@aldus-runtime/regression 0.1.0

package/dist/corpus.d.ts

@@ -0,0 +1,184 @@
+/**
+ * The defect corpus (architecture contract §22 WP-10, §12.1, §12.3, §24).
+ *
+ * §12.1 permits an evaluator to become blocking "only after it is calibrated against
+ * human-labeled examples", and §24 requires that "a representative defect corpus is executed
+ * during regression testing". This module defines what such a corpus is.
+ *
+ * Two shapes here are deliberately open rather than enumerated:
+ *
+ * - **Finding categories.** §12.3 presents its taxonomy as "for example", and §4.2 keeps
+ *   adopter concepts out of the runtime. A category is an opaque caller-supplied string.
+ * - **Severity levels.** Likewise named by the caller. The *weights* live in the promotion
+ *   policy, so an adopter can say a missed pronunciation error costs less than a missed
+ *   unsupported claim without Core deciding that for them.
+ *
+ * The human label is the oracle (§12 level 4). Nothing in this module treats an evaluator's
+ * output as ground truth.
+ *
+ * Zod is the single source of truth; TypeScript types are inferred (ADR-0002).
+ */
+import { z } from "zod";
+/**
+ * A finding category, structured per contract §12.3.
+ *
+ * An OPEN string, never a Core-defined enum. §12.3 introduces its taxonomy with "for example",
+ * and §4.2 forbids Core from owning an adopter's diagnosis vocabulary. Do not narrow this to a
+ * union.
+ */
+export declare const findingCategory: z.ZodString;
+/**
+ * A severity level, named by the caller.
+ *
+ * An OPEN string for the same reason as {@link findingCategory}. Ordering and weighting are
+ * supplied by the promotion policy, not inferred from the name — a package that decided
+ * `"critical"` outranks `"major"` would be guessing at an adopter's scale.
+ */
+export declare const severityLevel: z.ZodString;
+/**
+ * Scope dimensions a case belongs to (contract §12.1 "show, host, voice, model, and
+ * script-form scope").
+ *
+ * `Record<string, string>`, consistent with Knowledge Pack scope (§9.2, ADR-0006). §12.1's list
+ * is illustrative and §4.2 forbids naming a provider, so dimensions stay caller-supplied.
+ */
+export declare const scopeDimensions: z.ZodRecord<z.ZodString, z.ZodString>;
+/** @see scopeDimensions */
+export type ScopeDimensions = z.infer<typeof scopeDimensions>;
+/** One defect a human labeller identified in a case. */
+export declare const humanFindingSchema: z.ZodObject<{
+    category: z.ZodString;
+    severity: z.ZodString;
+    note: z.ZodOptional<z.ZodString>;
+}, z.core.$strip>;
+/** @see humanFindingSchema */
+export type HumanFinding = z.infer<typeof humanFindingSchema>;
+/**
+ * One labelled case: an input, and what a human said about it.
+ *
+ * The case does not carry the input itself. §8.1 makes artifacts addressable by ID and hash, and
+ * a corpus that embedded audio or scripts would be unreviewable and would risk carrying private
+ * source material into a test fixture (§19.2). `subjectRef` points at the subject; what it
+ * points into is the caller's business.
+ */
+export declare const defectCaseSchema: z.ZodObject<{
+    caseId: z.ZodString;
+    subjectRef: z.ZodString;
+    scope: z.ZodRecord<z.ZodString, z.ZodString>;
+    defective: z.ZodBoolean;
+    findings: z.ZodArray<z.ZodObject<{
+        category: z.ZodString;
+        severity: z.ZodString;
+        note: z.ZodOptional<z.ZodString>;
+    }, z.core.$strip>>;
+    severity: z.ZodOptional<z.ZodString>;
+    correctionOnFlag: z.ZodOptional<z.ZodString>;
+    labelledBy: z.ZodObject<{
+        kind: z.ZodEnum<{
+            agent: "agent";
+            human: "human";
+            system: "system";
+            worker: "worker";
+        }>;
+        id: z.ZodString;
+        displayName: z.ZodOptional<z.ZodString>;
+        backendId: z.ZodOptional<z.ZodString>;
+        sessionRef: z.ZodOptional<z.ZodString>;
+    }, z.core.$strip>;
+    labelledAt: z.ZodISODateTime;
+}, z.core.$strip>;
+/** @see defectCaseSchema */
+export type DefectCase = z.infer<typeof defectCaseSchema>;
+/** A named, versioned set of labelled cases. */
+export declare const defectCorpusSchema: z.ZodObject<{
+    schemaVersion: z.ZodString;
+    corpusId: z.ZodString;
+    description: z.ZodOptional<z.ZodString>;
+    cases: z.ZodArray<z.ZodObject<{
+        caseId: z.ZodString;
+        subjectRef: z.ZodString;
+        scope: z.ZodRecord<z.ZodString, z.ZodString>;
+        defective: z.ZodBoolean;
+        findings: z.ZodArray<z.ZodObject<{
+            category: z.ZodString;
+            severity: z.ZodString;
+            note: z.ZodOptional<z.ZodString>;
+        }, z.core.$strip>>;
+        severity: z.ZodOptional<z.ZodString>;
+        correctionOnFlag: z.ZodOptional<z.ZodString>;
+        labelledBy: z.ZodObject<{
+            kind: z.ZodEnum<{
+                agent: "agent";
+                human: "human";
+                system: "system";
+                worker: "worker";
+            }>;
+            id: z.ZodString;
+            displayName: z.ZodOptional<z.ZodString>;
+            backendId: z.ZodOptional<z.ZodString>;
+            sessionRef: z.ZodOptional<z.ZodString>;
+        }, z.core.$strip>;
+        labelledAt: z.ZodISODateTime;
+    }, z.core.$strip>>;
+}, z.core.$strip>;
+/** @see defectCorpusSchema */
+export type DefectCorpus = z.infer<typeof defectCorpusSchema>;
+/** One finding an evaluator reported. */
+export declare const evaluatorFindingSchema: z.ZodObject<{
+    category: z.ZodString;
+    severity: z.ZodOptional<z.ZodString>;
+    confidence: z.ZodOptional<z.ZodNumber>;
+}, z.core.$strip>;
+/** @see evaluatorFindingSchema */
+export type EvaluatorFinding = z.infer<typeof evaluatorFindingSchema>;
+/** What an evaluator said about one case. */
+export declare const evaluatorOutcomeSchema: z.ZodObject<{
+    caseId: z.ZodString;
+    flagged: z.ZodBoolean;
+    findings: z.ZodArray<z.ZodObject<{
+        category: z.ZodString;
+        severity: z.ZodOptional<z.ZodString>;
+        confidence: z.ZodOptional<z.ZodNumber>;
+    }, z.core.$strip>>;
+}, z.core.$strip>;
+/** @see evaluatorOutcomeSchema */
+export type EvaluatorOutcome = z.infer<typeof evaluatorOutcomeSchema>;
+/**
+ * One evaluator's outcomes over one corpus.
+ *
+ * This package does **not** run evaluators (contract §22 WP-10 scope). A run is produced
+ * elsewhere and handed here for comparison.
+ */
+export declare const evaluatorRunSchema: z.ZodObject<{
+    schemaVersion: z.ZodString;
+    evaluatorId: z.ZodString;
+    evaluatorVersion: z.ZodString;
+    corpusId: z.ZodString;
+    outcomes: z.ZodArray<z.ZodObject<{
+        caseId: z.ZodString;
+        flagged: z.ZodBoolean;
+        findings: z.ZodArray<z.ZodObject<{
+            category: z.ZodString;
+            severity: z.ZodOptional<z.ZodString>;
+            confidence: z.ZodOptional<z.ZodNumber>;
+        }, z.core.$strip>>;
+    }, z.core.$strip>>;
+    executedAt: z.ZodISODateTime;
+}, z.core.$strip>;
+/** @see evaluatorRunSchema */
+export type EvaluatorRun = z.infer<typeof evaluatorRunSchema>;
+/** Schema version this package stamps on records it constructs. */
+export declare const REGRESSION_SCHEMA_VERSION = "1.2";
+/**
+ * Validate a corpus and check the uniqueness constraint Zod cannot express.
+ *
+ * @throws {AldusError} `ALDUS_CORPUS_MALFORMED` or `ALDUS_CORPUS_DUPLICATE_CASE`.
+ */
+export declare function parseDefectCorpus(input: unknown): DefectCorpus;
+/**
+ * Validate an evaluator run and check uniqueness.
+ *
+ * @throws {AldusError} `ALDUS_CORPUS_MALFORMED` or `ALDUS_OUTCOME_DUPLICATE`.
+ */
+export declare function parseEvaluatorRun(input: unknown): EvaluatorRun;
+//# sourceMappingURL=corpus.d.ts.map

package/dist/corpus.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"corpus.d.ts","sourceRoot":"","sources":["../src/corpus.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAGH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAOxB;;;;;;GAMG;AACH,eAAO,MAAM,eAAe,aAA6B,CAAC;AAE1D;;;;;;GAMG;AACH,eAAO,MAAM,aAAa,aAA6B,CAAC;AAExD;;;;;;GAMG;AACH,eAAO,MAAM,eAAe,uCAAmE,CAAC;AAEhG,2BAA2B;AAC3B,MAAM,MAAM,eAAe,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,eAAe,CAAC,CAAC;AAE9D,wDAAwD;AACxD,eAAO,MAAM,kBAAkB;;;;iBASuB,CAAC;AAEvD,8BAA8B;AAC9B,MAAM,MAAM,YAAY,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,kBAAkB,CAAC,CAAC;AAE9D;;;;;;;GAOG;AACH,eAAO,MAAM,gBAAgB;;;;;;;;;;;;;;;;;;;;;;;;;iBA2DzB,CAAC;AAEL,4BAA4B;AAC5B,MAAM,MAAM,UAAU,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,gBAAgB,CAAC,CAAC;AAE1D,gDAAgD;AAChD,eAAO,MAAM,kBAAkB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAkB3B,CAAC;AAEL,8BAA8B;AAC9B,MAAM,MAAM,YAAY,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,kBAAkB,CAAC,CAAC;AAE9D,yCAAyC;AACzC,eAAO,MAAM,sBAAsB;;;;iBAS2B,CAAC;AAE/D,kCAAkC;AAClC,MAAM,MAAM,gBAAgB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,sBAAsB,CAAC,CAAC;AAEtE,6CAA6C;AAC7C,eAAO,MAAM,sBAAsB;;;;;;;;iBAS2B,CAAC;AAE/D,kCAAkC;AAClC,MAAM,MAAM,gBAAgB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,sBAAsB,CAAC,CAAC;AAEtE;;;;;GAKG;AACH,eAAO,MAAM,kBAAkB;;;;;;;;;;;;;;;iBAsB3B,CAAC;AAEL,8BAA8B;AAC9B,MAAM,MAAM,YAAY,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,kBAAkB,CAAC,CAAC;AAE9D,mEAAmE;AACnE,eAAO,MAAM,yBAAyB,QAAiB,CAAC;AAExD;;;;GAIG;AACH,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,OAAO,GAAG,YAAY,CAgC9D;AAED;;;;GAIG;AACH,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,OAAO,GAAG,YAAY,CA8B9D"}

package/dist/corpus.js

@@ -0,0 +1,254 @@
+/**
+ * The defect corpus (architecture contract §22 WP-10, §12.1, §12.3, §24).
+ *
+ * §12.1 permits an evaluator to become blocking "only after it is calibrated against
+ * human-labeled examples", and §24 requires that "a representative defect corpus is executed
+ * during regression testing". This module defines what such a corpus is.
+ *
+ * Two shapes here are deliberately open rather than enumerated:
+ *
+ * - **Finding categories.** §12.3 presents its taxonomy as "for example", and §4.2 keeps
+ *   adopter concepts out of the runtime. A category is an opaque caller-supplied string.
+ * - **Severity levels.** Likewise named by the caller. The *weights* live in the promotion
+ *   policy, so an adopter can say a missed pronunciation error costs less than a missed
+ *   unsupported claim without Core deciding that for them.
+ *
+ * The human label is the oracle (§12 level 4). Nothing in this module treats an evaluator's
+ * output as ground truth.
+ *
+ * Zod is the single source of truth; TypeScript types are inferred (ADR-0002).
+ */
+import { actorRefSchema, SCHEMA_VERSION } from "@aldus-runtime/core";
+import { z } from "zod";
+import { RegressionErrorCodes, regressionError } from "./errors.js";
+/** A short opaque identifier. */
+const identifier = z.string().min(1).max(200);
+/**
+ * A finding category, structured per contract §12.3.
+ *
+ * An OPEN string, never a Core-defined enum. §12.3 introduces its taxonomy with "for example",
+ * and §4.2 forbids Core from owning an adopter's diagnosis vocabulary. Do not narrow this to a
+ * union.
+ */
+export const findingCategory = z.string().min(1).max(200);
+/**
+ * A severity level, named by the caller.
+ *
+ * An OPEN string for the same reason as {@link findingCategory}. Ordering and weighting are
+ * supplied by the promotion policy, not inferred from the name — a package that decided
+ * `"critical"` outranks `"major"` would be guessing at an adopter's scale.
+ */
+export const severityLevel = z.string().min(1).max(100);
+/**
+ * Scope dimensions a case belongs to (contract §12.1 "show, host, voice, model, and
+ * script-form scope").
+ *
+ * `Record<string, string>`, consistent with Knowledge Pack scope (§9.2, ADR-0006). §12.1's list
+ * is illustrative and §4.2 forbids naming a provider, so dimensions stay caller-supplied.
+ */
+export const scopeDimensions = z.record(z.string().min(1).max(100), z.string().min(1).max(200));
+/** One defect a human labeller identified in a case. */
+export const humanFindingSchema = z
+    .object({
+    /** Category from the adopter's taxonomy (contract §12.3). */
+    category: findingCategory,
+    /** Severity from the adopter's scale; weighted by the promotion policy. */
+    severity: severityLevel,
+    /** What the labeller saw. Prose, never parsed. */
+    note: z.string().max(4000).optional(),
+})
+    .meta({ id: "HumanFinding", title: "HumanFinding" });
+/**
+ * One labelled case: an input, and what a human said about it.
+ *
+ * The case does not carry the input itself. §8.1 makes artifacts addressable by ID and hash, and
+ * a corpus that embedded audio or scripts would be unreviewable and would risk carrying private
+ * source material into a test fixture (§19.2). `subjectRef` points at the subject; what it
+ * points into is the caller's business.
+ */
+export const defectCaseSchema = z
+    .object({
+    /** Identity of this case within its corpus. */
+    caseId: identifier,
+    /** Reference to the material under test — an artifact ID, hash, or adopter-defined locator. */
+    subjectRef: z.string().min(1).max(1024),
+    /** Scope this case belongs to (contract §12.1). */
+    scope: scopeDimensions,
+    /**
+     * Whether a human judged the subject defective. **This is the oracle** (§12 level 4).
+     *
+     * Stored separately from `findings` so a case can be labelled clean explicitly. An empty
+     * findings array on a case nobody reviewed is not the same claim as "a human looked and
+     * found nothing", and conflating them would inflate the true-negative count with unreviewed
+     * material.
+     */
+    defective: z.boolean(),
+    /** What the labeller found. Empty when `defective` is false. */
+    findings: z.array(humanFindingSchema).max(256),
+    /**
+     * Overall severity of the case, used for severity-weighted metrics (contract §12.1).
+     *
+     * Required when `defective`, absent otherwise — enforced by refinement below.
+     */
+    severity: severityLevel.optional(),
+    /**
+     * Correction class this case would trigger if a blocking evaluator flagged it.
+     *
+     * §12.1 names "asymmetric harm caused by unnecessary automatic correction" as its own
+     * consideration, and §12.4 makes the repair layer explicit — regenerating one TTS segment
+     * and revising narration with cascading approval invalidation are not the same act. This
+     * names which one a flag would trigger here; the policy assigns the harm weight. Absent
+     * means the policy's default class applies.
+     */
+    correctionOnFlag: identifier.optional(),
+    /** Who labelled the case (contract §19.2: a decision without an actor is not a decision). */
+    labelledBy: actorRefSchema,
+    /** When the label was recorded. */
+    labelledAt: z.iso.datetime({ offset: true }),
+})
+    .refine((value) => value.defective === (value.severity !== undefined), {
+    message: "a defective case must carry a severity, and a clean case must not (architecture contract §12.1 requires severity-weighted metrics)",
+    path: ["severity"],
+})
+    .refine((value) => value.defective || value.findings.length === 0, {
+    message: "a case labelled clean must carry no findings",
+    path: ["findings"],
+})
+    .meta({
+    id: "DefectCase",
+    title: "DefectCase",
+    description: "One human-labelled case in a defect corpus (architecture contract §12.1, §24). The human " +
+        "label is the oracle; nothing treats an evaluator's output as ground truth. `category` and " +
+        "`severity` are open strings because §12.3's taxonomy is illustrative and §4.2 keeps " +
+        "adopter vocabularies out of the runtime. ADDITIONAL CONSTRAINTS NOT EXPRESSIBLE IN JSON " +
+        "SCHEMA: a defective case must carry a severity and a clean case must not, and a clean " +
+        "case must carry no findings.",
+});
+/** A named, versioned set of labelled cases. */
+export const defectCorpusSchema = z
+    .object({
+    /** Schema version of this record (ADR-0003). */
+    schemaVersion: z.string().regex(/^(0|[1-9]\d*)\.(0|[1-9]\d*)$/),
+    /** Identity of this corpus. */
+    corpusId: identifier,
+    /** Human-readable purpose. */
+    description: z.string().max(2000).optional(),
+    /** The labelled cases. */
+    cases: z.array(defectCaseSchema).max(100_000),
+})
+    .meta({
+    id: "DefectCorpus",
+    title: "DefectCorpus",
+    description: "A defect corpus: human-labelled cases an evaluator is calibrated against (architecture " +
+        "contract §12.1, §24). ADDITIONAL CONSTRAINT NOT EXPRESSIBLE IN JSON SCHEMA: `caseId` " +
+        "must be unique within the corpus.",
+});
+/** One finding an evaluator reported. */
+export const evaluatorFindingSchema = z
+    .object({
+    /** Category the evaluator assigned, from the same taxonomy as {@link humanFindingSchema}. */
+    category: findingCategory,
+    /** Severity the evaluator assigned, where it assigns one. */
+    severity: severityLevel.optional(),
+    /** Evaluator confidence in `[0, 1]`, where it reports one. */
+    confidence: z.number().min(0).max(1).optional(),
+})
+    .meta({ id: "EvaluatorFinding", title: "EvaluatorFinding" });
+/** What an evaluator said about one case. */
+export const evaluatorOutcomeSchema = z
+    .object({
+    /** Case this outcome reports on. */
+    caseId: identifier,
+    /** Whether the evaluator flagged the subject. */
+    flagged: z.boolean(),
+    /** What it reported. Empty when not flagged. */
+    findings: z.array(evaluatorFindingSchema).max(256),
+})
+    .meta({ id: "EvaluatorOutcome", title: "EvaluatorOutcome" });
+/**
+ * One evaluator's outcomes over one corpus.
+ *
+ * This package does **not** run evaluators (contract §22 WP-10 scope). A run is produced
+ * elsewhere and handed here for comparison.
+ */
+export const evaluatorRunSchema = z
+    .object({
+    /** Schema version of this record (ADR-0003). */
+    schemaVersion: z.string().regex(/^(0|[1-9]\d*)\.(0|[1-9]\d*)$/),
+    /** Which evaluator produced these outcomes. */
+    evaluatorId: identifier,
+    /** Version of the evaluator. Metrics from different versions are not comparable. */
+    evaluatorVersion: identifier,
+    /** Corpus the outcomes are against. */
+    corpusId: identifier,
+    /** The outcomes. */
+    outcomes: z.array(evaluatorOutcomeSchema).max(100_000),
+    /** When the run was executed. */
+    executedAt: z.iso.datetime({ offset: true }),
+})
+    .meta({
+    id: "EvaluatorRun",
+    title: "EvaluatorRun",
+    description: "One evaluator's outcomes over one defect corpus (architecture contract §22 WP-10). This " +
+        "package consumes runs; it does not execute evaluators. ADDITIONAL CONSTRAINT NOT " +
+        "EXPRESSIBLE IN JSON SCHEMA: `caseId` must be unique within `outcomes`.",
+});
+/** Schema version this package stamps on records it constructs. */
+export const REGRESSION_SCHEMA_VERSION = SCHEMA_VERSION;
+/**
+ * Validate a corpus and check the uniqueness constraint Zod cannot express.
+ *
+ * @throws {AldusError} `ALDUS_CORPUS_MALFORMED` or `ALDUS_CORPUS_DUPLICATE_CASE`.
+ */
+export function parseDefectCorpus(input) {
+    const result = defectCorpusSchema.safeParse(input);
+    if (!result.success) {
+        // Paths and issue codes only — never the received value (contract §19.2, ADR-0002).
+        throw regressionError(RegressionErrorCodes.CORPUS_MALFORMED, "The defect corpus does not validate.", {
+            category: "validation",
+            details: {
+                issues: result.error.issues.map((issue) => ({
+                    path: issue.path.join("."),
+                    code: issue.code,
+                })),
+            },
+        });
+    }
+    const seen = new Set();
+    for (const entry of result.data.cases) {
+        if (seen.has(entry.caseId)) {
+            throw regressionError(RegressionErrorCodes.CORPUS_DUPLICATE_CASE, `Corpus "${result.data.corpusId}" contains more than one case with id "${entry.caseId}". ` +
+                "Duplicate ids would let one case be counted twice and silently reweight every metric.", { category: "validation", details: { corpusId: result.data.corpusId } });
+        }
+        seen.add(entry.caseId);
+    }
+    return result.data;
+}
+/**
+ * Validate an evaluator run and check uniqueness.
+ *
+ * @throws {AldusError} `ALDUS_CORPUS_MALFORMED` or `ALDUS_OUTCOME_DUPLICATE`.
+ */
+export function parseEvaluatorRun(input) {
+    const result = evaluatorRunSchema.safeParse(input);
+    if (!result.success) {
+        throw regressionError(RegressionErrorCodes.CORPUS_MALFORMED, "The evaluator run does not validate.", {
+            category: "validation",
+            details: {
+                issues: result.error.issues.map((issue) => ({
+                    path: issue.path.join("."),
+                    code: issue.code,
+                })),
+            },
+        });
+    }
+    const seen = new Set();
+    for (const outcome of result.data.outcomes) {
+        if (seen.has(outcome.caseId)) {
+            throw regressionError(RegressionErrorCodes.OUTCOME_DUPLICATE, `Evaluator run for "${result.data.evaluatorId}" reports twice on case "${outcome.caseId}".`, { category: "validation", details: { evaluatorId: result.data.evaluatorId } });
+        }
+        seen.add(outcome.caseId);
+    }
+    return result.data;
+}
+//# sourceMappingURL=corpus.js.map

package/dist/corpus.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"corpus.js","sourceRoot":"","sources":["../src/corpus.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAEH,OAAO,EAAE,cAAc,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AACrE,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAAE,oBAAoB,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AAEpE,iCAAiC;AACjC,MAAM,UAAU,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;AAE9C;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,eAAe,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;AAE1D;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,aAAa,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;AAExD;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,eAAe,GAAG,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC;AAKhG,wDAAwD;AACxD,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC;KAChC,MAAM,CAAC;IACN,6DAA6D;IAC7D,QAAQ,EAAE,eAAe;IACzB,2EAA2E;IAC3E,QAAQ,EAAE,aAAa;IACvB,kDAAkD;IAClD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,QAAQ,EAAE;CACtC,CAAC;KACD,IAAI,CAAC,EAAE,EAAE,EAAE,cAAc,EAAE,KAAK,EAAE,cAAc,EAAE,CAAC,CAAC;AAKvD;;;;;;;GAOG;AACH,MAAM,CAAC,MAAM,gBAAgB,GAAG,CAAC;KAC9B,MAAM,CAAC;IACN,+CAA+C;IAC/C,MAAM,EAAE,UAAU;IAClB,+FAA+F;IAC/F,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC;IACvC,mDAAmD;IACnD,KAAK,EAAE,eAAe;IACtB;;;;;;;OAOG;IACH,SAAS,EAAE,CAAC,CAAC,OAAO,EAAE;IACtB,gEAAgE;IAChE,QAAQ,EAAE,CAAC,CAAC,KAAK,CAAC,kBAAkB,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC;IAC9C;;;;OAIG;IACH,QAAQ,EAAE,aAAa,CAAC,QAAQ,EAAE;IAClC;;;;;;;;OAQG;IACH,gBAAgB,EAAE,UAAU,CAAC,QAAQ,EAAE;IACvC,6FAA6F;IAC7F,UAAU,EAAE,cAAc;IAC1B,mCAAmC;IACnC,UAAU,EAAE,CAAC,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC;CAC7C,CAAC;KACD,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,SAAS,KAAK,CAAC,KAAK,CAAC,QAAQ,KAAK,SAAS,CAAC,EAAE;IACrE,OAAO,EACL,oIAAoI;IACtI,IAAI,EAAE,CAAC,UAAU,CAAC;CACnB,CAAC;KACD,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,SAAS,IAAI,KAAK,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE;IACjE,OAAO,EAAE,8CAA8C;IACvD,IAAI,EAAE,CAAC,UAAU,CAAC;CACnB,CAAC;KACD,IAAI,CAAC;IACJ,EAAE,EAAE,YAAY;IAChB,KAAK,EAAE,YAAY;IACnB,WAAW,EACT,2FAA2F;QAC3F,4FAA4F;QAC5F,sFAAsF;QACtF,0FAA0F;QAC1F,wFAAwF;QACxF,8BAA8B;CACjC,CAAC,CAAC;AAKL,gDAAgD;AAChD,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC;KAChC,MAAM,CAAC;IACN,gDAAgD;IAChD,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,KAAK,CAAC,8BAA8B,CAAC;IAC/D,+BAA+B;IAC/B,QAAQ,EAAE,UAAU;IACpB,8BAA8B;IAC9B,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,QAAQ,EAAE;IAC5C,0BAA0B;IAC1B,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,gBAAgB,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC;CAC9C,CAAC;KACD,IAAI,CAAC;IACJ,EAAE,EAAE,cAAc;IAClB,KAAK,EAAE,cAAc;IACrB,WAAW,EACT,yFAAyF;QACzF,uFAAuF;QACvF,mCAAmC;CACtC,CAAC,CAAC;AAKL,yCAAyC;AACzC,MAAM,CAAC,MAAM,sBAAsB,GAAG,CAAC;KACpC,MAAM,CAAC;IACN,6FAA6F;IAC7F,QAAQ,EAAE,eAAe;IACzB,6DAA6D;IAC7D,QAAQ,EAAE,aAAa,CAAC,QAAQ,EAAE;IAClC,8DAA8D;IAC9D,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;CAChD,CAAC;KACD,IAAI,CAAC,EAAE,EAAE,EAAE,kBAAkB,EAAE,KAAK,EAAE,kBAAkB,EAAE,CAAC,CAAC;AAK/D,6CAA6C;AAC7C,MAAM,CAAC,MAAM,sBAAsB,GAAG,CAAC;KACpC,MAAM,CAAC;IACN,oCAAoC;IACpC,MAAM,EAAE,UAAU;IAClB,iDAAiD;IACjD,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE;IACpB,gDAAgD;IAChD,QAAQ,EAAE,CAAC,CAAC,KAAK,CAAC,sBAAsB,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC;CACnD,CAAC;KACD,IAAI,CAAC,EAAE,EAAE,EAAE,kBAAkB,EAAE,KAAK,EAAE,kBAAkB,EAAE,CAAC,CAAC;AAK/D;;;;;GAKG;AACH,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC;KAChC,MAAM,CAAC;IACN,gDAAgD;IAChD,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,KAAK,CAAC,8BAA8B,CAAC;IAC/D,+CAA+C;IAC/C,WAAW,EAAE,UAAU;IACvB,oFAAoF;IACpF,gBAAgB,EAAE,UAAU;IAC5B,uCAAuC;IACvC,QAAQ,EAAE,UAAU;IACpB,oBAAoB;IACpB,QAAQ,EAAE,CAAC,CAAC,KAAK,CAAC,sBAAsB,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC;IACtD,iCAAiC;IACjC,UAAU,EAAE,CAAC,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC;CAC7C,CAAC;KACD,IAAI,CAAC;IACJ,EAAE,EAAE,cAAc;IAClB,KAAK,EAAE,cAAc;IACrB,WAAW,EACT,0FAA0F;QAC1F,mFAAmF;QACnF,wEAAwE;CAC3E,CAAC,CAAC;AAKL,mEAAmE;AACnE,MAAM,CAAC,MAAM,yBAAyB,GAAG,cAAc,CAAC;AAExD;;;;GAIG;AACH,MAAM,UAAU,iBAAiB,CAAC,KAAc;IAC9C,MAAM,MAAM,GAAG,kBAAkB,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;IACnD,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;QACpB,oFAAoF;QACpF,MAAM,eAAe,CACnB,oBAAoB,CAAC,gBAAgB,EACrC,sCAAsC,EACtC;YACE,QAAQ,EAAE,YAAY;YACtB,OAAO,EAAE;gBACP,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;oBAC1C,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC;oBAC1B,IAAI,EAAE,KAAK,CAAC,IAAI;iBACjB,CAAC,CAAC;aACJ;SACF,CACF,CAAC;IACJ,CAAC;IAED,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAC;IAC/B,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,IAAI,CAAC,KAAK,EAAE,CAAC;QACtC,IAAI,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,EAAE,CAAC;YAC3B,MAAM,eAAe,CACnB,oBAAoB,CAAC,qBAAqB,EAC1C,WAAW,MAAM,CAAC,IAAI,CAAC,QAAQ,0CAA0C,KAAK,CAAC,MAAM,KAAK;gBACxF,uFAAuF,EACzF,EAAE,QAAQ,EAAE,YAAY,EAAE,OAAO,EAAE,EAAE,QAAQ,EAAE,MAAM,CAAC,IAAI,CAAC,QAAQ,EAAE,EAAE,CACxE,CAAC;QACJ,CAAC;QACD,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;IACzB,CAAC;IACD,OAAO,MAAM,CAAC,IAAI,CAAC;AACrB,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,iBAAiB,CAAC,KAAc;IAC9C,MAAM,MAAM,GAAG,kBAAkB,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;IACnD,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;QACpB,MAAM,eAAe,CACnB,oBAAoB,CAAC,gBAAgB,EACrC,sCAAsC,EACtC;YACE,QAAQ,EAAE,YAAY;YACtB,OAAO,EAAE;gBACP,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;oBAC1C,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC;oBAC1B,IAAI,EAAE,KAAK,CAAC,IAAI;iBACjB,CAAC,CAAC;aACJ;SACF,CACF,CAAC;IACJ,CAAC;IAED,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAC;IAC/B,KAAK,MAAM,OAAO,IAAI,MAAM,CAAC,IAAI,CAAC,QAAQ,EAAE,CAAC;QAC3C,IAAI,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;YAC7B,MAAM,eAAe,CACnB,oBAAoB,CAAC,iBAAiB,EACtC,sBAAsB,MAAM,CAAC,IAAI,CAAC,WAAW,4BAA4B,OAAO,CAAC,MAAM,IAAI,EAC3F,EAAE,QAAQ,EAAE,YAAY,EAAE,OAAO,EAAE,EAAE,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,WAAW,EAAE,EAAE,CAC9E,CAAC;QACJ,CAAC;QACD,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;IAC3B,CAAC;IACD,OAAO,MAAM,CAAC,IAAI,CAAC;AACrB,CAAC"}

package/dist/errors.d.ts

@@ -0,0 +1,44 @@
+/**
+ * Failures specific to the regression harness.
+ *
+ * Aldus Core deliberately keeps no central error-code registry, so a package can name a new
+ * failure without forking Core. These codes carry the same `ALDUS_` prefix and
+ * `SCREAMING_SNAKE_CASE` shape so production trace (contract §20) stays uniform across packages.
+ */
+import { AldusError, type ErrorCategory } from "@aldus-runtime/core";
+/** Error codes raised by the regression harness. */
+export declare const RegressionErrorCodes: {
+    /** A defect corpus contained a case that does not validate. */
+    readonly CORPUS_MALFORMED: "ALDUS_CORPUS_MALFORMED";
+    /** Two cases in one corpus share a `caseId`. */
+    readonly CORPUS_DUPLICATE_CASE: "ALDUS_CORPUS_DUPLICATE_CASE";
+    /** An evaluator outcome referenced a `caseId` the corpus does not contain. */
+    readonly OUTCOME_UNKNOWN_CASE: "ALDUS_OUTCOME_UNKNOWN_CASE";
+    /** Two outcomes in one run reported on the same `caseId`. */
+    readonly OUTCOME_DUPLICATE: "ALDUS_OUTCOME_DUPLICATE";
+    /**
+     * A case carries a severity the policy assigns no weight to.
+     *
+     * Refused rather than defaulted: silently weighting an unknown severity as zero would drop a
+     * missed defect out of the severity-weighted recall that contract §12.1 requires be
+     * considered, and the metric would still report a number.
+     */
+    readonly SEVERITY_UNWEIGHTED: "ALDUS_SEVERITY_UNWEIGHTED";
+    /** A correction class was referenced that the policy assigns no harm weight to. */
+    readonly CORRECTION_CLASS_UNWEIGHTED: "ALDUS_CORRECTION_CLASS_UNWEIGHTED";
+    /** A blind-spot record does not validate. */
+    readonly BLIND_SPOT_MALFORMED: "ALDUS_BLIND_SPOT_MALFORMED";
+    /** Two blind spots share a `blindSpotId`. */
+    readonly BLIND_SPOT_DUPLICATE: "ALDUS_BLIND_SPOT_DUPLICATE";
+    /** A promotion policy is internally inconsistent — a threshold outside its valid range. */
+    readonly POLICY_INVALID: "ALDUS_POLICY_INVALID";
+};
+/** @see RegressionErrorCodes */
+export type RegressionErrorCode = (typeof RegressionErrorCodes)[keyof typeof RegressionErrorCodes];
+/** Construct an {@link AldusError} with a regression-harness code. */
+export declare function regressionError(code: RegressionErrorCode, message: string, options: {
+    category: ErrorCategory;
+    retryable?: boolean;
+    details?: Record<string, unknown>;
+}): AldusError;
+//# sourceMappingURL=errors.d.ts.map

package/dist/errors.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"errors.d.ts","sourceRoot":"","sources":["../src/errors.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,EAAE,UAAU,EAAE,KAAK,aAAa,EAAE,MAAM,qBAAqB,CAAC;AAErE,oDAAoD;AACpD,eAAO,MAAM,oBAAoB;IAC/B,+DAA+D;aAC/D,gBAAgB,EAAE,wBAAwB;IAC1C,gDAAgD;aAChD,qBAAqB,EAAE,6BAA6B;IACpD,8EAA8E;aAC9E,oBAAoB,EAAE,4BAA4B;IAClD,6DAA6D;aAC7D,iBAAiB,EAAE,yBAAyB;IAC5C;;;;;;OAMG;aACH,mBAAmB,EAAE,2BAA2B;IAChD,mFAAmF;aACnF,2BAA2B,EAAE,mCAAmC;IAChE,6CAA6C;aAC7C,oBAAoB,EAAE,4BAA4B;IAClD,6CAA6C;aAC7C,oBAAoB,EAAE,4BAA4B;IAClD,2FAA2F;aAC3F,cAAc,EAAE,sBAAsB;CAC9B,CAAC;AAEX,gCAAgC;AAChC,MAAM,MAAM,mBAAmB,GAAG,CAAC,OAAO,oBAAoB,CAAC,CAAC,MAAM,OAAO,oBAAoB,CAAC,CAAC;AAEnG,sEAAsE;AACtE,wBAAgB,eAAe,CAC7B,IAAI,EAAE,mBAAmB,EACzB,OAAO,EAAE,MAAM,EACf,OAAO,EAAE;IAAE,QAAQ,EAAE,aAAa,CAAC;IAAC,SAAS,CAAC,EAAE,OAAO,CAAC;IAAC,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CAAE,GAC3F,UAAU,CAEZ"}

package/dist/errors.js

@@ -0,0 +1,40 @@
+/**
+ * Failures specific to the regression harness.
+ *
+ * Aldus Core deliberately keeps no central error-code registry, so a package can name a new
+ * failure without forking Core. These codes carry the same `ALDUS_` prefix and
+ * `SCREAMING_SNAKE_CASE` shape so production trace (contract §20) stays uniform across packages.
+ */
+import { AldusError } from "@aldus-runtime/core";
+/** Error codes raised by the regression harness. */
+export const RegressionErrorCodes = {
+    /** A defect corpus contained a case that does not validate. */
+    CORPUS_MALFORMED: "ALDUS_CORPUS_MALFORMED",
+    /** Two cases in one corpus share a `caseId`. */
+    CORPUS_DUPLICATE_CASE: "ALDUS_CORPUS_DUPLICATE_CASE",
+    /** An evaluator outcome referenced a `caseId` the corpus does not contain. */
+    OUTCOME_UNKNOWN_CASE: "ALDUS_OUTCOME_UNKNOWN_CASE",
+    /** Two outcomes in one run reported on the same `caseId`. */
+    OUTCOME_DUPLICATE: "ALDUS_OUTCOME_DUPLICATE",
+    /**
+     * A case carries a severity the policy assigns no weight to.
+     *
+     * Refused rather than defaulted: silently weighting an unknown severity as zero would drop a
+     * missed defect out of the severity-weighted recall that contract §12.1 requires be
+     * considered, and the metric would still report a number.
+     */
+    SEVERITY_UNWEIGHTED: "ALDUS_SEVERITY_UNWEIGHTED",
+    /** A correction class was referenced that the policy assigns no harm weight to. */
+    CORRECTION_CLASS_UNWEIGHTED: "ALDUS_CORRECTION_CLASS_UNWEIGHTED",
+    /** A blind-spot record does not validate. */
+    BLIND_SPOT_MALFORMED: "ALDUS_BLIND_SPOT_MALFORMED",
+    /** Two blind spots share a `blindSpotId`. */
+    BLIND_SPOT_DUPLICATE: "ALDUS_BLIND_SPOT_DUPLICATE",
+    /** A promotion policy is internally inconsistent — a threshold outside its valid range. */
+    POLICY_INVALID: "ALDUS_POLICY_INVALID",
+};
+/** Construct an {@link AldusError} with a regression-harness code. */
+export function regressionError(code, message, options) {
+    return new AldusError(code, message, options);
+}
+//# sourceMappingURL=errors.js.map

package/dist/errors.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"errors.js","sourceRoot":"","sources":["../src/errors.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,EAAE,UAAU,EAAsB,MAAM,qBAAqB,CAAC;AAErE,oDAAoD;AACpD,MAAM,CAAC,MAAM,oBAAoB,GAAG;IAClC,+DAA+D;IAC/D,gBAAgB,EAAE,wBAAwB;IAC1C,gDAAgD;IAChD,qBAAqB,EAAE,6BAA6B;IACpD,8EAA8E;IAC9E,oBAAoB,EAAE,4BAA4B;IAClD,6DAA6D;IAC7D,iBAAiB,EAAE,yBAAyB;IAC5C;;;;;;OAMG;IACH,mBAAmB,EAAE,2BAA2B;IAChD,mFAAmF;IACnF,2BAA2B,EAAE,mCAAmC;IAChE,6CAA6C;IAC7C,oBAAoB,EAAE,4BAA4B;IAClD,6CAA6C;IAC7C,oBAAoB,EAAE,4BAA4B;IAClD,2FAA2F;IAC3F,cAAc,EAAE,sBAAsB;CAC9B,CAAC;AAKX,sEAAsE;AACtE,MAAM,UAAU,eAAe,CAC7B,IAAyB,EACzB,OAAe,EACf,OAA4F;IAE5F,OAAO,IAAI,UAAU,CAAC,IAAI,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;AAChD,CAAC"}

package/dist/index.d.ts

@@ -0,0 +1,28 @@
+/**
+ * `@aldus-runtime/regression` — the regression harness (architecture contract §22 WP-10).
+ *
+ * §12.1 permits an evaluator to become blocking "only after it is calibrated against
+ * human-labeled examples". This package is what makes that an evidence-based decision instead of
+ * a judgement call: it holds the defect corpus, compares an evaluator's outcomes against human
+ * labels, computes the metrics §12.1 names, and reports whether the bar is met — **per scope**.
+ *
+ * It does **not** run evaluators. Outcomes are produced elsewhere and handed here.
+ *
+ * Two things are load-bearing and easy to lose in a refactor:
+ *
+ * - **Promotion is scoped.** No API here returns "this evaluator is promotable". §12.1 lists
+ *   show, host, voice, model, and script-form scope because calibration does not generalise.
+ * - **The aggregate never decides.** The whole-corpus figure is descriptive; no threshold is
+ *   applied to it (ADR-0010).
+ *
+ * @packageDocumentation
+ */
+export { REGRESSION_SCHEMA_VERSION, defectCaseSchema, defectCorpusSchema, evaluatorFindingSchema, evaluatorOutcomeSchema, evaluatorRunSchema, findingCategory, humanFindingSchema, parseDefectCorpus, parseEvaluatorRun, scopeDimensions, severityLevel, type DefectCase, type DefectCorpus, type EvaluatorFinding, type EvaluatorOutcome, type EvaluatorRun, type HumanFinding, type ScopeDimensions, } from "./corpus.js";
+export { WHOLE_CORPUS_SLICE, deriveScopeSelectors, observedDimensions, scopeKey, scopeLabel, scopeMatches, type ScopeSelector, } from "./scope.js";
+export { DEFAULT_CORRECTION_HARM_WEIGHTS, DEFAULT_PROMOTION_THRESHOLDS, assertPolicyValid, correctionHarm, defaultPromotionPolicy, severityWeight, type CorrectionHarmWeights, type PolicyOrigin, type PromotionPolicy, type PromotionThresholds, type SeverityWeights, } from "./policy.js";
+export { compareRun, type CaseComparison, type CaseVerdict, type CompareOptions, type ComparisonReport, type SliceMetrics, } from "./metrics.js";
+export { BLIND_SPOT_STATUSES, BlindSpotRegistry, blindSpotCoversCase, blindSpotSchema, type BlindSpot, type BlindSpotStatus, } from "./blindspots.js";
+export { assessPromotion, isPromotableEverywhereMeasured, type PromotionShortfall, type PromotionShortfallCode, type PromotionVerdict, type SliceVerdict, } from "./promotion.js";
+export { REPORT_CAVEAT, forbiddenClaimWords, renderPromotionReport } from "./report.js";
+export { RegressionErrorCodes, regressionError, type RegressionErrorCode } from "./errors.js";
+//# sourceMappingURL=index.d.ts.map

package/dist/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAGH,OAAO,EACL,yBAAyB,EACzB,gBAAgB,EAChB,kBAAkB,EAClB,sBAAsB,EACtB,sBAAsB,EACtB,kBAAkB,EAClB,eAAe,EACf,kBAAkB,EAClB,iBAAiB,EACjB,iBAAiB,EACjB,eAAe,EACf,aAAa,EACb,KAAK,UAAU,EACf,KAAK,YAAY,EACjB,KAAK,gBAAgB,EACrB,KAAK,gBAAgB,EACrB,KAAK,YAAY,EACjB,KAAK,YAAY,EACjB,KAAK,eAAe,GACrB,MAAM,aAAa,CAAC;AAGrB,OAAO,EACL,kBAAkB,EAClB,oBAAoB,EACpB,kBAAkB,EAClB,QAAQ,EACR,UAAU,EACV,YAAY,EACZ,KAAK,aAAa,GACnB,MAAM,YAAY,CAAC;AAGpB,OAAO,EACL,+BAA+B,EAC/B,4BAA4B,EAC5B,iBAAiB,EACjB,cAAc,EACd,sBAAsB,EACtB,cAAc,EACd,KAAK,qBAAqB,EAC1B,KAAK,YAAY,EACjB,KAAK,eAAe,EACpB,KAAK,mBAAmB,EACxB,KAAK,eAAe,GACrB,MAAM,aAAa,CAAC;AAGrB,OAAO,EACL,UAAU,EACV,KAAK,cAAc,EACnB,KAAK,WAAW,EAChB,KAAK,cAAc,EACnB,KAAK,gBAAgB,EACrB,KAAK,YAAY,GAClB,MAAM,cAAc,CAAC;AAGtB,OAAO,EACL,mBAAmB,EACnB,iBAAiB,EACjB,mBAAmB,EACnB,eAAe,EACf,KAAK,SAAS,EACd,KAAK,eAAe,GACrB,MAAM,iBAAiB,CAAC;AAGzB,OAAO,EACL,eAAe,EACf,8BAA8B,EAC9B,KAAK,kBAAkB,EACvB,KAAK,sBAAsB,EAC3B,KAAK,gBAAgB,EACrB,KAAK,YAAY,GAClB,MAAM,gBAAgB,CAAC;AAGxB,OAAO,EAAE,aAAa,EAAE,mBAAmB,EAAE,qBAAqB,EAAE,MAAM,aAAa,CAAC;AAGxF,OAAO,EAAE,oBAAoB,EAAE,eAAe,EAAE,KAAK,mBAAmB,EAAE,MAAM,aAAa,CAAC"}

package/dist/index.js

@@ -0,0 +1,36 @@
+/**
+ * `@aldus-runtime/regression` — the regression harness (architecture contract §22 WP-10).
+ *
+ * §12.1 permits an evaluator to become blocking "only after it is calibrated against
+ * human-labeled examples". This package is what makes that an evidence-based decision instead of
+ * a judgement call: it holds the defect corpus, compares an evaluator's outcomes against human
+ * labels, computes the metrics §12.1 names, and reports whether the bar is met — **per scope**.
+ *
+ * It does **not** run evaluators. Outcomes are produced elsewhere and handed here.
+ *
+ * Two things are load-bearing and easy to lose in a refactor:
+ *
+ * - **Promotion is scoped.** No API here returns "this evaluator is promotable". §12.1 lists
+ *   show, host, voice, model, and script-form scope because calibration does not generalise.
+ * - **The aggregate never decides.** The whole-corpus figure is descriptive; no threshold is
+ *   applied to it (ADR-0010).
+ *
+ * @packageDocumentation
+ */
+// --- Defect corpus (§12.1, §12.3, §24) --------------------------------------------------------
+export { REGRESSION_SCHEMA_VERSION, defectCaseSchema, defectCorpusSchema, evaluatorFindingSchema, evaluatorOutcomeSchema, evaluatorRunSchema, findingCategory, humanFindingSchema, parseDefectCorpus, parseEvaluatorRun, scopeDimensions, severityLevel, } from "./corpus.js";
+// --- Scope slicing (§12.1) --------------------------------------------------------------------
+export { WHOLE_CORPUS_SLICE, deriveScopeSelectors, observedDimensions, scopeKey, scopeLabel, scopeMatches, } from "./scope.js";
+// --- Promotion policy (§12.1, §25.9; ADR-0010) ------------------------------------------------
+export { DEFAULT_CORRECTION_HARM_WEIGHTS, DEFAULT_PROMOTION_THRESHOLDS, assertPolicyValid, correctionHarm, defaultPromotionPolicy, severityWeight, } from "./policy.js";
+// --- Comparison and metrics (§12.1) -----------------------------------------------------------
+export { compareRun, } from "./metrics.js";
+// --- Known blind spots (§12.1, §9.3) ----------------------------------------------------------
+export { BLIND_SPOT_STATUSES, BlindSpotRegistry, blindSpotCoversCase, blindSpotSchema, } from "./blindspots.js";
+// --- Promotion verdict (§12.1; ADR-0010) ------------------------------------------------------
+export { assessPromotion, isPromotableEverywhereMeasured, } from "./promotion.js";
+// --- Report rendering (§12) -------------------------------------------------------------------
+export { REPORT_CAVEAT, forbiddenClaimWords, renderPromotionReport } from "./report.js";
+// --- Errors -----------------------------------------------------------------------------------
+export { RegressionErrorCodes, regressionError } from "./errors.js";
+//# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,iGAAiG;AACjG,OAAO,EACL,yBAAyB,EACzB,gBAAgB,EAChB,kBAAkB,EAClB,sBAAsB,EACtB,sBAAsB,EACtB,kBAAkB,EAClB,eAAe,EACf,kBAAkB,EAClB,iBAAiB,EACjB,iBAAiB,EACjB,eAAe,EACf,aAAa,GAQd,MAAM,aAAa,CAAC;AAErB,iGAAiG;AACjG,OAAO,EACL,kBAAkB,EAClB,oBAAoB,EACpB,kBAAkB,EAClB,QAAQ,EACR,UAAU,EACV,YAAY,GAEb,MAAM,YAAY,CAAC;AAEpB,iGAAiG;AACjG,OAAO,EACL,+BAA+B,EAC/B,4BAA4B,EAC5B,iBAAiB,EACjB,cAAc,EACd,sBAAsB,EACtB,cAAc,GAMf,MAAM,aAAa,CAAC;AAErB,iGAAiG;AACjG,OAAO,EACL,UAAU,GAMX,MAAM,cAAc,CAAC;AAEtB,iGAAiG;AACjG,OAAO,EACL,mBAAmB,EACnB,iBAAiB,EACjB,mBAAmB,EACnB,eAAe,GAGhB,MAAM,iBAAiB,CAAC;AAEzB,iGAAiG;AACjG,OAAO,EACL,eAAe,EACf,8BAA8B,GAK/B,MAAM,gBAAgB,CAAC;AAExB,iGAAiG;AACjG,OAAO,EAAE,aAAa,EAAE,mBAAmB,EAAE,qBAAqB,EAAE,MAAM,aAAa,CAAC;AAExF,iGAAiG;AACjG,OAAO,EAAE,oBAAoB,EAAE,eAAe,EAA4B,MAAM,aAAa,CAAC"}