spawnfile 0.1.25 → 0.1.28

package/dist/training/cli/grader/report.d.ts

@@ -0,0 +1,26 @@
+import type { CalibrationFile, CalibrationMetrics, GoldenSource } from '../../graders/index.js';
+export interface QuestionDiagnostics {
+    candidates?: {
+        id: string;
+        ece: number | null;
+        brier: number | null;
+    }[];
+    missingRate: number;
+    boundedRate: number;
+    sourceMix: Record<GoldenSource, number>;
+    sourceAgreement?: Record<string, {
+        instances: number;
+        agreement: number;
+    }>;
+    bySource?: Partial<Record<GoldenSource, CalibrationMetrics>>;
+    permutation?: {
+        compared: number;
+        missing: number;
+        flipRate: number | null;
+    };
+}
+export declare function calibrationReport(file: CalibrationFile, diagnostics: Record<string, QuestionDiagnostics>, counts: {
+    excluded: number;
+    disputed: number;
+    unavailable: number;
+}): string;

package/dist/training/cli/grader/report.js

@@ -0,0 +1,42 @@
+const number = (value) => value == null ? '—' : value.toFixed(3);
+const safe = (value) => value.replaceAll('|', '\\|').replaceAll('\n', ' ');
+export function calibrationReport(file, diagnostics, counts) {
+    const lines = ['# Grader calibration', '',
+        'Metrics measure agreement with teacher consensus, not validated correctness, unless human- or editor-adjudicated labels are present.',
+        'Held-out lineages never choose models, parameters or bands. ECE/Brier intervals use 1,000 whole-lineage bootstrap resamples (90%).',
+        'Missing decisions remain in held-out counts and count wrong; Brier assigns loss 1 and ECE assigns maximum error to missing mass (conservative penalties, not imputed predictions).',
+        'Teacher-only backends report independent-labeler agreement; they do not claim ECE or calibrated probabilities.',
+        'Bounded Ollama scores use the conservative lower p(good): absent expected options get zero, absent unexpected options get the smallest candidate probability before normalization.',
+        'Model benchmark calibration is a prior, not evidence that this grader is calibrated. No labels are used as few-shot examples.',
+        '', `Excluded outputs: ${counts.excluded}; disputed instances: ${counts.disputed}; unavailable projections: ${counts.unavailable}.`, '',
+        '| Question | Status | Groups | n | Base rate | Accuracy/agreement | Brier | Baseline Brier | ECE [90% CI] | Band / rate | Missing / bounded | Human / editor / consensus / single |',
+        '|---|---|---:|---:|---:|---:|---:|---:|---|---|---|---|'];
+    for (const [id, result] of Object.entries(file.questions)) {
+        const m = result.metrics, d = diagnostics[id];
+        lines.push(`| ${safe(id)} | ${result.status} | ${m.groups} | ${m.instances} | ${number(m.baseRate)} | ${number(m.accuracy ?? m.agreement)} | ${number(m.brier)} | ${number(m.baseRateBrier)} | ${number(m.ece)} ${m.eceInterval ? `[${m.eceInterval.map(number).join(', ')}]` : ''} | ${result.band?.map(number).join('–') ?? '—'} / ${number(m.escalationRate)} | ${number(d.missingRate)} / ${number(d.boundedRate)} | ${d.sourceMix.human} / ${d.sourceMix.editor} / ${d.sourceMix.consensus} / ${d.sourceMix['single-teacher']} |`);
+    }
+    for (const [id, result] of Object.entries(file.questions)) {
+        const d = diagnostics[id];
+        lines.push('', `## ${safe(id)}`, '', ...result.failures.map(failure => `- ${failure}`));
+        lines.push(`- Selected model: ${safe(result.model ?? 'default')}.`);
+        for (const candidate of d.candidates ?? []) {
+            lines.push(`- Candidate ${safe(candidate.id)} held-out ECE=${number(candidate.ece)}, Brier=${number(candidate.brier)} (not used for selection).`);
+        }
+        for (const [source, metrics] of Object.entries(d.bySource ?? {})) {
+            lines.push(`- ${source}: n=${metrics.instances}, accuracy=${number(metrics.accuracy)}, Brier=${number(metrics.brier)}, ECE=${number(metrics.ece)}.`);
+        }
+        for (const [source, metrics] of Object.entries(d.sourceAgreement ?? {})) {
+            lines.push(`- ${source}: n=${metrics.instances}, independent agreement=${number(metrics.agreement)}; ECE not claimed.`);
+        }
+        if (result.teacherAccuracy) {
+            const t = result.teacherAccuracy;
+            lines.push(`- Teacher independent held-out accuracy: ${number(t.agreement)}; source: ${t.source}; groups: ${t.groups}; instances: ${t.instances}; good/bad: ${t.good}/${t.bad}.`, ...t.failures.map(failure => `- Teacher bar: ${safe(failure)}`), ...t.warnings.map(warning => `- ${safe(warning)}`));
+        }
+        if (d.permutation)
+            lines.push(`- Reversed option order: ${d.permutation.compared} compared, ${d.permutation.missing} missing; decision flip rate ${number(d.permutation.flipRate ?? undefined)}.`);
+        lines.push(`- False-pass rate: ${number(result.metrics.falsePassRate)}; cascade accuracy: ${number(result.metrics.cascadeAccuracy)}${result.metrics.cascadeInterval ? ` [90% CI ${result.metrics.cascadeInterval.map(number).join(', ')}]` : ''}.`);
+    }
+    lines.push('', 'Escalation calls are grouped by output/projection; per-output overflow counts wrong. Cost estimates require labeling calls of the same shape; unavailable measurements are unknown, never zero.', '');
+    return lines.join('\n');
+}
+//# sourceMappingURL=report.js.map

package/dist/training/cli/grader/report.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"report.js","sourceRoot":"","sources":["../../../../src/training/cli/grader/report.ts"],"names":[],"mappings":"AAWA,MAAM,MAAM,GAAG,CAAC,KAAgC,EAAU,EAAE,CAAC,KAAK,IAAI,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;AACpG,MAAM,IAAI,GAAG,CAAC,KAAa,EAAU,EAAE,CAAC,KAAK,CAAC,UAAU,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC,UAAU,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC;AAE3F,MAAM,UAAU,iBAAiB,CAAC,IAAqB,EAAE,WAAgD,EAAE,MAE1G;IACC,MAAM,KAAK,GAAG,CAAC,sBAAsB,EAAE,EAAE;QACvC,sIAAsI;QACtI,oIAAoI;QACpI,oLAAoL;QACpL,gHAAgH;QAChH,oLAAoL;QACpL,+HAA+H;QAC/H,EAAE,EAAE,qBAAqB,MAAM,CAAC,QAAQ,yBAAyB,MAAM,CAAC,QAAQ,8BAA8B,MAAM,CAAC,WAAW,GAAG,EAAE,EAAE;QACvI,qLAAqL;QACrL,yDAAyD,CAAC,CAAC;IAC7D,KAAK,MAAM,CAAC,EAAE,EAAE,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC;QAC1D,MAAM,CAAC,GAAG,MAAM,CAAC,OAAO,EAAE,CAAC,GAAG,WAAW,CAAC,EAAE,CAAE,CAAC;QAC/C,KAAK,CAAC,IAAI,CAAC,KAAK,IAAI,CAAC,EAAE,CAAC,MAAM,MAAM,CAAC,MAAM,MAAM,CAAC,CAAC,MAAM,MAAM,CAAC,CAAC,SAAS,MAAM,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,MAAM,CAAC,CAAC,CAAC,QAAQ,IAAI,CAAC,CAAC,SAAS,CAAC,MAAM,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,MAAM,CAAC,CAAC,CAAC,aAAa,CAAC,MAAM,MAAM,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,WAAW,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,MAAM,MAAM,CAAC,IAAI,EAAE,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,GAAG,MAAM,MAAM,CAAC,CAAC,CAAC,cAAc,CAAC,MAAM,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,MAAM,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC,SAAS,CAAC,KAAK,MAAM,CAAC,CAAC,SAAS,CAAC,MAAM,MAAM,CAAC,CAAC,SAAS,CAAC,SAAS,MAAM,CAAC,CAAC,SAAS,CAAC,gBAAgB,CAAC,IAAI,CAAC,CAAC;IAC1gB,CAAC;IACD,KAAK,MAAM,CAAC,EAAE,EAAE,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC;QAC1D,MAAM,CAAC,GAAG,WAAW,CAAC,EAAE,CAAE,CAAC;QAC3B,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,MAAM,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,EAAE,EAAE,GAAG,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC,KAAK,OAAO,EAAE,CAAC,CAAC,CAAC;QACxF,KAAK,CAAC,IAAI,CAAC,qBAAqB,IAAI,CAAC,MAAM,CAAC,KAAK,IAAI,SAAS,CAAC,GAAG,CAAC,CAAC;QACpE,KAAK,MAAM,SAAS,IAAI,CAAC,CAAC,UAAU,IAAI,EAAE,EAAE,CAAC;YAC3C,KAAK,CAAC,IAAI,CAAC,eAAe,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,iBAAiB,MAAM,CAAC,SAAS,CAAC,GAAG,CAAC,WAAW,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,4BAA4B,CAAC,CAAC;QACpJ,CAAC;QACD,KAAK,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ,IAAI,EAAE,CAAC,EAAE,CAAC;YACjE,KAAK,CAAC,IAAI,CAAC,KAAK,MAAM,OAAO,OAAO,CAAC,SAAS,cAAc,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,WAAW,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,SAAS,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACvJ,CAAC;QACD,KAAK,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,eAAe,IAAI,EAAE,CAAC,EAAE,CAAC;YACxE,KAAK,CAAC,IAAI,CAAC,KAAK,MAAM,OAAO,OAAO,CAAC,SAAS,2BAA2B,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,oBAAoB,CAAC,CAAC;QAC1H,CAAC;QACD,IAAI,MAAM,CAAC,eAAe,EAAE,CAAC;YAC3B,MAAM,CAAC,GAAG,MAAM,CAAC,eAAe,CAAC;YACjC,KAAK,CAAC,IAAI,CAAC,4CAA4C,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,aAAa,CAAC,CAAC,MAAM,aAAa,CAAC,CAAC,MAAM,gBAAgB,CAAC,CAAC,SAAS,eAAe,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,GAAG,GAAG,EAC9K,GAAG,CAAC,CAAC,QAAQ,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC,kBAAkB,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC,EAAE,GAAG,CAAC,CAAC,QAAQ,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC,KAAK,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,CAAC;QACzH,CAAC;QACD,IAAI,CAAC,CAAC,WAAW;YAAE,KAAK,CAAC,IAAI,CAAC,4BAA4B,CAAC,CAAC,WAAW,CAAC,QAAQ,cAAc,CAAC,CAAC,WAAW,CAAC,OAAO,gCAAgC,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,QAAQ,IAAI,SAAS,CAAC,GAAG,CAAC,CAAC;QACnM,KAAK,CAAC,IAAI,CAAC,sBAAsB,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,aAAa,CAAC,uBAAuB,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,eAAe,CAAC,GAAG,MAAM,CAAC,OAAO,CAAC,eAAe,CAAC,CAAC,CAAC,YAAY,MAAM,CAAC,OAAO,CAAC,eAAe,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;IACtP,CAAC;IACD,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,iMAAiM,EAAE,EAAE,CAAC,CAAC;IACtN,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC"}

package/dist/training/cli/grader/runtime.d.ts

@@ -0,0 +1,5 @@
+import { type GradingRuntime } from '../../graders/index.js';
+import { type GraderDependencies } from './backend.js';
+export declare function buildGradingRuntime(graderFile: string, dependencies?: GraderDependencies, signal?: AbortSignal): Promise<GradingRuntime>;
+/** Public grader-file dispatch used by native training/evaluation hosts. */
+export declare function buildExecutableGradingRuntime(file: string, dependencies?: GraderDependencies, signal?: AbortSignal): Promise<import("../../graders/atomic/types.js").AtomicRuntime | GradingRuntime>;

package/dist/training/cli/grader/runtime.js

@@ -0,0 +1,64 @@
+import { checkTrainingFilename } from '../../platform/index.js';
+import { canonicalWireId } from '../../contracts/index.js';
+import { parseYaml } from '../../graders/format/index.js';
+import { buildAtomicRuntime } from './atomic/index.js';
+import { readFile } from 'node:fs/promises';
+import { createIdentityCalibration, loadGrader, loadLabels, parseCalibrationJson, sha256, checkCalibration, createCalibrationIdentity, decisionModels, decisionModelIdentities, stableDigest } from '../../graders/index.js';
+import { validateStoredVotes } from '../../graders/labeling/index.js';
+import { DECISION_PROMPT_VERSION } from '../../inference/decision/index.js';
+import { graderBackends } from './backend.js';
+import { createGraderTeacher } from './label.js';
+import { readExclusions } from './exclusions.js';
+/** Endpoint tokens are shared across separately constructed grader routes. */
+const resources = new Map();
+export async function buildGradingRuntime(graderFile, dependencies = {}, signal) {
+    const source = await readFile(graderFile, 'utf8');
+    const grader = await loadGrader(graderFile);
+    const identityMode = grader.calibration === 'identity';
+    const identityBackends = identityMode ? graderBackends(grader, dependencies) : undefined;
+    const modelIdentities = identityBackends ? await decisionModelIdentities(grader, identityBackends, signal) : undefined;
+    const rawCalibration = identityMode ? createIdentityCalibration(grader, modelIdentities.default, DECISION_PROMPT_VERSION) : undefined;
+    const calibrationSource = rawCalibration ? JSON.stringify(rawCalibration) : await readFile(grader.calibration, 'utf8');
+    const calibration = parseCalibrationJson(calibrationSource);
+    const golden = identityMode ? { goldenDigest: calibration.identity.golden } : await loadLabels(grader.golden, grader);
+    if ('labels' in golden)
+        validateStoredVotes(grader, golden.labels, { resolved: true });
+    const exclusions = await readExclusions(calibration.identity.exclusions.map(item => item.dataset));
+    const backends = identityBackends ?? graderBackends(grader, dependencies);
+    const models = decisionModels(grader);
+    const backend = backends[models[0].id];
+    const reported = modelIdentities ?? await decisionModelIdentities(grader, backends, signal);
+    const identity = createCalibrationIdentity(grader, {
+        backendModel: grader.models ? stableDigest(reported) : reported.default,
+        ...(grader.models ? { modelIdentities: reported } : {}),
+        promptTemplateVersion: DECISION_PROMPT_VERSION, golden: golden.goldenDigest,
+        seed: calibration.identity.seed, exclusions: exclusions.identities
+    });
+    const check = checkCalibration(grader, calibration, identity);
+    if (!check.ok)
+        throw new Error(check.refusals.join('; '));
+    const definition = grader.teachers.find(teacher => teacher.id === grader.escalate.to);
+    const teacher = identityMode ? undefined : { provider: (dependencies.provider ?? createGraderTeacher)(definition), model: definition.model };
+    const routeResources = models.map(model => {
+        const endpoint = model.backend === 'teacher'
+            ? `teacher:${model.teacher}:${grader.teachers.find(teacher => teacher.id === model.teacher).model}`
+            : new URL(model.endpoint).href.replace(/\/$/, '');
+        if (!resources.has(endpoint))
+            resources.set(endpoint, {});
+        return resources.get(endpoint);
+    });
+    // Detect an edit between parsing and hashing rather than binding different bytes.
+    if (await readFile(graderFile, 'utf8') !== source)
+        throw new Error('Grader file changed while loading');
+    return { grader, calibration, backend, backends, ...(teacher ? { teacher } : {}),
+        goldenDigest: golden.goldenDigest, promptTemplateVersion: DECISION_PROMPT_VERSION,
+        resource: routeResources[0], resources: [...new Set(routeResources)], definitionDigest: sha256(source), calibrationDigest: identityMode ? sha256('identity') : sha256(calibrationSource) };
+}
+/** Public grader-file dispatch used by native training/evaluation hosts. */
+export async function buildExecutableGradingRuntime(file, dependencies = {}, signal) {
+    signal?.throwIfAborted();
+    checkTrainingFilename(file);
+    const value = parseYaml(await readFile(file, 'utf8'));
+    return canonicalWireId(value?.schema) === 'spawnfile.training.atomic-grader.v1' ? buildAtomicRuntime(file, dependencies) : buildGradingRuntime(file, dependencies, signal);
+}
+//# sourceMappingURL=runtime.js.map

package/dist/training/cli/grader/runtime.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"runtime.js","sourceRoot":"","sources":["../../../../src/training/cli/grader/runtime.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,qBAAqB,EAAE,MAAM,yBAAyB,CAAC;AAChE,OAAO,EAAE,eAAe,EAAE,MAAM,0BAA0B,CAAC;AAC3D,OAAO,EAAE,SAAS,EAAE,MAAM,+BAA+B,CAAC;AAC1D,OAAO,EAAE,kBAAkB,EAAE,MAAM,mBAAmB,CAAC;AACvD,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAC5C,OAAO,EAAE,yBAAyB,EAAE,UAAU,EAAE,UAAU,EAAE,oBAAoB,EAAE,MAAM,EAAE,gBAAgB,EAAE,yBAAyB,EACnI,cAAc,EAAE,uBAAuB,EAAE,YAAY,EAAuB,MAAM,wBAAwB,CAAC;AAC7G,OAAO,EAAE,mBAAmB,EAAE,MAAM,iCAAiC,CAAC;AACtE,OAAO,EAAE,uBAAuB,EAAE,MAAM,mCAAmC,CAAC;AAC5E,OAAO,EAAE,cAAc,EAA2B,MAAM,cAAc,CAAC;AACvE,OAAO,EAAE,mBAAmB,EAAE,MAAM,YAAY,CAAC;AACjD,OAAO,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AAEjD,8EAA8E;AAC9E,MAAM,SAAS,GAAG,IAAI,GAAG,EAAkB,CAAC;AAE5C,MAAM,CAAC,KAAK,UAAU,mBAAmB,CAAC,UAAkB,EAAE,eAAmC,EAAE,EAAE,MAAoB;IACvH,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC,CAAC;IAClD,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,UAAU,CAAC,CAAC;IAC5C,MAAM,YAAY,GAAG,MAAM,CAAC,WAAW,KAAK,UAAU,CAAC;IACvD,MAAM,gBAAgB,GAAG,YAAY,CAAC,CAAC,CAAC,cAAc,CAAC,MAAM,EAAE,YAAY,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACzF,MAAM,eAAe,GAAG,gBAAgB,CAAC,CAAC,CAAC,MAAM,uBAAuB,CAAC,MAAM,EAAE,gBAAgB,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACvH,MAAM,cAAc,GAAG,YAAY,CAAC,CAAC,CAAC,yBAAyB,CAAC,MAAM,EAAE,eAAgB,CAAC,OAAQ,EAAE,uBAAuB,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACxI,MAAM,iBAAiB,GAAG,cAAc,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC,MAAM,QAAQ,CAAC,MAAM,CAAC,WAAW,EAAE,MAAM,CAAC,CAAC;IACvH,MAAM,WAAW,GAAG,oBAAoB,CAAC,iBAAiB,CAAC,CAAC;IAC5D,MAAM,MAAM,GAAG,YAAY,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,WAAW,CAAC,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,MAAM,UAAU,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACtH,IAAI,QAAQ,IAAI,MAAM;QAAE,mBAAmB,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,CAAC;IACvF,MAAM,UAAU,GAAG,MAAM,cAAc,CAAC,WAAW,CAAC,QAAQ,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC;IACnG,MAAM,QAAQ,GAAG,gBAAgB,IAAI,cAAc,CAAC,MAAM,EAAE,YAAY,CAAC,CAAC;IAC1E,MAAM,MAAM,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC;IACtC,MAAM,OAAO,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAE,CAAC,EAAE,CAAE,CAAC;IACzC,MAAM,QAAQ,GAAG,eAAe,IAAI,MAAM,uBAAuB,CAAC,MAAM,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC;IAC5F,MAAM,QAAQ,GAAG,yBAAyB,CAAC,MAAM,EAAE;QACjD,YAAY,EAAE,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,YAAY,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,OAAQ;QACxE,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACvD,qBAAqB,EAAE,uBAAuB,EAAE,MAAM,EAAE,MAAM,CAAC,YAAY;QAC3E,IAAI,EAAE,WAAW,CAAC,QAAQ,CAAC,IAAI,EAAE,UAAU,EAAE,UAAU,CAAC,UAAU;KAAE,CAAC,CAAC;IACxE,MAAM,KAAK,GAAG,gBAAgB,CAAC,MAAM,EAAE,WAAW,EAAE,QAAQ,CAAC,CAAC;IAC9D,IAAI,CAAC,KAAK,CAAC,EAAE;QAAE,MAAM,IAAI,KAAK,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;IAC1D,MAAM,UAAU,GAAG,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,EAAE,KAAK,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAE,CAAC;IACvF,MAAM,OAAO,GAAG,YAAY,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAAC,YAAY,CAAC,QAAQ,IAAI,mBAAmB,CAAC,CAAC,UAAU,CAAC,EAAE,KAAK,EAAE,UAAU,CAAC,KAAK,EAAE,CAAC;IAC7I,MAAM,cAAc,GAAG,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE;QACxC,MAAM,QAAQ,GAAG,KAAK,CAAC,OAAO,KAAK,SAAS;YAC1C,CAAC,CAAC,WAAW,KAAK,CAAC,OAAO,IAAI,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,EAAE,KAAK,KAAK,CAAC,OAAO,CAAE,CAAC,KAAK,EAAE;YACpG,CAAC,CAAC,IAAI,GAAG,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;QACpD,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,QAAQ,CAAC;YAAE,SAAS,CAAC,GAAG,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;QAC1D,OAAO,SAAS,CAAC,GAAG,CAAC,QAAQ,CAAE,CAAC;IAClC,CAAC,CAAC,CAAC;IACH,kFAAkF;IAClF,IAAI,MAAM,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC,KAAK,MAAM;QAAE,MAAM,IAAI,KAAK,CAAC,mCAAmC,CAAC,CAAC;IACxG,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,OAAO,EAAE,QAAQ,EAAE,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC9E,YAAY,EAAE,MAAM,CAAC,YAAY,EAAE,qBAAqB,EAAE,uBAAuB;QACjF,QAAQ,EAAE,cAAc,CAAC,CAAC,CAAE,EAAE,SAAS,EAAE,CAAC,GAAG,IAAI,GAAG,CAAC,cAAc,CAAC,CAAC,EAAE,gBAAgB,EAAE,MAAM,CAAC,MAAM,CAAC,EAAE,iBAAiB,EAAE,YAAY,CAAC,CAAC,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,iBAAiB,CAAC,EAAE,CAAC;AAChM,CAAC;AAED,4EAA4E;AAC5E,MAAM,CAAC,KAAK,UAAU,6BAA6B,CAAC,IAAY,EAAE,eAAmC,EAAE,EAAE,MAAoB;IAC3H,MAAM,EAAE,cAAc,EAAE,CAAC;IACzB,qBAAqB,CAAC,IAAI,CAAC,CAAC;IAC5B,MAAM,KAAK,GAAG,SAAS,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,CAAwB,CAAC;IAC7E,OAAO,eAAe,CAAC,KAAK,EAAE,MAAM,CAAC,KAAK,qCAAqC,CAAC,CAAC,CAAC,kBAAkB,CAAC,IAAI,EAAE,YAAY,CAAC,CAAC,CAAC,CAAC,mBAAmB,CAAC,IAAI,EAAE,YAAY,EAAE,MAAM,CAAC,CAAC;AAC7K,CAAC"}

package/dist/training/cli/grader/validate.d.ts

@@ -0,0 +1,5 @@
+import { type GraderDependencies } from './backend.js';
+export declare function runValidate(graderFile: string, dependencies?: GraderDependencies, signal?: AbortSignal): Promise<{
+    status: 'valid';
+    calibration: boolean;
+}>;

package/dist/training/cli/grader/validate.js

@@ -0,0 +1,42 @@
+import { readFile } from 'node:fs/promises';
+import { loadGrader, loadLabels, parseCalibration, calibrationIdentitySchema, createCalibrationIdentity, checkCalibration, decisionModelIdentities, stableDigest } from '../../graders/index.js';
+import { validateStoredVotes } from '../../graders/labeling/index.js';
+import { DECISION_PROMPT_VERSION } from '../../inference/decision/index.js';
+import { graderBackends } from './backend.js';
+import { readExclusions } from './exclusions.js';
+export async function runValidate(graderFile, dependencies = {}, signal) {
+    const grader = await loadGrader(graderFile);
+    if (grader.calibration === 'identity')
+        return { status: 'valid', calibration: false };
+    const loaded = await loadLabels(grader.golden, grader);
+    validateStoredVotes(grader, loaded.labels, { resolved: true });
+    let source;
+    try {
+        source = await readFile(grader.calibration, 'utf8');
+    }
+    catch (error) {
+        if (error.code === 'ENOENT')
+            return { status: 'valid', calibration: false };
+        throw error;
+    }
+    const value = JSON.parse(source);
+    if (value !== null && typeof value === 'object' && 'identity' in value &&
+        !calibrationIdentitySchema.safeParse(value.identity).success) {
+        throw new Error('calibration file is stale or from an older version; re-run grader calibrate');
+    }
+    const calibration = parseCalibration(value);
+    const exclusions = await readExclusions(calibration.identity.exclusions.map(item => item.dataset));
+    const modelIdentities = await decisionModelIdentities(grader, graderBackends(grader, dependencies), signal);
+    const identity = createCalibrationIdentity(grader, {
+        backendModel: grader.models ? stableDigest(modelIdentities) : modelIdentities.default,
+        ...(grader.models ? { modelIdentities } : {}),
+        promptTemplateVersion: DECISION_PROMPT_VERSION, golden: loaded.goldenDigest,
+        seed: calibration.identity.seed, exclusions: exclusions.identities,
+    });
+    // Validation checks freshness; the calibration command owns the quality verdict.
+    const check = checkCalibration(grader, calibration, identity, { quality: false });
+    if (!check.ok)
+        throw new Error(check.refusals.join('\n'));
+    return { status: 'valid', calibration: true };
+}
+//# sourceMappingURL=validate.js.map

package/dist/training/cli/grader/validate.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"validate.js","sourceRoot":"","sources":["../../../../src/training/cli/grader/validate.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAC5C,OAAO,EAAE,UAAU,EAAE,UAAU,EAAE,gBAAgB,EAAE,yBAAyB,EAAE,yBAAyB,EAAE,gBAAgB,EAAE,uBAAuB,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AACjM,OAAO,EAAE,mBAAmB,EAAE,MAAM,iCAAiC,CAAC;AACtE,OAAO,EAAE,uBAAuB,EAAE,MAAM,mCAAmC,CAAC;AAC5E,OAAO,EAAE,cAAc,EAA2B,MAAM,cAAc,CAAC;AACvE,OAAO,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AAEjD,MAAM,CAAC,KAAK,UAAU,WAAW,CAAC,UAAkB,EAAE,eAAmC,EAAE,EAAE,MAAoB;IAC/G,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,UAAU,CAAC,CAAC;IAC5C,IAAI,MAAM,CAAC,WAAW,KAAK,UAAU;QAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,WAAW,EAAE,KAAK,EAAE,CAAC;IACtF,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACvD,mBAAmB,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,CAAC;IAC/D,IAAI,MAAc,CAAC;IACnB,IAAI,CAAC;QAAC,MAAM,GAAG,MAAM,QAAQ,CAAC,MAAM,CAAC,WAAW,EAAE,MAAM,CAAC,CAAC;IAAC,CAAC;IAC5D,OAAO,KAAK,EAAE,CAAC;QACb,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;YAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,WAAW,EAAE,KAAK,EAAE,CAAC;QACvG,MAAM,KAAK,CAAC;IACd,CAAC;IACD,MAAM,KAAK,GAAY,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;IAC1C,IAAI,KAAK,KAAK,IAAI,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,UAAU,IAAI,KAAK;QACpE,CAAC,yBAAyB,CAAC,SAAS,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,OAAO,EAAE,CAAC;QAC/D,MAAM,IAAI,KAAK,CAAC,6EAA6E,CAAC,CAAC;IACjG,CAAC;IACD,MAAM,WAAW,GAAG,gBAAgB,CAAC,KAAK,CAAC,CAAC;IAC5C,MAAM,UAAU,GAAG,MAAM,cAAc,CAAC,WAAW,CAAC,QAAQ,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC;IACnG,MAAM,eAAe,GAAG,MAAM,uBAAuB,CAAC,MAAM,EAAE,cAAc,CAAC,MAAM,EAAE,YAAY,CAAC,EAAE,MAAM,CAAC,CAAC;IAC5G,MAAM,QAAQ,GAAG,yBAAyB,CAAC,MAAM,EAAE;QACjD,YAAY,EAAE,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,YAAY,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,eAAe,CAAC,OAAQ;QACtF,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC7C,qBAAqB,EAAE,uBAAuB,EAAE,MAAM,EAAE,MAAM,CAAC,YAAY;QAC3E,IAAI,EAAE,WAAW,CAAC,QAAQ,CAAC,IAAI,EAAE,UAAU,EAAE,UAAU,CAAC,UAAU;KACnE,CAAC,CAAC;IACH,iFAAiF;IACjF,MAAM,KAAK,GAAG,gBAAgB,CAAC,MAAM,EAAE,WAAW,EAAE,QAAQ,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC,CAAC;IAClF,IAAI,CAAC,KAAK,CAAC,EAAE;QAAE,MAAM,IAAI,KAAK,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;IAC1D,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,WAAW,EAAE,IAAI,EAAE,CAAC;AAChD,CAAC"}

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

@@ -0,0 +1,6 @@
+export { main } from './commands.js';
+export type { CliIO, CliDependencies, InvocationIndex, SpawnfileCliIntegration, CliExecutionIdentityContext, NativeTrialServices } from './types.js';
+export type { SpawnfileTrainingContext } from './spawnfile/index.js';
+export { launch } from './launch.js';
+export type { TrainingMomentCase, TrainingMomentReceipt, TrainingMomentPreparation } from './spawnfile/index.js';
+export { spawnfileEnvironment } from './spawnfile/index.js';

package/dist/training/cli/index.js

@@ -0,0 +1,4 @@
+export { main } from './commands.js';
+export { launch } from './launch.js';
+export { spawnfileEnvironment } from './spawnfile/index.js';
+//# sourceMappingURL=index.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/training/cli/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAE,MAAM,eAAe,CAAC;AAGrC,OAAO,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC;AAGrC,OAAO,EAAE,oBAAoB,EAAE,MAAM,sBAAsB,CAAC"}

package/dist/training/cli/launch.d.ts

@@ -0,0 +1,3 @@
+import type { CliDependencies } from './types.js';
+/** Shared signal/exit handling for installed project integrations as well as the plain CLI. */
+export declare function launch(argv?: string[], dependencies?: CliDependencies): Promise<void>;

package/dist/training/cli/launch.js

@@ -0,0 +1,17 @@
+import { main } from './commands.js';
+/** Shared signal/exit handling for installed project integrations as well as the plain CLI. */
+export async function launch(argv = process.argv.slice(2), dependencies = {}) {
+    const controller = new AbortController();
+    const interrupt = () => controller.abort();
+    process.once('SIGINT', interrupt);
+    process.once('SIGTERM', interrupt);
+    try {
+        process.exitCode = await main(argv, { cwd: process.cwd(), signal: controller.signal,
+            stdout: (text) => { process.stdout.write(text); }, stderr: (text) => { process.stderr.write(text); } }, dependencies);
+    }
+    finally {
+        process.removeListener('SIGINT', interrupt);
+        process.removeListener('SIGTERM', interrupt);
+    }
+}
+//# sourceMappingURL=launch.js.map

package/dist/training/cli/launch.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"launch.js","sourceRoot":"","sources":["../../../src/training/cli/launch.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAE,MAAM,eAAe,CAAC;AAGrC,+FAA+F;AAC/F,MAAM,CAAC,KAAK,UAAU,MAAM,CAAC,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,eAAgC,EAAE;IAC3F,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;IACzC,MAAM,SAAS,GAAG,GAAS,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC;IACjD,OAAO,CAAC,IAAI,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC;IAAC,OAAO,CAAC,IAAI,CAAC,SAAS,EAAE,SAAS,CAAC,CAAC;IACtE,IAAI,CAAC;QACH,OAAO,CAAC,QAAQ,GAAG,MAAM,IAAI,CAAC,IAAI,EAAE,EAAE,GAAG,EAAE,OAAO,CAAC,GAAG,EAAE,EAAE,MAAM,EAAE,UAAU,CAAC,MAAM;YACjF,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,EAAE,YAAY,CAAC,CAAC;IAC1H,CAAC;YAAS,CAAC;QAAC,OAAO,CAAC,cAAc,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC;QAAC,OAAO,CAAC,cAAc,CAAC,SAAS,EAAE,SAAS,CAAC,CAAC;IAAC,CAAC;AAC1G,CAAC"}

package/dist/training/cli/native.d.ts

@@ -0,0 +1,235 @@
+import type { InferenceProviderRegistry } from '../contracts/index.js';
+import { type ResourceMapping } from '../candidates/index.js';
+import { type CliArguments } from './arguments.js';
+import { CliDatasets } from './datasets.js';
+import type { CliDependencies, CliIO } from './types.js';
+export declare function resourceMappings(args: CliArguments): ResourceMapping[];
+/** File selection only; deliberately has no runtime or inference construction. */
+export declare function prepareInputs(args: CliArguments, io: CliIO, executionIdentity?: string, subjects?: CliDependencies['subjects'], inferenceProviders?: InferenceProviderRegistry): Promise<{
+    context: undefined;
+    readiness: undefined;
+    runtime: string;
+    model: string;
+    sourceRoot: string;
+    out: string;
+    candidate: import("../candidates/types.js").CandidateManifest;
+    judges: Record<string, string>;
+    datasets: CliDatasets;
+    citationRepairs: Record<string, 0 | 1>;
+    maxTrials: number;
+    resources: {
+        [k: string]: string;
+    };
+}>;
+export declare function prepareCliInputs(args: CliArguments, io: CliIO, dependencies: CliDependencies): Promise<{
+    context: undefined;
+    readiness: undefined;
+    runtime: string;
+    model: string;
+    sourceRoot: string;
+    out: string;
+    candidate: import("../candidates/types.js").CandidateManifest;
+    judges: Record<string, string>;
+    datasets: CliDatasets;
+    citationRepairs: Record<string, 0 | 1>;
+    maxTrials: number;
+    resources: {
+        [k: string]: string;
+    };
+} | {
+    runtime: string;
+    model: string;
+    candidate: import("../candidates/types.js").CandidateManifest;
+    judges: Record<string, string>;
+    citationRepairs: Record<string, 0 | 1>;
+    sourceRoot: string;
+    out: string;
+    resources: {
+        [k: string]: string;
+    };
+    datasets: CliDatasets;
+    maxTrials: number;
+    context: {
+        version: "spawnfile.training-context.v1";
+        producer: {
+            package: "spawnfile";
+            version: string;
+        };
+        project: {
+            root: string;
+            manifest: string;
+            sourceDigest: string;
+        };
+        agent: {
+            id: string;
+            name: string;
+            source: string;
+            runtime: string;
+            engine: string | null;
+            model: {
+                provider: string;
+                name: string;
+                authMethod: string;
+            } | null;
+        };
+        sources: {
+            sourcePath: string;
+            destinationPath: string;
+            sha256: string;
+        }[];
+        documents: {
+            sourcePath: string;
+            destinationPath: string;
+            sha256: string;
+            role: string;
+        }[];
+        skills: {
+            sourcePath: string;
+            destinationPath: string;
+            sha256: string;
+            name: string;
+            ref: string;
+            requiresMcp: string[];
+        }[];
+        resources: {
+            id: string;
+            kind: "bundle" | "git" | "volume";
+            mount: string;
+            mode: "readonly" | "mutable";
+            sharing: "per_agent" | "team";
+            definitionDigest: string;
+            pin: string | null;
+        }[];
+        requirements: {
+            nativeCompilation: true;
+            isolatedPreparation: true;
+        };
+    };
+    readiness: {
+        runnable: boolean;
+        integration: {
+            id: string;
+            version: string;
+        } | null;
+        reason: string;
+    };
+}>;
+export declare function prepareNative(args: CliArguments, io: CliIO, dependencies: CliDependencies, timeoutMs: number, prepared?: Awaited<ReturnType<typeof prepareCliInputs>>, inferenceTimeoutMs?: number): Promise<{
+    adapter: import("../contracts/execution.js").SubjectAdapter;
+    inference: import("../contracts/execution.js").InferenceProvider | undefined;
+    prepareEnvironment: ((paths: Parameters<import("../experiments/index.js").TrialOptions["prepare"]>[0], signal: AbortSignal, context: import("../experiments/index.js").NativePreparationContext) => Promise<Omit<Awaited<ReturnType<import("../experiments/index.js").TrialOptions["prepare"]>>, "instructions">>) | undefined;
+    cacheRoots: string[];
+    subjectRoot: string | undefined;
+    closeInference: () => Promise<void>;
+    registry: Readonly<Record<string, import("../contracts/evaluation.js").RegisteredEvaluator>>;
+    evaluation: {
+        judges: Record<string, import("../contracts/evaluation.js").JudgeRoute>;
+        graders: Record<string, import("../graders/atomic/types.js").AtomicRuntime | import("../graders/index.js").GradingRuntime>;
+        embeddings: {};
+        readText: (path: string) => Promise<string>;
+    };
+    context: undefined;
+    readiness: undefined;
+    runtime: string;
+    model: string;
+    sourceRoot: string;
+    out: string;
+    candidate: import("../candidates/types.js").CandidateManifest;
+    judges: Record<string, string>;
+    datasets: CliDatasets;
+    citationRepairs: Record<string, 0 | 1>;
+    maxTrials: number;
+    resources: {
+        [k: string]: string;
+    };
+} | {
+    adapter: import("../contracts/execution.js").SubjectAdapter;
+    inference: import("../contracts/execution.js").InferenceProvider | undefined;
+    prepareEnvironment: ((paths: Parameters<import("../experiments/index.js").TrialOptions["prepare"]>[0], signal: AbortSignal, context: import("../experiments/index.js").NativePreparationContext) => Promise<Omit<Awaited<ReturnType<import("../experiments/index.js").TrialOptions["prepare"]>>, "instructions">>) | undefined;
+    cacheRoots: string[];
+    subjectRoot: string | undefined;
+    closeInference: () => Promise<void>;
+    registry: Readonly<Record<string, import("../contracts/evaluation.js").RegisteredEvaluator>>;
+    evaluation: {
+        judges: Record<string, import("../contracts/evaluation.js").JudgeRoute>;
+        graders: Record<string, import("../graders/atomic/types.js").AtomicRuntime | import("../graders/index.js").GradingRuntime>;
+        embeddings: {};
+        readText: (path: string) => Promise<string>;
+    };
+    runtime: string;
+    model: string;
+    candidate: import("../candidates/types.js").CandidateManifest;
+    judges: Record<string, string>;
+    citationRepairs: Record<string, 0 | 1>;
+    sourceRoot: string;
+    out: string;
+    resources: {
+        [k: string]: string;
+    };
+    datasets: CliDatasets;
+    maxTrials: number;
+    context: {
+        version: "spawnfile.training-context.v1";
+        producer: {
+            package: "spawnfile";
+            version: string;
+        };
+        project: {
+            root: string;
+            manifest: string;
+            sourceDigest: string;
+        };
+        agent: {
+            id: string;
+            name: string;
+            source: string;
+            runtime: string;
+            engine: string | null;
+            model: {
+                provider: string;
+                name: string;
+                authMethod: string;
+            } | null;
+        };
+        sources: {
+            sourcePath: string;
+            destinationPath: string;
+            sha256: string;
+        }[];
+        documents: {
+            sourcePath: string;
+            destinationPath: string;
+            sha256: string;
+            role: string;
+        }[];
+        skills: {
+            sourcePath: string;
+            destinationPath: string;
+            sha256: string;
+            name: string;
+            ref: string;
+            requiresMcp: string[];
+        }[];
+        resources: {
+            id: string;
+            kind: "bundle" | "git" | "volume";
+            mount: string;
+            mode: "readonly" | "mutable";
+            sharing: "per_agent" | "team";
+            definitionDigest: string;
+            pin: string | null;
+        }[];
+        requirements: {
+            nativeCompilation: true;
+            isolatedPreparation: true;
+        };
+    };
+    readiness: {
+        runnable: boolean;
+        integration: {
+            id: string;
+            version: string;
+        } | null;
+        reason: string;
+    };
+}>;

package/dist/training/cli/native.js

@@ -0,0 +1,83 @@
+import { judgeVersions, createTrainingInference, prepareJudgeCaches } from './training/index.js';
+import { createHash } from 'node:crypto';
+import { resolve } from 'node:path';
+import { resolveSubjectAdapter } from '../subjects/index.js';
+import { snapshotCandidate } from '../candidates/index.js';
+import { createEvaluationRegistry } from '../evaluation/index.js';
+import { count, judgeCitationRepairs, one, pairs } from './arguments.js';
+import { CliDatasets } from './datasets.js';
+import { spawnfileEnvironment } from './spawnfile/index.js';
+import { prepareSpawnfileInputs } from './spawnfile/index.js';
+export function resourceMappings(args) {
+    const instructions = args.flags.instructions ?? [];
+    if (!instructions.length)
+        throw new Error('At least one --instructions Markdown file is required');
+    const supporting = [...(args.flags.supporting ?? []), ...(args.flags.locked ?? [])];
+    const declared = [...instructions, ...supporting];
+    const editable = args.flags.editable ?? [];
+    if (new Set(declared).size !== declared.length || new Set(editable).size !== editable.length)
+        throw new Error('Resource selections must be unique');
+    if (editable.some((path) => !declared.includes(path) || args.flags.locked?.includes(path)))
+        throw new Error('--editable must select a declared, non-locked resource');
+    return declared.map((path) => {
+        if (!/\.(md|markdown)$/i.test(path))
+            throw new Error('The CLI maps Markdown files only; native project configuration requires its own integration');
+        return { id: `resource-${createHash('sha256').update(path).digest('hex').slice(0, 16)}`, sourcePath: path,
+            destinations: [{ kind: instructions.includes(path) ? 'instructions' : 'supporting-files', path }],
+            edit: editable.includes(path) ? 'whole-file' : 'locked' };
+    });
+}
+/** File selection only; deliberately has no runtime or inference construction. */
+export async function prepareInputs(args, io, executionIdentity, subjects, inferenceProviders) {
+    const runtime = one(args, 'runtime');
+    if (!runtime || !subjects || !Object.hasOwn(subjects, runtime))
+        throw new Error(`Supported native runtimes: ${Object.keys(subjects ?? {}).join(', ') || 'none registered'}`);
+    const model = one(args, 'model');
+    const sourceRoot = resolve(io.cwd, one(args, 'source-root', io.cwd));
+    const out = resolve(io.cwd, one(args, 'out', args.flags['dry-run'] ? '.runtime/training' : undefined));
+    const candidate = await snapshotCandidate({ sourceRoot, resources: resourceMappings(args) });
+    const judges = pairs(args.flags.judge ?? []);
+    const datasets = new CliDatasets(judges, io.cwd, { inferenceExecutionIdentity: executionIdentity, inferenceProviders }, judgeVersions(args.training, executionIdentity, Object.keys(judges), inferenceProviders), args.frozenDataset);
+    return { context: undefined, readiness: undefined, runtime, model, sourceRoot, out, candidate, judges, datasets,
+        citationRepairs: judgeCitationRepairs(args, judges),
+        maxTrials: count(args, 'max-trials', 20, 10000),
+        resources: Object.fromEntries(Object.entries(pairs(args.flags.resource ?? [])).map(([name, path]) => [name, resolve(io.cwd, path)])),
+    };
+}
+export async function prepareCliInputs(args, io, dependencies) {
+    return args.flags['spawnfile-context'] || dependencies.spawnfileContext !== undefined
+        ? await prepareSpawnfileInputs(args, io, dependencies.spawnfile, dependencies.spawnfileContext, dependencies.inferenceProviders) : await prepareInputs(args, io, dependencies.spawnfile?.inferenceExecutionIdentity, dependencies.subjects, dependencies.inferenceProviders);
+}
+export async function prepareNative(args, io, dependencies, timeoutMs, prepared, inferenceTimeoutMs = timeoutMs) {
+    const inputs = prepared ?? await prepareCliInputs(args, io, dependencies);
+    const { runtime, model, judges, datasets } = inputs;
+    const graders = await datasets.buildGraders(dependencies.grader, args.command === 'train');
+    // Native services are created once per experiment in its durable output state.
+    const services = inputs.context !== undefined ? (dependencies.createTrialServices?.(inputs.out) ?? {}) : undefined;
+    const adapter = inputs.context !== undefined
+        ? dependencies.spawnfile.createAdapter(inputs.context, { model, timeoutMs, ...services })
+        : dependencies.adapter?.(runtime, { model, timeoutMs })
+            ?? resolveSubjectAdapter(dependencies.subjects, runtime, { model, timeoutMs });
+    const configured = createTrainingInference(args, dependencies, inferenceTimeoutMs, judges, inputs.citationRepairs);
+    let cacheRoots;
+    try {
+        cacheRoots = await prepareJudgeCaches(args, configured.judges, inputs.out, dependencies.spawnfile?.judgeCacheRoot);
+    }
+    catch (error) {
+        await configured.close();
+        throw error;
+    }
+    const inference = args.training?.comparison ? undefined : configured.inference;
+    let prepareEnvironment;
+    if (inputs.context !== undefined) {
+        const context = inputs.context;
+        prepareEnvironment = spawnfileEnvironment(dependencies.spawnfile, context, services);
+    }
+    const subjectRoot = inputs.context !== undefined ? dependencies.spawnfile.subjectRoot : undefined;
+    return { ...inputs, adapter, inference, prepareEnvironment, cacheRoots, subjectRoot, closeInference: configured.close,
+        registry: createEvaluationRegistry(),
+        evaluation: { judges: configured.judges, graders,
+            embeddings: {}, readText: (path) => datasets.readCriterion(path) },
+    };
+}
+//# sourceMappingURL=native.js.map

package/dist/training/cli/native.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"native.js","sourceRoot":"","sources":["../../../src/training/cli/native.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,aAAa,EAAE,uBAAuB,EAAE,kBAAkB,EAAE,MAAM,qBAAqB,CAAC;AACjG,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,qBAAqB,EAAE,MAAM,sBAAsB,CAAC;AAC7D,OAAO,EAAE,iBAAiB,EAAwB,MAAM,wBAAwB,CAAC;AACjF,OAAO,EAAE,wBAAwB,EAAE,MAAM,wBAAwB,CAAC;AAClE,OAAO,EAAE,KAAK,EAAE,oBAAoB,EAAE,GAAG,EAAE,KAAK,EAAqB,MAAM,gBAAgB,CAAC;AAC5F,OAAO,EAAE,WAAW,EAAE,MAAM,eAAe,CAAC;AAE5C,OAAO,EAAE,oBAAoB,EAAE,MAAM,sBAAsB,CAAC;AAC5D,OAAO,EAAE,sBAAsB,EAAE,MAAM,sBAAsB,CAAC;AAG9D,MAAM,UAAU,gBAAgB,CAAC,IAAkB;IACjD,MAAM,YAAY,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,IAAI,EAAE,CAAC;IACnD,IAAI,CAAC,YAAY,CAAC,MAAM;QAAE,MAAM,IAAI,KAAK,CAAC,uDAAuD,CAAC,CAAC;IACnG,MAAM,UAAU,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,UAAU,IAAI,EAAE,CAAC,EAAE,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,CAAC;IACpF,MAAM,QAAQ,GAAG,CAAC,GAAG,YAAY,EAAE,GAAG,UAAU,CAAC,CAAC;IAClD,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,IAAI,EAAE,CAAC;IAC3C,IAAI,IAAI,GAAG,CAAC,QAAQ,CAAC,CAAC,IAAI,KAAK,QAAQ,CAAC,MAAM,IAAI,IAAI,GAAG,CAAC,QAAQ,CAAC,CAAC,IAAI,KAAK,QAAQ,CAAC,MAAM;QAAE,MAAM,IAAI,KAAK,CAAC,oCAAoC,CAAC,CAAC;IACpJ,IAAI,QAAQ,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,QAAQ,CAAC,IAAI,CAAC,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,wDAAwD,CAAC,CAAC;IACtK,OAAO,QAAQ,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;QAC3B,IAAI,CAAC,mBAAmB,CAAC,IAAI,CAAC,IAAI,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,6FAA6F,CAAC,CAAC;QACpJ,OAAO,EAAE,EAAE,EAAE,YAAY,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,EAAE,EAAE,UAAU,EAAE,IAAI;YACvG,YAAY,EAAE,CAAC,EAAE,IAAI,EAAE,YAAY,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,cAAuB,CAAC,CAAC,CAAC,kBAA2B,EAAE,IAAI,EAAE,CAAC;YACnH,IAAI,EAAE,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,YAAqB,CAAC,CAAC,CAAC,QAAiB,EAAE,CAAC;IAChF,CAAC,CAAC,CAAC;AACL,CAAC;AAED,kFAAkF;AAClF,MAAM,CAAC,KAAK,UAAU,aAAa,CAAC,IAAkB,EAAE,EAAS,EAAE,iBAA0B,EAAE,QAAsC,EAAE,kBAA8C;IACnL,MAAM,OAAO,GAAG,GAAG,CAAC,IAAI,EAAE,SAAS,CAAC,CAAC;IACrC,IAAI,CAAC,OAAO,IAAI,CAAC,QAAQ,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,QAAQ,EAAE,OAAO,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,8BAA8B,MAAM,CAAC,IAAI,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,iBAAiB,EAAE,CAAC,CAAC;IAC7K,MAAM,KAAK,GAAG,GAAG,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;IACjC,MAAM,UAAU,GAAG,OAAO,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,CAAC,IAAI,EAAE,aAAa,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACrE,MAAM,GAAG,GAAG,OAAO,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,CAAC,IAAI,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,mBAAmB,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC;IACvG,MAAM,SAAS,GAAG,MAAM,iBAAiB,CAAC,EAAE,UAAU,EAAE,SAAS,EAAE,gBAAgB,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAC7F,MAAM,MAAM,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,IAAI,EAAE,CAAC,CAAC;IAC7C,MAAM,QAAQ,GAAG,IAAI,WAAW,CAAC,MAAM,EAAE,EAAE,CAAC,GAAG,EAAE,EAAE,0BAA0B,EAAE,iBAAiB,EAAE,kBAAkB,EAAE,EAAE,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE,iBAAiB,EAAE,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,kBAAkB,CAAC,EAAE,IAAI,CAAC,aAAa,CAAC,CAAC;IACtO,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,SAAS,EAAE,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE,UAAU,EAAE,GAAG,EAAE,SAAS,EAAE,MAAM,EAAE,QAAQ;QAC7G,eAAe,EAAE,oBAAoB,CAAC,IAAI,EAAE,MAAM,CAAC;QACnD,SAAS,EAAE,KAAK,CAAC,IAAI,EAAE,YAAY,EAAE,EAAE,EAAE,KAAK,CAAC;QAC/C,SAAS,EAAE,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,OAAO,CAAC,EAAE,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC;KACrI,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,gBAAgB,CAAC,IAAkB,EAAE,EAAS,EAAE,YAA6B;IACjG,OAAO,IAAI,CAAC,KAAK,CAAC,mBAAmB,CAAC,IAAI,YAAY,CAAC,gBAAgB,KAAK,SAAS;QACnF,CAAC,CAAC,MAAM,sBAAsB,CAAC,IAAI,EAAE,EAAE,EAAE,YAAY,CAAC,SAAS,EAAE,YAAY,CAAC,gBAAgB,EAAE,YAAY,CAAC,kBAAkB,CAAC,CAAC,CAAC,CAAC,MAAM,aAAa,CAAC,IAAI,EAAE,EAAE,EAAE,YAAY,CAAC,SAAS,EAAE,0BAA0B,EAAE,YAAY,CAAC,QAAQ,EAAE,YAAY,CAAC,kBAAkB,CAAC,CAAC;AACjR,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,aAAa,CAAC,IAAkB,EAAE,EAAS,EAAE,YAA6B,EAAE,SAAiB,EACjH,QAAuD,EAAE,kBAAkB,GAAG,SAAS;IACvF,MAAM,MAAM,GAAG,QAAQ,IAAI,MAAM,gBAAgB,CAAC,IAAI,EAAE,EAAE,EAAE,YAAY,CAAC,CAAC;IAC1E,MAAM,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,GAAG,MAAM,CAAC;IACpD,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAC,YAAY,CAAC,YAAY,CAAC,MAAM,EAAE,IAAI,CAAC,OAAO,KAAK,OAAO,CAAC,CAAC;IAC3F,+EAA+E;IAC/E,MAAM,QAAQ,GAAG,MAAM,CAAC,OAAO,KAAK,SAAS,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,mBAAmB,EAAE,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACnH,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,KAAK,SAAS;QAC1C,CAAC,CAAC,YAAY,CAAC,SAAU,CAAC,aAAa,CAAC,MAAM,CAAC,OAAO,EAAE,EAAE,KAAK,EAAE,SAAS,EAAE,GAAG,QAAQ,EAAE,CAAC;QAC1F,CAAC,CAAC,YAAY,CAAC,OAAO,EAAE,CAAC,OAAO,EAAE,EAAE,KAAK,EAAE,SAAS,EAAE,CAAC;eAClD,qBAAqB,CAAC,YAAY,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,KAAK,EAAE,SAAS,EAAE,CAAC,CAAC;IACnF,MAAM,UAAU,GAAG,uBAAuB,CAAC,IAAI,EAAE,YAAY,EAAE,kBAAkB,EAAE,MAAM,EAAE,MAAM,CAAC,eAAe,CAAC,CAAC;IACnH,IAAI,UAA0D,CAAC;IAC/D,IAAI,CAAC;QAAC,UAAU,GAAG,MAAM,kBAAkB,CAAC,IAAI,EAAE,UAAU,CAAC,MAAM,EAAE,MAAM,CAAC,GAAG,EAAE,YAAY,CAAC,SAAS,EAAE,cAAc,CAAC,CAAC;IAAC,CAAC;IAC3H,OAAO,KAAK,EAAE,CAAC;QAAC,MAAM,UAAU,CAAC,KAAK,EAAE,CAAC;QAAC,MAAM,KAAK,CAAC;IAAC,CAAC;IACxD,MAAM,SAAS,GAAG,IAAI,CAAC,QAAQ,EAAE,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,CAAC,SAAS,CAAC;IAC/E,IAAI,kBAA4D,CAAC;IACjE,IAAI,MAAM,CAAC,OAAO,KAAK,SAAS,EAAE,CAAC;QACjC,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC;QAC/B,kBAAkB,GAAG,oBAAoB,CAAC,YAAY,CAAC,SAAU,EAAE,OAAO,EAAE,QAAS,CAAC,CAAC;IACzF,CAAC;IACD,MAAM,WAAW,GAAG,MAAM,CAAC,OAAO,KAAK,SAAS,CAAC,CAAC,CAAC,YAAY,CAAC,SAAU,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC;IACnG,OAAO,EAAE,GAAG,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,kBAAkB,EAAE,UAAU,EAAE,WAAW,EAAE,cAAc,EAAE,UAAU,CAAC,KAAK;QACnH,QAAQ,EAAE,wBAAwB,EAAE;QACpC,UAAU,EAAE,EAAE,MAAM,EAAE,UAAU,CAAC,MAAM,EAAE,OAAO;YAC9C,UAAU,EAAE,EAAE,EAAE,QAAQ,EAAE,CAAC,IAAY,EAAE,EAAE,CAAC,QAAQ,CAAC,aAAa,CAAC,IAAI,CAAC,EAAE;KAC7E,CAAC;AACJ,CAAC"}

package/dist/training/cli/output.d.ts

@@ -0,0 +1,21 @@
+import type { CandidateManifest } from '../candidates/index.js';
+import { type ExperimentStore } from '../storage/index.js';
+import type { InvocationIndex } from './types.js';
+export declare function withOutput<T>(root: string, operation: (store: ExperimentStore) => Promise<T>, resume?: boolean): Promise<T>;
+export declare function persistResult(options: {
+    root: string;
+    store: ExperimentStore;
+    id: string;
+    command: 'evaluate' | 'train';
+    datasetId: string;
+    planDigest: string;
+    candidate: CandidateManifest;
+    report: unknown;
+    trialIds: string[];
+    optimizerBudget?: {
+        maxMetricCalls: number;
+        maxProposals: number;
+    };
+    outcome?: 'failed';
+    replaceIndex?: boolean;
+}): Promise<InvocationIndex>;