@sanity/ailf 0.1.26 → 0.1.28

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

@@ -89,6 +89,8 @@ export interface ResolvedConfig {
     studioOriginOverride?: string;
     /** Sanity document filter args */
     sanityDocumentArgs?: string[];
+    /** Report ID that triggered this re-run (flows to provenance.lineage.rerunOf) */
+    sourceReportId?: string;
     /** Disable release-aware auto-scoping (evaluate all tasks even when perspective is set) */
     noAutoScope: boolean;
     /** Before option for comparison */

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

@@ -68,6 +68,7 @@ export declare const PipelineRequestSchema: z.ZodObject<{
         "origin-only": "origin-only";
     }>>;
     source: z.ZodOptional<z.ZodString>;
+    sourceReportId: z.ZodOptional<z.ZodString>;
     taskMode: z.ZodOptional<z.ZodEnum<{
         "content-lake": "content-lake";
         yaml: "yaml";

package/dist/_vendor/ailf-core/schemas/pipeline-request.js

@@ -80,6 +80,7 @@ export const PipelineRequestSchema = z.object({
     readiness: z.boolean().optional(),
     searchMode: z.enum(["off", "open", "origin-only"]).optional(),
     source: z.string().optional(),
+    sourceReportId: z.string().optional(),
     taskMode: z.enum(["content-lake", "yaml", "inline"]).optional(),
     tasks: z.array(z.string()).optional(),
     urls: z.array(z.string().url()).optional(),

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

@@ -438,6 +438,12 @@ export interface PipelineState {
     evalFingerprint?: string;
     /** Promptfoo share URLs produced by RunEvalStep, consumed by PublishReportStep */
     promptfooUrls?: PromptfooUrlEntry[];
+    /**
+     * Eval modes that were satisfied by a remote cache hit (score-summary.json
+     * was restored from the Content Lake). Produced by RunEvalStep, consumed by
+     * CalculateScoresStep to skip re-calculation when all required modes are cached.
+     */
+    remoteCacheHits?: Set<string>;
     /**
      * Release auto-scope metadata. Set by FetchDocsStep when a perspective
      * is active and release impact identifies affected documents.
@@ -1026,6 +1032,28 @@ export interface ReportAutoScope {
         removed: number;
     };
 }
+/**
+ * Typed relationships between reports. Each field is optional and
+ * independent — populated only when that relationship exists.
+ *
+ * Stored at `provenance.lineage` in the report document.
+ *
+ * @see docs/design-docs/report-store/domain-model.md
+ */
+export interface ReportLineage {
+    /**
+     * This report was explicitly compared against another report.
+     * Set when auto-compare selects a specific baseline or when the user
+     * requests comparison against a named report.
+     */
+    comparedAgainst?: ReportId;
+    /**
+     * This report was explicitly re-run from another report.
+     * The re-run has the same EvalScope (mode, areas, perspective, etc.)
+     * but measures the current state of docs/models/tasks.
+     */
+    rerunOf?: ReportId;
+}
 /** Full provenance metadata for an evaluation report */
 export interface ReportProvenance {
     /** Which feature areas were evaluated */
@@ -1049,6 +1077,8 @@ export interface ReportProvenance {
     };
     /** Grader model used for scoring */
     graderModel: string;
+    /** Typed relationships with other reports (re-run, comparison) */
+    lineage?: ReportLineage;
     /** Evaluation mode */
     mode: EvalMode;
     /** Models under evaluation */

package/dist/adapters/api-client/progress.js

@@ -36,7 +36,7 @@ export function createProgressDisplay() {
                 line = `⏳ [queued] Waiting for runner... (${elapsed})`;
                 break;
             case "running": {
-                if (job.progress) {
+                if (job.progress?.step && job.progress.current && job.progress.total) {
                     const { step, current, total } = job.progress;
                     line = `⏳ [running] Step ${current}/${total}: ${step} (${elapsed})`;
                 }

package/dist/commands/publish.js

@@ -24,7 +24,7 @@ import { fileURLToPath } from "url";
 import { Command } from "commander";
 import { createAppContext } from "../composition-root.js";
 import { buildProvenance, } from "../pipeline/provenance.js";
-import { generateReportId } from "../report-store.js";
+import { generateReportId, } from "../report-store.js";
 import { withRetry } from "../sinks/retry.js";
 const __dirname = dirname(fileURLToPath(import.meta.url));
 const ROOT = resolve(__dirname, "..", "..");
@@ -154,9 +154,17 @@ async function runPublishCommand(summaryPath, opts) {
         }
     }
     // Auto-compare against most recent comparable baseline
-    const comparison = opts.dryRun || !store
+    const autoCompareResult = opts.dryRun || !store
         ? null
         : await store.autoCompare(summary, provenance, now);
+    const comparison = autoCompareResult?.comparison ?? null;
+    // Record which report we compared against in lineage
+    if (autoCompareResult) {
+        provenance.lineage = {
+            ...provenance.lineage,
+            comparedAgainst: autoCompareResult.baselineReportId,
+        };
+    }
     const reportId = generateReportId();
     const report = {
         comparison: comparison ?? undefined,

package/dist/orchestration/steps/calculate-scores-step.d.ts

@@ -4,10 +4,10 @@
  * Calls calculateAndWriteScores() from pipeline/calculate-scores.ts with
  * typed options derived from AppContext. No env bridge needed.
  */
-import type { AppContext, PipelineStep, StepResult, ValidationIssue } from "../../_vendor/ailf-core/index.d.ts";
+import type { AppContext, PipelineState, PipelineStep, StepResult, ValidationIssue } from "../../_vendor/ailf-core/index.d.ts";
 export declare class CalculateScoresStep implements PipelineStep {
     readonly name = "calculate-scores";
     check(): ValidationIssue[];
-    execute(ctx: AppContext): Promise<StepResult>;
+    execute(ctx: AppContext, state: PipelineState): Promise<StepResult>;
     cacheInputs(ctx: AppContext): string[];
 }

package/dist/orchestration/steps/calculate-scores-step.js

@@ -5,6 +5,7 @@
  * typed options derived from AppContext. No env bridge needed.
  */
 import { join } from "path";
+import { FULL_MODE_SUBMODES } from "../../_vendor/ailf-shared/index.js";
 import { getStepInputPaths } from "../../pipeline/cache.js";
 import { calculateAndWriteScores } from "../../pipeline/calculate-scores.js";
 import { checkResultsExist, checkScoreSummaryValid, } from "../../pipeline/checks.js";
@@ -16,8 +17,29 @@ export class CalculateScoresStep {
     check() {
         return [];
     }
-    async execute(ctx) {
+    async execute(ctx, state) {
         const start = Date.now();
+        // When all required eval modes were satisfied by remote cache hits,
+        // score-summary.json was already restored from the cached report.
+        // Skip re-calculation — the raw eval-results files don't exist.
+        if (state.remoteCacheHits?.size) {
+            const requiredModes = ctx.config.mode === "full"
+                ? [...FULL_MODE_SUBMODES]
+                : [ctx.config.mode];
+            const allCached = requiredModes.every((m) => state.remoteCacheHits.has(m));
+            if (allCached) {
+                // Verify the restored score-summary.json is valid
+                const summaryIssues = checkScoreSummaryValid(ctx.config.rootDir);
+                const summaryErrors = summaryIssues.filter((i) => i.severity === "error");
+                if (summaryErrors.length === 0) {
+                    return {
+                        reason: "Remote cache hit — score-summary.json restored from cached report",
+                        status: "skipped",
+                    };
+                }
+                // If the summary is invalid, fall through to normal calculation
+            }
+        }
         const primaryMode = ctx.config.mode === "full"
             ? "baseline"
             : ctx.config.mode;

package/dist/orchestration/steps/publish-report-step.js

@@ -80,10 +80,19 @@ export class PublishReportStep {
         const now = new Date().toISOString();
         const reportId = generateReportId();
         const durationMs = Date.now() - this.pipelineStart;
-        // Auto-compare against most recent comparable baseline
-        const comparison = ctx.reportStore
+        // Auto-compare against most recent comparable baseline.
+        // Returns the comparison + baseline report ID for lineage tracking.
+        const autoCompareResult = ctx.reportStore
             ? (await ctx.reportStore.autoCompare(summary, provenance, now))
             : null;
+        const comparison = autoCompareResult?.comparison ?? null;
+        // Record which report we compared against in lineage
+        if (autoCompareResult) {
+            provenance.lineage = {
+                ...provenance.lineage,
+                comparedAgainst: autoCompareResult.baselineReportId,
+            };
+        }
         const report = {
             comparison: comparison ?? undefined,
             completedAt: now,
@@ -168,6 +177,7 @@ function buildProvenanceInput(summary, ctx, options, autoScope) {
         rootDir: ctx.config.rootDir,
         sanityDocumentIds,
         source,
+        sourceReportId: ctx.config.sourceReportId,
         taskIds,
     };
 }

package/dist/orchestration/steps/run-eval-step.js

@@ -102,6 +102,10 @@ export class RunEvalStep {
             ctx.reportStore) {
             const remoteCacheResult = await checkRemoteCache(evalFingerprint, ctx.reportStore, rootDir);
             if (remoteCacheResult) {
+                // Record the cache hit so CalculateScoresStep can skip when all
+                // required eval modes were satisfied from the remote cache.
+                state.remoteCacheHits ??= new Set();
+                state.remoteCacheHits.add(this.mode);
                 return {
                     durationMs: Date.now() - start,
                     status: "success",

package/dist/pipeline/map-request-to-config.js

@@ -55,6 +55,7 @@ export function mapRequestToConfig(request, rootDir) {
         promptfooUrl: undefined,
         studioOriginOverride: undefined,
         sanityDocumentArgs: undefined,
+        sourceReportId: request.sourceReportId,
         beforeOption: undefined,
         repoTasksPath: undefined,
         callerGit: request.callerGit,

package/dist/pipeline/provenance.d.ts

@@ -41,6 +41,8 @@ export interface ProvenanceInput {
     promptfooUrls?: PromptfooUrlEntry[];
     /** Path to the package root (for reading models.yaml) */
     rootDir: string;
+    /** Report ID that triggered this re-run (becomes lineage.rerunOf) */
+    sourceReportId?: string;
     /** Sanity document IDs targeted */
     sanityDocumentIds?: string[];
     /** Resolved documentation source */

package/dist/pipeline/provenance.js

@@ -35,6 +35,10 @@ export function buildProvenance(input) {
             sha: input.callerGit.sha ?? "unknown",
         }
         : detectGitMetadata();
+    // Build lineage from explicit relationships
+    const lineage = input.sourceReportId
+        ? { rerunOf: input.sourceReportId }
+        : undefined;
     return {
         areas: input.areas,
         autoScope: input.autoScope,
@@ -42,6 +46,7 @@ export function buildProvenance(input) {
         evalFingerprint: input.evalFingerprint,
         git,
         graderModel: models.grader.id,
+        lineage,
         mode: input.mode,
         models: models.models.map((m) => ({ id: m.id, label: m.label })),
         promptfooUrl: input.promptfooUrl,

package/dist/report-store.d.ts

@@ -16,6 +16,16 @@
  */
 import type { SanityClient } from "@sanity/client";
 import type { ComparisonReport, ISOTimestamp, LineageQuery, Report, ReportId, ReportProvenance, ScoreSummary } from "./pipeline/types.js";
+/**
+ * Result of an auto-comparison, bundling the ComparisonReport with the
+ * baseline report ID so the caller can record lineage (comparedAgainst).
+ */
+export interface AutoCompareResult {
+    /** The report ID of the baseline used for comparison */
+    baselineReportId: ReportId;
+    /** The computed comparison report */
+    comparison: ComparisonReport;
+}
 export interface ReportStoreOptions {
     /** Override the Sanity client (for testing) */
     client?: SanityClient;
@@ -33,9 +43,17 @@ export declare class ReportStore {
      * Auto-compare: find the most recent comparable report and compute
      * a ComparisonReport using the existing compare() primitive.
      *
-     * @returns The comparison report, or null if no baseline found or on error
+     * Baseline selection order:
+     * 1. Explicit lineage — if `provenance.lineage.rerunOf` is set,
+     *    compare against that specific report (deterministic re-run comparison)
+     * 2. Fuzzy matching — most recent report with same mode + source name
+     *
+     * Returns the comparison plus the baseline report ID so the caller
+     * can record `provenance.lineage.comparedAgainst`.
+     *
+     * @returns The comparison result with baseline ID, or null if no baseline found
      */
-    autoCompare(currentSummary: ScoreSummary, provenance: ReportProvenance, completedAt: ISOTimestamp): Promise<ComparisonReport | null>;
+    autoCompare(currentSummary: ScoreSummary, provenance: ReportProvenance, completedAt: ISOTimestamp): Promise<AutoCompareResult | null>;
     /**
      * Find a report by its evaluation fingerprint (cross-environment cache lookup).
      *

package/dist/report-store.js

@@ -41,19 +41,43 @@ export class ReportStore {
      * Auto-compare: find the most recent comparable report and compute
      * a ComparisonReport using the existing compare() primitive.
      *
-     * @returns The comparison report, or null if no baseline found or on error
+     * Baseline selection order:
+     * 1. Explicit lineage — if `provenance.lineage.rerunOf` is set,
+     *    compare against that specific report (deterministic re-run comparison)
+     * 2. Fuzzy matching — most recent report with same mode + source name
+     *
+     * Returns the comparison plus the baseline report ID so the caller
+     * can record `provenance.lineage.comparedAgainst`.
+     *
+     * @returns The comparison result with baseline ID, or null if no baseline found
      */
     async autoCompare(currentSummary, provenance, completedAt) {
-        const baseline = await this.findComparableBaseline({
-            before: completedAt,
-            mode: provenance.mode,
-            source: { name: provenance.source.name },
-        });
+        // 1. Prefer explicit lineage source (deterministic re-run comparison)
+        const rerunSourceId = provenance.lineage?.rerunOf;
+        let baseline = null;
+        if (rerunSourceId) {
+            baseline = await this.read(rerunSourceId);
+            if (baseline) {
+                console.log(`  🔗 Comparing against lineage source: ${rerunSourceId}`);
+            }
+            else {
+                console.warn(`  ⚠️  Lineage source ${rerunSourceId} not found, falling back to fuzzy match`);
+            }
+        }
+        // 2. Fall back to fuzzy matching
+        if (!baseline) {
+            baseline = await this.findComparableBaseline({
+                before: completedAt,
+                mode: provenance.mode,
+                source: { name: provenance.source.name },
+            });
+        }
         if (!baseline) {
             return null;
         }
         try {
-            return compare(baseline.summary, currentSummary);
+            const comparison = compare(baseline.summary, currentSummary);
+            return { baselineReportId: baseline.id, comparison };
         }
         catch (error) {
             console.warn(`  ⚠️  Auto-comparison failed: ${error instanceof Error ? error.message : String(error)}`);

package/dist/webhook/eval-request-handler.d.ts

@@ -68,6 +68,8 @@ export interface EvalRequestPayload {
     requestedAt: string;
     /** User ID who requested */
     requestedBy?: string;
+    /** Report ID that triggered this re-run (if any) */
+    sourceReportId?: string;
     /** Request status */
     status: string;
     /** Publish tag */

package/dist/webhook/eval-request-handler.js

@@ -180,6 +180,9 @@ async function dispatchGitHubEval(repo, payload, config) {
             ...(hasAreas ? { areas: payload.areas } : {}),
             ...(payload.debug ? { debug: true } : {}),
             ...(payload.tag ? { publishTag: payload.tag } : {}),
+            ...(payload.sourceReportId
+                ? { sourceReportId: payload.sourceReportId }
+                : {}),
         },
         event_type: "external-eval",
     };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sanity/ailf",
-  "version": "0.1.26",
+  "version": "0.1.28",
   "private": false,
   "publishConfig": {
     "access": "restricted"