spawnfile 0.1.24 → 0.1.27

package/dist/training/evaluation/mechanical/tools.js

@@ -0,0 +1,39 @@
+import { capturedCalls, completeToolScope } from '../../capture/index.js';
+import { result } from '../logic/index.js';
+export async function evaluateTool(check, context) {
+    const { snapshot } = context;
+    const complete = completeToolScope(snapshot.closed, snapshot.coverage);
+    const calls = capturedCalls(snapshot.events).filter(call => {
+        if (call.tool !== check.tool)
+            return false;
+        if (check.phase === 'dispatch' && !call.dispatch)
+            return false;
+        if ((check.phase === 'completion' || check.outcome) && !call.completion)
+            return false;
+        if (check.outcome === 'succeeded')
+            return call.success === true;
+        if (check.outcome === 'failed')
+            return call.success === false;
+        return Boolean(call.dispatch || call.completion);
+    });
+    const count = calls.length;
+    const evidence = calls.map(call => ({ kind: 'event', id: (call.dispatch ?? call.completion).id }));
+    if (check.eval === 'tool-called')
+        return count > 0
+            ? { ...result('pass', `Observed ${count} matching calls`), evidence, score: 1 }
+            : result(complete ? 'fail' : 'unknown', 'No matching call in declared observation scope', snapshot.closed);
+    if (check.eval !== 'tool-count')
+        throw new Error(`Unknown tool evaluator ${check.eval}`);
+    const low = typeof check.exactly === 'number' ? check.exactly : typeof check.at_least === 'number' ? check.at_least : 0;
+    const high = typeof check.exactly === 'number' ? check.exactly : typeof check.at_most === 'number' ? check.at_most : Infinity;
+    // Counts grow monotonically: an exceeded upper bound cannot be repaired.
+    if (count > high)
+        return { ...result('fail', `${count} calls exceed ${high}`), evidence, score: 0 };
+    if (high === Infinity && count >= low)
+        return { ...result('pass', `${count} calls meet ${low}`), evidence, score: 1 };
+    if (!complete)
+        return { ...result('unknown', `Observed ${count}; count scope is not complete`, snapshot.closed), evidence };
+    return { ...result(count >= low ? 'pass' : 'fail', `${count} calls in [${low}, ${high}]`),
+        evidence, score: count >= low ? 1 : 0 };
+}
+//# sourceMappingURL=tools.js.map

package/dist/training/evaluation/mechanical/tools.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"tools.js","sourceRoot":"","sources":["../../../../src/training/evaluation/mechanical/tools.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,aAAa,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAC;AAC1E,OAAO,EAAE,MAAM,EAAE,MAAM,mBAAmB,CAAC;AAE3C,MAAM,CAAC,KAAK,UAAU,YAAY,CAAC,KAAgB,EAAE,OAA0B;IAC7E,MAAM,EAAE,QAAQ,EAAE,GAAG,OAAO,CAAC;IAC7B,MAAM,QAAQ,GAAG,iBAAiB,CAAC,QAAQ,CAAC,MAAM,EAAE,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACvE,MAAM,KAAK,GAAG,aAAa,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE;QACzD,IAAI,IAAI,CAAC,IAAI,KAAK,KAAK,CAAC,IAAI;YAAE,OAAO,KAAK,CAAC;QAC3C,IAAI,KAAK,CAAC,KAAK,KAAK,UAAU,IAAI,CAAC,IAAI,CAAC,QAAQ;YAAE,OAAO,KAAK,CAAC;QAC/D,IAAI,CAAC,KAAK,CAAC,KAAK,KAAK,YAAY,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU;YAAE,OAAO,KAAK,CAAC;QACtF,IAAI,KAAK,CAAC,OAAO,KAAK,WAAW;YAAE,OAAO,IAAI,CAAC,OAAO,KAAK,IAAI,CAAC;QAChE,IAAI,KAAK,CAAC,OAAO,KAAK,QAAQ;YAAE,OAAO,IAAI,CAAC,OAAO,KAAK,KAAK,CAAC;QAC9D,OAAO,OAAO,CAAC,IAAI,CAAC,QAAQ,IAAI,IAAI,CAAC,UAAU,CAAC,CAAC;IACnD,CAAC,CAAC,CAAC;IACH,MAAM,KAAK,GAAG,KAAK,CAAC,MAAM,CAAC;IAC3B,MAAM,QAAQ,GAAG,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,OAAgB,EAAE,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,IAAI,IAAI,CAAC,UAAU,CAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;IAC7G,IAAI,KAAK,CAAC,IAAI,KAAK,aAAa;QAAE,OAAO,KAAK,GAAG,CAAC;YAChD,CAAC,CAAC,EAAE,GAAG,MAAM,CAAC,MAAM,EAAE,YAAY,KAAK,iBAAiB,CAAC,EAAE,QAAQ,EAAE,KAAK,EAAE,CAAC,EAAE;YAC/E,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,SAAS,EAAE,gDAAgD,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;IAC7G,IAAI,KAAK,CAAC,IAAI,KAAK,YAAY;QAAE,MAAM,IAAI,KAAK,CAAC,0BAA0B,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC;IACzF,MAAM,GAAG,GAAG,OAAO,KAAK,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,KAAK,CAAC,QAAQ,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;IACxH,MAAM,IAAI,GAAG,OAAO,KAAK,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,KAAK,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ,CAAC;IAC9H,yEAAyE;IACzE,IAAI,KAAK,GAAG,IAAI;QAAE,OAAO,EAAE,GAAG,MAAM,CAAC,MAAM,EAAE,GAAG,KAAK,iBAAiB,IAAI,EAAE,CAAC,EAAE,QAAQ,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;IACpG,IAAI,IAAI,KAAK,QAAQ,IAAI,KAAK,IAAI,GAAG;QAAE,OAAO,EAAE,GAAG,MAAM,CAAC,MAAM,EAAE,GAAG,KAAK,eAAe,GAAG,EAAE,CAAC,EAAE,QAAQ,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;IACtH,IAAI,CAAC,QAAQ;QAAE,OAAO,EAAE,GAAG,MAAM,CAAC,SAAS,EAAE,YAAY,KAAK,+BAA+B,EAAE,QAAQ,CAAC,MAAM,CAAC,EAAE,QAAQ,EAAE,CAAC;IAC5H,OAAO,EAAE,GAAG,MAAM,CAAC,KAAK,IAAI,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,GAAG,KAAK,cAAc,GAAG,KAAK,IAAI,GAAG,CAAC;QACvF,QAAQ,EAAE,KAAK,EAAE,KAAK,IAAI,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;AAC5C,CAAC"}

package/dist/training/evaluation/registry.d.ts

@@ -0,0 +1,4 @@
+import type { EvaluationRegistry, RegisteredEvaluator } from '../contracts/evaluation.js';
+/** Custom evaluators always take a slot, whatever kind they declare. */
+export declare function isBuiltinMechanical(entry: RegisteredEvaluator | undefined): boolean;
+export declare function createEvaluationRegistry(custom?: EvaluationRegistry): EvaluationRegistry;

package/dist/training/evaluation/registry.js

@@ -0,0 +1,64 @@
+import { builtinEvaluatorDefinitions } from '../cases/index.js';
+import { evaluateSequence, evaluateTextCheck, evaluateText, evaluateTokenEfficiency, evaluateTool } from './mechanical/index.js';
+import { evaluateGrader } from './grader.js';
+import { evaluateRubric } from './semantic/index.js';
+import { evaluateSimilarity } from './embedding/index.js';
+/** Captured at construction: only these entries may run without an inference slot. */
+const builtinMechanical = new WeakSet();
+/** Custom evaluators always take a slot, whatever kind they declare. */
+export function isBuiltinMechanical(entry) {
+    return entry !== undefined && builtinMechanical.has(entry);
+}
+export function createEvaluationRegistry(custom = {}) {
+    const functions = {
+        'word-count': evaluateTextCheck, 'values-in-source': evaluateTextCheck,
+        equals: evaluateText, 'contains-any': evaluateText, 'json-schema': evaluateText,
+        'tool-called': evaluateTool, 'tool-count': evaluateTool, 'tool-sequence': evaluateSequence,
+        'token-efficiency': evaluateTokenEfficiency,
+        grader: evaluateGrader,
+        'llm-rubric': evaluateRubric, 'embedding-similarity': evaluateSimilarity,
+    };
+    const builtins = Object.fromEntries(Object.entries(functions).map(([id, evaluate]) => [id,
+        { definition: builtinEvaluatorDefinitions[id], evaluate }]));
+    // No builtin reads sibling results. Mechanical checks are pure over the snapshot; a
+    // judge's shared cache and an embedding provider serialize the checks that use them.
+    const named = (routes, name) => typeof name === 'string' && Object.hasOwn(routes, name) ? routes[name] : undefined;
+    for (const entry of Object.values(builtins))
+        entry.concurrency = () => [];
+    builtins['llm-rubric'].concurrency = (check, context) => {
+        const cache = named(context.judges, check.judge)?.cache;
+        return cache ? [cache] : [];
+    };
+    builtins.grader.concurrency = (check, context) => {
+        const route = named(context.graders ?? {}, check.grader);
+        const resources = route ? [...route.resources ?? [route.resource]] : [];
+        function collect(node) {
+            if (node.eval) {
+                const nested = (builtins[node.eval] ?? custom[node.eval])?.concurrency?.(node, context);
+                if (!nested)
+                    return false;
+                resources.push(...nested);
+                return true;
+            }
+            const children = node.all ?? node.any ?? (node.not ? [node.not] : undefined);
+            return children !== undefined && children.every(collect);
+        }
+        const binding = check.grader_binding;
+        return (binding?.gates ?? []).every(gate => collect(gate.check)) ? resources : undefined;
+    };
+    builtins['embedding-similarity'].concurrency = (check, context) => {
+        const provider = named(context.embeddings, check.provider);
+        return provider ? [provider] : [];
+    };
+    for (const entry of Object.values(builtins)) {
+        if (entry.definition.kind === 'mechanical')
+            builtinMechanical.add(entry);
+        Object.freeze(entry);
+    }
+    for (const id of Object.keys(custom)) {
+        if (Object.hasOwn(builtins, id))
+            throw new Error(`Cannot replace builtin evaluator ${id}`);
+    }
+    return Object.freeze({ ...builtins, ...custom });
+}
+//# sourceMappingURL=registry.js.map

package/dist/training/evaluation/registry.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"registry.js","sourceRoot":"","sources":["../../../src/training/evaluation/registry.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,2BAA2B,EAAE,MAAM,mBAAmB,CAAC;AAEhE,OAAO,EAAE,gBAAgB,EAAE,iBAAiB,EAAE,YAAY,EAAE,uBAAuB,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAC;AACjI,OAAO,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAC7C,OAAO,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AACrD,OAAO,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAE1D,sFAAsF;AACtF,MAAM,iBAAiB,GAAG,IAAI,OAAO,EAAuB,CAAC;AAE7D,wEAAwE;AACxE,MAAM,UAAU,mBAAmB,CAAC,KAAsC;IACxE,OAAO,KAAK,KAAK,SAAS,IAAI,iBAAiB,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;AAC7D,CAAC;AAED,MAAM,UAAU,wBAAwB,CAAC,SAA6B,EAAE;IACtE,MAAM,SAAS,GAAoD;QACjE,YAAY,EAAE,iBAAiB,EAAE,kBAAkB,EAAE,iBAAiB;QACtE,MAAM,EAAE,YAAY,EAAE,cAAc,EAAE,YAAY,EAAE,aAAa,EAAE,YAAY;QAC/E,aAAa,EAAE,YAAY,EAAE,YAAY,EAAE,YAAY,EAAE,eAAe,EAAE,gBAAgB;QAC1F,kBAAkB,EAAE,uBAAuB;QAC3C,MAAM,EAAE,cAAc;QACtB,YAAY,EAAE,cAAc,EAAE,sBAAsB,EAAE,kBAAkB;KACzE,CAAC;IACF,MAAM,QAAQ,GAAwC,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,QAAQ,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE;QAC5H,EAAE,UAAU,EAAE,2BAA2B,CAAC,EAAE,CAAE,EAAE,QAAQ,EAAE,CAAC,CAAC,CAAC,CAAC;IAChE,oFAAoF;IACpF,qFAAqF;IACrF,MAAM,KAAK,GAAG,CAAI,MAAmC,EAAE,IAAa,EAAE,EAAE,CACtE,OAAO,IAAI,KAAK,QAAQ,IAAI,MAAM,CAAC,MAAM,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACrF,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC;QAAE,KAAK,CAAC,WAAW,GAAG,GAAG,EAAE,CAAC,EAAE,CAAC;IAC1E,QAAQ,CAAC,YAAY,CAAE,CAAC,WAAW,GAAG,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;QACvD,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,MAAM,EAAE,KAAK,CAAC,KAAK,CAAC,EAAE,KAAK,CAAC;QACxD,OAAO,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;IAC9B,CAAC,CAAC;IACF,QAAQ,CAAC,MAAO,CAAC,WAAW,GAAG,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;QAChD,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,OAAO,IAAI,EAAE,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;QACzD,MAAM,SAAS,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,SAAS,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;QACxE,SAAS,OAAO,CAAC,IAAe;YAC9B,IAAI,IAAI,CAAC,IAAI,EAAE,CAAC;gBACd,MAAM,MAAM,GAAG,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;gBACxF,IAAI,CAAC,MAAM;oBAAE,OAAO,KAAK,CAAC;gBAC1B,SAAS,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,CAAC;gBAC1B,OAAO,IAAI,CAAC;YACd,CAAC;YACD,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,IAAI,IAAI,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC;YAC7E,OAAO,QAAQ,KAAK,SAAS,IAAI,QAAQ,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QAC3D,CAAC;QACD,MAAM,OAAO,GAAG,KAAK,CAAC,cAA2C,CAAC;QAClE,OAAO,CAAC,OAAO,EAAE,KAAK,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC;IAC3F,CAAC,CAAC;IACF,QAAQ,CAAC,sBAAsB,CAAE,CAAC,WAAW,GAAG,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE;QACjE,MAAM,QAAQ,GAAG,KAAK,CAAC,OAAO,CAAC,UAAU,EAAE,KAAK,CAAC,QAAQ,CAAC,CAAC;QAC3D,OAAO,QAAQ,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;IACpC,CAAC,CAAC;IACF,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC;QAC5C,IAAI,KAAK,CAAC,UAAU,CAAC,IAAI,KAAK,YAAY;YAAE,iBAAiB,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;QACzE,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACvB,CAAC;IACD,KAAK,MAAM,EAAE,IAAI,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC;QACrC,IAAI,MAAM,CAAC,MAAM,CAAC,QAAQ,EAAE,EAAE,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,oCAAoC,EAAE,EAAE,CAAC,CAAC;IAC7F,CAAC;IACD,OAAO,MAAM,CAAC,MAAM,CAAC,EAAE,GAAG,QAAQ,EAAE,GAAG,MAAM,EAAE,CAAC,CAAC;AACnD,CAAC"}

package/dist/training/evaluation/repair.d.ts

@@ -0,0 +1,14 @@
+import type { Stage } from '../contracts/dataset.js';
+import type { EvaluationRegistry, EvaluationResult, PlanEvaluation } from '../contracts/evaluation.js';
+/** Evaluator-owned originals must come from an integrity-verified capture. */
+export interface PlanMeasurementRepair {
+    originalStages: readonly Stage[];
+    original: PlanEvaluation;
+    /** Structural paths, independent of optional nested check IDs. */
+    errorPaths: readonly string[];
+}
+export type CheckReuse = (path: string) => EvaluationResult | undefined;
+/** Selective repair never authorizes a different plan or rerolls a valid grade. */
+export declare function measurementRepair(stages: readonly Stage[], repair: PlanMeasurementRepair, registry: EvaluationRegistry): CheckReuse;
+/** Admit a repair before spending its durable recovery reservation. */
+export declare function validateMeasurementRepair(stages: readonly Stage[], repair: PlanMeasurementRepair, registry: EvaluationRegistry): void;

package/dist/training/evaluation/repair.js

@@ -0,0 +1,73 @@
+import { isDeepStrictEqual } from 'node:util';
+/** Selective repair never authorizes a different plan or rerolls a valid grade. */
+export function measurementRepair(stages, repair, registry) {
+    if (!isDeepStrictEqual(stages, repair.originalStages))
+        throw new Error('Measurement repair changed the original plan');
+    if (!repair.errorPaths.length || new Set(repair.errorPaths).size !== repair.errorPaths.length) {
+        throw new Error('Measurement repair requires unique error leaf paths');
+    }
+    const selected = new Set(repair.errorPaths);
+    const found = new Set();
+    const reusable = new Map();
+    const visit = (node, prior, path) => {
+        if (!prior || prior.finality !== 'final' || (node.id !== undefined && node.id !== prior.id)) {
+            throw new Error('Measurement repair original check binding is invalid');
+        }
+        if (prior.status === 'pass' || prior.status === 'fail') {
+            if ([...selected].some(target => target === path || target.startsWith(path + '/'))) {
+                throw new Error('Measurement repair cannot reroll a resolved pass/fail check');
+            }
+            reusable.set(path, structuredClone(prior));
+            return false;
+        }
+        if (selected.has(path)) {
+            if (!node.eval || registry[node.eval]?.definition.kind !== 'semantic' || prior.status !== 'error'
+                || !prior.judgeCalls?.some(call => call.outcome !== 'completed')) {
+                throw new Error('Measurement repair may target only failed semantic inference leaves');
+            }
+            found.add(path);
+            return true;
+        }
+        const operator = node.all ? 'all' : node.any ? 'any' : node.not ? 'not' : undefined;
+        const children = node.all ?? node.any ?? (node.not ? [node.not] : []);
+        let changed = false;
+        if (children.length) {
+            if (prior.children?.length !== children.length)
+                throw new Error('Measurement repair original child bindings differ');
+            children.forEach((child, index) => {
+                const childPath = `${path}/${operator}${operator === 'not' ? '' : `/${index}`}`;
+                changed = visit(child, prior.children[index], childPath) || changed;
+            });
+        }
+        if (!changed)
+            reusable.set(path, structuredClone(prior));
+        return changed;
+    };
+    if (repair.original.stages.length !== stages.length)
+        throw new Error('Measurement repair original stage bindings differ');
+    stages.forEach((stage, index) => {
+        const original = repair.original.stages[index];
+        if (stage.id !== original.id || stage.checks.length !== original.checks.length) {
+            throw new Error('Measurement repair original stage bindings differ');
+        }
+        // Eligibility and gates are recomputed from check references. Evaluator calls
+        // outside declared checks have no separately captured leaf to repair safely.
+        const referencesOnly = (node) => !node.eval &&
+            (node.all ?? node.any ?? (node.not ? [node.not] : [])).every(referencesOnly);
+        if ((stage.when && !referencesOnly(stage.when)) || (stage.gate && !referencesOnly(stage.gate))) {
+            throw new Error('Measurement repair requires reference-only stage dependencies and gates');
+        }
+        stage.checks.forEach((check, position) => visit(check, original.checks[position], `/stages/${index}/checks/${position}`));
+    });
+    if (found.size !== selected.size)
+        throw new Error('Measurement repair names an unknown error leaf');
+    return path => {
+        const prior = reusable.get(path);
+        return prior === undefined ? undefined : structuredClone(prior);
+    };
+}
+/** Admit a repair before spending its durable recovery reservation. */
+export function validateMeasurementRepair(stages, repair, registry) {
+    measurementRepair(stages, repair, registry);
+}
+//# sourceMappingURL=repair.js.map

package/dist/training/evaluation/repair.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"repair.js","sourceRoot":"","sources":["../../../src/training/evaluation/repair.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,iBAAiB,EAAE,MAAM,WAAW,CAAC;AAa9C,mFAAmF;AACnF,MAAM,UAAU,iBAAiB,CAAC,MAAwB,EAAE,MAA6B,EACvF,QAA4B;IAC5B,IAAI,CAAC,iBAAiB,CAAC,MAAM,EAAE,MAAM,CAAC,cAAc,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,8CAA8C,CAAC,CAAC;IACvH,IAAI,CAAC,MAAM,CAAC,UAAU,CAAC,MAAM,IAAI,IAAI,GAAG,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,UAAU,CAAC,MAAM,EAAE,CAAC;QAC9F,MAAM,IAAI,KAAK,CAAC,qDAAqD,CAAC,CAAC;IACzE,CAAC;IACD,MAAM,QAAQ,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC;IAC5C,MAAM,KAAK,GAAG,IAAI,GAAG,EAAU,CAAC;IAChC,MAAM,QAAQ,GAAG,IAAI,GAAG,EAA4B,CAAC;IACrD,MAAM,KAAK,GAAG,CAAC,IAAe,EAAE,KAAuB,EAAE,IAAY,EAAW,EAAE;QAChF,IAAI,CAAC,KAAK,IAAI,KAAK,CAAC,QAAQ,KAAK,OAAO,IAAI,CAAC,IAAI,CAAC,EAAE,KAAK,SAAS,IAAI,IAAI,CAAC,EAAE,KAAK,KAAK,CAAC,EAAE,CAAC,EAAE,CAAC;YAC5F,MAAM,IAAI,KAAK,CAAC,sDAAsD,CAAC,CAAC;QAC1E,CAAC;QACD,IAAI,KAAK,CAAC,MAAM,KAAK,MAAM,IAAI,KAAK,CAAC,MAAM,KAAK,MAAM,EAAE,CAAC;YACvD,IAAI,CAAC,GAAG,QAAQ,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,KAAK,IAAI,IAAI,MAAM,CAAC,UAAU,CAAC,IAAI,GAAG,GAAG,CAAC,CAAC,EAAE,CAAC;gBACnF,MAAM,IAAI,KAAK,CAAC,6DAA6D,CAAC,CAAC;YACjF,CAAC;YACD,QAAQ,CAAC,GAAG,CAAC,IAAI,EAAE,eAAe,CAAC,KAAK,CAAC,CAAC,CAAC;YAC3C,OAAO,KAAK,CAAC;QACf,CAAC;QACD,IAAI,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;YACvB,IAAI,CAAC,IAAI,CAAC,IAAI,IAAI,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,UAAU,CAAC,IAAI,KAAK,UAAU,IAAI,KAAK,CAAC,MAAM,KAAK,OAAO;mBAC1F,CAAC,KAAK,CAAC,UAAU,EAAE,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,OAAO,KAAK,WAAW,CAAC,EAAE,CAAC;gBACrE,MAAM,IAAI,KAAK,CAAC,qEAAqE,CAAC,CAAC;YACzF,CAAC;YACD,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;YAChB,OAAO,IAAI,CAAC;QACd,CAAC;QACD,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;QACpF,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,IAAI,IAAI,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QACtE,IAAI,OAAO,GAAG,KAAK,CAAC;QACpB,IAAI,QAAQ,CAAC,MAAM,EAAE,CAAC;YACpB,IAAI,KAAK,CAAC,QAAQ,EAAE,MAAM,KAAK,QAAQ,CAAC,MAAM;gBAAE,MAAM,IAAI,KAAK,CAAC,mDAAmD,CAAC,CAAC;YACrH,QAAQ,CAAC,OAAO,CAAC,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE;gBAChC,MAAM,SAAS,GAAG,GAAG,IAAI,IAAI,QAAQ,GAAG,QAAQ,KAAK,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,EAAE,EAAE,CAAC;gBAChF,OAAO,GAAG,KAAK,CAAC,KAAK,EAAE,KAAK,CAAC,QAAS,CAAC,KAAK,CAAE,EAAE,SAAS,CAAC,IAAI,OAAO,CAAC;YACxE,CAAC,CAAC,CAAC;QACL,CAAC;QACD,IAAI,CAAC,OAAO;YAAE,QAAQ,CAAC,GAAG,CAAC,IAAI,EAAE,eAAe,CAAC,KAAK,CAAC,CAAC,CAAC;QACzD,OAAO,OAAO,CAAC;IACjB,CAAC,CAAC;IACF,IAAI,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,MAAM,KAAK,MAAM,CAAC,MAAM;QAAE,MAAM,IAAI,KAAK,CAAC,mDAAmD,CAAC,CAAC;IAC1H,MAAM,CAAC,OAAO,CAAC,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE;QAC9B,MAAM,QAAQ,GAAG,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAE,CAAC;QAChD,IAAI,KAAK,CAAC,EAAE,KAAK,QAAQ,CAAC,EAAE,IAAI,KAAK,CAAC,MAAM,CAAC,MAAM,KAAK,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC;YAC/E,MAAM,IAAI,KAAK,CAAC,mDAAmD,CAAC,CAAC;QACvE,CAAC;QACD,8EAA8E;QAC9E,6EAA6E;QAC7E,MAAM,cAAc,GAAG,CAAC,IAAe,EAAW,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI;YAC7D,CAAC,IAAI,CAAC,GAAG,IAAI,IAAI,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,cAAc,CAAC,CAAC;QAC/E,IAAI,CAAC,KAAK,CAAC,IAAI,IAAI,CAAC,cAAc,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,IAAI,CAAC,cAAc,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC;YAC/F,MAAM,IAAI,KAAK,CAAC,yEAAyE,CAAC,CAAC;QAC7F,CAAC;QACD,KAAK,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,KAAK,EAAE,QAAQ,EAAE,EAAE,CAAC,KAAK,CAAC,KAAK,EAAE,QAAQ,CAAC,MAAM,CAAC,QAAQ,CAAE,EAAE,WAAW,KAAK,WAAW,QAAQ,EAAE,CAAC,CAAC,CAAC;IAC7H,CAAC,CAAC,CAAC;IACH,IAAI,KAAK,CAAC,IAAI,KAAK,QAAQ,CAAC,IAAI;QAAE,MAAM,IAAI,KAAK,CAAC,gDAAgD,CAAC,CAAC;IACpG,OAAO,IAAI,CAAC,EAAE;QACZ,MAAM,KAAK,GAAG,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QACjC,OAAO,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,eAAe,CAAC,KAAK,CAAC,CAAC;IAClE,CAAC,CAAC;AACJ,CAAC;AAED,uEAAuE;AACvE,MAAM,UAAU,yBAAyB,CAAC,MAAwB,EAAE,MAA6B,EAC/F,QAA4B;IAC5B,iBAAiB,CAAC,MAAM,EAAE,MAAM,EAAE,QAAQ,CAAC,CAAC;AAC9C,CAAC"}

package/dist/training/evaluation/schedule.d.ts

@@ -0,0 +1,21 @@
+import type { EvaluationResult } from '../contracts/evaluation.js';
+/** Two covers the paired semantic judges of one stage without widening provider fan-out. */
+export declare const defaultCheckConcurrency = 2;
+export declare const maxCheckConcurrency = 8;
+export interface StageCheckTask {
+    /** Exclusive resources for an independent check; undefined means it must see every earlier result. */
+    resources: readonly object[] | undefined;
+    /** False only for work with no inference spend; it starts without holding a bounded slot. */
+    slot?: boolean;
+    run(): Promise<EvaluationResult>;
+}
+export declare function checkConcurrency(value: number | undefined): number;
+/**
+ * Start independent checks early, but commit every result in declaration order.
+ * Dependent checks wait for all earlier commits. Checks start in declaration order
+ * and checks sharing a resource never overlap. A slot and its resources are handed
+ * on only after the holder's commit outcome is known; commits are ordered, so every
+ * earlier outcome is known too. After a commit fails no run starts, and work already
+ * started settles before the failure is rethrown.
+ */
+export declare function runStageChecks(tasks: readonly StageCheckTask[], limit: number, commit: (value: EvaluationResult, index: number) => Promise<void>): Promise<void>;

package/dist/training/evaluation/schedule.js

@@ -0,0 +1,98 @@
+/** Two covers the paired semantic judges of one stage without widening provider fan-out. */
+export const defaultCheckConcurrency = 2;
+export const maxCheckConcurrency = 8;
+export function checkConcurrency(value) {
+    const limit = value ?? defaultCheckConcurrency;
+    if (!Number.isSafeInteger(limit) || limit < 1 || limit > maxCheckConcurrency)
+        throw new Error('Invalid evaluation check concurrency');
+    return limit;
+}
+/**
+ * Start independent checks early, but commit every result in declaration order.
+ * Dependent checks wait for all earlier commits. Checks start in declaration order
+ * and checks sharing a resource never overlap. A slot and its resources are handed
+ * on only after the holder's commit outcome is known; commits are ordered, so every
+ * earlier outcome is known too. After a commit fails no run starts, and work already
+ * started settles before the failure is rethrown.
+ */
+export async function runStageChecks(tasks, limit, commit) {
+    let active = 0;
+    const slots = [];
+    const acquire = async () => {
+        if (active < limit) {
+            active++;
+            return;
+        }
+        await new Promise(resolve => slots.push(resolve));
+    };
+    const release = () => {
+        const next = slots.shift();
+        if (next)
+            next();
+        else
+            active--;
+    };
+    const holders = new Map();
+    let failure;
+    let committed = Promise.resolve();
+    let startTurn = Promise.resolve();
+    const settle = (index, value, slotted) => {
+        committed = committed.then(async () => {
+            try {
+                const evaluated = await value;
+                if (!failure)
+                    await commit(evaluated, index);
+            }
+            catch (error) {
+                failure ??= { error };
+            }
+            finally {
+                if (slotted)
+                    release();
+            }
+        });
+        return committed;
+    };
+    for (const [index, task] of tasks.entries()) {
+        if (task.resources === undefined) {
+            await committed;
+            if (failure)
+                break;
+            settle(index, task.run(), false);
+            await committed;
+            continue;
+        }
+        if (failure)
+            break;
+        const blockers = [...new Set(task.resources)].map(resource => holders.get(resource)).filter(value => value !== undefined);
+        const slotted = task.slot !== false;
+        const previous = startTurn;
+        let started;
+        startTurn = new Promise(resolve => { started = resolve; });
+        const running = (async () => {
+            try {
+                await previous;
+                await Promise.all(blockers);
+                if (slotted)
+                    await acquire();
+            }
+            finally {
+                started();
+            }
+            if (failure)
+                throw new Error('Stage evaluation stopped after an earlier commit failure');
+            return task.run();
+        })();
+        // The ordered commit chain observes a rejection later; mark it handled now.
+        running.catch(() => undefined);
+        // Resource successors wait for this commit outcome, not merely for the run.
+        const outcome = settle(index, running, slotted);
+        for (const resource of task.resources)
+            holders.set(resource, outcome);
+    }
+    // Every started run is awaited by the ordered commit chain, including after a failure.
+    await committed;
+    if (failure)
+        throw failure.error;
+}
+//# sourceMappingURL=schedule.js.map

package/dist/training/evaluation/schedule.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"schedule.js","sourceRoot":"","sources":["../../../src/training/evaluation/schedule.ts"],"names":[],"mappings":"AAEA,4FAA4F;AAC5F,MAAM,CAAC,MAAM,uBAAuB,GAAG,CAAC,CAAC;AACzC,MAAM,CAAC,MAAM,mBAAmB,GAAG,CAAC,CAAC;AAUrC,MAAM,UAAU,gBAAgB,CAAC,KAAyB;IACxD,MAAM,KAAK,GAAG,KAAK,IAAI,uBAAuB,CAAC;IAC/C,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,CAAC,IAAI,KAAK,GAAG,mBAAmB;QAAE,MAAM,IAAI,KAAK,CAAC,sCAAsC,CAAC,CAAC;IACtI,OAAO,KAAK,CAAC;AACf,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,CAAC,KAAK,UAAU,cAAc,CAAC,KAAgC,EAAE,KAAa,EAClF,MAAiE;IACjE,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,MAAM,KAAK,GAAmB,EAAE,CAAC;IACjC,MAAM,OAAO,GAAG,KAAK,IAAI,EAAE;QACzB,IAAI,MAAM,GAAG,KAAK,EAAE,CAAC;YAAC,MAAM,EAAE,CAAC;YAAC,OAAO;QAAC,CAAC;QACzC,MAAM,IAAI,OAAO,CAAO,OAAO,CAAC,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC;IAC1D,CAAC,CAAC;IACF,MAAM,OAAO,GAAG,GAAG,EAAE;QACnB,MAAM,IAAI,GAAG,KAAK,CAAC,KAAK,EAAE,CAAC;QAC3B,IAAI,IAAI;YAAE,IAAI,EAAE,CAAC;;YAAM,MAAM,EAAE,CAAC;IAClC,CAAC,CAAC;IACF,MAAM,OAAO,GAAG,IAAI,GAAG,EAA4B,CAAC;IACpD,IAAI,OAAuC,CAAC;IAC5C,IAAI,SAAS,GAAG,OAAO,CAAC,OAAO,EAAE,CAAC;IAClC,IAAI,SAAS,GAAG,OAAO,CAAC,OAAO,EAAE,CAAC;IAClC,MAAM,MAAM,GAAG,CAAC,KAAa,EAAE,KAAgC,EAAE,OAAgB,EAAE,EAAE;QACnF,SAAS,GAAG,SAAS,CAAC,IAAI,CAAC,KAAK,IAAI,EAAE;YACpC,IAAI,CAAC;gBACH,MAAM,SAAS,GAAG,MAAM,KAAK,CAAC;gBAC9B,IAAI,CAAC,OAAO;oBAAE,MAAM,MAAM,CAAC,SAAS,EAAE,KAAK,CAAC,CAAC;YAC/C,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBAAC,OAAO,KAAK,EAAE,KAAK,EAAE,CAAC;YAAC,CAAC;oBAClC,CAAC;gBAAC,IAAI,OAAO;oBAAE,OAAO,EAAE,CAAC;YAAC,CAAC;QACrC,CAAC,CAAC,CAAC;QACH,OAAO,SAAS,CAAC;IACnB,CAAC,CAAC;IACF,KAAK,MAAM,CAAC,KAAK,EAAE,IAAI,CAAC,IAAI,KAAK,CAAC,OAAO,EAAE,EAAE,CAAC;QAC5C,IAAI,IAAI,CAAC,SAAS,KAAK,SAAS,EAAE,CAAC;YACjC,MAAM,SAAS,CAAC;YAChB,IAAI,OAAO;gBAAE,MAAM;YACnB,MAAM,CAAC,KAAK,EAAE,IAAI,CAAC,GAAG,EAAE,EAAE,KAAK,CAAC,CAAC;YACjC,MAAM,SAAS,CAAC;YAChB,SAAS;QACX,CAAC;QACD,IAAI,OAAO;YAAE,MAAM;QACnB,MAAM,QAAQ,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC;QAC1H,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,KAAK,KAAK,CAAC;QACpC,MAAM,QAAQ,GAAG,SAAS,CAAC;QAC3B,IAAI,OAAoB,CAAC;QACzB,SAAS,GAAG,IAAI,OAAO,CAAO,OAAO,CAAC,EAAE,GAAG,OAAO,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC;QACjE,MAAM,OAAO,GAAG,CAAC,KAAK,IAAI,EAAE;YAC1B,IAAI,CAAC;gBACH,MAAM,QAAQ,CAAC;gBACf,MAAM,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;gBAC5B,IAAI,OAAO;oBAAE,MAAM,OAAO,EAAE,CAAC;YAC/B,CAAC;oBAAS,CAAC;gBAAC,OAAO,EAAE,CAAC;YAAC,CAAC;YACxB,IAAI,OAAO;gBAAE,MAAM,IAAI,KAAK,CAAC,0DAA0D,CAAC,CAAC;YACzF,OAAO,IAAI,CAAC,GAAG,EAAE,CAAC;QACpB,CAAC,CAAC,EAAE,CAAC;QACL,4EAA4E;QAC5E,OAAO,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC,CAAC;QAC/B,4EAA4E;QAC5E,MAAM,OAAO,GAAG,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;QAChD,KAAK,MAAM,QAAQ,IAAI,IAAI,CAAC,SAAS;YAAE,OAAO,CAAC,GAAG,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;IACxE,CAAC;IACD,uFAAuF;IACvF,MAAM,SAAS,CAAC;IAChB,IAAI,OAAO;QAAE,MAAM,OAAO,CAAC,KAAK,CAAC;AACnC,CAAC"}

package/dist/training/evaluation/selectors.d.ts

@@ -0,0 +1,22 @@
+import type { EvaluationSnapshot, EvidenceReference } from '../contracts/evaluation.js';
+export interface Selection {
+    found: boolean;
+    value?: unknown;
+    evidence: EvidenceReference[];
+}
+export declare function textReference(kind: 'input' | 'output' | 'sidecar', id: string, text: string): EvidenceReference;
+export declare function ownPath(value: unknown, parts: readonly string[]): unknown;
+export declare function selectEvidence(selector: string, snapshot: EvaluationSnapshot): Selection;
+/**
+ * Whether an empty selection is an absence the trial observed, rather than an
+ * observation it could not make. Only an artifact is collected by the host at
+ * closure, so only its absence is something the run actually looked at: an empty
+ * `output.text` is itself incomplete coverage and can never certify anything. The
+ * trial must also be closed over a subject execution that is completed, host-settled and
+ * stop-proven, which is what `subjectUsage.settled` attests. Token accounting is a
+ * separate claim (`status`): a retried attempt can leave usage partial without making
+ * the closed artifact collection any less observed. A legacy capture without the
+ * attestation, like every uncertain run, is never an observed absence.
+ */
+export declare function observedAbsence(selector: string, snapshot: EvaluationSnapshot): boolean;
+export declare function evidenceText(value: unknown): string;

package/dist/training/evaluation/selectors.js

@@ -0,0 +1,61 @@
+import { createHash } from 'node:crypto';
+export function textReference(kind, id, text) {
+    return { kind, id, digest: `sha256:${createHash('sha256').update(text).digest('hex')}` };
+}
+export function ownPath(value, parts) {
+    let current = value;
+    for (const part of parts) {
+        if (part === '__proto__' || part === 'constructor' || part === 'prototype' ||
+            typeof current !== 'object' || current === null || !Object.hasOwn(current, part))
+            return undefined;
+        current = current[part];
+    }
+    return current;
+}
+export function selectEvidence(selector, snapshot) {
+    if (selector === 'output.text') {
+        const output = snapshot.coverage.filter(row => row.channel === 'output');
+        const found = output.length > 0 && output.every(row => row.completeness === 'complete');
+        return { found, ...(found ? { value: snapshot.output } : {}), evidence: found
+                ? [textReference('output', `${snapshot.trialId}/${snapshot.recordId}/output`, snapshot.output)] : [] };
+    }
+    if (selector === 'input:files')
+        return { found: true, value: snapshot.inputFiles,
+            evidence: Object.entries(snapshot.inputFiles).map(([id, value]) => textReference('input', id, value)) };
+    if (selector.startsWith('context:')) {
+        const value = ownPath(snapshot.context, [selector.slice(8)]);
+        return { found: value !== undefined, value, evidence: value === undefined ? []
+                : [textReference('sidecar', selector, evidenceText(value))] };
+    }
+    const match = /^artifact:([^#]+)(?:#(.*))?$/.exec(selector);
+    if (!match)
+        throw new Error(`Unsupported evidence selector ${selector}`);
+    const artifact = snapshot.artifacts[match[1]];
+    if (!artifact)
+        return { found: false, evidence: [] };
+    const pointer = match[2];
+    if (pointer && (!pointer.startsWith('/') || /~[^01]/.test(pointer) || pointer.endsWith('~'))) {
+        throw new Error('Invalid JSON pointer');
+    }
+    const value = pointer ? ownPath(artifact.value, pointer.slice(1).split('/').map(part => part.replace(/~1/g, '/').replace(/~0/g, '~'))) : artifact.value;
+    return { found: value !== undefined, value,
+        evidence: [{ kind: 'artifact', id: artifact.checkpointId, digest: artifact.digest }] };
+}
+/**
+ * Whether an empty selection is an absence the trial observed, rather than an
+ * observation it could not make. Only an artifact is collected by the host at
+ * closure, so only its absence is something the run actually looked at: an empty
+ * `output.text` is itself incomplete coverage and can never certify anything. The
+ * trial must also be closed over a subject execution that is completed, host-settled and
+ * stop-proven, which is what `subjectUsage.settled` attests. Token accounting is a
+ * separate claim (`status`): a retried attempt can leave usage partial without making
+ * the closed artifact collection any less observed. A legacy capture without the
+ * attestation, like every uncertain run, is never an observed absence.
+ */
+export function observedAbsence(selector, snapshot) {
+    return selector.startsWith('artifact:') && snapshot.closed && snapshot.subjectUsage?.settled === true;
+}
+export function evidenceText(value) {
+    return typeof value === 'string' ? value : JSON.stringify(value);
+}
+//# sourceMappingURL=selectors.js.map

package/dist/training/evaluation/selectors.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"selectors.js","sourceRoot":"","sources":["../../../src/training/evaluation/selectors.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAKzC,MAAM,UAAU,aAAa,CAAC,IAAoC,EAAE,EAAU,EAAE,IAAY;IAC1F,OAAO,EAAE,IAAI,EAAE,EAAE,EAAE,MAAM,EAAE,UAAU,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,EAAE,CAAC;AAC3F,CAAC;AAED,MAAM,UAAU,OAAO,CAAC,KAAc,EAAE,KAAwB;IAC9D,IAAI,OAAO,GAAG,KAAK,CAAC;IACpB,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,IAAI,IAAI,KAAK,WAAW,IAAI,IAAI,KAAK,aAAa,IAAI,IAAI,KAAK,WAAW;YACtE,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,IAAI,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,OAAO,EAAE,IAAI,CAAC;YAAE,OAAO,SAAS,CAAC;QACvG,OAAO,GAAI,OAAmC,CAAC,IAAI,CAAC,CAAC;IACvD,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,QAAgB,EAAE,QAA4B;IAC3E,IAAI,QAAQ,KAAK,aAAa,EAAE,CAAC;QAC/B,MAAM,MAAM,GAAG,QAAQ,CAAC,QAAQ,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC;QACzE,MAAM,KAAK,GAAG,MAAM,CAAC,MAAM,GAAG,CAAC,IAAI,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,YAAY,KAAK,UAAU,CAAC,CAAC;QACxF,OAAO,EAAE,KAAK,EAAE,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,QAAQ,EAAE,KAAK;gBAC3E,CAAC,CAAC,CAAC,aAAa,CAAC,QAAQ,EAAE,GAAG,QAAQ,CAAC,OAAO,IAAI,QAAQ,CAAC,QAAQ,SAAS,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;IAC3G,CAAC;IACD,IAAI,QAAQ,KAAK,aAAa;QAAE,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,QAAQ,CAAC,UAAU;YAC9E,QAAQ,EAAE,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,aAAa,CAAC,OAAO,EAAE,EAAE,EAAE,KAAK,CAAC,CAAC,EAAE,CAAC;IAC1G,IAAI,QAAQ,CAAC,UAAU,CAAC,UAAU,CAAC,EAAE,CAAC;QACpC,MAAM,KAAK,GAAG,OAAO,CAAC,QAAQ,CAAC,OAAO,EAAE,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QAC7D,OAAO,EAAE,KAAK,EAAE,KAAK,KAAK,SAAS,EAAE,KAAK,EAAE,QAAQ,EAAE,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE;gBAC5E,CAAC,CAAC,CAAC,aAAa,CAAC,SAAS,EAAE,QAAQ,EAAE,YAAY,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC;IAClE,CAAC;IACD,MAAM,KAAK,GAAG,8BAA8B,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IAC5D,IAAI,CAAC,KAAK;QAAE,MAAM,IAAI,KAAK,CAAC,iCAAiC,QAAQ,EAAE,CAAC,CAAC;IACzE,MAAM,QAAQ,GAAG,QAAQ,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,CAAE,CAAC,CAAC;IAC/C,IAAI,CAAC,QAAQ;QAAE,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC;IACrD,MAAM,OAAO,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;IACzB,IAAI,OAAO,IAAI,CAAC,CAAC,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC;QAC7F,MAAM,IAAI,KAAK,CAAC,sBAAsB,CAAC,CAAC;IAC1C,CAAC;IACD,MAAM,KAAK,GAAG,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,KAAK,EAC5C,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC;IAC1G,OAAO,EAAE,KAAK,EAAE,KAAK,KAAK,SAAS,EAAE,KAAK;QACxC,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,UAAU,EAAE,EAAE,EAAE,QAAQ,CAAC,YAAY,EAAE,MAAM,EAAE,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,CAAC;AAC3F,CAAC;AAED;;;;;;;;;;GAUG;AACH,MAAM,UAAU,eAAe,CAAC,QAAgB,EAAE,QAA4B;IAC5E,OAAO,QAAQ,CAAC,UAAU,CAAC,WAAW,CAAC,IAAI,QAAQ,CAAC,MAAM,IAAI,QAAQ,CAAC,YAAY,EAAE,OAAO,KAAK,IAAI,CAAC;AACxG,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,KAAc;IACzC,OAAO,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;AACnE,CAAC"}

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

@@ -0,0 +1,12 @@
+import type { JudgeCall } from '../../contracts/evaluation.js';
+import { type InferenceDiagnostic, type InferenceAttempt, type InferenceFailureKind } from '../../contracts/execution.js';
+/** Expand transport attempts into actual calls so retries cannot disappear from costs. */
+export declare function routedJudgeCalls(base: JudgeCall, attempts: readonly InferenceAttempt[]): JudgeCall[];
+/** The provider is a trusted host boundary; only its typed metadata is projected. */
+export declare function failedInferenceAttempts(error: unknown): readonly InferenceAttempt[];
+/** Only a trusted provider's typed discriminator; never infer policy from prose. */
+export declare function inferenceFailureKind(error: unknown): InferenceFailureKind | undefined;
+/** Validate and project only named accounting fields from a typed direct failure. */
+export declare function directFailureEvidence(error: unknown): Partial<JudgeCall>;
+/** Project only bounded trusted diagnostic fields; never forward arbitrary error objects. */
+export declare function directFailureDiagnostic(error: unknown): InferenceDiagnostic | undefined;

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

@@ -0,0 +1,71 @@
+import { inferenceDiagnosticCodes } from '../../contracts/execution.js';
+/** Expand transport attempts into actual calls so retries cannot disappear from costs. */
+export function routedJudgeCalls(base, attempts) {
+    return attempts.map((attempt, index) => ({
+        ...base, id: attempt.id, providerId: attempt.providerId,
+        requestedModel: attempt.requestedModel, responseModel: attempt.responseModel,
+        responseRoute: attempt.responseRoute, usage: structuredClone(attempt.usage),
+        outcome: attempt.outcome === 'failed' ? 'generation-error' : base.outcome,
+        ...(attempt.failureKind ? { failureKind: attempt.failureKind } : {}),
+        ...(index ? { fallbackOf: attempts[0].id } : {}),
+    }));
+}
+/** The provider is a trusted host boundary; only its typed metadata is projected. */
+export function failedInferenceAttempts(error) {
+    if (!(error instanceof Error) || !('attempts' in error) || !Array.isArray(error.attempts))
+        return [];
+    return error.attempts;
+}
+/** Only a trusted provider's typed discriminator; never infer policy from prose. */
+export function inferenceFailureKind(error) {
+    if (!(error instanceof Error) || !('kind' in error))
+        return undefined;
+    const kinds = ['quota', 'transport', 'unavailable', 'authentication', 'auth-stale',
+        'protocol', 'invalid-response', 'cancelled', 'timeout', 'unsupported', 'unknown'];
+    return kinds.includes(error.kind) ? error.kind : undefined;
+}
+/** Validate and project only named accounting fields from a typed direct failure. */
+export function directFailureEvidence(error) {
+    const failureKind = inferenceFailureKind(error);
+    if (!failureKind)
+        return {};
+    const value = error;
+    const text = (field) => typeof value[field] === 'string' && value[field].trim() ? value[field] : null;
+    const raw = value.usage;
+    let usage = null;
+    if (raw && typeof raw === 'object' && !Array.isArray(raw) && 'source' in raw && typeof raw.source === 'string' && raw.source.trim()) {
+        const counts = raw;
+        const count = (field) => typeof counts[field] === 'number' && Number.isSafeInteger(counts[field]) && counts[field] >= 0 ? counts[field] : null;
+        usage = { inputTokens: count('inputTokens'), cachedInputTokens: count('cachedInputTokens'),
+            outputTokens: count('outputTokens'), reasoningTokens: count('reasoningTokens'), source: raw.source };
+        for (const field of ['cacheWriteInputTokens', 'cacheWrite5mInputTokens', 'cacheWrite1hInputTokens']) {
+            if (Object.hasOwn(counts, field))
+                usage[field] = count(field);
+        }
+    }
+    return { failureKind, responseModel: text('responseModel'), responseRoute: text('responseRoute'), usage };
+}
+/** Project only bounded trusted diagnostic fields; never forward arbitrary error objects. */
+export function directFailureDiagnostic(error) {
+    if (!inferenceFailureKind(error) || !(error instanceof Error) || !('diagnostic' in error))
+        return undefined;
+    const value = error.diagnostic;
+    if (!value || typeof value !== 'object' || !('code' in value) || !inferenceDiagnosticCodes.includes(value.code))
+        return undefined;
+    const code = value.code;
+    if (!('response' in value))
+        return { code };
+    const response = value.response;
+    if (code !== 'response-schema' || !response || typeof response !== 'object')
+        return undefined;
+    const raw = response;
+    if (!Number.isSafeInteger(raw.bytes) || raw.bytes < 0 || typeof raw.sha256 !== 'string' || !/^sha256:[a-f0-9]{64}$/.test(raw.sha256))
+        return undefined;
+    if (raw.omitted === true && raw.bytes > 65_536 && raw.text === undefined) {
+        return { code, response: { bytes: raw.bytes, sha256: raw.sha256, omitted: true } };
+    }
+    if (raw.omitted !== false || typeof raw.text !== 'string' || Buffer.byteLength(raw.text) !== raw.bytes || raw.bytes > 65_536)
+        return undefined;
+    return { code, response: { bytes: raw.bytes, sha256: raw.sha256, omitted: false, text: raw.text } };
+}
+//# sourceMappingURL=attempts.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"attempts.js","sourceRoot":"","sources":["../../../../src/training/evaluation/semantic/attempts.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,wBAAwB,EAA0F,MAAM,8BAA8B,CAAC;AAEhK,0FAA0F;AAC1F,MAAM,UAAU,gBAAgB,CAAC,IAAe,EAAE,QAAqC;IACrF,OAAO,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,KAAK,EAAE,EAAE,CAAC,CAAC;QACvC,GAAG,IAAI,EAAE,EAAE,EAAE,OAAO,CAAC,EAAE,EAAE,UAAU,EAAE,OAAO,CAAC,UAAU;QACvD,cAAc,EAAE,OAAO,CAAC,cAAc,EAAE,aAAa,EAAE,OAAO,CAAC,aAAa;QAC5E,aAAa,EAAE,OAAO,CAAC,aAAa,EAAE,KAAK,EAAE,eAAe,CAAC,OAAO,CAAC,KAAK,CAAC;QAC3E,OAAO,EAAE,OAAO,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,kBAAkB,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO;QACzE,GAAG,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,OAAO,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACpE,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,QAAQ,CAAC,CAAC,CAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAClD,CAAC,CAAC,CAAC;AACN,CAAC;AAED,qFAAqF;AACrF,MAAM,UAAU,uBAAuB,CAAC,KAAc;IACpD,IAAI,CAAC,CAAC,KAAK,YAAY,KAAK,CAAC,IAAI,CAAC,CAAC,UAAU,IAAI,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,QAAQ,CAAC;QAAE,OAAO,EAAE,CAAC;IACrG,OAAO,KAAK,CAAC,QAA8B,CAAC;AAC9C,CAAC;AAED,oFAAoF;AACpF,MAAM,UAAU,oBAAoB,CAAC,KAAc;IACjD,IAAI,CAAC,CAAC,KAAK,YAAY,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,IAAI,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IACtE,MAAM,KAAK,GAAuB,CAAC,OAAO,EAAE,WAAW,EAAE,aAAa,EAAE,gBAAgB,EAAE,YAAY;QACpG,UAAU,EAAE,kBAAkB,EAAE,WAAW,EAAE,SAAS,EAAE,aAAa,EAAE,SAAS,CAAC,CAAC;IACpF,OAAO,KAAK,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,IAA4B,CAAC,CAAC,CAAC,SAAS,CAAC;AACrF,CAAC;AAED,qFAAqF;AACrF,MAAM,UAAU,qBAAqB,CAAC,KAAc;IAClD,MAAM,WAAW,GAAG,oBAAoB,CAAC,KAAK,CAAC,CAAC;IAChD,IAAI,CAAC,WAAW;QAAE,OAAO,EAAE,CAAC;IAC5B,MAAM,KAAK,GAAG,KAAwC,CAAC;IACvD,MAAM,IAAI,GAAG,CAAC,KAAa,EAAE,EAAE,CAAC,OAAO,KAAK,CAAC,KAAK,CAAC,KAAK,QAAQ,IAAI,KAAK,CAAC,KAAK,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAC9G,MAAM,GAAG,GAAG,KAAK,CAAC,KAAK,CAAC;IACxB,IAAI,KAAK,GAAiB,IAAI,CAAC;IAC/B,IAAI,GAAG,IAAI,OAAO,GAAG,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,IAAI,QAAQ,IAAI,GAAG,IAAI,OAAO,GAAG,CAAC,MAAM,KAAK,QAAQ,IAAI,GAAG,CAAC,MAAM,CAAC,IAAI,EAAE,EAAE,CAAC;QACpI,MAAM,MAAM,GAAG,GAA8B,CAAC;QAC9C,MAAM,KAAK,GAAG,CAAC,KAAa,EAAiB,EAAE,CAAC,OAAO,MAAM,CAAC,KAAK,CAAC,KAAK,QAAQ,IAAI,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;QACtK,KAAK,GAAG,EAAE,WAAW,EAAE,KAAK,CAAC,aAAa,CAAC,EAAE,iBAAiB,EAAE,KAAK,CAAC,mBAAmB,CAAC;YACxF,YAAY,EAAE,KAAK,CAAC,cAAc,CAAC,EAAE,eAAe,EAAE,KAAK,CAAC,iBAAiB,CAAC,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,CAAC;QACvG,KAAK,MAAM,KAAK,IAAI,CAAC,uBAAuB,EAAE,yBAAyB,EAAE,yBAAyB,CAAU,EAAE,CAAC;YAC7G,IAAI,MAAM,CAAC,MAAM,CAAC,MAAM,EAAE,KAAK,CAAC;gBAAE,KAAK,CAAC,KAAK,CAAC,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC;QAChE,CAAC;IACH,CAAC;IACD,OAAO,EAAE,WAAW,EAAE,aAAa,EAAE,IAAI,CAAC,eAAe,CAAC,EAAE,aAAa,EAAE,IAAI,CAAC,eAAe,CAAC,EAAE,KAAK,EAAE,CAAC;AAC5G,CAAC;AAED,6FAA6F;AAC7F,MAAM,UAAU,uBAAuB,CAAC,KAAc;IACpD,IAAI,CAAC,oBAAoB,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,KAAK,YAAY,KAAK,CAAC,IAAI,CAAC,CAAC,YAAY,IAAI,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAC5G,MAAM,KAAK,GAAG,KAAK,CAAC,UAAU,CAAC;IAC/B,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,CAAC,MAAM,IAAI,KAAK,CAAC,IAAI,CAAC,wBAAwB,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAa,CAAC;QAAE,OAAO,SAAS,CAAC;IAC3I,MAAM,IAAI,GAAG,KAAK,CAAC,IAAmC,CAAC;IACvD,IAAI,CAAC,CAAC,UAAU,IAAI,KAAK,CAAC;QAAE,OAAO,EAAE,IAAI,EAAE,CAAC;IAC5C,MAAM,QAAQ,GAAG,KAAK,CAAC,QAAQ,CAAC;IAChC,IAAI,IAAI,KAAK,iBAAiB,IAAI,CAAC,QAAQ,IAAI,OAAO,QAAQ,KAAK,QAAQ;QAAE,OAAO,SAAS,CAAC;IAC9F,MAAM,GAAG,GAAG,QAAmC,CAAC;IAChD,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,GAAG,CAAC,KAAK,CAAC,IAAK,GAAG,CAAC,KAAgB,GAAG,CAAC,IAAI,OAAO,GAAG,CAAC,MAAM,KAAK,QAAQ,IAAI,CAAC,uBAAuB,CAAC,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC;QAAE,OAAO,SAAS,CAAC;IACnK,IAAI,GAAG,CAAC,OAAO,KAAK,IAAI,IAAK,GAAG,CAAC,KAAgB,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;QACrF,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,KAAK,EAAE,GAAG,CAAC,KAAe,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,EAAE,CAAC;IAC/F,CAAC;IACD,IAAI,GAAG,CAAC,OAAO,KAAK,KAAK,IAAI,OAAO,GAAG,CAAC,IAAI,KAAK,QAAQ,IAAI,MAAM,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,KAAK,IAAK,GAAG,CAAC,KAAgB,GAAG,MAAM;QAAE,OAAO,SAAS,CAAC;IAC3J,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,KAAK,EAAE,GAAG,CAAC,KAAe,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,GAAG,CAAC,IAAI,EAAE,EAAE,CAAC;AAChH,CAAC"}

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

@@ -0,0 +1,9 @@
+import type { CachedJudgment, JudgeCache } from '../../contracts/evaluation.js';
+/** Invocation-local LRU. A hit is reused evidence, never another inference or trial. */
+export declare function createJudgeCache({ maxEntries }: {
+    maxEntries: number;
+}): Omit<JudgeCache, 'get' | 'put'> & {
+    get(key: string): CachedJudgment | undefined;
+    put(key: string, value: CachedJudgment): void;
+};
+export declare function judgmentCacheKey(identity: string, system: string, prompt: string, minimum: number, scope?: string): string;

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

@@ -0,0 +1,40 @@
+import { JUDGMENT_CACHE_DOMAIN } from '../../contracts/index.js';
+import { createHash } from 'node:crypto';
+/** Invocation-local LRU. A hit is reused evidence, never another inference or trial. */
+export function createJudgeCache({ maxEntries }) {
+    if (!Number.isSafeInteger(maxEntries) || maxEntries < 1 || maxEntries > 10000)
+        throw new Error('Invalid judge cache entry bound');
+    const entries = new Map();
+    let bytes = 0;
+    const size = (key, value) => Buffer.byteLength(key + JSON.stringify(value));
+    return {
+        get(key) {
+            const found = entries.get(key);
+            if (!found)
+                return undefined;
+            entries.delete(key);
+            entries.set(key, found);
+            return structuredClone(found);
+        },
+        put(key, value) {
+            const old = entries.get(key);
+            if (old)
+                bytes -= size(key, old);
+            entries.delete(key);
+            const length = size(key, value);
+            if (length > 32 * 1024 * 1024)
+                return;
+            entries.set(key, structuredClone(value));
+            bytes += length;
+            while (entries.size > maxEntries || bytes > 32 * 1024 * 1024) {
+                const first = entries.keys().next().value;
+                bytes -= size(first, entries.get(first));
+                entries.delete(first);
+            }
+        },
+    };
+}
+export function judgmentCacheKey(identity, system, prompt, minimum, scope = '') {
+    return `sha256:${createHash('sha256').update(JSON.stringify([JUDGMENT_CACHE_DOMAIN, scope, identity, system, prompt, minimum])).digest('hex')}`;
+}
+//# sourceMappingURL=cache.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"cache.js","sourceRoot":"","sources":["../../../../src/training/evaluation/semantic/cache.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,qBAAqB,EAAE,MAAM,0BAA0B,CAAC;AACjE,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAGzC,wFAAwF;AACxF,MAAM,UAAU,gBAAgB,CAAC,EAAE,UAAU,EAA0B;IAIrE,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,UAAU,CAAC,IAAI,UAAU,GAAG,CAAC,IAAI,UAAU,GAAG,KAAK;QAAE,MAAM,IAAI,KAAK,CAAC,iCAAiC,CAAC,CAAC;IAClI,MAAM,OAAO,GAAG,IAAI,GAAG,EAA0B,CAAC;IAClD,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,MAAM,IAAI,GAAG,CAAC,GAAW,EAAE,KAAqB,EAAE,EAAE,CAAC,MAAM,CAAC,UAAU,CAAC,GAAG,GAAG,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,CAAC;IACpG,OAAO;QACL,GAAG,CAAC,GAAG;YACL,MAAM,KAAK,GAAG,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;YAC/B,IAAI,CAAC,KAAK;gBAAE,OAAO,SAAS,CAAC;YAC7B,OAAO,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAAC,OAAO,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;YAC7C,OAAO,eAAe,CAAC,KAAK,CAAC,CAAC;QAChC,CAAC;QACD,GAAG,CAAC,GAAG,EAAE,KAAK;YACZ,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;YAC7B,IAAI,GAAG;gBAAE,KAAK,IAAI,IAAI,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC;YACjC,OAAO,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YACpB,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;YAChC,IAAI,MAAM,GAAG,EAAE,GAAG,IAAI,GAAG,IAAI;gBAAE,OAAO;YACtC,OAAO,CAAC,GAAG,CAAC,GAAG,EAAE,eAAe,CAAC,KAAK,CAAC,CAAC,CAAC;YACzC,KAAK,IAAI,MAAM,CAAC;YAChB,OAAO,OAAO,CAAC,IAAI,GAAG,UAAU,IAAI,KAAK,GAAG,EAAE,GAAG,IAAI,GAAG,IAAI,EAAE,CAAC;gBAC7D,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,EAAE,CAAC,IAAI,EAAE,CAAC,KAAM,CAAC;gBAC3C,KAAK,IAAI,IAAI,CAAC,KAAK,EAAE,OAAO,CAAC,GAAG,CAAC,KAAK,CAAE,CAAC,CAAC;gBAC1C,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;YACxB,CAAC;QACH,CAAC;KACF,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,gBAAgB,CAAC,QAAgB,EAAE,MAAc,EAAE,MAAc,EAAE,OAAe,EAAE,KAAK,GAAG,EAAE;IAC5G,OAAO,UAAU,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,qBAAqB,EAAE,KAAK,EAAE,QAAQ,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC;AAClJ,CAAC"}

package/dist/training/evaluation/semantic/evidence-units.d.ts

@@ -0,0 +1,14 @@
+export interface EvidenceUnit {
+    id: string;
+    source: string;
+    pointer: string;
+    chunk: number;
+    digest: string;
+    value: string | number | boolean | null | unknown[] | Record<string, never>;
+}
+export interface EvidencePacket {
+    schema: 'spawnfile.training.evidence-units.v1';
+    units: EvidenceUnit[];
+}
+/** Lossless, deterministic primitive leaves. IDs attest pointers, never factual support. */
+export declare function evidenceUnits(sources: Readonly<Record<string, unknown>>): EvidencePacket;