@sanity/ailf 2.8.0 → 3.0.0

package/dist/_vendor/ailf-core/batch-signing.d.ts

@@ -0,0 +1,64 @@
+/**
+ * Batch signed-URL types (D0033 / W0052 M3).
+ *
+ * Contract shared between the API Gateway and Studio for the two batch
+ * endpoints that amortise signing over many artifacts in a single round-trip:
+ *
+ *   POST /v1/runs/:runId/artifacts/batch/read-urls
+ *   POST /v1/runs/:runId/artifacts/batch/upload-urls
+ *
+ * The batch shape is intentionally symmetric: request carries the set of
+ * artifact types and (for per-entry layouts) the entry keys; response carries
+ * signed URLs keyed by type and entry key. For bulk layouts the response key
+ * is the empty string, since bulk artifacts have no entry dimension.
+ *
+ * Validation is all-or-nothing per AC 4 — a single malformed entry key, an
+ * unknown artifact type, or a missing keys list for a per-entry type causes
+ * the whole request to 400 with zero signed URLs emitted.
+ *
+ * @see docs/design-docs/unified-run-artifacts.md § M3
+ * @see docs/decisions/D0033-unified-run-anchored-artifact-capture.md
+ */
+import type { ArtifactType } from "./artifact-registry.js";
+/** The empty entry-key sentinel used for bulk responses. */
+export declare const BULK_ENTRY_KEY: "";
+/**
+ * Batch request body.
+ *
+ * - `types` — the artifact types to sign. Must be non-empty. Unknown types
+ *   cause a 400.
+ * - `keys` — per-type arrays of entry keys. Required for per-entry layouts;
+ *   omitted (or an empty array) for bulk layouts. Extra keys for a bulk type
+ *   are rejected; missing keys for a per-entry type are rejected.
+ */
+export interface BatchSignRequest {
+    readonly types: readonly ArtifactType[];
+    readonly keys?: Partial<Record<ArtifactType, readonly string[]>>;
+}
+/** One signed read URL in a batch read response. */
+export interface BatchSignedReadUrl {
+    readonly url: string;
+    readonly path: string;
+    readonly expiresIn: number;
+}
+/** One signed upload URL in a batch upload response. */
+export interface BatchSignedUploadUrl {
+    readonly url: string;
+    readonly method: "PUT";
+    readonly path: string;
+    readonly expiresIn: number;
+    readonly requiredHeaders: Readonly<Record<string, string>>;
+}
+/**
+ * Batch response. Outer map is keyed by artifact type; inner map is keyed by
+ * entry key (empty string for bulk layouts). All types requested in the
+ * request body are present in the response, even when `keys[type]` was empty.
+ */
+export interface BatchSignReadResponse {
+    readonly bucket: string;
+    readonly urls: Partial<Record<ArtifactType, Readonly<Record<string, BatchSignedReadUrl>>>>;
+}
+export interface BatchSignUploadResponse {
+    readonly bucket: string;
+    readonly urls: Partial<Record<ArtifactType, Readonly<Record<string, BatchSignedUploadUrl>>>>;
+}

package/dist/_vendor/ailf-core/batch-signing.js

@@ -0,0 +1,23 @@
+/**
+ * Batch signed-URL types (D0033 / W0052 M3).
+ *
+ * Contract shared between the API Gateway and Studio for the two batch
+ * endpoints that amortise signing over many artifacts in a single round-trip:
+ *
+ *   POST /v1/runs/:runId/artifacts/batch/read-urls
+ *   POST /v1/runs/:runId/artifacts/batch/upload-urls
+ *
+ * The batch shape is intentionally symmetric: request carries the set of
+ * artifact types and (for per-entry layouts) the entry keys; response carries
+ * signed URLs keyed by type and entry key. For bulk layouts the response key
+ * is the empty string, since bulk artifacts have no entry dimension.
+ *
+ * Validation is all-or-nothing per AC 4 — a single malformed entry key, an
+ * unknown artifact type, or a missing keys list for a per-entry type causes
+ * the whole request to 400 with zero signed URLs emitted.
+ *
+ * @see docs/design-docs/unified-run-artifacts.md § M3
+ * @see docs/decisions/D0033-unified-run-anchored-artifact-capture.md
+ */
+/** The empty entry-key sentinel used for bulk responses. */
+export const BULK_ENTRY_KEY = "";

package/dist/_vendor/ailf-core/index.d.ts

@@ -16,8 +16,9 @@ export * from "./ports/index.js";
 export * from "./services/index.js";
 export * from "./examples/index.js";
 export * from "./artifact-registry.js";
+export * from "./batch-signing.js";
 export { defineConfig, defineFeatures, defineModeBase, defineModels, definePricingTable, definePreset, definePrompts, defineRubrics, defineSchedules, defineSinks, defineSources, defineTask, defineThresholds, } from "./config-helpers.js";
 export type { PricingEntry, PromptEntry, SourceEntry, } from "./config-helpers.js";
 export { env } from "./env-helper.js";
-export { NoOpArtifactCollector } from "./artifact-capture/noop-collector.js";
-export { NoOpArtifactWriter } from "./ports/artifact-writer.js";
+export { NoOpArtifactWriter, NotImplementedError, } from "./ports/artifact-writer.js";
+export { assoc, type AssocContext } from "./artifact-capture/association.js";

package/dist/_vendor/ailf-core/index.js

@@ -16,10 +16,11 @@ export * from "./ports/index.js";
 export * from "./services/index.js";
 export * from "./examples/index.js";
 export * from "./artifact-registry.js";
+export * from "./batch-signing.js";
 // ---------------------------------------------------------------------------
 // Architecture overhaul — Phase 0 helpers
 // ---------------------------------------------------------------------------
 export { defineConfig, defineFeatures, defineModeBase, defineModels, definePricingTable, definePreset, definePrompts, defineRubrics, defineSchedules, defineSinks, defineSources, defineTask, defineThresholds, } from "./config-helpers.js";
 export { env } from "./env-helper.js";
-export { NoOpArtifactCollector } from "./artifact-capture/noop-collector.js";
-export { NoOpArtifactWriter } from "./ports/artifact-writer.js";
+export { NoOpArtifactWriter, NotImplementedError, } from "./ports/artifact-writer.js";
+export { assoc } from "./artifact-capture/association.js";

package/dist/_vendor/ailf-core/ports/artifact-writer.d.ts

@@ -1,26 +1,35 @@
 /**
  * Port: ArtifactWriter — writes run artifacts + the run manifest to external storage.
  *
- * Replaces the older `ArtifactUploader` port from D0030. Differences:
- *   - Paths anchor to `RunId` (not `ReportId`) via the registry's `objectPath`.
- *   - Supports both `"bulk"` and `"per-entry"` layouts.
- *   - A dedicated `writeManifest()` method for the run manifest at
- *     `runs/{runId}/manifest.json`.
+ * D0033 / W0049 unifies the writer API around a single caller-facing method:
  *
- * Producer steps call writer methods directly with the artifact type from
- * `ARTIFACT_REGISTRY`. Path construction, schema validation, and entry-key
- * sanitization live in the registry, not the call site.
+ *   - `emit(type, association, payload)` — the canonical write. Dispatch on
+ *     `descriptor.layout` is internal; callers never pick a shape.
+ *   - `appendNdjson(type, association, rows)` — streaming-append variant used
+ *     only by the `traces` artifact. Semantics differ from `emit` (repeated
+ *     append vs. single-shot write) so it gets its own method rather than an
+ *     overload.
+ *   - `writeManifest(runId, manifest)` — writes the run manifest at
+ *     `runs/{runId}/manifest.json`.
+ *   - `writeBulk` / `writePerEntry` — @deprecated legacy surface retained for
+ *     producer code that has not migrated to `emit`. Removal scheduled for
+ *     W0052 (see `docs/decisions/D0033-unified-run-anchored-artifact-capture.md`).
  *
  * @see docs/decisions/D0032-run-anchored-artifact-store.md
+ * @see docs/decisions/D0033-unified-run-anchored-artifact-capture.md
  * @see packages/core/src/artifact-registry.ts
  */
 import type { ArtifactType } from "../artifact-registry.js";
-import type { RunId } from "../types/branded-ids.js";
+import type { AssociationValues, RunId } from "../types/branded-ids.js";
 import type { ArtifactRef, RunManifest } from "../types/index.js";
 /**
  * An entry in a per-entry upload. The `key` is the wire-format identifier
  * (e.g. `{taskId}::{modelId}` for testOutputs); the writer sanitizes it into
- * the filename using the registry's `parseEntryKey`.
+ * the filename using the registry's `parseEntryKey` or direct path building.
+ *
+ * @deprecated Use `ArtifactWriter.emit()` with `AssociationValues` instead.
+ * This type is retained for producers still on the legacy
+ * `writeBulk`/`writePerEntry` path; removal scheduled for W0052.
  */
 export interface ArtifactEntry<TData = unknown> {
     key: string;
@@ -28,29 +37,59 @@ export interface ArtifactEntry<TData = unknown> {
 }
 export interface ArtifactWriter {
     /**
-     * Write a bulk artifact — one JSON object per (runId, type).
+     * Write a single artifact. The descriptor's `layout` determines whether
+     * this produces a bulk object (`runs/{runId}/{slug}.{ext}`) or a per-entry
+     * object (`runs/{runId}/{slug}/{entryKey}.{ext}`).
      *
-     * @returns An `ArtifactRef` pointing at `runs/{runId}/{slug}.json`, or
-     *          `null` when upload is skipped or fails (P5: non-blocking).
+     * For per-entry descriptors, `association` must carry the axis values the
+     * descriptor's `formatEntryKey` consumes. For bulk descriptors, only `run`
+     * is required.
+     *
+     * @returns `ArtifactRef` on success, or `null` when upload is skipped or
+     *          fails (P5: non-blocking).
      */
-    writeBulk(type: ArtifactType, runId: RunId, data: unknown): Promise<ArtifactRef | null>;
+    emit<T extends ArtifactType>(type: T, association: AssociationValues, payload: unknown): Promise<ArtifactRef | null>;
     /**
-     * Write a per-entry artifact — one JSON object per entry, all under
-     * `runs/{runId}/{slug}/`.
+     * Append NDJSON rows to a per-entry artifact. Used only by `traces`, whose
+     * per-entry payload is itself unbounded during production. The writer
+     * buffers rows keyed on (type, entryKey) and flushes to numbered part
+     * objects; the backend composes parts into the final object at trial
+     * completion.
      *
-     * The returned `ArtifactRef.entries` inlines the catalog so consumers
-     * can render drill-down state without a second listing call.
+     * @throws NotImplementedError on writers that don't support streaming
+     *         appends (e.g. the API-gateway-backed writer; see W0052).
      */
-    writePerEntry(type: ArtifactType, runId: RunId, entries: readonly ArtifactEntry[]): Promise<ArtifactRef | null>;
+    appendNdjson<T extends ArtifactType>(type: T, association: AssociationValues, rows: readonly unknown[]): Promise<ArtifactRef | null>;
     /**
      * Write the run manifest to `runs/{runId}/manifest.json`. Single-writer
      * per run; subsequent publishes may rewrite to append `reportIds[]`.
      */
     writeManifest(runId: RunId, manifest: RunManifest): Promise<ArtifactRef | null>;
+    /**
+     * @deprecated Use `emit()` with `AssociationValues` instead. Retained for
+     * producers still on the legacy path; removal scheduled for W0052.
+     */
+    writeBulk(type: ArtifactType, runId: RunId, data: unknown): Promise<ArtifactRef | null>;
+    /**
+     * @deprecated Use `emit()` per entry instead. Retained for producers still
+     * on the legacy path; removal scheduled for W0052.
+     */
+    writePerEntry(type: ArtifactType, runId: RunId, entries: readonly ArtifactEntry[]): Promise<ArtifactRef | null>;
+}
+/**
+ * Thrown by writers that can't satisfy a method — e.g. an
+ * `ApiGatewayArtifactWriter` cannot implement `appendNdjson` until the batch
+ * signing endpoint (W0052) lands. Callers should treat this as an explicit
+ * failure rather than a silent no-op so the gap surfaces in logs.
+ */
+export declare class NotImplementedError extends Error {
+    constructor(message: string);
 }
 /** No-op writer — every method returns null. Used when no storage is configured. */
 export declare class NoOpArtifactWriter implements ArtifactWriter {
+    emit(): Promise<null>;
+    appendNdjson(): Promise<null>;
+    writeManifest(): Promise<null>;
     writeBulk(): Promise<null>;
     writePerEntry(): Promise<null>;
-    writeManifest(): Promise<null>;
 }

package/dist/_vendor/ailf-core/ports/artifact-writer.js

@@ -1,28 +1,51 @@
 /**
  * Port: ArtifactWriter — writes run artifacts + the run manifest to external storage.
  *
- * Replaces the older `ArtifactUploader` port from D0030. Differences:
- *   - Paths anchor to `RunId` (not `ReportId`) via the registry's `objectPath`.
- *   - Supports both `"bulk"` and `"per-entry"` layouts.
- *   - A dedicated `writeManifest()` method for the run manifest at
- *     `runs/{runId}/manifest.json`.
+ * D0033 / W0049 unifies the writer API around a single caller-facing method:
  *
- * Producer steps call writer methods directly with the artifact type from
- * `ARTIFACT_REGISTRY`. Path construction, schema validation, and entry-key
- * sanitization live in the registry, not the call site.
+ *   - `emit(type, association, payload)` — the canonical write. Dispatch on
+ *     `descriptor.layout` is internal; callers never pick a shape.
+ *   - `appendNdjson(type, association, rows)` — streaming-append variant used
+ *     only by the `traces` artifact. Semantics differ from `emit` (repeated
+ *     append vs. single-shot write) so it gets its own method rather than an
+ *     overload.
+ *   - `writeManifest(runId, manifest)` — writes the run manifest at
+ *     `runs/{runId}/manifest.json`.
+ *   - `writeBulk` / `writePerEntry` — @deprecated legacy surface retained for
+ *     producer code that has not migrated to `emit`. Removal scheduled for
+ *     W0052 (see `docs/decisions/D0033-unified-run-anchored-artifact-capture.md`).
  *
  * @see docs/decisions/D0032-run-anchored-artifact-store.md
+ * @see docs/decisions/D0033-unified-run-anchored-artifact-capture.md
  * @see packages/core/src/artifact-registry.ts
  */
+/**
+ * Thrown by writers that can't satisfy a method — e.g. an
+ * `ApiGatewayArtifactWriter` cannot implement `appendNdjson` until the batch
+ * signing endpoint (W0052) lands. Callers should treat this as an explicit
+ * failure rather than a silent no-op so the gap surfaces in logs.
+ */
+export class NotImplementedError extends Error {
+    constructor(message) {
+        super(message);
+        this.name = "NotImplementedError";
+    }
+}
 /** No-op writer — every method returns null. Used when no storage is configured. */
 export class NoOpArtifactWriter {
-    async writeBulk() {
+    async emit() {
         return null;
     }
-    async writePerEntry() {
+    async appendNdjson() {
         return null;
     }
     async writeManifest() {
         return null;
     }
+    async writeBulk() {
+        return null;
+    }
+    async writePerEntry() {
+        return null;
+    }
 }

package/dist/_vendor/ailf-core/ports/context.d.ts

@@ -13,7 +13,6 @@
  */
 import type { RunId } from "../types/branded-ids.js";
 import type { DebugOptions, EvalMode, PluginRegistry } from "../types/index.js";
-import type { ArtifactCollector } from "./artifact-collector.js";
 import type { ArtifactWriter } from "./artifact-writer.js";
 import type { CacheStore } from "./cache-store.js";
 import type { DocFetcher } from "./doc-fetcher.js";
@@ -153,18 +152,19 @@ export interface ResolvedConfig {
     apiKey?: string;
     /** External preset file paths or npm package names to load */
     presets?: string[];
-    /** Whether artifact capture is enabled for this run (default: false) */
-    captureEnabled?: boolean;
-    /** Base directory for capture output (default: results/captures/) */
-    captureDir?: string;
-    /** Whether to compress capture output to tar.gz (default: true) */
-    captureCompress?: boolean;
-    /** Whether to include mode-specific extra artifacts (default: true) */
-    captureExtras?: boolean;
-    /** GCS bucket for capture upload (enables GCS decorator when set) */
-    captureGcsBucket?: string;
-    /** GCS object prefix for capture uploads (default: "captures/") */
-    captureGcsPrefix?: string;
+    /**
+     * D0033 unified artifact surface. Consumed by the writer factory to
+     * decide whether to attach a writer at all, where it writes to, and
+     * what to skip. Legacy `capture*` fields were retired in W0052.
+     */
+    /** Disables all artifact writers — `--no-artifacts`. */
+    artifactsDisabled?: boolean;
+    /** Root directory for local artifact output — `--artifacts-dir`. */
+    artifactsDir?: string;
+    /** Run writers in dry-run mode — `--artifacts-dry-run`. */
+    artifactsDryRun?: boolean;
+    /** Comma-separated artifact types to skip — `--artifacts-exclude`. */
+    artifactsExclude?: readonly string[];
     /**
      * GCS bucket for report artifact uploads. Defaults to "ailf-artifacts"
      * at the composition root — only set this to override (e.g., self-hosted
@@ -198,12 +198,15 @@ export interface ResolvedConfig {
  * Created per-test by createTestContext().
  */
 export interface AppContext {
-    /** Artifact writer — writes run artifacts + manifest to GCS (D0032) */
-    readonly artifactWriter?: ArtifactWriter;
+    /**
+     * Artifact writer — writes run artifacts + manifest to local fs (D0033
+     * M4: always on) and optionally to GCS (D0032, layered via
+     * FanoutArtifactWriter). Required post-W0050 — the composition root
+     * always provides a writer (NoOpArtifactWriter when `--no-artifacts`).
+     */
+    readonly artifactWriter: ArtifactWriter;
     /** Evaluation caching (filesystem + optional Content Lake fallback) */
     readonly cache?: CacheStore;
-    /** Artifact capture collector (no-op when --capture is not set) */
-    readonly collector: ArtifactCollector;
     /** Resolved pipeline configuration */
     readonly config: ResolvedConfig;
     /** Documentation context fetcher */

package/dist/_vendor/ailf-core/ports/index.d.ts

@@ -4,10 +4,8 @@
  * Ports define the contracts between the domain kernel and the outside world.
  * Adapters (in packages/eval) implement these interfaces.
  */
-export type { ArtifactCollector, CaptureFlushResult, CaptureManifest, CaptureManifestEntry, } from "./artifact-collector.js";
 export type { ArtifactEntry, ArtifactWriter } from "./artifact-writer.js";
 export { NoOpArtifactWriter } from "./artifact-writer.js";
-export type { ArtifactContentDiff, CaptureDiffReport, ComparisonMode, ComparisonOptions, InventoryDiff, JsonDiffEntry, MetadataComparison, ScoreComparison, SecurityScan, TimingComparison, } from "./capture-comparator.js";
 export type { CacheEntryMetadata, CacheKey, CacheLookupResult, CacheRecordInput, CacheStore, } from "./cache-store.js";
 export type { ConfigSource } from "./config-source.js";
 export type { AppContext, ReportSinkPort, ReportStorePort, ResolvedConfig, } from "./context.js";

package/dist/_vendor/ailf-core/schemas/pipeline.d.ts

@@ -67,10 +67,10 @@ export declare const FeatureSchema: z.ZodObject<{
     id: z.ZodString;
     name: z.ZodString;
     priority: z.ZodEnum<{
-        critical: "critical";
-        high: "high";
-        medium: "medium";
         low: "low";
+        medium: "medium";
+        high: "high";
+        critical: "critical";
     }>;
     sections: z.ZodArray<z.ZodString>;
     status: z.ZodEnum<{
@@ -91,10 +91,10 @@ export declare const FeatureRegistrySchema: z.ZodObject<{
         id: z.ZodString;
         name: z.ZodString;
         priority: z.ZodEnum<{
-            critical: "critical";
-            high: "high";
-            medium: "medium";
             low: "low";
+            medium: "medium";
+            high: "high";
+            critical: "critical";
         }>;
         sections: z.ZodArray<z.ZodString>;
         status: z.ZodEnum<{

package/dist/_vendor/ailf-core/services/index.d.ts

@@ -11,3 +11,4 @@ export { classifyRubric, detectFeatureArea, extractDimensions, extractUrlMetadat
 export { formatComparisonMarkdown, formatComparisonTable, } from "./comparison-formatters.js";
 export { aggregateAreas, aggregateDimensions, computeEnsembleScore, computeTaskScore, normalizeScore, type AggregationStrategy, type AreaScore, type AssertionScore, type DimensionScore, type EnsembleGradingConfig, type GraderTransitionConfig, type TaskScore, type TaskScoreOptions, } from "./scoring-engine.js";
 export { extractModelName, extractProvider, mergeConfig, modelMatchesMode, resolveModelVariants, } from "./config-helpers.js";
+export { buildSlimReportSummary } from "./slim-report-summary.js";

package/dist/_vendor/ailf-core/services/index.js

@@ -11,3 +11,4 @@ export { classifyRubric, detectFeatureArea, extractDimensions, extractUrlMetadat
 export { formatComparisonMarkdown, formatComparisonTable, } from "./comparison-formatters.js";
 export { aggregateAreas, aggregateDimensions, computeEnsembleScore, computeTaskScore, normalizeScore, } from "./scoring-engine.js";
 export { extractModelName, extractProvider, mergeConfig, modelMatchesMode, resolveModelVariants, } from "./config-helpers.js";
+export { buildSlimReportSummary } from "./slim-report-summary.js";

package/dist/_vendor/ailf-core/services/slim-report-summary.d.ts

@@ -0,0 +1,31 @@
+/**
+ * slim-report-summary.ts
+ *
+ * Pure transformer: a full `ScoreSummary` (the shape of `score-summary.json`)
+ * into the slim `ReportSummary` that Phase C of D0033 publishes on the
+ * Report Content-Lake document. Inlined prose and long arrays are replaced
+ * with `id` references to external artifacts (graderJudgments, failureModes,
+ * gapReport, traces).
+ *
+ * **The `id` principle**: every slim reference carries the manifest entry
+ * key of the external artifact it points at, produced by the descriptor's
+ * own `formatEntryKey(axes)`. Studio looks the id up in `Report.artifactManifest`
+ * to get the preview, and hydrates the full payload on drill-down.
+ *
+ * @see docs/decisions/D0033-unified-run-anchored-artifact-capture.md (§ M7)
+ * @see docs/work-items/W0051-report-slim-down-manifest-preview-hooks.json
+ */
+import type { ReportSummary, ScoreSummary } from "../types/index.js";
+/**
+ * Transform a full pipeline `ScoreSummary` into its slim Report counterpart.
+ *
+ * Each of the four heavy fields is reshaped independently; everything else
+ * flows through untouched via structural spread. Pure function — the input
+ * summary is not mutated.
+ *
+ * @param mode The evaluation mode (used to populate the `mode` axis on slim
+ *             judgment / failure-mode ids). `score-summary.json` carries
+ *             the mode in `evaluationMode` but the publisher supplies it
+ *             explicitly to keep the helper independent of that field.
+ */
+export declare function buildSlimReportSummary(summary: ScoreSummary, mode: string): ReportSummary;