAAE,OAAO,CAAC;CACnC;AACD,wBAAgB,kBAAkB,CAAC,GAAG,EAAE,mBAAmB,GAAG,OAAO,CAEpE;AAED,4EAA2E;AAC3E,MAAM,WAAW,WAAW;IAC3B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,eAAe,EAAE,MAAM,CAAC;IACjC,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;CAC9B;AACD,wBAAgB,WAAW,CAAC,CAAC,EAAE,WAAW,GAAG,MAAM,CAElD;AACD,wBAAgB,UAAU,CACzB,OAAO,EAAE,WAAW,EACpB,OAAO,EAAE,SAAS;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,WAAW,EAAE,OAAO,CAAC;IAAC,mBAAmB,EAAE,MAAM,CAAC;IAAC,cAAc,EAAE,MAAM,CAAA;CAAE,EAAE,EAC9G,cAAc,EAAE,MAAM,GACpB,OAAO,CAUT;AAED,+DAA8D;AAC9D,wBAAgB,cAAc,CAAC,kBAAkB,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,GAAG,MAAM,CAKnG","sourcesContent":["/**\n * Algorithm D — meta-verification: checking that a check detects the defect\n * it claims to cover. Spec §7.\n *\n * Obligation-scoped negative controls are evaluated over VALID, non-equivalent\n * mutants only. Environment failures, compile-broken mutants, and timeouts are\n * never counted as semantic detections. An empty denominator is `unknown`,\n * never a score. This is an empirical detection rate for this mutant set,\n * not the probability the code is correct.\n */\nimport { ensure, finite, integer, lexical, member, text, unique } from \"./validation.ts\";\n\nexport type MutantClass =\n\t| \"killed-intended\"\n\t| \"survived\"\n\t| \"equivalent\"\n\t| \"invalid-compile\"\n\t| \"environment-failure\"\n\t| \"timeout\"\n\t| \"killed-unintended\";\nexport interface MutantRecord {\n\treadonly mutantId: string;\n\treadonly obligationId: string;\n\t/** The host classifies the outcome; this module never guesses. */\n\treadonly classification: MutantClass;\n\t/** Whether the intended obligation violation was what the check caught. */\n\treadonly intendedViolation: boolean;\n}\nexport interface VerifierEvaluation {\n\treadonly obligationId: string;\n\t/** D_v over valid non-equivalent mutants; \"unknown\" when empty. */\n\treadonly detectionRate: number | \"unknown\";\n\treadonly evaluated: number;\n\treadonly excluded: Readonly<Record<Exclude<MutantClass, \"killed-intended\" | \"survived\">, number>>;\n\treadonly survivingMutantIds: readonly string[];\n\treadonly blindSpots: readonly string[];\n}\n\nconst MUTANT_BOUND = 1024;\n\nexport function evaluateVerifier(\n\tobligationId: string,\n\tmutants: readonly MutantRecord[],\n\tchecksValidOnHealthy: boolean,\n): VerifierEvaluation {\n\ttext(obligationId, \"obligationId\", 256);\n\tensure(mutants.length <= MUTANT_BOUND, \"too many mutants\");\n\tunique(\n\t\tmutants.map((m) => m.mutantId),\n\t\t\"mutant ids\",\n\t);\n\tconst scoped = mutants.filter((m) => m.obligationId === obligationId);\n\tconst excluded: Record<Exclude<MutantClass, \"killed-intended\" | \"survived\">, number> = {\n\t\tequivalent: 0,\n\t\t\"invalid-compile\": 0,\n\t\t\"environment-failure\": 0,\n\t\ttimeout: 0,\n\t\t\"killed-unintended\": 0,\n\t};\n\tlet detected = 0;\n\tconst survivors: string[] = [];\n\tconst blindSpots: string[] = [];\n\tfor (const m of scoped) {\n\t\tmember(\n\t\t\tm.classification,\n\t\t\t[\n\t\t\t\t\"killed-intended\",\n\t\t\t\t\"survived\",\n\t\t\t\t\"equivalent\",\n\t\t\t\t\"invalid-compile\",\n\t\t\t\t\"environment-failure\",\n\t\t\t\t\"timeout\",\n\t\t\t\t\"killed-unintended\",\n\t\t\t],\n\t\t\t\"mutant classification\",\n\t\t);\n\t\tif (m.classification === \"killed-intended\" && m.intendedViolation) {\n\t\t\tdetected += 1;\n\t\t} else if (m.classification === \"survived\") {\n\t\t\tsurvivors.push(m.mutantId);\n\t\t\tblindSpots.push(`undetected-violation:${m.mutantId}`);\n\t\t} else {\n\t\t\texcluded[m.classification as keyof typeof excluded] += 1;\n\t\t\tif (m.classification === \"invalid-compile\" || m.classification === \"environment-failure\") {\n\t\t\t\tblindSpots.push(`uninformative-mutant:${m.mutantId}`);\n\t\t\t}\n\t\t}\n\t}\n\tif (!checksValidOnHealthy) blindSpots.push(\"check-invalid-on-healthy-candidate\");\n\tconst denominator = detected + survivors.length;\n\treturn {\n\t\tobligationId,\n\t\tdetectionRate: denominator === 0 ? \"unknown\" : detected / denominator,\n\t\tevaluated: denominator,\n\t\texcluded,\n\t\tsurvivingMutantIds: survivors.sort(lexical),\n\t\tblindSpots: blindSpots.sort(lexical),\n\t};\n}\n\n/**\n * §7.4 metamorphic relation: Pre(x,T) ⟹ R(f(x), f(T(x))). The relation must\n * be a declared required property of this program — never a universal rule.\n */\nexport interface MetamorphicRelation {\n\treadonly relationId: string;\n\treadonly description: string;\n\treadonly precondition: string;\n\treadonly requiredProperty: boolean;\n}\nexport function relationApplicable(rel: MetamorphicRelation): boolean {\n\treturn rel.requiredProperty;\n}\n\n/** §10.1 progress key: same key + no new adopted evidence = repetition. */\nexport interface ProgressKey {\n\treadonly goalScope: string;\n\treadonly candidateFamily: string;\n\treadonly errorClass: string;\n\treadonly approach: string;\n\treadonly checkMethod: string;\n\treadonly sourceFamily: string;\n}\nexport function progressKey(k: ProgressKey): string {\n\treturn [k.goalScope, k.candidateFamily, k.errorClass, k.approach, k.checkMethod, k.sourceFamily].join(\"\\u001f\");\n}\nexport function isStagnant(\n\tcurrent: ProgressKey,\n\thistory: readonly { key: string; newEvidence: boolean; resolvedObligations: number; newValidChecks: number }[],\n\tmaxRepetitions: number,\n): boolean {\n\tinteger(maxRepetitions, \"maxRepetitions\", 64);\n\tconst key = progressKey(current);\n\tlet repeats = 0;\n\tfor (const entry of history) {\n\t\tif (entry.key === key && !entry.newEvidence && entry.resolvedObligations === 0 && entry.newValidChecks === 0) {\n\t\t\trepeats += 1;\n\t\t}\n\t}\n\treturn repeats >= maxRepetitions;\n}\n\n/** §10.2 net strategy-switch score with hysteresis margin. */\nexport function netSwitchValue(estimatedLossDelta: number, cost: number, switchCost: number): number {\n\tfinite(estimatedLossDelta, \"estimatedLossDelta\");\n\tfinite(cost, \"cost\");\n\tfinite(switchCost, \"switchCost\");\n\treturn estimatedLossDelta - cost - switchCost;\n}\n"]}

package/dist/metacognition/verifier.js

@@ -0,0 +1,88 @@
+/**
+ * Algorithm D — meta-verification: checking that a check detects the defect
+ * it claims to cover. Spec §7.
+ *
+ * Obligation-scoped negative controls are evaluated over VALID, non-equivalent
+ * mutants only. Environment failures, compile-broken mutants, and timeouts are
+ * never counted as semantic detections. An empty denominator is `unknown`,
+ * never a score. This is an empirical detection rate for this mutant set,
+ * not the probability the code is correct.
+ */
+import { ensure, finite, integer, lexical, member, text, unique } from "./validation.js";
+const MUTANT_BOUND = 1024;
+export function evaluateVerifier(obligationId, mutants, checksValidOnHealthy) {
+    text(obligationId, "obligationId", 256);
+    ensure(mutants.length <= MUTANT_BOUND, "too many mutants");
+    unique(mutants.map((m) => m.mutantId), "mutant ids");
+    const scoped = mutants.filter((m) => m.obligationId === obligationId);
+    const excluded = {
+        equivalent: 0,
+        "invalid-compile": 0,
+        "environment-failure": 0,
+        timeout: 0,
+        "killed-unintended": 0,
+    };
+    let detected = 0;
+    const survivors = [];
+    const blindSpots = [];
+    for (const m of scoped) {
+        member(m.classification, [
+            "killed-intended",
+            "survived",
+            "equivalent",
+            "invalid-compile",
+            "environment-failure",
+            "timeout",
+            "killed-unintended",
+        ], "mutant classification");
+        if (m.classification === "killed-intended" && m.intendedViolation) {
+            detected += 1;
+        }
+        else if (m.classification === "survived") {
+            survivors.push(m.mutantId);
+            blindSpots.push(`undetected-violation:${m.mutantId}`);
+        }
+        else {
+            excluded[m.classification] += 1;
+            if (m.classification === "invalid-compile" || m.classification === "environment-failure") {
+                blindSpots.push(`uninformative-mutant:${m.mutantId}`);
+            }
+        }
+    }
+    if (!checksValidOnHealthy)
+        blindSpots.push("check-invalid-on-healthy-candidate");
+    const denominator = detected + survivors.length;
+    return {
+        obligationId,
+        detectionRate: denominator === 0 ? "unknown" : detected / denominator,
+        evaluated: denominator,
+        excluded,
+        survivingMutantIds: survivors.sort(lexical),
+        blindSpots: blindSpots.sort(lexical),
+    };
+}
+export function relationApplicable(rel) {
+    return rel.requiredProperty;
+}
+export function progressKey(k) {
+    return [k.goalScope, k.candidateFamily, k.errorClass, k.approach, k.checkMethod, k.sourceFamily].join("\u001f");
+}
+export function isStagnant(current, history, maxRepetitions) {
+    integer(maxRepetitions, "maxRepetitions", 64);
+    const key = progressKey(current);
+    let repeats = 0;
+    for (const entry of history) {
+        if (entry.key === key && !entry.newEvidence && entry.resolvedObligations === 0 && entry.newValidChecks === 0) {
+            repeats += 1;
+        }
+    }
+    return repeats >= maxRepetitions;
+}
+/** §10.2 net strategy-switch score with hysteresis margin. */
+export function netSwitchValue(estimatedLossDelta, cost, switchCost) {
+    finite(estimatedLossDelta, "estimatedLossDelta");
+    finite(cost, "cost");
+    finite(switchCost, "switchCost");
+    return estimatedLossDelta - cost - switchCost;
+}
+//# sourceMappingURL=verifier.js.map

package/dist/metacognition/verifier.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"verifier.js","sourceRoot":"","sources":["../../src/metacognition/verifier.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AACH,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,iBAAiB,CAAC;AA4BzF,MAAM,YAAY,GAAG,IAAI,CAAC;AAE1B,MAAM,UAAU,gBAAgB,CAC/B,YAAoB,EACpB,OAAgC,EAChC,oBAA6B,EACR;IACrB,IAAI,CAAC,YAAY,EAAE,cAAc,EAAE,GAAG,CAAC,CAAC;IACxC,MAAM,CAAC,OAAO,CAAC,MAAM,IAAI,YAAY,EAAE,kBAAkB,CAAC,CAAC;IAC3D,MAAM,CACL,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,EAC9B,YAAY,CACZ,CAAC;IACF,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,YAAY,KAAK,YAAY,CAAC,CAAC;IACtE,MAAM,QAAQ,GAAyE;QACtF,UAAU,EAAE,CAAC;QACb,iBAAiB,EAAE,CAAC;QACpB,qBAAqB,EAAE,CAAC;QACxB,OAAO,EAAE,CAAC;QACV,mBAAmB,EAAE,CAAC;KACtB,CAAC;IACF,IAAI,QAAQ,GAAG,CAAC,CAAC;IACjB,MAAM,SAAS,GAAa,EAAE,CAAC;IAC/B,MAAM,UAAU,GAAa,EAAE,CAAC;IAChC,KAAK,MAAM,CAAC,IAAI,MAAM,EAAE,CAAC;QACxB,MAAM,CACL,CAAC,CAAC,cAAc,EAChB;YACC,iBAAiB;YACjB,UAAU;YACV,YAAY;YACZ,iBAAiB;YACjB,qBAAqB;YACrB,SAAS;YACT,mBAAmB;SACnB,EACD,uBAAuB,CACvB,CAAC;QACF,IAAI,CAAC,CAAC,cAAc,KAAK,iBAAiB,IAAI,CAAC,CAAC,iBAAiB,EAAE,CAAC;YACnE,QAAQ,IAAI,CAAC,CAAC;QACf,CAAC;aAAM,IAAI,CAAC,CAAC,cAAc,KAAK,UAAU,EAAE,CAAC;YAC5C,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC;YAC3B,UAAU,CAAC,IAAI,CAAC,wBAAwB,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC;QACvD,CAAC;aAAM,CAAC;YACP,QAAQ,CAAC,CAAC,CAAC,cAAuC,CAAC,IAAI,CAAC,CAAC;YACzD,IAAI,CAAC,CAAC,cAAc,KAAK,iBAAiB,IAAI,CAAC,CAAC,cAAc,KAAK,qBAAqB,EAAE,CAAC;gBAC1F,UAAU,CAAC,IAAI,CAAC,wBAAwB,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC;YACvD,CAAC;QACF,CAAC;IACF,CAAC;IACD,IAAI,CAAC,oBAAoB;QAAE,UAAU,CAAC,IAAI,CAAC,oCAAoC,CAAC,CAAC;IACjF,MAAM,WAAW,GAAG,QAAQ,GAAG,SAAS,CAAC,MAAM,CAAC;IAChD,OAAO;QACN,YAAY;QACZ,aAAa,EAAE,WAAW,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,QAAQ,GAAG,WAAW;QACrE,SAAS,EAAE,WAAW;QACtB,QAAQ;QACR,kBAAkB,EAAE,SAAS,CAAC,IAAI,CAAC,OAAO,CAAC;QAC3C,UAAU,EAAE,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC;KACpC,CAAC;AAAA,CACF;AAYD,MAAM,UAAU,kBAAkB,CAAC,GAAwB,EAAW;IACrE,OAAO,GAAG,CAAC,gBAAgB,CAAC;AAAA,CAC5B;AAWD,MAAM,UAAU,WAAW,CAAC,CAAc,EAAU;IACnD,OAAO,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC,eAAe,EAAE,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC,YAAY,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;AAAA,CAChH;AACD,MAAM,UAAU,UAAU,CACzB,OAAoB,EACpB,OAA8G,EAC9G,cAAsB,EACZ;IACV,OAAO,CAAC,cAAc,EAAE,gBAAgB,EAAE,EAAE,CAAC,CAAC;IAC9C,MAAM,GAAG,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC;IACjC,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC7B,IAAI,KAAK,CAAC,GAAG,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,WAAW,IAAI,KAAK,CAAC,mBAAmB,KAAK,CAAC,IAAI,KAAK,CAAC,cAAc,KAAK,CAAC,EAAE,CAAC;YAC9G,OAAO,IAAI,CAAC,CAAC;QACd,CAAC;IACF,CAAC;IACD,OAAO,OAAO,IAAI,cAAc,CAAC;AAAA,CACjC;AAED,+DAA8D;AAC9D,MAAM,UAAU,cAAc,CAAC,kBAA0B,EAAE,IAAY,EAAE,UAAkB,EAAU;IACpG,MAAM,CAAC,kBAAkB,EAAE,oBAAoB,CAAC,CAAC;IACjD,MAAM,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IACrB,MAAM,CAAC,UAAU,EAAE,YAAY,CAAC,CAAC;IACjC,OAAO,kBAAkB,GAAG,IAAI,GAAG,UAAU,CAAC;AAAA,CAC9C","sourcesContent":["/**\n * Algorithm D — meta-verification: checking that a check detects the defect\n * it claims to cover. Spec §7.\n *\n * Obligation-scoped negative controls are evaluated over VALID, non-equivalent\n * mutants only. Environment failures, compile-broken mutants, and timeouts are\n * never counted as semantic detections. An empty denominator is `unknown`,\n * never a score. This is an empirical detection rate for this mutant set,\n * not the probability the code is correct.\n */\nimport { ensure, finite, integer, lexical, member, text, unique } from \"./validation.ts\";\n\nexport type MutantClass =\n\t| \"killed-intended\"\n\t| \"survived\"\n\t| \"equivalent\"\n\t| \"invalid-compile\"\n\t| \"environment-failure\"\n\t| \"timeout\"\n\t| \"killed-unintended\";\nexport interface MutantRecord {\n\treadonly mutantId: string;\n\treadonly obligationId: string;\n\t/** The host classifies the outcome; this module never guesses. */\n\treadonly classification: MutantClass;\n\t/** Whether the intended obligation violation was what the check caught. */\n\treadonly intendedViolation: boolean;\n}\nexport interface VerifierEvaluation {\n\treadonly obligationId: string;\n\t/** D_v over valid non-equivalent mutants; \"unknown\" when empty. */\n\treadonly detectionRate: number | \"unknown\";\n\treadonly evaluated: number;\n\treadonly excluded: Readonly<Record<Exclude<MutantClass, \"killed-intended\" | \"survived\">, number>>;\n\treadonly survivingMutantIds: readonly string[];\n\treadonly blindSpots: readonly string[];\n}\n\nconst MUTANT_BOUND = 1024;\n\nexport function evaluateVerifier(\n\tobligationId: string,\n\tmutants: readonly MutantRecord[],\n\tchecksValidOnHealthy: boolean,\n): VerifierEvaluation {\n\ttext(obligationId, \"obligationId\", 256);\n\tensure(mutants.length <= MUTANT_BOUND, \"too many mutants\");\n\tunique(\n\t\tmutants.map((m) => m.mutantId),\n\t\t\"mutant ids\",\n\t);\n\tconst scoped = mutants.filter((m) => m.obligationId === obligationId);\n\tconst excluded: Record<Exclude<MutantClass, \"killed-intended\" | \"survived\">, number> = {\n\t\tequivalent: 0,\n\t\t\"invalid-compile\": 0,\n\t\t\"environment-failure\": 0,\n\t\ttimeout: 0,\n\t\t\"killed-unintended\": 0,\n\t};\n\tlet detected = 0;\n\tconst survivors: string[] = [];\n\tconst blindSpots: string[] = [];\n\tfor (const m of scoped) {\n\t\tmember(\n\t\t\tm.classification,\n\t\t\t[\n\t\t\t\t\"killed-intended\",\n\t\t\t\t\"survived\",\n\t\t\t\t\"equivalent\",\n\t\t\t\t\"invalid-compile\",\n\t\t\t\t\"environment-failure\",\n\t\t\t\t\"timeout\",\n\t\t\t\t\"killed-unintended\",\n\t\t\t],\n\t\t\t\"mutant classification\",\n\t\t);\n\t\tif (m.classification === \"killed-intended\" && m.intendedViolation) {\n\t\t\tdetected += 1;\n\t\t} else if (m.classification === \"survived\") {\n\t\t\tsurvivors.push(m.mutantId);\n\t\t\tblindSpots.push(`undetected-violation:${m.mutantId}`);\n\t\t} else {\n\t\t\texcluded[m.classification as keyof typeof excluded] += 1;\n\t\t\tif (m.classification === \"invalid-compile\" || m.classification === \"environment-failure\") {\n\t\t\t\tblindSpots.push(`uninformative-mutant:${m.mutantId}`);\n\t\t\t}\n\t\t}\n\t}\n\tif (!checksValidOnHealthy) blindSpots.push(\"check-invalid-on-healthy-candidate\");\n\tconst denominator = detected + survivors.length;\n\treturn {\n\t\tobligationId,\n\t\tdetectionRate: denominator === 0 ? \"unknown\" : detected / denominator,\n\t\tevaluated: denominator,\n\t\texcluded,\n\t\tsurvivingMutantIds: survivors.sort(lexical),\n\t\tblindSpots: blindSpots.sort(lexical),\n\t};\n}\n\n/**\n * §7.4 metamorphic relation: Pre(x,T) ⟹ R(f(x), f(T(x))). The relation must\n * be a declared required property of this program — never a universal rule.\n */\nexport interface MetamorphicRelation {\n\treadonly relationId: string;\n\treadonly description: string;\n\treadonly precondition: string;\n\treadonly requiredProperty: boolean;\n}\nexport function relationApplicable(rel: MetamorphicRelation): boolean {\n\treturn rel.requiredProperty;\n}\n\n/** §10.1 progress key: same key + no new adopted evidence = repetition. */\nexport interface ProgressKey {\n\treadonly goalScope: string;\n\treadonly candidateFamily: string;\n\treadonly errorClass: string;\n\treadonly approach: string;\n\treadonly checkMethod: string;\n\treadonly sourceFamily: string;\n}\nexport function progressKey(k: ProgressKey): string {\n\treturn [k.goalScope, k.candidateFamily, k.errorClass, k.approach, k.checkMethod, k.sourceFamily].join(\"\\u001f\");\n}\nexport function isStagnant(\n\tcurrent: ProgressKey,\n\thistory: readonly { key: string; newEvidence: boolean; resolvedObligations: number; newValidChecks: number }[],\n\tmaxRepetitions: number,\n): boolean {\n\tinteger(maxRepetitions, \"maxRepetitions\", 64);\n\tconst key = progressKey(current);\n\tlet repeats = 0;\n\tfor (const entry of history) {\n\t\tif (entry.key === key && !entry.newEvidence && entry.resolvedObligations === 0 && entry.newValidChecks === 0) {\n\t\t\trepeats += 1;\n\t\t}\n\t}\n\treturn repeats >= maxRepetitions;\n}\n\n/** §10.2 net strategy-switch score with hysteresis margin. */\nexport function netSwitchValue(estimatedLossDelta: number, cost: number, switchCost: number): number {\n\tfinite(estimatedLossDelta, \"estimatedLossDelta\");\n\tfinite(cost, \"cost\");\n\tfinite(switchCost, \"switchCost\");\n\treturn estimatedLossDelta - cost - switchCost;\n}\n"]}

package/dist/modes/interactive/components/assistant-message.d.ts

@@ -10,11 +10,13 @@ export declare class AssistantMessageComponent extends Container {
     private hiddenThinkingLabel;
     private lastMessage?;
     private hasToolCalls;
+    private readonly contentViews;
     constructor(message?: AssistantMessage, hideThinkingBlock?: boolean, markdownTheme?: MarkdownTheme, hiddenThinkingLabel?: string);
     invalidate(): void;
     setHideThinkingBlock(hide: boolean): void;
     setHiddenThinkingLabel(label: string): void;
     render(width: number): string[];
+    private contentView;
     updateContent(message: AssistantMessage): void;
 }
 //# sourceMappingURL=assistant-message.d.ts.map

package/dist/modes/interactive/components/assistant-message.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"assistant-message.d.ts","sourceRoot":"","sources":["../../../../src/modes/interactive/components/assistant-message.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,QAAQ,CAAC;AAC/C,OAAO,EAAE,SAAS,EAAY,KAAK,aAAa,EAAgB,MAAM,SAAS,CAAC;AAOhF;;GAEG;AACH,qBAAa,yBAA0B,SAAQ,SAAS;IACvD,OAAO,CAAC,gBAAgB,CAAY;IACpC,OAAO,CAAC,iBAAiB,CAAU;IACnC,OAAO,CAAC,aAAa,CAAgB;IACrC,OAAO,CAAC,mBAAmB,CAAS;IACpC,OAAO,CAAC,WAAW,CAAC,CAAmB;IACvC,OAAO,CAAC,YAAY,CAAS;IAE7B,YACC,OAAO,CAAC,EAAE,gBAAgB,EAC1B,iBAAiB,UAAQ,EACzB,aAAa,GAAE,aAAkC,EACjD,mBAAmB,SAAgB,EAenC;IAEQ,UAAU,IAAI,IAAI,CAK1B;IAED,oBAAoB,CAAC,IAAI,EAAE,OAAO,GAAG,IAAI,CAKxC;IAED,sBAAsB,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI,CAK1C;IAEQ,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,EAAE,CASvC;IAED,aAAa,CAAC,OAAO,EAAE,gBAAgB,GAAG,IAAI,CAyE7C;CACD","sourcesContent":["import type { AssistantMessage } from \"omk-ai\";\nimport { Container, Markdown, type MarkdownTheme, Spacer, Text } from \"omk-tui\";\nimport { getMarkdownTheme, theme } from \"../theme/theme.ts\";\n\nconst OSC133_ZONE_START = \"\\x1b]133;A\\x07\";\nconst OSC133_ZONE_END = \"\\x1b]133;B\\x07\";\nconst OSC133_ZONE_FINAL = \"\\x1b]133;C\\x07\";\n\n/**\n * Component that renders a complete assistant message\n */\nexport class AssistantMessageComponent extends Container {\n\tprivate contentContainer: Container;\n\tprivate hideThinkingBlock: boolean;\n\tprivate markdownTheme: MarkdownTheme;\n\tprivate hiddenThinkingLabel: string;\n\tprivate lastMessage?: AssistantMessage;\n\tprivate hasToolCalls = false;\n\n\tconstructor(\n\t\tmessage?: AssistantMessage,\n\t\thideThinkingBlock = false,\n\t\tmarkdownTheme: MarkdownTheme = getMarkdownTheme(),\n\t\thiddenThinkingLabel = \"Thinking...\",\n\t) {\n\t\tsuper();\n\n\t\tthis.hideThinkingBlock = hideThinkingBlock;\n\t\tthis.markdownTheme = markdownTheme;\n\t\tthis.hiddenThinkingLabel = hiddenThinkingLabel;\n\n\t\t// Container for text/thinking content\n\t\tthis.contentContainer = new Container();\n\t\tthis.addChild(this.contentContainer);\n\n\t\tif (message) {\n\t\t\tthis.updateContent(message);\n\t\t}\n\t}\n\n\toverride invalidate(): void {\n\t\tsuper.invalidate();\n\t\tif (this.lastMessage) {\n\t\t\tthis.updateContent(this.lastMessage);\n\t\t}\n\t}\n\n\tsetHideThinkingBlock(hide: boolean): void {\n\t\tthis.hideThinkingBlock = hide;\n\t\tif (this.lastMessage) {\n\t\t\tthis.updateContent(this.lastMessage);\n\t\t}\n\t}\n\n\tsetHiddenThinkingLabel(label: string): void {\n\t\tthis.hiddenThinkingLabel = label;\n\t\tif (this.lastMessage) {\n\t\t\tthis.updateContent(this.lastMessage);\n\t\t}\n\t}\n\n\toverride render(width: number): string[] {\n\t\tconst lines = super.render(width);\n\t\tif (this.hasToolCalls || lines.length === 0) {\n\t\t\treturn lines;\n\t\t}\n\n\t\tlines[0] = OSC133_ZONE_START + lines[0];\n\t\tlines[lines.length - 1] = OSC133_ZONE_END + OSC133_ZONE_FINAL + lines[lines.length - 1];\n\t\treturn lines;\n\t}\n\n\tupdateContent(message: AssistantMessage): void {\n\t\tthis.lastMessage = message;\n\n\t\t// Clear content container\n\t\tthis.contentContainer.clear();\n\n\t\tconst hasVisibleContent = message.content.some(\n\t\t\t(c) => (c.type === \"text\" && c.text.trim()) || (c.type === \"thinking\" && c.thinking.trim()),\n\t\t);\n\n\t\tif (hasVisibleContent) {\n\t\t\tthis.contentContainer.addChild(new Spacer(1));\n\t\t}\n\n\t\t// Render content in order\n\t\tfor (let i = 0; i < message.content.length; i++) {\n\t\t\tconst content = message.content[i];\n\t\t\tif (content.type === \"text\" && content.text.trim()) {\n\t\t\t\t// Assistant text messages with no background - trim the text\n\t\t\t\t// Set paddingY=0 to avoid extra spacing before tool executions\n\t\t\t\tthis.contentContainer.addChild(new Markdown(content.text.trim(), 1, 0, this.markdownTheme));\n\t\t\t} else if (content.type === \"thinking\" && content.thinking.trim()) {\n\t\t\t\t// Add spacing only when another visible assistant content block follows.\n\t\t\t\t// This avoids a superfluous blank line before separately-rendered tool execution blocks.\n\t\t\t\tconst hasVisibleContentAfter = message.content\n\t\t\t\t\t.slice(i + 1)\n\t\t\t\t\t.some((c) => (c.type === \"text\" && c.text.trim()) || (c.type === \"thinking\" && c.thinking.trim()));\n\n\t\t\t\tif (this.hideThinkingBlock) {\n\t\t\t\t\t// Show static thinking label when hidden\n\t\t\t\t\tthis.contentContainer.addChild(\n\t\t\t\t\t\tnew Text(theme.italic(theme.fg(\"thinkingText\", this.hiddenThinkingLabel)), 1, 0),\n\t\t\t\t\t);\n\t\t\t\t\tif (hasVisibleContentAfter) {\n\t\t\t\t\t\tthis.contentContainer.addChild(new Spacer(1));\n\t\t\t\t\t}\n\t\t\t\t} else {\n\t\t\t\t\t// Thinking traces in thinkingText color, italic\n\t\t\t\t\tthis.contentContainer.addChild(\n\t\t\t\t\t\tnew Markdown(content.thinking.trim(), 1, 0, this.markdownTheme, {\n\t\t\t\t\t\t\tcolor: (text: string) => theme.fg(\"thinkingText\", text),\n\t\t\t\t\t\t\titalic: true,\n\t\t\t\t\t\t}),\n\t\t\t\t\t);\n\t\t\t\t\tif (hasVisibleContentAfter) {\n\t\t\t\t\t\tthis.contentContainer.addChild(new Spacer(1));\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// Check if aborted - show after partial content\n\t\t// But only if there are no tool calls (tool execution components will show the error)\n\t\tconst hasToolCalls = message.content.some((c) => c.type === \"toolCall\");\n\t\tthis.hasToolCalls = hasToolCalls;\n\t\tif (!hasToolCalls) {\n\t\t\tif (message.stopReason === \"aborted\") {\n\t\t\t\tconst abortMessage =\n\t\t\t\t\tmessage.errorMessage && message.errorMessage !== \"Request was aborted\"\n\t\t\t\t\t\t? message.errorMessage\n\t\t\t\t\t\t: \"Operation aborted\";\n\t\t\t\tif (hasVisibleContent) {\n\t\t\t\t\tthis.contentContainer.addChild(new Spacer(1));\n\t\t\t\t} else {\n\t\t\t\t\tthis.contentContainer.addChild(new Spacer(1));\n\t\t\t\t}\n\t\t\t\tthis.contentContainer.addChild(new Text(theme.fg(\"error\", abortMessage), 1, 0));\n\t\t\t} else if (message.stopReason === \"error\") {\n\t\t\t\tconst errorMsg = message.errorMessage || \"Unknown error\";\n\t\t\t\tthis.contentContainer.addChild(new Spacer(1));\n\t\t\t\tthis.contentContainer.addChild(new Text(theme.fg(\"error\", `Error: ${errorMsg}`), 1, 0));\n\t\t\t}\n\t\t}\n\t}\n}\n"]}
+{"version":3,"file":"assistant-message.d.ts","sourceRoot":"","sources":["../../../../src/modes/interactive/components/assistant-message.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,QAAQ,CAAC;AAC/C,OAAO,EAAE,SAAS,EAAY,KAAK,aAAa,EAAgB,MAAM,SAAS,CAAC;AAchF;;GAEG;AACH,qBAAa,yBAA0B,SAAQ,SAAS;IACvD,OAAO,CAAC,gBAAgB,CAAY;IACpC,OAAO,CAAC,iBAAiB,CAAU;IACnC,OAAO,CAAC,aAAa,CAAgB;IACrC,OAAO,CAAC,mBAAmB,CAAS;IACpC,OAAO,CAAC,WAAW,CAAC,CAAmB;IACvC,OAAO,CAAC,YAAY,CAAS;IAC7B,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAkC;IAE/D,YACC,OAAO,CAAC,EAAE,gBAAgB,EAC1B,iBAAiB,UAAQ,EACzB,aAAa,GAAE,aAAkC,EACjD,mBAAmB,SAAgB,EAenC;IAEQ,UAAU,IAAI,IAAI,CAO1B;IAED,oBAAoB,CAAC,IAAI,EAAE,OAAO,GAAG,IAAI,CAKxC;IAED,sBAAsB,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI,CAK1C;IAEQ,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,EAAE,CASvC;IAED,OAAO,CAAC,WAAW;IA2BnB,aAAa,CAAC,OAAO,EAAE,gBAAgB,GAAG,IAAI,CAsE7C;CACD","sourcesContent":["import type { AssistantMessage } from \"omk-ai\";\nimport { Container, Markdown, type MarkdownTheme, Spacer, Text } from \"omk-tui\";\nimport { getMarkdownTheme, theme } from \"../theme/theme.ts\";\n\ntype ContentViewKind = \"text\" | \"thinking\" | \"hidden-thinking\";\ninterface ContentView {\n\tkind: ContentViewKind;\n\ttext: string;\n\tcomponent: Markdown | Text;\n}\n\nconst OSC133_ZONE_START = \"\\x1b]133;A\\x07\";\nconst OSC133_ZONE_END = \"\\x1b]133;B\\x07\";\nconst OSC133_ZONE_FINAL = \"\\x1b]133;C\\x07\";\n\n/**\n * Component that renders a complete assistant message\n */\nexport class AssistantMessageComponent extends Container {\n\tprivate contentContainer: Container;\n\tprivate hideThinkingBlock: boolean;\n\tprivate markdownTheme: MarkdownTheme;\n\tprivate hiddenThinkingLabel: string;\n\tprivate lastMessage?: AssistantMessage;\n\tprivate hasToolCalls = false;\n\tprivate readonly contentViews = new Map<number, ContentView>();\n\n\tconstructor(\n\t\tmessage?: AssistantMessage,\n\t\thideThinkingBlock = false,\n\t\tmarkdownTheme: MarkdownTheme = getMarkdownTheme(),\n\t\thiddenThinkingLabel = \"Thinking...\",\n\t) {\n\t\tsuper();\n\n\t\tthis.hideThinkingBlock = hideThinkingBlock;\n\t\tthis.markdownTheme = markdownTheme;\n\t\tthis.hiddenThinkingLabel = hiddenThinkingLabel;\n\n\t\t// Container for text/thinking content\n\t\tthis.contentContainer = new Container();\n\t\tthis.addChild(this.contentContainer);\n\n\t\tif (message) {\n\t\t\tthis.updateContent(message);\n\t\t}\n\t}\n\n\toverride invalidate(): void {\n\t\tsuper.invalidate();\n\t\t// Theme changes also invalidate pre-styled labels and Markdown style prefixes.\n\t\tthis.contentViews.clear();\n\t\tif (this.lastMessage) {\n\t\t\tthis.updateContent(this.lastMessage);\n\t\t}\n\t}\n\n\tsetHideThinkingBlock(hide: boolean): void {\n\t\tthis.hideThinkingBlock = hide;\n\t\tif (this.lastMessage) {\n\t\t\tthis.updateContent(this.lastMessage);\n\t\t}\n\t}\n\n\tsetHiddenThinkingLabel(label: string): void {\n\t\tthis.hiddenThinkingLabel = label;\n\t\tif (this.lastMessage) {\n\t\t\tthis.updateContent(this.lastMessage);\n\t\t}\n\t}\n\n\toverride render(width: number): string[] {\n\t\tconst lines = super.render(width);\n\t\tif (this.hasToolCalls || lines.length === 0) {\n\t\t\treturn lines;\n\t\t}\n\n\t\tlines[0] = OSC133_ZONE_START + lines[0];\n\t\tlines[lines.length - 1] = OSC133_ZONE_END + OSC133_ZONE_FINAL + lines[lines.length - 1];\n\t\treturn lines;\n\t}\n\n\tprivate contentView(index: number, kind: ContentViewKind, text: string): Markdown | Text {\n\t\tconst previous = this.contentViews.get(index);\n\t\tif (previous?.kind === kind) {\n\t\t\tif (previous.text !== text) {\n\t\t\t\tprevious.component.setText(\n\t\t\t\t\tkind === \"hidden-thinking\" ? theme.italic(theme.fg(\"thinkingText\", text)) : text,\n\t\t\t\t);\n\t\t\t\tprevious.text = text;\n\t\t\t}\n\t\t\treturn previous.component;\n\t\t}\n\t\tconst component =\n\t\t\tkind === \"hidden-thinking\"\n\t\t\t\t? new Text(theme.italic(theme.fg(\"thinkingText\", text)), 1, 0)\n\t\t\t\t: new Markdown(\n\t\t\t\t\t\ttext,\n\t\t\t\t\t\t1,\n\t\t\t\t\t\t0,\n\t\t\t\t\t\tthis.markdownTheme,\n\t\t\t\t\t\tkind === \"thinking\"\n\t\t\t\t\t\t\t? { color: (value: string) => theme.fg(\"thinkingText\", value), italic: true }\n\t\t\t\t\t\t\t: undefined,\n\t\t\t\t\t);\n\t\tthis.contentViews.set(index, { kind, text, component });\n\t\treturn component;\n\t}\n\n\tupdateContent(message: AssistantMessage): void {\n\t\tthis.lastMessage = message;\n\n\t\t// Clear content container\n\t\tthis.contentContainer.clear();\n\n\t\tconst hasVisibleContent = message.content.some(\n\t\t\t(c) => (c.type === \"text\" && c.text.trim()) || (c.type === \"thinking\" && c.thinking.trim()),\n\t\t);\n\n\t\tif (hasVisibleContent) {\n\t\t\tthis.contentContainer.addChild(new Spacer(1));\n\t\t}\n\n\t\tconst usedViews = new Set<number>();\n\t\tconst visibleAfter = new Array<boolean>(message.content.length);\n\t\tlet laterVisible = false;\n\t\tfor (let i = message.content.length - 1; i >= 0; i--) {\n\t\t\tvisibleAfter[i] = laterVisible;\n\t\t\tconst part = message.content[i];\n\t\t\tlaterVisible ||=\n\t\t\t\t(part.type === \"text\" && part.text.trim().length > 0) ||\n\t\t\t\t(part.type === \"thinking\" && part.thinking.trim().length > 0);\n\t\t}\n\n\t\t// Keep unchanged block render caches without relying on message identity.\n\t\tfor (let i = 0; i < message.content.length; i++) {\n\t\t\tconst content = message.content[i];\n\t\t\tif (content.type === \"text\" && content.text.trim()) {\n\t\t\t\tusedViews.add(i);\n\t\t\t\tthis.contentContainer.addChild(this.contentView(i, \"text\", content.text.trim()));\n\t\t\t} else if (content.type === \"thinking\" && content.thinking.trim()) {\n\t\t\t\tusedViews.add(i);\n\t\t\t\tthis.contentContainer.addChild(\n\t\t\t\t\tthis.contentView(\n\t\t\t\t\t\ti,\n\t\t\t\t\t\tthis.hideThinkingBlock ? \"hidden-thinking\" : \"thinking\",\n\t\t\t\t\t\tthis.hideThinkingBlock ? this.hiddenThinkingLabel : content.thinking.trim(),\n\t\t\t\t\t),\n\t\t\t\t);\n\t\t\t\t// Tool execution blocks are rendered separately and need no trailing spacer.\n\t\t\t\tif (visibleAfter[i]) this.contentContainer.addChild(new Spacer(1));\n\t\t\t}\n\t\t}\n\t\tfor (const index of this.contentViews.keys()) {\n\t\t\tif (!usedViews.has(index)) this.contentViews.delete(index);\n\t\t}\n\n\t\t// Check if aborted - show after partial content\n\t\t// But only if there are no tool calls (tool execution components will show the error)\n\t\tconst hasToolCalls = message.content.some((c) => c.type === \"toolCall\");\n\t\tthis.hasToolCalls = hasToolCalls;\n\t\tif (!hasToolCalls) {\n\t\t\tif (message.stopReason === \"aborted\") {\n\t\t\t\tconst abortMessage =\n\t\t\t\t\tmessage.errorMessage && message.errorMessage !== \"Request was aborted\"\n\t\t\t\t\t\t? message.errorMessage\n\t\t\t\t\t\t: \"Operation aborted\";\n\t\t\t\tif (hasVisibleContent) {\n\t\t\t\t\tthis.contentContainer.addChild(new Spacer(1));\n\t\t\t\t} else {\n\t\t\t\t\tthis.contentContainer.addChild(new Spacer(1));\n\t\t\t\t}\n\t\t\t\tthis.contentContainer.addChild(new Text(theme.fg(\"error\", abortMessage), 1, 0));\n\t\t\t} else if (message.stopReason === \"error\") {\n\t\t\t\tconst errorMsg = message.errorMessage || \"Unknown error\";\n\t\t\t\tthis.contentContainer.addChild(new Spacer(1));\n\t\t\t\tthis.contentContainer.addChild(new Text(theme.fg(\"error\", `Error: ${errorMsg}`), 1, 0));\n\t\t\t}\n\t\t}\n\t}\n}\n"]}

package/dist/modes/interactive/components/assistant-message.js

@@ -13,6 +13,7 @@ export class AssistantMessageComponent extends Container {
     hiddenThinkingLabel;
     lastMessage;
     hasToolCalls = false;
+    contentViews = new Map();
     constructor(message, hideThinkingBlock = false, markdownTheme = getMarkdownTheme(), hiddenThinkingLabel = "Thinking...") {
         super();
         this.hideThinkingBlock = hideThinkingBlock;
@@ -27,6 +28,8 @@ export class AssistantMessageComponent extends Container {
     }
     invalidate() {
         super.invalidate();
+        // Theme changes also invalidate pre-styled labels and Markdown style prefixes.
+        this.contentViews.clear();
         if (this.lastMessage) {
             this.updateContent(this.lastMessage);
         }
@@ -52,6 +55,23 @@ export class AssistantMessageComponent extends Container {
         lines[lines.length - 1] = OSC133_ZONE_END + OSC133_ZONE_FINAL + lines[lines.length - 1];
         return lines;
     }
+    contentView(index, kind, text) {
+        const previous = this.contentViews.get(index);
+        if (previous?.kind === kind) {
+            if (previous.text !== text) {
+                previous.component.setText(kind === "hidden-thinking" ? theme.italic(theme.fg("thinkingText", text)) : text);
+                previous.text = text;
+            }
+            return previous.component;
+        }
+        const component = kind === "hidden-thinking"
+            ? new Text(theme.italic(theme.fg("thinkingText", text)), 1, 0)
+            : new Markdown(text, 1, 0, this.markdownTheme, kind === "thinking"
+                ? { color: (value) => theme.fg("thinkingText", value), italic: true }
+                : undefined);
+        this.contentViews.set(index, { kind, text, component });
+        return component;
+    }
     updateContent(message) {
         this.lastMessage = message;
         // Clear content container
@@ -60,39 +80,35 @@ export class AssistantMessageComponent extends Container {
         if (hasVisibleContent) {
             this.contentContainer.addChild(new Spacer(1));
         }
-        // Render content in order
+        const usedViews = new Set();
+        const visibleAfter = new Array(message.content.length);
+        let laterVisible = false;
+        for (let i = message.content.length - 1; i >= 0; i--) {
+            visibleAfter[i] = laterVisible;
+            const part = message.content[i];
+            laterVisible ||=
+                (part.type === "text" && part.text.trim().length > 0) ||
+                    (part.type === "thinking" && part.thinking.trim().length > 0);
+        }
+        // Keep unchanged block render caches without relying on message identity.
         for (let i = 0; i < message.content.length; i++) {
             const content = message.content[i];
             if (content.type === "text" && content.text.trim()) {
-                // Assistant text messages with no background - trim the text
-                // Set paddingY=0 to avoid extra spacing before tool executions
-                this.contentContainer.addChild(new Markdown(content.text.trim(), 1, 0, this.markdownTheme));
+                usedViews.add(i);
+                this.contentContainer.addChild(this.contentView(i, "text", content.text.trim()));
             }
             else if (content.type === "thinking" && content.thinking.trim()) {
-                // Add spacing only when another visible assistant content block follows.
-                // This avoids a superfluous blank line before separately-rendered tool execution blocks.
-                const hasVisibleContentAfter = message.content
-                    .slice(i + 1)
-                    .some((c) => (c.type === "text" && c.text.trim()) || (c.type === "thinking" && c.thinking.trim()));
-                if (this.hideThinkingBlock) {
-                    // Show static thinking label when hidden
-                    this.contentContainer.addChild(new Text(theme.italic(theme.fg("thinkingText", this.hiddenThinkingLabel)), 1, 0));
-                    if (hasVisibleContentAfter) {
-                        this.contentContainer.addChild(new Spacer(1));
-                    }
-                }
-                else {
-                    // Thinking traces in thinkingText color, italic
-                    this.contentContainer.addChild(new Markdown(content.thinking.trim(), 1, 0, this.markdownTheme, {
-                        color: (text) => theme.fg("thinkingText", text),
-                        italic: true,
-                    }));
-                    if (hasVisibleContentAfter) {
-                        this.contentContainer.addChild(new Spacer(1));
-                    }
-                }
+                usedViews.add(i);
+                this.contentContainer.addChild(this.contentView(i, this.hideThinkingBlock ? "hidden-thinking" : "thinking", this.hideThinkingBlock ? this.hiddenThinkingLabel : content.thinking.trim()));
+                // Tool execution blocks are rendered separately and need no trailing spacer.
+                if (visibleAfter[i])
+                    this.contentContainer.addChild(new Spacer(1));
             }
         }
+        for (const index of this.contentViews.keys()) {
+            if (!usedViews.has(index))
+                this.contentViews.delete(index);
+        }
         // Check if aborted - show after partial content
         // But only if there are no tool calls (tool execution components will show the error)
         const hasToolCalls = message.content.some((c) => c.type === "toolCall");