spawnfile 0.1.24 → 0.1.27

package/dist/training/experiments/training/held-out.d.ts

@@ -0,0 +1,24 @@
+import type { CandidateManifest } from '../../candidates/index.js';
+import type { TrialReport } from '../trials/index.js';
+export interface HeldOutScore {
+    candidateId: string;
+    meanScore: number;
+    trialIds: string[];
+    evaluationIds?: string[];
+    /** Aligned with trialIds/evaluationIds; present only for repeated tests to preserve single-pass report bytes. */
+    trialScores?: number[];
+    /** Sample standard deviation / sqrt(n); null when n < 2. Descriptive only. */
+    standardError?: number | null;
+}
+/** Run only frozen candidates, bounded by provisioned concurrency; drain on failure. */
+export declare function evaluateHeldOut(options: {
+    candidates: readonly CandidateManifest[];
+    caseIds: readonly string[];
+    concurrency: number;
+    testRepeats?: number;
+    trial(candidate: CandidateManifest, caseId: string, repeat: number): Promise<TrialReport>;
+    measurement(report: TrialReport): {
+        id: string;
+        score: number;
+    };
+}): Promise<HeldOutScore[]>;

package/dist/training/experiments/training/held-out.js

@@ -0,0 +1,41 @@
+import { testRepeatCount } from '../../contracts/training-config.js';
+/** Run only frozen candidates, bounded by provisioned concurrency; drain on failure. */
+export async function evaluateHeldOut(options) {
+    const repeats = testRepeatCount(options.testRepeats);
+    const jobs = options.candidates.flatMap(candidate => options.caseIds.flatMap(caseId => Array.from({ length: repeats }, (_, repeat) => ({ candidate, caseId, repeat }))));
+    const reports = new Array(jobs.length);
+    const measurements = new Array(jobs.length);
+    let next = 0, failed = false;
+    let failure;
+    const worker = async () => {
+        while (!failed && next < jobs.length) {
+            const index = next++;
+            const job = jobs[index];
+            try {
+                reports[index] = await options.trial(job.candidate, job.caseId, job.repeat);
+                measurements[index] = options.measurement(reports[index]);
+            }
+            catch (error) {
+                if (!failed)
+                    failure = error;
+                failed = true;
+            }
+        }
+    };
+    await Promise.all(Array.from({ length: Math.min(options.concurrency, jobs.length) }, worker));
+    if (failed)
+        throw failure;
+    const batchSize = options.caseIds.length * repeats;
+    return options.candidates.map((candidate, index) => {
+        const batch = reports.slice(index * batchSize, (index + 1) * batchSize);
+        const measured = measurements.slice(index * batchSize, (index + 1) * batchSize);
+        const trialScores = measured.map(item => item.score);
+        const meanScore = trialScores.reduce((sum, score) => sum + score, 0) / batch.length;
+        const standardError = batch.length < 2 ? null
+            : Math.sqrt(trialScores.reduce((sum, score) => sum + (score - meanScore) ** 2, 0) / (batch.length - 1) / batch.length);
+        return { candidateId: candidate.id, meanScore,
+            trialIds: batch.map(report => report.id), evaluationIds: measured.map(item => item.id),
+            ...(repeats === 1 ? {} : { trialScores, standardError }) };
+    });
+}
+//# sourceMappingURL=held-out.js.map

package/dist/training/experiments/training/held-out.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"held-out.js","sourceRoot":"","sources":["../../../../src/training/experiments/training/held-out.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,oCAAoC,CAAC;AAerE,wFAAwF;AACxF,MAAM,CAAC,KAAK,UAAU,eAAe,CAAC,OAOrC;IACC,MAAM,OAAO,GAAG,eAAe,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC;IACrD,MAAM,IAAI,GAAG,OAAO,CAAC,UAAU,CAAC,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CACpF,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE,CAAC,CAAC,EAAE,MAAM,EAAE,EAAE,CAAC,CAAC,EAAE,SAAS,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IACpF,MAAM,OAAO,GAAkB,IAAI,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IACtD,MAAM,YAAY,GAAoC,IAAI,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAC7E,IAAI,IAAI,GAAG,CAAC,EAAE,MAAM,GAAG,KAAK,CAAC;IAC7B,IAAI,OAAgB,CAAC;IACrB,MAAM,MAAM,GAAG,KAAK,IAAI,EAAE;QACxB,OAAO,CAAC,MAAM,IAAI,IAAI,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC;YACrC,MAAM,KAAK,GAAG,IAAI,EAAE,CAAC;YACrB,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAE,CAAC;YACzB,IAAI,CAAC;gBACH,OAAO,CAAC,KAAK,CAAC,GAAG,MAAM,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,SAAS,EAAE,GAAG,CAAC,MAAM,EAAE,GAAG,CAAC,MAAM,CAAC,CAAC;gBAC5E,YAAY,CAAC,KAAK,CAAC,GAAG,OAAO,CAAC,WAAW,CAAC,OAAO,CAAC,KAAK,CAAE,CAAC,CAAC;YAC7D,CAAC;YACD,OAAO,KAAK,EAAE,CAAC;gBAAC,IAAI,CAAC,MAAM;oBAAE,OAAO,GAAG,KAAK,CAAC;gBAAC,MAAM,GAAG,IAAI,CAAC;YAAC,CAAC;QAChE,CAAC;IACH,CAAC,CAAC;IACF,MAAM,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,WAAW,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,MAAM,CAAC,CAAC,CAAC;IAC9F,IAAI,MAAM;QAAE,MAAM,OAAO,CAAC;IAC1B,MAAM,SAAS,GAAG,OAAO,CAAC,OAAO,CAAC,MAAM,GAAG,OAAO,CAAC;IACnD,OAAO,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,KAAK,EAAE,EAAE;QACjD,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,GAAG,SAAS,EAAE,CAAC,KAAK,GAAG,CAAC,CAAC,GAAG,SAAS,CAAC,CAAC;QACxE,MAAM,QAAQ,GAAG,YAAY,CAAC,KAAK,CAAC,KAAK,GAAG,SAAS,EAAE,CAAC,KAAK,GAAG,CAAC,CAAC,GAAG,SAAS,CAAC,CAAC;QAChF,MAAM,WAAW,GAAG,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACrD,MAAM,SAAS,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,EAAE,CAAC,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC;QACpF,MAAM,aAAa,GAAG,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI;YAC3C,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,KAAK,GAAG,SAAS,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC;QACzH,OAAO,EAAE,WAAW,EAAE,SAAS,CAAC,EAAE,EAAE,SAAS;YAC3C,QAAQ,EAAE,KAAK,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,EAAE,CAAC,EAAE,aAAa,EAAE,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC;YACtF,GAAG,CAAC,OAAO,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,aAAa,EAAE,CAAC,EAAE,CAAC;IAC/D,CAAC,CAAC,CAAC;AACL,CAAC"}

package/dist/training/experiments/training/index.d.ts

@@ -0,0 +1,7 @@
+export { trainAgent, type TrainingOptions, type TrainingReport } from './training.js';
+export { partitionTraining, resolveTrainingPartitions } from './partition.js';
+export { trialScore } from './scoring.js';
+export { freezeCriteria } from './criteria.js';
+export { TrainingCheckpoint, trainingDigest } from './checkpoint.js';
+export { recoverMeasurement, recoverStoredReport } from './recovery.js';
+export type { HeldOutScore } from './held-out.js';

package/dist/training/experiments/training/index.js

@@ -0,0 +1,7 @@
+export { trainAgent } from './training.js';
+export { partitionTraining, resolveTrainingPartitions } from './partition.js';
+export { trialScore } from './scoring.js';
+export { freezeCriteria } from './criteria.js';
+export { TrainingCheckpoint, trainingDigest } from './checkpoint.js';
+export { recoverMeasurement, recoverStoredReport } from './recovery.js';
+//# sourceMappingURL=index.js.map

package/dist/training/experiments/training/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../../src/training/experiments/training/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAA6C,MAAM,eAAe,CAAC;AACtF,OAAO,EAAE,iBAAiB,EAAE,yBAAyB,EAAE,MAAM,gBAAgB,CAAC;AAC9E,OAAO,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAC1C,OAAO,EAAE,cAAc,EAAE,MAAM,eAAe,CAAC;AAC/C,OAAO,EAAE,kBAAkB,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AACrE,OAAO,EAAE,kBAAkB,EAAE,mBAAmB,EAAE,MAAM,eAAe,CAAC"}

package/dist/training/experiments/training/partition.d.ts

@@ -0,0 +1,20 @@
+/** One train YAML can provide an explicitly frozen internal validation partition. */
+export declare function partitionTraining<T extends {
+    id: string;
+    records: Array<{
+        group?: string;
+    }>;
+}>(dataset: T, selectedGroups?: readonly string[]): {
+    train: T;
+    validation: T;
+};
+/** Explicit authored validation retains its own dataset identity and records. */
+export declare function resolveTrainingPartitions<T extends {
+    id: string;
+    records: Array<{
+        group?: string;
+    }>;
+}>(train: T, validation?: T, groups?: readonly string[]): {
+    train: T;
+    validation: T;
+};

package/dist/training/experiments/training/partition.js

@@ -0,0 +1,24 @@
+/** One train YAML can provide an explicitly frozen internal validation partition. */
+export function partitionTraining(dataset, selectedGroups) {
+    const groups = [...new Set(dataset.records.map(record => record.group))];
+    if (groups.some(group => !group) || groups.length < 2)
+        throw new Error('Training requires at least two nonempty evidence groups');
+    const sorted = groups.sort();
+    const holdout = selectedGroups ?? sorted.slice(-Math.max(1, Math.floor(sorted.length * 0.2)));
+    if (!holdout.length || new Set(holdout).size !== holdout.length || holdout.some(group => !sorted.includes(group))) {
+        throw new Error('Validation groups must be unique existing training groups');
+    }
+    const train = dataset.records.filter(record => !holdout.includes(record.group));
+    const validation = dataset.records.filter(record => holdout.includes(record.group));
+    if (!train.length)
+        throw new Error('Validation cannot consume every training group');
+    return { train: { ...dataset, id: `${dataset.id}/train`, records: train },
+        validation: { ...dataset, id: `${dataset.id}/validation`, records: validation } };
+}
+/** Explicit authored validation retains its own dataset identity and records. */
+export function resolveTrainingPartitions(train, validation, groups) {
+    if (validation && groups)
+        throw new Error('Explicit validation and validation groups are mutually exclusive');
+    return validation ? { train, validation } : partitionTraining(train, groups);
+}
+//# sourceMappingURL=partition.js.map

package/dist/training/experiments/training/partition.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"partition.js","sourceRoot":"","sources":["../../../../src/training/experiments/training/partition.ts"],"names":[],"mappings":"AACA,qFAAqF;AACrF,MAAM,UAAU,iBAAiB,CAA+D,OAAU,EAAE,cAAkC;IAG5I,MAAM,MAAM,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;IACzE,IAAI,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,yDAAyD,CAAC,CAAC;IAClI,MAAM,MAAM,GAAI,MAAmB,CAAC,IAAI,EAAE,CAAC;IAC3C,MAAM,OAAO,GAAG,cAAc,IAAI,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC;IAC9F,IAAI,CAAC,OAAO,CAAC,MAAM,IAAI,IAAI,GAAG,CAAC,OAAO,CAAC,CAAC,IAAI,KAAK,OAAO,CAAC,MAAM,IAAI,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC;QAClH,MAAM,IAAI,KAAK,CAAC,2DAA2D,CAAC,CAAC;IAC/E,CAAC;IACD,MAAM,KAAK,GAAG,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAM,CAAC,CAAC,CAAC;IACjF,MAAM,UAAU,GAAG,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAM,CAAC,CAAC,CAAC;IACrF,IAAI,CAAC,KAAK,CAAC,MAAM;QAAE,MAAM,IAAI,KAAK,CAAC,gDAAgD,CAAC,CAAC;IACrF,OAAO,EAAE,KAAK,EAAE,EAAE,GAAG,OAAO,EAAE,EAAE,EAAE,GAAG,OAAO,CAAC,EAAE,QAAQ,EAAE,OAAO,EAAE,KAAK,EAAE;QACvE,UAAU,EAAE,EAAE,GAAG,OAAO,EAAE,EAAE,EAAE,GAAG,OAAO,CAAC,EAAE,aAAa,EAAE,OAAO,EAAE,UAAU,EAAE,EAAE,CAAC;AACtF,CAAC;AAED,iFAAiF;AACjF,MAAM,UAAU,yBAAyB,CACvC,KAAQ,EAAE,UAAc,EAAE,MAA0B;IAEpD,IAAI,UAAU,IAAI,MAAM;QAAE,MAAM,IAAI,KAAK,CAAC,kEAAkE,CAAC,CAAC;IAC9G,OAAO,UAAU,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC,CAAC,iBAAiB,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;AAC/E,CAAC"}

package/dist/training/experiments/training/recovery.d.ts

@@ -0,0 +1,17 @@
+import type { ResolvedDataset } from '../../contracts/dataset.js';
+import { type RescoreReport } from '../replay/index.js';
+import type { StoredTrialOptions, TrialReport } from '../trials/index.js';
+import type { TrainingOptions } from './training.js';
+/** Only a completed native capture with failed final measurement is eligible. */
+export declare function recoverMeasurement(input: {
+    options: Pick<TrainingOptions, 'baseline' | 'native'>;
+    dataset: ResolvedDataset;
+    report: TrialReport;
+    previous?: RescoreReport;
+    evaluation: StoredTrialOptions['evaluation'];
+    signal?: AbortSignal;
+    pendingRescoreId?: string;
+    reserve(): Promise<string>;
+}): Promise<RescoreReport | undefined>;
+/** Restore only an integrity-checked terminal report; pending work is never presumed stopped. */
+export declare function recoverStoredReport(options: Pick<TrainingOptions, 'baseline' | 'native'>, id: string, phase: TrialReport['phase'], candidateId: string, expected?: TrialReport): Promise<TrialReport>;

package/dist/training/experiments/training/recovery.js

@@ -0,0 +1,54 @@
+import { isDeepStrictEqual } from 'node:util';
+import { capturedTrial, rescoreStoredTrial } from '../replay/index.js';
+const retryable = new Set(['quota', 'transport', 'unavailable', 'timeout', 'cancelled', 'invalid-response']);
+function transientJudgeFailure(check) {
+    if (check.children?.length)
+        return check.children.filter(child => child.status === 'error').every(transientJudgeFailure);
+    const last = check.judgeCalls?.at(-1);
+    return last?.outcome === 'generation-error' && retryable.has(last.failureKind ?? 'unknown');
+}
+/** Only a completed native capture with failed final measurement is eligible. */
+export async function recoverMeasurement(input) {
+    const { report, options } = input;
+    let previous = input.previous;
+    if (input.pendingRescoreId && input.pendingRescoreId !== previous?.id) {
+        const replay = await options.native.store.readReplay(report.id, { audience: 'evaluator',
+            phase: report.phase === 'train' ? 'training' : report.phase, frozenCandidateIds: [options.baseline.id, report.candidateId] });
+        const settled = replay.evaluations.find(entry => entry.id === input.pendingRescoreId)?.report;
+        if (settled)
+            previous = settled;
+    }
+    if (previous?.status === 'pass' || previous?.status === 'fail')
+        return previous;
+    if (report.status === 'pass' || report.status === 'fail')
+        return undefined;
+    if (report.execution !== 'completed' || report.hostWork !== 'settled' || report.stop?.state !== 'acknowledged') {
+        throw new Error(`Cannot resume incomplete native capture ${report.id}; no successful execution or cleanup may be assumed`);
+    }
+    if (report.status !== 'error' || report.evaluation.status !== 'error') {
+        throw new Error(`Cannot resume ${report.status} measurement; missing evidence is not an inference retry`);
+    }
+    const errors = Object.values((previous ?? report).evaluation.checks).filter(check => check.status === 'error');
+    if (!errors.length || !errors.every(transientJudgeFailure))
+        throw new Error('Measurement recovery requires classified transient judge failures');
+    const id = await input.reserve();
+    return rescoreStoredTrial({ store: options.native.store, runId: report.id,
+        access: { audience: 'evaluator', phase: report.phase === 'train' ? 'training' : report.phase,
+            ...(report.phase === 'test' ? { frozenCandidateIds: [options.baseline.id, report.candidateId] } : {}) },
+        id, dataset: input.dataset, registry: options.native.registry,
+        evaluation: input.evaluation, signal: input.signal });
+}
+/** Restore only an integrity-checked terminal report; pending work is never presumed stopped. */
+export async function recoverStoredReport(options, id, phase, candidateId, expected) {
+    const replay = await options.native.store.readReplay(id, { audience: 'evaluator', phase: phase === 'train' ? 'training' : phase,
+        ...(phase === 'test' ? { frozenCandidateIds: [options.baseline.id, candidateId] } : {}) });
+    const { report } = capturedTrial(replay);
+    if (report.candidateId !== candidateId || report.phase !== phase || report.hostWork !== 'settled'
+        || report.stop?.state !== 'acknowledged' || report.stop.quiescence !== 'confirmed') {
+        throw new Error(`Pending native trial ${id} lacks verified terminal execution and shutdown; recovery is blocked`);
+    }
+    if (expected && !isDeepStrictEqual(report, JSON.parse(JSON.stringify(expected))))
+        throw new Error('Cached native report differs from immutable trial evidence');
+    return report;
+}
+//# sourceMappingURL=recovery.js.map

package/dist/training/experiments/training/recovery.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"recovery.js","sourceRoot":"","sources":["../../../../src/training/experiments/training/recovery.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,iBAAiB,EAAE,MAAM,WAAW,CAAC;AAG9C,OAAO,EAAE,aAAa,EAAE,kBAAkB,EAAsB,MAAM,oBAAoB,CAAC;AAI3F,MAAM,SAAS,GAAG,IAAI,GAAG,CAAC,CAAC,OAAO,EAAE,WAAW,EAAE,aAAa,EAAE,SAAS,EAAE,WAAW,EAAE,kBAAkB,CAAC,CAAC,CAAC;AAC7G,SAAS,qBAAqB,CAAC,KAAuB;IACpD,IAAI,KAAK,CAAC,QAAQ,EAAE,MAAM;QAAE,OAAO,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,MAAM,KAAK,OAAO,CAAC,CAAC,KAAK,CAAC,qBAAqB,CAAC,CAAC;IACzH,MAAM,IAAI,GAAG,KAAK,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IACtC,OAAO,IAAI,EAAE,OAAO,KAAK,kBAAkB,IAAI,SAAS,CAAC,GAAG,CAAC,IAAI,CAAC,WAAW,IAAI,SAAS,CAAC,CAAC;AAC9F,CAAC;AAED,iFAAiF;AACjF,MAAM,CAAC,KAAK,UAAU,kBAAkB,CAAC,KASxC;IACC,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,GAAG,KAAK,CAAC;IAClC,IAAI,QAAQ,GAAG,KAAK,CAAC,QAAQ,CAAC;IAC9B,IAAI,KAAK,CAAC,gBAAgB,IAAI,KAAK,CAAC,gBAAgB,KAAK,QAAQ,EAAE,EAAE,EAAE,CAAC;QACtE,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,UAAU,CAAC,MAAM,CAAC,EAAE,EAAE,EAAE,QAAQ,EAAE,WAAW;YACrF,KAAK,EAAE,MAAM,CAAC,KAAK,KAAK,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,EAAE,kBAAkB,EAAE,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC,WAAW,CAAC,EAAE,CAAC,CAAC;QAChI,MAAM,OAAO,GAAG,MAAM,CAAC,WAAW,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,KAAK,CAAC,gBAAgB,CAAC,EAAE,MAAM,CAAC;QAC9F,IAAI,OAAO;YAAE,QAAQ,GAAG,OAAwB,CAAC;IACnD,CAAC;IACD,IAAI,QAAQ,EAAE,MAAM,KAAK,MAAM,IAAI,QAAQ,EAAE,MAAM,KAAK,MAAM;QAAE,OAAO,QAAQ,CAAC;IAChF,IAAI,MAAM,CAAC,MAAM,KAAK,MAAM,IAAI,MAAM,CAAC,MAAM,KAAK,MAAM;QAAE,OAAO,SAAS,CAAC;IAC3E,IAAI,MAAM,CAAC,SAAS,KAAK,WAAW,IAAI,MAAM,CAAC,QAAQ,KAAK,SAAS,IAAI,MAAM,CAAC,IAAI,EAAE,KAAK,KAAK,cAAc,EAAE,CAAC;QAC/G,MAAM,IAAI,KAAK,CAAC,2CAA2C,MAAM,CAAC,EAAE,qDAAqD,CAAC,CAAC;IAC7H,CAAC;IACD,IAAI,MAAM,CAAC,MAAM,KAAK,OAAO,IAAI,MAAM,CAAC,UAAU,CAAC,MAAM,KAAK,OAAO,EAAE,CAAC;QACtE,MAAM,IAAI,KAAK,CAAC,iBAAiB,MAAM,CAAC,MAAM,0DAA0D,CAAC,CAAC;IAC5G,CAAC;IACD,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,QAAQ,IAAI,MAAM,CAAC,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,MAAM,KAAK,OAAO,CAAC,CAAC;IAC/G,IAAI,CAAC,MAAM,CAAC,MAAM,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,qBAAqB,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,mEAAmE,CAAC,CAAC;IACjJ,MAAM,EAAE,GAAG,MAAM,KAAK,CAAC,OAAO,EAAE,CAAC;IACjC,OAAO,kBAAkB,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,MAAM,CAAC,KAAK,EAAE,KAAK,EAAE,MAAM,CAAC,EAAE;QACvE,MAAM,EAAE,EAAE,QAAQ,EAAE,WAAW,EAAE,KAAK,EAAE,MAAM,CAAC,KAAK,KAAK,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK;YAC1F,GAAG,CAAC,MAAM,CAAC,KAAK,KAAK,MAAM,CAAC,CAAC,CAAC,EAAE,kBAAkB,EAAE,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC,WAAW,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE;QACzG,EAAE,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,QAAQ,EAAE,OAAO,CAAC,MAAM,CAAC,QAAQ;QAC7D,UAAU,EAAE,KAAK,CAAC,UAAU,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC;AAC1D,CAAC;AAED,iGAAiG;AACjG,MAAM,CAAC,KAAK,UAAU,mBAAmB,CAAC,OAAqD,EAAE,EAAU,EAAE,KAA2B,EAAE,WAAmB,EAC3J,QAAsB;IACtB,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,UAAU,CAAC,EAAE,EAAE,EAAE,QAAQ,EAAE,WAAW,EAAE,KAAK,EAAE,KAAK,KAAK,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,KAAK;QAC7H,GAAG,CAAC,KAAK,KAAK,MAAM,CAAC,CAAC,CAAC,EAAE,kBAAkB,EAAE,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,EAAE,WAAW,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;IAC7F,MAAM,EAAE,MAAM,EAAE,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC;IACzC,IAAI,MAAM,CAAC,WAAW,KAAK,WAAW,IAAI,MAAM,CAAC,KAAK,KAAK,KAAK,IAAI,MAAM,CAAC,QAAQ,KAAK,SAAS;WAC1F,MAAM,CAAC,IAAI,EAAE,KAAK,KAAK,cAAc,IAAI,MAAM,CAAC,IAAI,CAAC,UAAU,KAAK,WAAW,EAAE,CAAC;QACvF,MAAM,IAAI,KAAK,CAAC,wBAAwB,EAAE,sEAAsE,CAAC,CAAC;IACpH,CAAC;IACD,IAAI,QAAQ,IAAI,CAAC,iBAAiB,CAAC,MAAM,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,4DAA4D,CAAC,CAAC;IAChK,OAAO,MAAM,CAAC;AAChB,CAAC"}

package/dist/training/experiments/training/retry.d.ts

@@ -0,0 +1,14 @@
+import { type TrialReport } from '../trials/index.js';
+export interface TrialRetry {
+    failedTrialId: string;
+    replacementTrialId: string;
+    cause: string;
+    attempt: number;
+}
+/** Sequential replacements stay inside the original callback/held-out worker. */
+export declare function retryTrial(input: {
+    key: string;
+    maximum: number;
+    signal?: AbortSignal;
+    execute(key: string, retry?: Omit<TrialRetry, 'replacementTrialId'>): Promise<TrialReport>;
+}): Promise<TrialReport>;

package/dist/training/experiments/training/retry.js

@@ -0,0 +1,16 @@
+import { hasSubjectBehavior, hasUnadmittedExecution } from '../trials/index.js';
+/** Sequential replacements stay inside the original callback/held-out worker. */
+export async function retryTrial(input) {
+    let report = await input.execute(input.key);
+    for (let attempt = 1; report.execution === 'failed' && report.failureKind === 'native-infrastructure' &&
+        !hasSubjectBehavior(report) && (!report.collectionFailed || hasUnadmittedExecution(report.turns)) && attempt <= input.maximum; attempt++) {
+        input.signal?.throwIfAborted();
+        // An abandoned native process must never overlap its replacement.
+        if (report.hostWork !== 'settled' || report.stop?.state !== 'acknowledged' || report.stop.quiescence !== 'confirmed')
+            break;
+        const retry = { failedTrialId: report.id, cause: report.error ?? 'Native infrastructure failure', attempt };
+        report = await input.execute(`trial-retry:${JSON.stringify([input.key, attempt])}`, retry);
+    }
+    return report;
+}
+//# sourceMappingURL=retry.js.map

package/dist/training/experiments/training/retry.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"retry.js","sourceRoot":"","sources":["../../../../src/training/experiments/training/retry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,kBAAkB,EAAE,sBAAsB,EAAoB,MAAM,oBAAoB,CAAC;AASlG,iFAAiF;AACjF,MAAM,CAAC,KAAK,UAAU,UAAU,CAAC,KAKhC;IACC,IAAI,MAAM,GAAG,MAAM,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IAC5C,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,MAAM,CAAC,SAAS,KAAK,QAAQ,IAAI,MAAM,CAAC,WAAW,KAAK,uBAAuB;QACnG,CAAC,kBAAkB,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,gBAAgB,IAAI,sBAAsB,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,OAAO,IAAI,KAAK,CAAC,OAAO,EAAE,OAAO,EAAE,EAAE,CAAC;QAC3I,KAAK,CAAC,MAAM,EAAE,cAAc,EAAE,CAAC;QAC/B,kEAAkE;QAClE,IAAI,MAAM,CAAC,QAAQ,KAAK,SAAS,IAAI,MAAM,CAAC,IAAI,EAAE,KAAK,KAAK,cAAc,IAAI,MAAM,CAAC,IAAI,CAAC,UAAU,KAAK,WAAW;YAAE,MAAM;QAC5H,MAAM,KAAK,GAAG,EAAE,aAAa,EAAE,MAAM,CAAC,EAAE,EAAE,KAAK,EAAE,MAAM,CAAC,KAAK,IAAI,+BAA+B,EAAE,OAAO,EAAE,CAAC;QAC5G,MAAM,GAAG,MAAM,KAAK,CAAC,OAAO,CAAC,eAAe,IAAI,CAAC,SAAS,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC,EAAE,EAAE,KAAK,CAAC,CAAC;IAC7F,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC"}

package/dist/training/experiments/training/scoring.d.ts

@@ -0,0 +1,4 @@
+import type { TrainingScoring } from '../../contracts/training-config.js';
+import type { TrialReport } from '../trials/index.js';
+/** A missing or interrupted measurement is not a cheap zero-quality observation. */
+export declare function trialScore(report: Pick<TrialReport, 'id' | 'execution' | 'hostWork' | 'stop' | 'decisiveFailure' | 'status' | 'preDeferralStatus' | 'evaluation'>, scoring?: TrainingScoring): number;

package/dist/training/experiments/training/scoring.js

@@ -0,0 +1,42 @@
+function atomicMeasurementFailed(check) {
+    return Boolean(check.atomicFailure) || check.atomicGrading?.status === 'error' || Boolean(check.atomicGrading?.failure) ||
+        Boolean(check.children?.some(atomicMeasurementFailed));
+}
+/** A missing or interrupted measurement is not a cheap zero-quality observation. */
+export function trialScore(report, scoring) {
+    if (report.execution === 'failed')
+        throw new Error(`Trial ${report.id} failed in infrastructure`);
+    if (report.hostWork !== 'settled' || report.stop?.state !== 'acknowledged')
+        throw new Error(`Trial ${report.id} has incomplete infrastructure cleanup`);
+    const decisive = report.decisiveFailure?.status === 'fail' && report.decisiveFailure.finality === 'final' && report.status === 'fail';
+    // A closed execution whose delivery stayed deferred is a behavioural outcome the
+    // optimizer must see, not a broken machine — but only with the quiescence proof, and
+    // never as a pass: an unanswered delivery is not completed training evidence.
+    const deferred = report.execution === 'deferred' && report.stop?.quiescence === 'confirmed';
+    // So is a subject that ran out of a budget it declared itself, under the same proof.
+    const limited = report.execution === 'limit-exceeded' && report.stop?.quiescence === 'confirmed';
+    if (report.execution !== 'completed' && !decisive && !deferred && !limited)
+        throw new Error(`Trial ${report.id} has incomplete native execution`);
+    if (report.status !== 'pass' && report.status !== 'fail')
+        throw new Error(`Trial ${report.id} has unscorable ${report.status} evidence`);
+    if (deferred && report.status !== 'fail')
+        throw new Error(`Trial ${report.id} cannot pass on a deferred delivery`);
+    if (limited && report.status !== 'fail')
+        throw new Error(`Trial ${report.id} cannot pass on an exhausted declared limit`);
+    // Logical gates may mask errors; failed atomic measurements still have no behavioral score.
+    const checks = [...Object.values(report.evaluation.checks ?? {}),
+        ...report.evaluation.stages.flatMap(stage => [...stage.checks, stage.gate])];
+    if (checks.some(atomicMeasurementFailed))
+        throw new Error(`Trial ${report.id} has unscorable atomic grading measurement`);
+    // The trial owner combines every actual gate before adding the deferral witness.
+    // Legacy captures have no such provenance and conservatively retain zero reward.
+    const deferredReward = scoring?.deferred === 'stage-score' && deferred &&
+        report.preDeferralStatus === 'pass' && !decisive;
+    if (report.status === 'fail' && !deferredReward)
+        return 0;
+    const score = report.evaluation.stages.at(-1)?.score;
+    if (score && score.missingWeight > 1e-9)
+        throw new Error('Trial scoring coverage is incomplete');
+    return score?.lower ?? (deferredReward ? 0 : 1);
+}
+//# sourceMappingURL=scoring.js.map

package/dist/training/experiments/training/scoring.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"scoring.js","sourceRoot":"","sources":["../../../../src/training/experiments/training/scoring.ts"],"names":[],"mappings":"AAIA,SAAS,uBAAuB,CAAC,KAAuB;IACtD,OAAO,OAAO,CAAC,KAAK,CAAC,aAAa,CAAC,IAAI,KAAK,CAAC,aAAa,EAAE,MAAM,KAAK,OAAO,IAAI,OAAO,CAAC,KAAK,CAAC,aAAa,EAAE,OAAO,CAAC;QACrH,OAAO,CAAC,KAAK,CAAC,QAAQ,EAAE,IAAI,CAAC,uBAAuB,CAAC,CAAC,CAAC;AAC3D,CAAC;AAED,oFAAoF;AACpF,MAAM,UAAU,UAAU,CACxB,MAAuI,EACvI,OAAyB;IAEzB,IAAI,MAAM,CAAC,SAAS,KAAK,QAAQ;QAAE,MAAM,IAAI,KAAK,CAAC,SAAS,MAAM,CAAC,EAAE,2BAA2B,CAAC,CAAC;IAClG,IAAI,MAAM,CAAC,QAAQ,KAAK,SAAS,IAAI,MAAM,CAAC,IAAI,EAAE,KAAK,KAAK,cAAc;QAAE,MAAM,IAAI,KAAK,CAAC,SAAS,MAAM,CAAC,EAAE,wCAAwC,CAAC,CAAC;IACxJ,MAAM,QAAQ,GAAG,MAAM,CAAC,eAAe,EAAE,MAAM,KAAK,MAAM,IAAI,MAAM,CAAC,eAAe,CAAC,QAAQ,KAAK,OAAO,IAAI,MAAM,CAAC,MAAM,KAAK,MAAM,CAAC;IACtI,iFAAiF;IACjF,qFAAqF;IACrF,8EAA8E;IAC9E,MAAM,QAAQ,GAAG,MAAM,CAAC,SAAS,KAAK,UAAU,IAAI,MAAM,CAAC,IAAI,EAAE,UAAU,KAAK,WAAW,CAAC;IAC5F,qFAAqF;IACrF,MAAM,OAAO,GAAG,MAAM,CAAC,SAAS,KAAK,gBAAgB,IAAI,MAAM,CAAC,IAAI,EAAE,UAAU,KAAK,WAAW,CAAC;IACjG,IAAI,MAAM,CAAC,SAAS,KAAK,WAAW,IAAI,CAAC,QAAQ,IAAI,CAAC,QAAQ,IAAI,CAAC,OAAO;QAAE,MAAM,IAAI,KAAK,CAAC,SAAS,MAAM,CAAC,EAAE,kCAAkC,CAAC,CAAC;IAClJ,IAAI,MAAM,CAAC,MAAM,KAAK,MAAM,IAAI,MAAM,CAAC,MAAM,KAAK,MAAM;QAAE,MAAM,IAAI,KAAK,CAAC,SAAS,MAAM,CAAC,EAAE,mBAAmB,MAAM,CAAC,MAAM,WAAW,CAAC,CAAC;IACzI,IAAI,QAAQ,IAAI,MAAM,CAAC,MAAM,KAAK,MAAM;QAAE,MAAM,IAAI,KAAK,CAAC,SAAS,MAAM,CAAC,EAAE,qCAAqC,CAAC,CAAC;IACnH,IAAI,OAAO,IAAI,MAAM,CAAC,MAAM,KAAK,MAAM;QAAE,MAAM,IAAI,KAAK,CAAC,SAAS,MAAM,CAAC,EAAE,6CAA6C,CAAC,CAAC;IAC1H,4FAA4F;IAC5F,MAAM,MAAM,GAAG,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,MAAM,IAAI,EAAE,CAAC;QAC9D,GAAG,MAAM,CAAC,UAAU,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAC/E,IAAI,MAAM,CAAC,IAAI,CAAC,uBAAuB,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,SAAS,MAAM,CAAC,EAAE,4CAA4C,CAAC,CAAC;IAC1H,iFAAiF;IACjF,iFAAiF;IACjF,MAAM,cAAc,GAAG,OAAO,EAAE,QAAQ,KAAK,aAAa,IAAI,QAAQ;QACpE,MAAM,CAAC,iBAAiB,KAAK,MAAM,IAAI,CAAC,QAAQ,CAAC;IACnD,IAAI,MAAM,CAAC,MAAM,KAAK,MAAM,IAAI,CAAC,cAAc;QAAE,OAAO,CAAC,CAAC;IAC1D,MAAM,KAAK,GAAG,MAAM,CAAC,UAAU,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,KAAK,CAAC;IACrD,IAAI,KAAK,IAAI,KAAK,CAAC,aAAa,GAAG,IAAI;QAAE,MAAM,IAAI,KAAK,CAAC,sCAAsC,CAAC,CAAC;IACjG,OAAO,KAAK,EAAE,KAAK,IAAI,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AAClD,CAAC"}

package/dist/training/experiments/training/training.d.ts

@@ -0,0 +1,85 @@
+import { type TrialRetry } from './retry.js';
+import type { TrainingScoring } from '../../contracts/training-config.js';
+import type { TrialSlot } from '../../contracts/trial-slot.js';
+import type { RescoreReport } from '../replay/index.js';
+import type { CandidateManifest } from '../../candidates/index.js';
+import type { ResolvedDataset } from '../../contracts/dataset.js';
+import type { InferenceProvider } from '../../contracts/execution.js';
+import { type DspyOptimizationOptions, type DspyResult } from '../../optimization/index.js';
+import type { TrialReport } from '../trials/index.js';
+import { type StoredTrialOptions } from '../trials/index.js';
+import { type HeldOutScore } from './held-out.js';
+import type { CostReport, TrainingCostPlan } from '../costs/index.js';
+export interface TrainingOptions {
+    trialRetries?: number;
+    testRepeats?: number;
+    scoring?: TrainingScoring;
+    concurrency?: number;
+    slotsFile?: string;
+    slots?: readonly TrialSlot[];
+    id: string;
+    baseline: CandidateManifest;
+    train: ResolvedDataset;
+    validation?: ResolvedDataset;
+    validationGroups?: readonly string[];
+    test: ResolvedDataset;
+    native: Omit<StoredTrialOptions, 'id' | 'phase' | 'record' | 'candidate' | 'datasetDigest' | 'datasetId' | 'planDigest'>;
+    inference: {
+        provider: InferenceProvider;
+        model: string;
+    };
+    bridge: DspyOptimizationOptions['bridge'];
+    budget: DspyOptimizationOptions['budget'];
+    proposal?: DspyOptimizationOptions['proposal'];
+    search?: DspyOptimizationOptions['search'];
+    checkpoint?: {
+        directory: string;
+        resume: boolean;
+        identity: string;
+        maxRecoveryAttempts?: number;
+        fork?: NonNullable<DspyOptimizationOptions['checkpoint']>['fork'] & {
+            trainingDigest: string;
+            trainingIdentity: string;
+            receiptDigest: string;
+        };
+    };
+    seed?: number;
+    signal?: AbortSignal;
+    onProgress?: (event: {
+        type: string;
+        payload: unknown;
+    }) => Promise<void>;
+}
+export interface TrainingReport {
+    scoring?: TrainingScoring;
+    schema: 'spawnfile.training.training.v1';
+    id: string;
+    baseline: CandidateManifest;
+    selected: CandidateManifest;
+    /** Elapsed host time for this invocation, including restoration and sealed testing. */
+    durationMs?: number;
+    optimizer: DspyResult;
+    partitions: {
+        train: string[];
+        validation: string[];
+        test: string[];
+    };
+    trials: TrialReport[];
+    heldOut: HeldOutScore[];
+    /** Separate measurements over an original native capture, never additional trials. */
+    rescores?: RescoreReport[];
+    /** Failed infrastructure captures and their separately persisted replacements. */
+    retries?: TrialRetry[];
+    /** Optional host reporting, never optimizer feedback or a subscription invoice. */
+    costs?: CostReport;
+    costPlan?: TrainingCostPlan;
+    lineage?: {
+        kind: 'measurement-repair';
+        parentExperimentId: string;
+        parentExecutionIdentity: string;
+        executionIdentity: string;
+        receiptDigest: string;
+        inheritedTrialIds: string[];
+    };
+}
+export declare function trainAgent(options: TrainingOptions): Promise<TrainingReport>;

package/dist/training/experiments/training/training.js

@@ -0,0 +1,208 @@
+import { retryTrial } from './retry.js';
+import { testRepeatCount, trialRetryCount } from '../../contracts/training-config.js';
+import { MutationQueue } from '../../platform/mutations/index.js';
+import { loadTrialSlots } from '../concurrency/index.js';
+import { repairMeasurement } from '../repair/index.js';
+import { recoverMeasurement, recoverStoredReport } from './recovery.js';
+import { TrainingCheckpoint } from './checkpoint.js';
+import { randomUUID } from 'node:crypto';
+import { candidateComponents, deriveCandidate, validateCandidate } from '../../candidates/index.js';
+import { validateDatasetSplit } from '../../cases/index.js';
+import { InferenceFailure } from '../../inference/index.js';
+import { createDspyGeneration, runDspyOptimization, readDspyCheckpointEvidence } from '../../optimization/index.js';
+import { resolveTrainingPartitions } from './partition.js';
+import { runStoredTrial } from '../trials/index.js';
+import { trialScore } from './scoring.js';
+import { freezeCriteria } from './criteria.js';
+import { trainingFeedback } from './feedback.js';
+import { evaluateHeldOut } from './held-out.js';
+export async function trainAgent(options) {
+    const started = performance.now();
+    const trialRetries = trialRetryCount(options.trialRetries);
+    const testRepeats = testRepeatCount(options.testRepeats);
+    if ('maxRunRequests' in options)
+        throw new Error('Run-wide provider request admission is not implemented; maxRunRequests cannot be enforced');
+    const { concurrency, pool, slots } = await loadTrialSlots(options);
+    const progress = options.onProgress;
+    const mutations = new MutationQueue();
+    options = { ...options, concurrency, slots,
+        onProgress: progress ? event => mutations.run(() => progress(event)) : undefined };
+    validateCandidate(options.baseline);
+    const partition = resolveTrainingPartitions(options.train, options.validation, options.validationGroups);
+    validateDatasetSplit(partition.train, options.test, partition.validation);
+    const criteria = new Map();
+    for (const [phase, dataset] of [['train', partition.train], ['validation', partition.validation], ['test', options.test]]) {
+        criteria.set(dataset, await freezeCriteria(dataset, options.native.evaluation, options.native.registry, `${phase}:${dataset.id}:${dataset.sourceDigest}:${dataset.sharedPlanDigest}`));
+    }
+    const components = candidateComponents(options.baseline);
+    const baselineComponents = Object.fromEntries(components.map(component => [component.id, component.text]));
+    const editScope = options.search ? Object.fromEntries(components.map(component => [component.id, component.heading ?? 'whole-file'])) : undefined;
+    if (!Object.keys(baselineComponents).length)
+        throw new Error('No declared editable components');
+    const checkpoint = await TrainingCheckpoint.open(options, [partition.train, partition.validation, options.test]);
+    const trials = [...checkpoint?.reports ?? []];
+    const retries = new Map((checkpoint?.retries ?? []).map(retry => [retry.replacementTrialId, retry]));
+    const rescores = [...checkpoint?.rescores ?? []];
+    const inheritedTrialIds = checkpoint?.inheritedTrialIds ?? [];
+    for (const report of trials)
+        await options.onProgress?.({ type: 'trial.restored', payload: report });
+    for (const report of rescores)
+        await options.onProgress?.({ type: 'trial.rescore.restored', payload: report });
+    const measured = (report) => rescores.findLast(rescore => rescore.runId === report.id) ?? report;
+    let generations = 0;
+    const generate = createDspyGeneration(options.inference.provider, options.inference.model);
+    const candidates = new Map();
+    const key = (components) => JSON.stringify(Object.keys(components).sort().map(name => [name, components[name]]));
+    candidates.set(key(baselineComponents), options.baseline);
+    const candidateFor = (components) => {
+        if (Object.keys(components).length !== Object.keys(baselineComponents).length || Object.keys(baselineComponents).some(id => !Object.hasOwn(components, id))) {
+            throw new Error('Optimizer changed the editable component set');
+        }
+        const identity = key(components);
+        let candidate = candidates.get(identity);
+        if (!candidate) {
+            candidate = deriveCandidate(options.baseline, options.baseline, components);
+            candidates.set(identity, candidate);
+        }
+        return candidate;
+    };
+    const attemptTrial = async (dataset, caseId, phase, candidate, signal, callbackId, round, retry) => {
+        const record = dataset.records.find(record => record.id === caseId);
+        if (!record)
+            throw new Error(`Case ${caseId} is not in ${phase}`);
+        const key = callbackId ?? `test:${candidate.id}:${caseId}`;
+        const entry = await checkpoint?.begin(key, retry);
+        const id = entry?.id ?? randomUUID();
+        if (retry) {
+            const receipt = { ...retry, replacementTrialId: id };
+            retries.set(id, receipt);
+            await options.onProgress?.({ type: 'trial.retry', payload: receipt });
+        }
+        if (entry?.restored) {
+            const report = await recoverStoredReport(options, entry.id, phase, candidate.id, entry.report);
+            if (!entry.report) {
+                await checkpoint.complete(key, report);
+                entry.report = report;
+                trials.push(report);
+            }
+        }
+        if (entry?.report) {
+            if (trialRetries && entry.report.execution === 'failed')
+                return entry.report;
+            const recovery = { options, dataset, report: entry.report, previous: entry.rescores?.at(-1),
+                evaluation: criteria.get(dataset), signal: signal ?? options.signal,
+                pendingRescoreId: entry.pendingRescoreId,
+                reserve: () => checkpoint.reserveRecovery(options.checkpoint?.maxRecoveryAttempts ?? 2, key) };
+            const rescore = options.checkpoint?.fork && checkpoint.inheritedTrialIds.includes(entry.report.id)
+                ? await repairMeasurement(recovery, options.checkpoint.fork.receiptDigest)
+                : await recoverMeasurement(recovery);
+            if (rescore && rescore !== entry.rescores?.at(-1)) {
+                await checkpoint.rescore(key, rescore);
+                rescores.push(rescore);
+                await options.onProgress?.({ type: 'trial.rescored', payload: rescore });
+            }
+            await options.onProgress?.({ type: 'trial.reused', payload: { id: entry.report.id, caseId, phase, candidateId: candidate.id } });
+            return entry.report;
+        }
+        await options.onProgress?.({ type: 'trial.queued', payload: { caseId, phase, candidateId: candidate.id } });
+        const lease = await pool?.acquire(signal ?? options.signal);
+        let confirmed = false;
+        let report;
+        let failure;
+        try {
+            report = await runStoredTrial({ ...options.native,
+                ...(options.scoring?.deferred === 'stage-score' ? { retainDeferredStatus: true } : {}), slot: lease?.slot, id, phase, round, record: structuredClone(record), candidate,
+                datasetId: dataset.id, datasetDigest: dataset.sourceDigest, planDigest: dataset.sharedPlanDigest,
+                signal: signal ?? options.signal, evaluation: criteria.get(dataset),
+                protectedPaths: [...options.native.protectedPaths ?? [], ...options.checkpoint ? [options.checkpoint.directory] : []],
+                observeEvent: async (event) => {
+                    await options.native.observeEvent?.(event);
+                    await options.onProgress?.({ type: 'trial.progress', payload: { trialId: id, caseId, phase, candidateId: candidate.id, event } });
+                } });
+            confirmed = report.hostWork === 'settled' && report.stop?.state === 'acknowledged' && report.stop.quiescence === 'confirmed';
+            await checkpoint?.complete(key, report);
+        }
+        catch (error) {
+            failure = error;
+            throw error;
+        }
+        finally {
+            lease?.release(confirmed, quarantineCause(id, report, failure));
+        }
+        trials.push(report);
+        await options.onProgress?.({ type: 'trial.finished', payload: report });
+        return report;
+    };
+    const trial = (dataset, caseId, phase, candidate, signal, callbackId, round) => {
+        if (!trialRetries)
+            return attemptTrial(dataset, caseId, phase, candidate, signal, callbackId, round);
+        const combined = AbortSignal.any([signal, options.signal].filter((value) => value !== undefined));
+        return retryTrial({ key: callbackId ?? `test:${candidate.id}:${caseId}`, maximum: trialRetries, signal: combined,
+            execute: (key, retry) => {
+                combined.throwIfAborted();
+                return attemptTrial(dataset, caseId, phase, candidate, combined, key, round, retry);
+            } });
+    };
+    await options.onProgress?.({ type: 'training.started', payload: { id: options.id, baselineId: options.baseline.id,
+            train: partition.train.records.map(record => record.id), validation: partition.validation.records.map(record => record.id), budget: options.budget } });
+    if (checkpoint?.optimizer && options.checkpoint) {
+        const restored = await readDspyCheckpointEvidence({ id: options.id, components: baselineComponents,
+            trainCaseIds: partition.train.records.map(record => record.id), validationCaseIds: partition.validation.records.map(record => record.id),
+            concurrency, budget: options.budget, seed: options.seed, proposal: options.proposal, search: options.search, editScope, checkpoint: options.checkpoint });
+        await options.onProgress?.({ type: 'optimizer.restored', payload: restored });
+    }
+    const optimizer = checkpoint?.optimizer ?? await runDspyOptimization({ id: options.id, components: baselineComponents,
+        trainCaseIds: partition.train.records.map(record => record.id), validationCaseIds: partition.validation.records.map(record => record.id),
+        concurrency, bridge: options.bridge, checkpoint: options.checkpoint, proposal: options.proposal, search: options.search, editScope,
+        onRestore: async (evidence) => { await options.onProgress?.({ type: 'optimizer.restored', payload: evidence }); }, budget: options.budget, seed: options.seed, signal: options.signal,
+        validateComponents: components => { candidateFor(components); },
+        generate: async (request, signal) => {
+            const attempt = ++generations;
+            await options.onProgress?.({ type: 'optimizer.generation.started', payload: { attempt } });
+            let response;
+            try {
+                response = await generate(request, signal);
+            }
+            catch (error) {
+                await options.onProgress?.({ type: 'optimizer.generation.failed', payload: { attempt,
+                        requestedModel: options.inference.model, model: null, route: null, usage: null, outcome: 'generation-error',
+                        ...(error instanceof InferenceFailure ? { attempts: error.attempts } : {}) } });
+                throw error;
+            }
+            await options.onProgress?.({ type: 'optimizer.generation.completed', payload: { attempt, model: response.model,
+                    modelBasis: response.modelBasis, route: response.route, usage: response.usage,
+                    ...(response.attempts ? { attempts: response.attempts } : {}) } });
+            return response;
+        },
+        evaluate: async (request, signal, callbackId) => {
+            const dataset = request.split === 'train' ? partition.train : partition.validation;
+            const report = await trial(dataset, request.caseId, request.split, candidateFor(request.components), signal, callbackId, request.round);
+            const record = dataset.records.find(record => record.id === request.caseId);
+            return { score: trialScore(measured(report), options.scoring), captureId: report.id, ...trainingFeedback(record, report, measured(report).evaluation) };
+        } });
+    await checkpoint?.freeze(optimizer);
+    const selected = candidateFor(optimizer.components);
+    // Freeze selection BEFORE the sealed test can execute or become feedback.
+    const frozen = [...new Map([options.baseline, selected].map(candidate => [candidate.id, candidate])).values()];
+    await options.onProgress?.({ type: 'selection.frozen', payload: { selectedId: selected.id, baselineId: options.baseline.id, optimizer } });
+    const heldOut = await evaluateHeldOut({ candidates: frozen, caseIds: options.test.records.map(record => record.id), concurrency, testRepeats,
+        trial: (candidate, caseId, repeat) => trial(options.test, caseId, 'test', candidate, undefined, repeat === 0 ? undefined : `test-repeat:${JSON.stringify([candidate.id, caseId, repeat])}`),
+        measurement: report => ({ id: measured(report).id, score: trialScore(measured(report), options.scoring) }) });
+    const report = { schema: 'spawnfile.training.training.v1', id: options.id, durationMs: performance.now() - started, baseline: options.baseline, selected, optimizer,
+        ...(options.scoring?.deferred === 'stage-score' ? { scoring: { ...options.scoring } } : {}),
+        partitions: { train: partition.train.records.map(record => record.id), validation: partition.validation.records.map(record => record.id),
+            test: options.test.records.map(record => record.id) }, trials, heldOut, ...(retries.size ? { retries: [...retries.values()] } : {}), ...(rescores.length ? { rescores } : {}),
+        ...(options.checkpoint?.fork ? { lineage: { kind: 'measurement-repair',
+                parentExperimentId: options.checkpoint.fork.id, parentExecutionIdentity: options.checkpoint.fork.identity,
+                executionIdentity: options.checkpoint.identity, receiptDigest: options.checkpoint.fork.receiptDigest, inheritedTrialIds } } : {}) };
+    await options.onProgress?.({ type: 'training.completed', payload: report });
+    return report;
+}
+/** Why a slot lease was not confirmed closed, recorded with its quarantine. */
+function quarantineCause(id, report, failure) {
+    if (!report)
+        return `trial ${id} threw: ${failure instanceof Error ? failure.message : String(failure)}`;
+    const stop = report.stop;
+    return `trial ${id} hostWork=${report.hostWork} stop=${stop?.state ?? 'missing'}/${stop?.quiescence ?? 'missing'}: ${stop?.reason ?? report.error ?? 'no reason'}`;
+}
+//# sourceMappingURL=training.js.map