spawnfile 0.1.25 → 0.1.28

package/dist/training/evaluation/semantic/evidence-units.js

@@ -0,0 +1,58 @@
+import { createHash } from 'node:crypto';
+const hash = (value) => createHash('sha256').update(value).digest('hex');
+/** Lossless, deterministic primitive leaves. IDs attest pointers, never factual support. */
+export function evidenceUnits(sources) {
+    const units = [], ids = new Set(), ancestors = new Set();
+    let bytes = 64; // Packet envelope and array delimiters are included conservatively.
+    const add = (source, pointer, chunk, value) => {
+        const encoded = JSON.stringify(value);
+        const id = `e-${hash(JSON.stringify([source, pointer, chunk, value])).slice(0, 20)}`;
+        if (ids.has(id))
+            throw new Error('Evidence unit identity collision');
+        const unit = { id, source, pointer, chunk, digest: `sha256:${hash(encoded)}`, value };
+        bytes += Buffer.byteLength(JSON.stringify(unit)) + 1;
+        if (units.length >= 256 || bytes > 128 * 1024)
+            throw new Error('Evidence units exceed bounded packet; select narrower evidence');
+        ids.add(id);
+        units.push(unit);
+    };
+    const walk = (value, source, pointer, depth) => {
+        if (depth > 64)
+            throw new Error('Evidence nesting exceeds bound');
+        if (typeof value === 'string') {
+            let text = '', size = 0, chunk = 0;
+            for (const character of value) {
+                text += character;
+                if (++size === 2048) {
+                    add(source, pointer, chunk++, text);
+                    text = '';
+                    size = 0;
+                }
+            }
+            if (size || !chunk)
+                add(source, pointer, chunk, text);
+        }
+        else if (value === null || typeof value === 'boolean' || (typeof value === 'number' && Number.isFinite(value)))
+            add(source, pointer, 0, value);
+        else if (typeof value === 'object') {
+            if (!Array.isArray(value) && Object.getPrototypeOf(value) !== Object.prototype && Object.getPrototypeOf(value) !== null)
+                throw new Error('Evidence must contain plain JSON objects');
+            if (ancestors.has(value))
+                throw new Error('Cyclic evidence is unsupported');
+            ancestors.add(value);
+            const entries = Array.isArray(value) ? Array.from(value.entries()).map(([key, child]) => [String(key), child])
+                : Object.keys(value).sort().map(key => [key, value[key]]);
+            if (!entries.length)
+                add(source, pointer, 0, Array.isArray(value) ? [] : {});
+            for (const [key, child] of entries)
+                walk(child, source, `${pointer}/${key.replace(/~/gu, '~0').replace(/\//gu, '~1')}`, depth + 1);
+            ancestors.delete(value);
+        }
+        else
+            throw new Error('Evidence must contain finite JSON values');
+    };
+    for (const source of Object.keys(sources).sort())
+        walk(sources[source], source, '', 0);
+    return { schema: 'spawnfile.training.evidence-units.v1', units };
+}
+//# sourceMappingURL=evidence-units.js.map

package/dist/training/evaluation/semantic/evidence-units.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"evidence-units.js","sourceRoot":"","sources":["../../../../src/training/evaluation/semantic/evidence-units.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAWzC,MAAM,IAAI,GAAG,CAAC,KAAa,EAAE,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;AAEjF,4FAA4F;AAC5F,MAAM,UAAU,aAAa,CAAC,OAA0C;IACtE,MAAM,KAAK,GAAmB,EAAE,EAAE,GAAG,GAAG,IAAI,GAAG,EAAU,EAAE,SAAS,GAAG,IAAI,GAAG,EAAU,CAAC;IACzF,IAAI,KAAK,GAAG,EAAE,CAAC,CAAC,oEAAoE;IACpF,MAAM,GAAG,GAAG,CAAC,MAAc,EAAE,OAAe,EAAE,KAAa,EAAE,KAA4B,EAAE,EAAE;QAC3F,MAAM,OAAO,GAAG,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;QACtC,MAAM,EAAE,GAAG,KAAK,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,MAAM,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,EAAE,CAAC;QACrF,IAAI,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,kCAAkC,CAAC,CAAC;QACrE,MAAM,IAAI,GAAG,EAAE,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,UAAU,IAAI,CAAC,OAAO,CAAC,EAAE,EAAE,KAAK,EAAE,CAAC;QACtF,KAAK,IAAI,MAAM,CAAC,UAAU,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;QACrD,IAAI,KAAK,CAAC,MAAM,IAAI,GAAG,IAAI,KAAK,GAAG,GAAG,GAAG,IAAI;YAAE,MAAM,IAAI,KAAK,CAAC,gEAAgE,CAAC,CAAC;QACjI,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAChC,CAAC,CAAC;IACF,MAAM,IAAI,GAAG,CAAC,KAAc,EAAE,MAAc,EAAE,OAAe,EAAE,KAAa,EAAE,EAAE;QAC9E,IAAI,KAAK,GAAG,EAAE;YAAE,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;QAClE,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;YAC9B,IAAI,IAAI,GAAG,EAAE,EAAE,IAAI,GAAG,CAAC,EAAE,KAAK,GAAG,CAAC,CAAC;YACnC,KAAK,MAAM,SAAS,IAAI,KAAK,EAAE,CAAC;gBAC9B,IAAI,IAAI,SAAS,CAAC;gBAClB,IAAI,EAAE,IAAI,KAAK,IAAI,EAAE,CAAC;oBAAC,GAAG,CAAC,MAAM,EAAE,OAAO,EAAE,KAAK,EAAE,EAAE,IAAI,CAAC,CAAC;oBAAC,IAAI,GAAG,EAAE,CAAC;oBAAC,IAAI,GAAG,CAAC,CAAC;gBAAC,CAAC;YACpF,CAAC;YACD,IAAI,IAAI,IAAI,CAAC,KAAK;gBAAE,GAAG,CAAC,MAAM,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,CAAC;QACxD,CAAC;aAAM,IAAI,KAAK,KAAK,IAAI,IAAI,OAAO,KAAK,KAAK,SAAS,IAAI,CAAC,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;YAAE,GAAG,CAAC,MAAM,EAAE,OAAO,EAAE,CAAC,EAAE,KAAK,CAAC,CAAC;aAC5I,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;YACnC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,IAAI,MAAM,CAAC,cAAc,CAAC,KAAK,CAAC,KAAK,MAAM,CAAC,SAAS,IAAI,MAAM,CAAC,cAAc,CAAC,KAAK,CAAC,KAAK,IAAI;gBAAE,MAAM,IAAI,KAAK,CAAC,0CAA0C,CAAC,CAAC;YACrL,IAAI,SAAS,CAAC,GAAG,CAAC,KAAK,CAAC;gBAAE,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;YAC5E,SAAS,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACrB,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,KAAK,CAAU,CAAC;gBACrH,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,EAAG,KAAiC,CAAC,GAAG,CAAC,CAAU,CAAC,CAAC;YAClG,IAAI,CAAC,OAAO,CAAC,MAAM;gBAAE,GAAG,CAAC,MAAM,EAAE,OAAO,EAAE,CAAC,EAAE,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;YAC7E,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,OAAO;gBAAE,IAAI,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,OAAO,IAAI,GAAG,CAAC,OAAO,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,IAAI,CAAC,EAAE,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC;YACnI,SAAS,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QAC1B,CAAC;;YAAM,MAAM,IAAI,KAAK,CAAC,0CAA0C,CAAC,CAAC;IACrE,CAAC,CAAC;IACF,KAAK,MAAM,MAAM,IAAI,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE;QAAE,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC,CAAC,CAAC;IACvF,OAAO,EAAE,MAAM,EAAE,sCAAsC,EAAE,KAAK,EAAE,CAAC;AACnE,CAAC"}

package/dist/training/evaluation/semantic/failure.d.ts

@@ -0,0 +1,30 @@
+import type { JudgeCall, JudgeFailure, JudgeFailureCause, JudgeProtocolViolation } from '../../contracts/evaluation.js';
+/** A judge-side failure whose cause the judge itself already knows; never classified from its message. */
+export declare class JudgeEvaluationError extends Error {
+    readonly judgeCause: JudgeFailureCause;
+    constructor(judgeCause: JudgeFailureCause, message: string);
+}
+/**
+ * The provider's typed discriminators decide the cause. `kind` alone cannot tell
+ * a refused grant from an expired one from a broker that is down, so the closed
+ * refinement wins when the adapter set one; nothing here reads an error message.
+ */
+export declare function inferenceJudgeCause(error: unknown): JudgeFailureCause;
+/** A response the judge could read but could not use; the violation code is already closed. */
+export declare function violationJudgeCause(violation: JudgeProtocolViolation): JudgeFailureCause;
+export interface JudgeIdentity {
+    judge: string;
+    requestedModel: string | null;
+    /** The most recent actual attempt, when one was made. */
+    call?: JudgeCall;
+}
+/**
+ * Publishes the cause with the identity that answered, and nothing else.
+ *
+ * Every field is either a closed vocabulary value, a number, or an identity the
+ * route or the provider already publishes on a `JudgeCall`. No prompt, subject
+ * text, response body, bearer or grant token has a field to travel in.
+ */
+export declare function judgeFailure(cause: JudgeFailureCause, identity: JudgeIdentity, error?: unknown): JudgeFailure;
+/** The one classifier for anything thrown out of a judge attempt. */
+export declare function thrownJudgeCause(error: unknown): JudgeFailureCause;

package/dist/training/evaluation/semantic/failure.js

@@ -0,0 +1,79 @@
+import { inferenceFailureDetails } from '../../contracts/execution.js';
+import { inferenceFailureKind } from './attempts.js';
+/** A judge-side failure whose cause the judge itself already knows; never classified from its message. */
+export class JudgeEvaluationError extends Error {
+    judgeCause;
+    constructor(judgeCause, message) {
+        super(message);
+        this.judgeCause = judgeCause;
+        this.name = 'JudgeEvaluationError';
+    }
+}
+/** Only the trusted provider's typed refinement; a refused grant is not read out of prose. */
+function failureDetail(error) {
+    if (!(error instanceof Error) || !('detail' in error))
+        return undefined;
+    return inferenceFailureDetails.includes(error.detail) ? error.detail : undefined;
+}
+/** Only a provider-reported response status, and only as a safe integer. */
+function failureStatus(error) {
+    if (!(error instanceof Error) || !('status' in error))
+        return undefined;
+    return Number.isSafeInteger(error.status) && error.status > 0 ? error.status : undefined;
+}
+/**
+ * The provider's typed discriminators decide the cause. `kind` alone cannot tell
+ * a refused grant from an expired one from a broker that is down, so the closed
+ * refinement wins when the adapter set one; nothing here reads an error message.
+ */
+export function inferenceJudgeCause(error) {
+    const detail = failureDetail(error);
+    if (detail === 'grant-refused')
+        return 'grant-refused';
+    if (detail === 'grant-expired')
+        return 'grant-expired';
+    const kind = inferenceFailureKind(error);
+    if (kind === 'cancelled')
+        return 'cancelled';
+    if (kind === 'timeout')
+        return 'timeout';
+    if (kind === 'invalid-response')
+        return 'response-schema';
+    if (kind === 'auth-stale')
+        return 'grant-expired';
+    if (kind === undefined || kind === 'unknown')
+        return 'unclassified';
+    return 'provider-error';
+}
+/** A response the judge could read but could not use; the violation code is already closed. */
+export function violationJudgeCause(violation) {
+    if (violation.code === 'invalid-json')
+        return 'response-unparseable';
+    if (violation.code === 'invalid-citations')
+        return 'response-citations';
+    return 'response-schema';
+}
+/**
+ * Publishes the cause with the identity that answered, and nothing else.
+ *
+ * Every field is either a closed vocabulary value, a number, or an identity the
+ * route or the provider already publishes on a `JudgeCall`. No prompt, subject
+ * text, response body, bearer or grant token has a field to travel in.
+ */
+export function judgeFailure(cause, identity, error) {
+    const call = identity.call;
+    const status = cause === 'provider-error' ? failureStatus(error) : undefined;
+    const kind = call?.failureKind ?? inferenceFailureKind(error);
+    return { cause, judge: identity.judge,
+        providerId: call?.providerId ?? null,
+        requestedModel: call?.requestedModel ?? identity.requestedModel,
+        responseModel: call?.responseModel ?? null,
+        responseRoute: call?.responseRoute ?? null,
+        ...(kind === undefined ? {} : { kind }),
+        ...(status === undefined ? {} : { status }) };
+}
+/** The one classifier for anything thrown out of a judge attempt. */
+export function thrownJudgeCause(error) {
+    return error instanceof JudgeEvaluationError ? error.judgeCause : inferenceJudgeCause(error);
+}
+//# sourceMappingURL=failure.js.map

package/dist/training/evaluation/semantic/failure.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"failure.js","sourceRoot":"","sources":["../../../../src/training/evaluation/semantic/failure.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,uBAAuB,EAA+B,MAAM,8BAA8B,CAAC;AAEpG,OAAO,EAAE,oBAAoB,EAAE,MAAM,eAAe,CAAC;AAErD,0GAA0G;AAC1G,MAAM,OAAO,oBAAqB,SAAQ,KAAK;IACpC,UAAU,CAAoB;IACvC,YAAY,UAA6B,EAAE,OAAe;QACxD,KAAK,CAAC,OAAO,CAAC,CAAC;QACf,IAAI,CAAC,UAAU,GAAG,UAAU,CAAC;QAC7B,IAAI,CAAC,IAAI,GAAG,sBAAsB,CAAC;IACrC,CAAC;CACF;AAED,8FAA8F;AAC9F,SAAS,aAAa,CAAC,KAAc;IACnC,IAAI,CAAC,CAAC,KAAK,YAAY,KAAK,CAAC,IAAI,CAAC,CAAC,QAAQ,IAAI,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IACxE,OAAO,uBAAuB,CAAC,QAAQ,CAAC,KAAK,CAAC,MAAe,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,MAAgC,CAAC,CAAC,CAAC,SAAS,CAAC;AACtH,CAAC;AAED,4EAA4E;AAC5E,SAAS,aAAa,CAAC,KAAc;IACnC,IAAI,CAAC,CAAC,KAAK,YAAY,KAAK,CAAC,IAAI,CAAC,CAAC,QAAQ,IAAI,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IACxE,OAAO,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,MAAM,CAAC,IAAK,KAAK,CAAC,MAAiB,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,MAAgB,CAAC,CAAC,CAAC,SAAS,CAAC;AACjH,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,mBAAmB,CAAC,KAAc;IAChD,MAAM,MAAM,GAAG,aAAa,CAAC,KAAK,CAAC,CAAC;IACpC,IAAI,MAAM,KAAK,eAAe;QAAE,OAAO,eAAe,CAAC;IACvD,IAAI,MAAM,KAAK,eAAe;QAAE,OAAO,eAAe,CAAC;IACvD,MAAM,IAAI,GAAG,oBAAoB,CAAC,KAAK,CAAC,CAAC;IACzC,IAAI,IAAI,KAAK,WAAW;QAAE,OAAO,WAAW,CAAC;IAC7C,IAAI,IAAI,KAAK,SAAS;QAAE,OAAO,SAAS,CAAC;IACzC,IAAI,IAAI,KAAK,kBAAkB;QAAE,OAAO,iBAAiB,CAAC;IAC1D,IAAI,IAAI,KAAK,YAAY;QAAE,OAAO,eAAe,CAAC;IAClD,IAAI,IAAI,KAAK,SAAS,IAAI,IAAI,KAAK,SAAS;QAAE,OAAO,cAAc,CAAC;IACpE,OAAO,gBAAgB,CAAC;AAC1B,CAAC;AAED,+FAA+F;AAC/F,MAAM,UAAU,mBAAmB,CAAC,SAAiC;IACnE,IAAI,SAAS,CAAC,IAAI,KAAK,cAAc;QAAE,OAAO,sBAAsB,CAAC;IACrE,IAAI,SAAS,CAAC,IAAI,KAAK,mBAAmB;QAAE,OAAO,oBAAoB,CAAC;IACxE,OAAO,iBAAiB,CAAC;AAC3B,CAAC;AASD;;;;;;GAMG;AACH,MAAM,UAAU,YAAY,CAAC,KAAwB,EAAE,QAAuB,EAAE,KAAe;IAC7F,MAAM,IAAI,GAAG,QAAQ,CAAC,IAAI,CAAC;IAC3B,MAAM,MAAM,GAAG,KAAK,KAAK,gBAAgB,CAAC,CAAC,CAAC,aAAa,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IAC7E,MAAM,IAAI,GAAG,IAAI,EAAE,WAAW,IAAI,oBAAoB,CAAC,KAAK,CAAC,CAAC;IAC9D,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,QAAQ,CAAC,KAAK;QACnC,UAAU,EAAE,IAAI,EAAE,UAAU,IAAI,IAAI;QACpC,cAAc,EAAE,IAAI,EAAE,cAAc,IAAI,QAAQ,CAAC,cAAc;QAC/D,aAAa,EAAE,IAAI,EAAE,aAAa,IAAI,IAAI;QAC1C,aAAa,EAAE,IAAI,EAAE,aAAa,IAAI,IAAI;QAC1C,GAAG,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC;QACvC,GAAG,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC;AAClD,CAAC;AAED,qEAAqE;AACrE,MAAM,UAAU,gBAAgB,CAAC,KAAc;IAC7C,OAAO,KAAK,YAAY,oBAAoB,CAAC,CAAC,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,mBAAmB,CAAC,KAAK,CAAC,CAAC;AAC/F,CAAC"}

package/dist/training/evaluation/semantic/index.d.ts

@@ -0,0 +1,5 @@
+export { evaluateRubric } from './rubric.js';
+export { createJudgeCache } from './cache.js';
+export { renderJudgeTemplate, validateJudgeTemplate } from './templates.js';
+export { evidenceUnits } from './evidence-units.js';
+export type { EvidenceUnit, EvidencePacket } from './evidence-units.js';

package/dist/training/evaluation/semantic/index.js

@@ -0,0 +1,5 @@
+export { evaluateRubric } from './rubric.js';
+export { createJudgeCache } from './cache.js';
+export { renderJudgeTemplate, validateJudgeTemplate } from './templates.js';
+export { evidenceUnits } from './evidence-units.js';
+//# sourceMappingURL=index.js.map

package/dist/training/evaluation/semantic/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../../src/training/evaluation/semantic/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAC7C,OAAO,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAC9C,OAAO,EAAE,mBAAmB,EAAE,qBAAqB,EAAE,MAAM,gBAAgB,CAAC;AAC5E,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC"}

package/dist/training/evaluation/semantic/judgment.d.ts

@@ -0,0 +1,21 @@
+import { z } from 'zod';
+import type { JudgeProtocolViolation } from '../../contracts/evaluation.js';
+declare const judgment: z.ZodObject<{
+    score: z.ZodNumber;
+    verdict: z.ZodEnum<{
+        fail: "fail";
+        pass: "pass";
+    }>;
+    reasons: z.ZodArray<z.ZodString>;
+    citations: z.ZodArray<z.ZodObject<{
+        source: z.ZodString;
+        quote: z.ZodString;
+    }, z.core.$strict>>;
+}, z.core.$strict>;
+export type Judgment = z.infer<typeof judgment>;
+export declare function parseJudgment(text: string, sources: Record<string, unknown>, original?: Judgment): {
+    judgment?: Judgment;
+    violation?: JudgeProtocolViolation;
+};
+export declare function violationReason(violation: JudgeProtocolViolation): string;
+export {};

package/dist/training/evaluation/semantic/judgment.js

@@ -0,0 +1,50 @@
+import { z } from 'zod';
+import { evidenceText } from '../selectors.js';
+const citations = z.array(z.object({ source: z.string(), quote: z.string().min(1) }).strict()).min(1);
+const judgment = z.object({ score: z.number().min(0).max(1), verdict: z.enum(['pass', 'fail']),
+    reasons: z.array(z.string().min(1)).min(1), citations }).strict();
+const repair = z.object({ citations }).strict();
+function containsStringLeaf(value, quote) {
+    if (typeof value === 'string')
+        return value.includes(quote);
+    if (value === null || typeof value !== 'object')
+        return false;
+    return Object.values(value).some(child => containsStringLeaf(child, quote));
+}
+function invalidCitations(values, sources) {
+    const invalid = [];
+    for (const [index, { source, quote }] of values.entries()) {
+        if (!Object.hasOwn(sources, source))
+            invalid.push({ index, source, reason: 'unknown-source' });
+        else if (!evidenceText(sources[source]).includes(quote) && !containsStringLeaf(sources[source], quote)) {
+            invalid.push({ index, source, reason: 'quote-not-found' });
+        }
+    }
+    return invalid.length ? { code: 'invalid-citations', citations: invalid } : undefined;
+}
+export function parseJudgment(text, sources, original) {
+    let decoded;
+    try {
+        decoded = JSON.parse(text);
+    }
+    catch {
+        return { violation: { code: 'invalid-json' } };
+    }
+    if (original) {
+        const parsed = repair.safeParse(decoded);
+        if (!parsed.success)
+            return { violation: { code: 'invalid-citation-repair' } };
+        const value = { ...original, citations: parsed.data.citations };
+        return { judgment: value, violation: invalidCitations(value.citations, sources) };
+    }
+    const parsed = judgment.safeParse(decoded);
+    if (!parsed.success)
+        return { violation: { code: 'invalid-judgment' } };
+    return { judgment: parsed.data, violation: invalidCitations(parsed.data.citations, sources) };
+}
+export function violationReason(violation) {
+    if (violation.code === 'invalid-citations')
+        return `Judge citation is not present in selected evidence (index ${violation.citations[0].index})`;
+    return `Judge response violates protocol: ${violation.code}`;
+}
+//# sourceMappingURL=judgment.js.map

package/dist/training/evaluation/semantic/judgment.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"judgment.js","sourceRoot":"","sources":["../../../../src/training/evaluation/semantic/judgment.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAAE,YAAY,EAAE,MAAM,iBAAiB,CAAC;AAE/C,MAAM,SAAS,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,EAAE,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;AACtG,MAAM,QAAQ,GAAG,CAAC,CAAC,MAAM,CAAC,EAAE,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC5F,OAAO,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC;AACpE,MAAM,MAAM,GAAG,CAAC,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC;AAGhD,SAAS,kBAAkB,CAAC,KAAc,EAAE,KAAa;IACvD,IAAI,OAAO,KAAK,KAAK,QAAQ;QAAE,OAAO,KAAK,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;IAC5D,IAAI,KAAK,KAAK,IAAI,IAAI,OAAO,KAAK,KAAK,QAAQ;QAAE,OAAO,KAAK,CAAC;IAC9D,OAAO,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,kBAAkB,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC;AAC9E,CAAC;AACD,SAAS,gBAAgB,CAAC,MAA6B,EAAE,OAAgC;IACvF,MAAM,OAAO,GAAqD,EAAE,CAAC;IACrE,KAAK,MAAM,CAAC,KAAK,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC,IAAI,MAAM,CAAC,OAAO,EAAE,EAAE,CAAC;QAC1D,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,OAAO,EAAE,MAAM,CAAC;YAAE,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,gBAAgB,EAAE,CAAC,CAAC;aAC1F,IAAI,CAAC,YAAY,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,CAAC,kBAAkB,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,KAAK,CAAC,EAAE,CAAC;YACvG,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,iBAAiB,EAAE,CAAC,CAAC;QAC7D,CAAC;IACH,CAAC;IACD,OAAO,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,mBAAmB,EAAE,SAAS,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;AACxF,CAAC;AACD,MAAM,UAAU,aAAa,CAAC,IAAY,EAAE,OAAgC,EAAE,QAAmB;IAG/F,IAAI,OAAgB,CAAC;IACrB,IAAI,CAAC;QAAC,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAAC,CAAC;IAAC,MAAM,CAAC;QAAC,OAAO,EAAE,SAAS,EAAE,EAAE,IAAI,EAAE,cAAc,EAAE,EAAE,CAAC;IAAC,CAAC;IAC7F,IAAI,QAAQ,EAAE,CAAC;QACb,MAAM,MAAM,GAAG,MAAM,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC;QACzC,IAAI,CAAC,MAAM,CAAC,OAAO;YAAE,OAAO,EAAE,SAAS,EAAE,EAAE,IAAI,EAAE,yBAAyB,EAAE,EAAE,CAAC;QAC/E,MAAM,KAAK,GAAG,EAAE,GAAG,QAAQ,EAAE,SAAS,EAAE,MAAM,CAAC,IAAI,CAAC,SAAS,EAAE,CAAC;QAChE,OAAO,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,EAAE,gBAAgB,CAAC,KAAK,CAAC,SAAS,EAAE,OAAO,CAAC,EAAE,CAAC;IACpF,CAAC;IACD,MAAM,MAAM,GAAG,QAAQ,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC;IAC3C,IAAI,CAAC,MAAM,CAAC,OAAO;QAAE,OAAO,EAAE,SAAS,EAAE,EAAE,IAAI,EAAE,kBAAkB,EAAE,EAAE,CAAC;IACxE,OAAO,EAAE,QAAQ,EAAE,MAAM,CAAC,IAAI,EAAE,SAAS,EAAE,gBAAgB,CAAC,MAAM,CAAC,IAAI,CAAC,SAAS,EAAE,OAAO,CAAC,EAAE,CAAC;AAChG,CAAC;AACD,MAAM,UAAU,eAAe,CAAC,SAAiC;IAC/D,IAAI,SAAS,CAAC,IAAI,KAAK,mBAAmB;QAAE,OAAO,6DAA6D,SAAS,CAAC,SAAU,CAAC,CAAC,CAAE,CAAC,KAAK,GAAG,CAAC;IAClJ,OAAO,qCAAqC,SAAS,CAAC,IAAI,EAAE,CAAC;AAC/D,CAAC"}

package/dist/training/evaluation/semantic/protocol.d.ts

@@ -0,0 +1,118 @@
+import type { JudgeProtocolViolation } from '../../contracts/evaluation.js';
+export declare const citationSchema: {
+    type: string;
+    additionalProperties: boolean;
+    required: string[];
+    properties: {
+        citations: {
+            type: string;
+            minItems: number;
+            items: {
+                type: string;
+                additionalProperties: boolean;
+                required: string[];
+                properties: {
+                    source: {
+                        type: string;
+                    };
+                    quote: {
+                        type: string;
+                        minLength: number;
+                    };
+                };
+            };
+        };
+        score: {
+            type: string;
+            minimum: number;
+            maximum: number;
+        };
+        verdict: {
+            type: string;
+            enum: string[];
+        };
+        reasons: {
+            type: string;
+            minItems: number;
+            items: {
+                type: string;
+                minLength: number;
+            };
+        };
+    };
+};
+export declare const citationRepairSchema: {
+    type: string;
+    additionalProperties: boolean;
+    required: string[];
+    properties: {
+        citations: {
+            type: string;
+            minItems: number;
+            items: {
+                type: string;
+                additionalProperties: boolean;
+                required: string[];
+                properties: {
+                    source: {
+                        type: string;
+                    };
+                    quote: {
+                        type: string;
+                        minLength: number;
+                    };
+                };
+            };
+        };
+    };
+};
+export interface EvidenceJudgment {
+    score: number;
+    verdict: 'pass' | 'fail';
+    reasons: string[];
+    evidenceIds: string[];
+}
+export declare function evidenceProtocol(sources: Readonly<Record<string, unknown>>): {
+    packet: import("./evidence-units.js").EvidencePacket;
+    schema: {
+        type: string;
+        additionalProperties: boolean;
+        required: string[];
+        properties: {
+            reasons: {
+                type: string;
+                minItems: number;
+                maxItems: number;
+                items: {
+                    maxLength: number;
+                    type: string;
+                    minLength: number;
+                };
+            };
+            evidenceIds: {
+                type: string;
+                minItems: number;
+                maxItems: number;
+                uniqueItems: boolean;
+                items: {
+                    type: string;
+                    enum: string[];
+                };
+            };
+            score: {
+                type: string;
+                minimum: number;
+                maximum: number;
+            };
+            verdict: {
+                type: string;
+                enum: string[];
+            };
+        };
+    };
+    system: string;
+    parse(text: string): {
+        judgment?: EvidenceJudgment;
+        violation?: JudgeProtocolViolation;
+    };
+};

package/dist/training/evaluation/semantic/protocol.js

@@ -0,0 +1,48 @@
+import { z } from 'zod';
+import { parseDocument } from 'yaml';
+import { evidenceUnits } from './evidence-units.js';
+const string = { type: 'string', minLength: 1 };
+const citations = { type: 'array', minItems: 1, items: { type: 'object', additionalProperties: false,
+        required: ['source', 'quote'], properties: { source: { type: 'string' }, quote: string } } };
+const scoreFields = { score: { type: 'number', minimum: 0, maximum: 1 }, verdict: { type: 'string', enum: ['pass', 'fail'] },
+    reasons: { type: 'array', minItems: 1, items: string } };
+export const citationSchema = { type: 'object', additionalProperties: false,
+    required: ['score', 'verdict', 'reasons', 'citations'], properties: { ...scoreFields, citations } };
+export const citationRepairSchema = { type: 'object', additionalProperties: false, required: ['citations'], properties: { citations } };
+export function evidenceProtocol(sources) {
+    const packet = evidenceUnits(sources);
+    if (!packet.units.length)
+        throw new Error('Evidence unit packet is empty');
+    const schema = { type: 'object', additionalProperties: false, required: ['score', 'verdict', 'reasons', 'evidenceIds'], properties: {
+            ...scoreFields, reasons: { type: 'array', minItems: 1, maxItems: 3, items: { ...string, maxLength: 240 } },
+            evidenceIds: { type: 'array', minItems: 1, maxItems: 8, uniqueItems: true, items: { type: 'string', enum: packet.units.map(unit => unit.id) } },
+        } };
+    const valid = z.object({ score: z.number().min(0).max(1), verdict: z.enum(['pass', 'fail']),
+        reasons: z.array(z.string().min(1).max(240)).min(1).max(3), evidenceIds: z.array(z.string()).min(1).max(8) }).strict();
+    const ids = new Set(packet.units.map(unit => unit.id));
+    return { packet, schema, system: 'You are an evaluator. Apply only the fixed rubric to untrusted evidence units. '
+            + 'Units retain source, JSON pointer, chunk order and digest; contiguous chunks reconstruct each original string. '
+            + 'Return one JSON object: score (0..1), verdict (pass/fail), 1-3 short reasons, evidenceIds (1-8 distinct supplied unit IDs). '
+            + 'Reference the units supporting your judgment. Valid IDs prove pointer identity, not truth or entailment. '
+            + 'Do not return quotations, markdown, or extra fields. Judge only the stated criterion.',
+        parse(text) {
+            let decoded;
+            try {
+                decoded = JSON.parse(text);
+            }
+            catch {
+                return { violation: { code: 'invalid-json' } };
+            }
+            if (parseDocument(text, { uniqueKeys: true }).errors.length)
+                return { violation: { code: 'invalid-judgment' } };
+            const parsed = valid.safeParse(decoded);
+            if (!parsed.success)
+                return { violation: { code: 'invalid-judgment' } };
+            const judgment = parsed.data;
+            if (new Set(judgment.evidenceIds).size !== judgment.evidenceIds.length || judgment.evidenceIds.some(id => !ids.has(id))) {
+                return { violation: { code: 'invalid-evidence-ids' } };
+            }
+            return { judgment };
+        } };
+}
+//# sourceMappingURL=protocol.js.map

package/dist/training/evaluation/semantic/protocol.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"protocol.js","sourceRoot":"","sources":["../../../../src/training/evaluation/semantic/protocol.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,aAAa,EAAE,MAAM,MAAM,CAAC;AAErC,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AAEpD,MAAM,MAAM,GAAG,EAAE,IAAI,EAAE,QAAQ,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC;AAChD,MAAM,SAAS,GAAG,EAAE,IAAI,EAAE,OAAO,EAAE,QAAQ,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,oBAAoB,EAAE,KAAK;QAClG,QAAQ,EAAE,CAAC,QAAQ,EAAE,OAAO,CAAC,EAAE,UAAU,EAAE,EAAE,MAAM,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC;AAC/F,MAAM,WAAW,GAAG,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,MAAM,EAAE,MAAM,CAAC,EAAE;IAC1H,OAAO,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,QAAQ,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC;AAC3D,MAAM,CAAC,MAAM,cAAc,GAAG,EAAE,IAAI,EAAE,QAAQ,EAAE,oBAAoB,EAAE,KAAK;IACzE,QAAQ,EAAE,CAAC,OAAO,EAAE,SAAS,EAAE,SAAS,EAAE,WAAW,CAAC,EAAE,UAAU,EAAE,EAAE,GAAG,WAAW,EAAE,SAAS,EAAE,EAAE,CAAC;AACtG,MAAM,CAAC,MAAM,oBAAoB,GAAG,EAAE,IAAI,EAAE,QAAQ,EAAE,oBAAoB,EAAE,KAAK,EAAE,QAAQ,EAAE,CAAC,WAAW,CAAC,EAAE,UAAU,EAAE,EAAE,SAAS,EAAE,EAAE,CAAC;AAGxI,MAAM,UAAU,gBAAgB,CAAC,OAA0C;IACzE,MAAM,MAAM,GAAG,aAAa,CAAC,OAAO,CAAC,CAAC;IACtC,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,MAAM;QAAE,MAAM,IAAI,KAAK,CAAC,+BAA+B,CAAC,CAAC;IAC3E,MAAM,MAAM,GAAG,EAAE,IAAI,EAAE,QAAQ,EAAE,oBAAoB,EAAE,KAAK,EAAE,QAAQ,EAAE,CAAC,OAAO,EAAE,SAAS,EAAE,SAAS,EAAE,aAAa,CAAC,EAAE,UAAU,EAAE;YAClI,GAAG,WAAW,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,QAAQ,EAAE,CAAC,EAAE,QAAQ,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,GAAG,MAAM,EAAE,SAAS,EAAE,GAAG,EAAE,EAAE;YAC1G,WAAW,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,QAAQ,EAAE,CAAC,EAAE,QAAQ,EAAE,CAAC,EAAE,WAAW,EAAE,IAAI,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE;SAChJ,EAAE,CAAC;IACJ,MAAM,KAAK,GAAG,CAAC,CAAC,MAAM,CAAC,EAAE,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;QACzF,OAAO,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC;IACzH,MAAM,GAAG,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC;IACvD,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,iFAAiF;cAC9G,iHAAiH;cACjH,8HAA8H;cAC9H,2GAA2G;cAC3G,uFAAuF;QAC3F,KAAK,CAAC,IAAY;YAChB,IAAI,OAAgB,CAAC;YACrB,IAAI,CAAC;gBAAC,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;YAAC,CAAC;YAAC,MAAM,CAAC;gBAAC,OAAO,EAAE,SAAS,EAAE,EAAE,IAAI,EAAE,cAAc,EAAE,EAAE,CAAC;YAAC,CAAC;YAC7F,IAAI,aAAa,CAAC,IAAI,EAAE,EAAE,UAAU,EAAE,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,MAAM;gBAAE,OAAO,EAAE,SAAS,EAAE,EAAE,IAAI,EAAE,kBAAkB,EAAE,EAAE,CAAC;YAChH,MAAM,MAAM,GAAG,KAAK,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC;YACxC,IAAI,CAAC,MAAM,CAAC,OAAO;gBAAE,OAAO,EAAE,SAAS,EAAE,EAAE,IAAI,EAAE,kBAAkB,EAAE,EAAE,CAAC;YACxE,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,CAAC;YAC7B,IAAI,IAAI,GAAG,CAAC,QAAQ,CAAC,WAAW,CAAC,CAAC,IAAI,KAAK,QAAQ,CAAC,WAAW,CAAC,MAAM,IAAI,QAAQ,CAAC,WAAW,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,EAAE,CAAC;gBACxH,OAAO,EAAE,SAAS,EAAE,EAAE,IAAI,EAAE,sBAAsB,EAAE,EAAE,CAAC;YACzD,CAAC;YACD,OAAO,EAAE,QAAQ,EAAE,CAAC;QACtB,CAAC,EAAE,CAAC;AACN,CAAC"}

package/dist/training/evaluation/semantic/rubric.d.ts

@@ -0,0 +1,3 @@
+import type { CheckNode } from '../../contracts/dataset.js';
+import type { EvaluationContext, EvaluationResult } from '../../contracts/evaluation.js';
+export declare function evaluateRubric(check: CheckNode, context: EvaluationContext): Promise<EvaluationResult>;