@sanity/ailf 6.1.2 → 7.0.0

package/dist/_vendor/ailf-shared/glossary.js

@@ -0,0 +1,330 @@
+/**
+ * glossary.ts
+ *
+ * Centralized metric glossary used by Studio and dashboard alike. Each entry
+ * is keyed by a canonical slug. The value carries:
+ *
+ *   - `label`: the canonical user-facing label, e.g., "Overall Score". Used
+ *     by the dashboard's `<MetricLabel>` and may be reused as table-header
+ *     copy where space allows.
+ *   - `long`: full description for tooltips and help drawers. Copy was
+ *     ported verbatim from the Studio-era `packages/studio/src/glossary.ts`.
+ *
+ * A `short` field is a planned future addition for tight contexts where the
+ * `long` description is too verbose. Adding it later is a non-breaking
+ * schema change.
+ *
+ * @see docs/design-docs/scenario-matrix/evaluation-ceiling.md (three reference points)
+ * @see docs/architecture.md (scoring model)
+ */
+export const GLOSSARY = {
+    // -- Overview stats -------------------------------------------------------
+    overallScore: {
+        label: "Overall Score",
+        long: "A weighted average across all feature areas, using the gold scoring profile: Task Completion (50%), Code Correctness (25%), and Doc Coverage (25%).",
+    },
+    docLift: {
+        label: "Doc Lift",
+        long: "How much the docs help, compared to the model's training data alone. Calculated as ceiling minus floor, where ceiling includes Doc Coverage and floor does not. Higher is better.",
+    },
+    actualScore: {
+        label: "Actual Score",
+        long: "How well an AI agent scores when it has to find docs on its own through web search and page fetching. This is the real-world scenario. Only available in full mode.",
+    },
+    retrievalGap: {
+        label: "Retrieval Gap",
+        long: "The score lost because agents can't find or use all the relevant docs. Calculated as ceiling minus actual. Lower is better; zero means agents find everything.",
+    },
+    infraEfficiency: {
+        label: "Infra Efficiency",
+        long: "What percentage of the docs' potential quality actually reaches agents (actual ÷ ceiling). 100% means agents find and use all relevant docs perfectly.",
+    },
+    // -- Three-layer decomposition columns ------------------------------------
+    floor: {
+        label: "Floor",
+        long: "Output-quality composite without documentation — Task Completion (60%) and Code Correctness (40%) only. Doc Coverage is excluded because it's undefined when no docs are provided. This tells you what the model already knows from its training data.",
+    },
+    ceiling: {
+        label: "Ceiling",
+        long: "Score with gold-standard docs injected directly into the prompt. This is the best the documentation can do.",
+    },
+    actual: {
+        label: "Actual",
+        long: "Score when an AI agent finds docs on its own through web search and page fetching. This is the real-world experience.",
+    },
+    retGap: {
+        label: "Ret. Gap",
+        long: "Quality lost to discoverability (ceiling minus actual). The gap between what the docs could deliver and what agents actually get.",
+    },
+    efficiency: {
+        label: "Efficiency",
+        long: "What fraction of the docs' quality reaches agents in practice (actual ÷ ceiling, shown as a percentage).",
+    },
+    invertedRetGap: {
+        label: "Inverted Retrieval Gap",
+        long: "⚠️ Inverted retrieval gap: agents that can't find the docs actually score higher, because the docs hurt performance. This usually means there's a doc quality problem.",
+    },
+    // -- Per-area score columns -----------------------------------------------
+    score: {
+        label: "Score",
+        long: "Ceiling composite for this feature area: Task Completion × 50% + Code Correctness × 25% + Doc Coverage × 25%.",
+    },
+    taskCompletion: {
+        label: "Task Completion",
+        long: "Can the LLM implement the requested feature? Graded 0–100.",
+    },
+    codeCorrectness: {
+        label: "Code Correctness",
+        long: "Is the generated code idiomatic, correct, and following best practices? Graded 0–100.",
+    },
+    docCoverage: {
+        label: "Doc Coverage",
+        long: "Did the docs provide the information needed to implement the feature? Graded 0–100. This dimension only contributes to the ceiling composite (with docs) — it's excluded from the floor composite because it's undefined without documentation.",
+    },
+    tests: {
+        label: "Tests",
+        long: "Number of test cases in this feature area.",
+    },
+    // -- Comparison deltas ----------------------------------------------------
+    overallDelta: {
+        label: "Overall Δ",
+        long: "Change in overall score between the two runs. Positive means the experiment scored higher.",
+    },
+    actualDelta: {
+        label: "Actual Δ",
+        long: "Change in actual (agent-retrieved) score between runs. Positive means agents did better.",
+    },
+    retGapDelta: {
+        label: "Ret. Gap Δ",
+        long: "Change in retrieval gap between runs. Negative is good here: it means the gap shrank and agents found more relevant docs.",
+    },
+    efficiencyDelta: {
+        label: "Efficiency Δ",
+        long: "Change in infrastructure efficiency between runs. Positive means agents are capturing more of the docs' potential.",
+    },
+    // -- Comparison table columns ---------------------------------------------
+    baseline: {
+        label: "Baseline",
+        long: "The reference run you're comparing against.",
+    },
+    experiment: {
+        label: "Experiment",
+        long: "The new run you're evaluating.",
+    },
+    delta: {
+        label: "Delta",
+        long: "Difference between experiment and baseline. Positive means improvement, negative means regression.",
+    },
+    change: {
+        label: "Change",
+        long: "Whether the change is meaningful: improved, regressed, or unchanged (within the noise threshold).",
+    },
+    // -- Grader judgments ------------------------------------------------------
+    lowScoringJudgments: {
+        label: "Low-Scoring Judgments",
+        long: "The grading model's explanations for tests that scored below 70/100.",
+    },
+    judgmentReason: {
+        label: "Judgment Reason",
+        long: "The grading model's natural language explanation of what went wrong.",
+    },
+    // -- Diagnostics overview ---------------------------------------------------
+    healthStrong: {
+        label: "Strong (80+)",
+        long: "Feature areas scoring 80 or above. The docs are working well for these features — AI agents produce correct, complete implementations.",
+    },
+    healthAttention: {
+        label: "Needs Attention (70–79)",
+        long: "Feature areas scoring 70–79. These are okay but could be improved — there may be gaps in specific dimensions like doc coverage or code correctness.",
+    },
+    healthWeak: {
+        label: "Weak (<70)",
+        long: "Feature areas scoring below 70. The docs are not providing enough support for AI agents to implement these features correctly.",
+    },
+    negativeDocLiftMetric: {
+        label: "Negative Doc Lift",
+        long: "Number of areas where the documentation actually hurts AI performance — the model scores higher without docs than with them. This usually means the docs contain outdated patterns or incorrect examples.",
+    },
+    weakAreas: {
+        label: "Weak Areas",
+        long: "Feature areas where the overall score is below 70. These need the most attention — low scores mean AI agents consistently struggle to implement these features.",
+    },
+    docsHurt: {
+        label: "Docs Hurt Performance",
+        long: "Areas where the floor score (no docs) is higher than the ceiling score (with docs). The documentation may be actively misleading the model. These docs should be reviewed.",
+    },
+    retrievalIssues: {
+        label: "Retrieval Issues",
+        long: "Areas where AI agents can find less than 70% of the available doc quality. The docs exist and are good, but agents can't discover them through search. Consider improving page titles, metadata, or search engine indexing.",
+    },
+    dimWeaknesses: {
+        label: "Dimension Weaknesses",
+        long: "Individual grading dimensions scoring below 50 within an area. These are the specific skills where AI agents fail most — task completion (can it build the feature?), code correctness (is the code right?), or doc coverage (did it use the docs?).",
+    },
+    efficiencyAnomalies: {
+        label: "Efficiency Anomalies",
+        long: "Areas where agent efficiency exceeds 100% — meaning agents perform better with self-found docs than with gold-standard docs injected directly. This can indicate doc quality issues (injected docs confuse the model) or agent memorization.",
+    },
+    docLiftWins: {
+        label: "Doc Lift Wins",
+        long: "Areas where documentation boosts AI performance by 5 or more points. Higher doc lift means the docs are providing crucial information that the model doesn't already know.",
+    },
+    retrievalExcellence: {
+        label: "Retrieval Excellence",
+        long: "Areas where AI agents successfully find and use at least 85% of the available doc quality through web search. Good retrieval means your docs are well-indexed and easy for agents to discover.",
+    },
+    // -- Model breakdown --------------------------------------------------------
+    modelBreakdown: {
+        label: "Model Breakdown",
+        long: "Break down scores by individual LLM model. The default 'All Models' view shows the cross-model average. Select a specific model to see how it performed independently — useful for spotting models that struggle with specific feature areas.",
+    },
+    // -- Strengths (positive diagnostics) ---------------------------------------
+    strengths: {
+        label: "Strengths",
+        long: "What's working well: high-scoring areas, dimensions where the docs are strong, and areas where AI agents successfully find and use the documentation.",
+    },
+    // -- Recommendations / gap analysis ----------------------------------------
+    recommendations: {
+        label: "Recommendations",
+        long: "Prioritized remediation plan from gap analysis. Each recommendation identifies a documentation problem, the affected feature area, and the estimated score lift from fixing it.",
+    },
+    totalPotentialLift: {
+        label: "Total Potential Lift",
+        long: "Aggregate potential score lift if all identified gaps were fixed. This is a conservative estimate — each gap targets the median of non-bottlenecked dimensions, not 100.",
+    },
+    failureMode: {
+        label: "Failure Mode",
+        long: "The type of failure the grader emitted. Cross-cutting modes apply to any dimension (api-error, model-limitation, false-floor, unclassified). Per-dimension extensions cover documentation problems (missing-docs, incorrect-docs, outdated-docs, poor-structure), spec adherence (spec-mismatch), tool use (tool-misuse, chaotic-process, missing-recovery), and knowledge probes (factual-error, incompleteness, currency-violation, hallucination).",
+    },
+    estimatedLift: {
+        label: "Estimated Lift",
+        long: "Estimated composite score improvement if this gap is fully fixed. Based on raising bottleneck dimensions to the median of non-bottlenecked dimensions.",
+    },
+    confidence: {
+        label: "Confidence",
+        long: "How confident we are in this diagnosis (D0049 ceiling-cross-check derivation). High = the grader's emitted failure mode agrees with the structural ceiling-decomposition signal. Medium = signals disagree (or the ceiling pattern is not informative for this score). Low = passing scores never classify; treat as absent.",
+    },
+    // -- Agent behavior --------------------------------------------------------
+    agentBehaviorOverview: {
+        label: "Agent Behavior",
+        long: "How AI agents interacted with your documentation during evaluation: what they searched for, which pages they visited, and how much time they spent on network requests.",
+    },
+    searchQueries: {
+        label: "Search Queries",
+        long: "The exact search queries agents used to find documentation. Helps you understand how agents discover your content and whether your docs appear for relevant queries.",
+    },
+    docSlugsVisited: {
+        label: "Unique Doc Slugs",
+        long: "Documentation page slugs that agents actually visited during evaluation. Compare against canonical docs to see if agents found the right pages.",
+    },
+    externalDomains: {
+        label: "External Domains",
+        long: "Non-Sanity domains that agents contacted during evaluation. High external domain counts may indicate agents couldn't find what they needed in your docs.",
+    },
+    avgDocPagesVisited: {
+        label: "Avg Pages Visited",
+        long: "Average number of documentation pages visited per test. Higher counts can mean agents need to consult many pages (complex task) or can't find the right one quickly.",
+    },
+    avgSearchesPerformed: {
+        label: "Avg Searches",
+        long: "Average number of web searches performed per test. High search counts can indicate docs are hard to discover through search engines.",
+    },
+    avgNetworkTimeMs: {
+        label: "Avg Network Time",
+        long: "Average time spent on network requests per test. Includes page fetches, search queries, and API calls.",
+    },
+    totalRequests: {
+        label: "Total Requests",
+        long: "Total number of HTTP requests the agent made during the test, including searches, page visits, and API calls.",
+    },
+    totalBytesDownloaded: {
+        label: "Total Bytes Downloaded",
+        long: "Total bytes downloaded by the agent. Large downloads may indicate the agent is fetching many pages or very large documents.",
+    },
+    // -- Dimension deltas -----------------------------------------------------
+    dimTaskCompletion: {
+        label: "Task Completion Δ",
+        long: "Change in task completion between runs. Positive means implementations are more complete.",
+    },
+    dimCodeCorrectness: {
+        label: "Code Correctness Δ",
+        long: "Change in code correctness between runs. Positive means better code quality.",
+    },
+    dimDocCoverage: {
+        label: "Doc Coverage Δ",
+        long: "Change in doc coverage between runs. Positive means the docs are providing more useful information.",
+    },
+    // -- Per-area trend delta ----------------------------------------------------
+    areaDelta: {
+        label: "Area Δ",
+        long: "Score change for this area compared to the previous evaluation run.",
+    },
+    // -- Source values -----------------------------------------------------------
+    sourceProduction: {
+        label: "Production",
+        long: "Production source — docs fetched from the live production dataset. Scores reflect what real users and AI agents experience today.",
+    },
+    sourceBranch: {
+        label: "Branch",
+        long: "Branch source — docs fetched from a branch or draft dataset. Use this to preview how content changes affect scores before publishing.",
+    },
+    sourceLocal: {
+        label: "Local",
+        long: "Local source — docs fetched from local files or a local dev server. Useful for testing doc changes before pushing.",
+    },
+    // -- Report list columns ----------------------------------------------------
+    reportScore: {
+        label: "Score",
+        long: "The overall ceiling composite for this evaluation run: Task Completion (50%), Code Correctness (25%), and Doc Coverage (25%), averaged across all feature areas.",
+    },
+    reportMode: {
+        label: "Mode",
+        long: "The evaluation mode determines which reference points are measured. Different modes test different aspects of how AI agents interact with documentation.",
+    },
+    reportTrigger: {
+        label: "Trigger",
+        long: "What initiated this evaluation run. Knowing the trigger helps you understand whether a score change was from a content edit, a code deploy, or a scheduled check.",
+    },
+    // -- Mode values -----------------------------------------------------------
+    modeBaseline: {
+        label: "Baseline",
+        long: "Baseline mode — tests the model with gold-standard docs injected directly. Measures ceiling performance (best the docs can do).",
+    },
+    modeFull: {
+        label: "Full",
+        long: "Full mode — runs baseline + agentic. Compares ceiling (injected docs) against actual (agent-retrieved docs) to measure retrieval gap and infrastructure efficiency.",
+    },
+    modeAgentic: {
+        label: "Agentic",
+        long: "Agentic mode — the AI agent finds docs on its own via web search. Measures real-world performance: can agents actually discover and use your documentation?",
+    },
+    modeObserved: {
+        label: "Observed",
+        long: "Observed mode — records how agents interact with docs without scoring. Captures search queries, pages visited, and browsing patterns for analysis.",
+    },
+    modeDebug: {
+        label: "Debug",
+        long: "Debug mode — a diagnostic run for pipeline development. May use non-standard configurations or limited task sets.",
+    },
+    // -- Trigger values --------------------------------------------------------
+    triggerManual: {
+        label: "Manual",
+        long: "Manually triggered — someone ran the evaluation pipeline by hand, either locally or via the Studio UI.",
+    },
+    triggerCi: {
+        label: "CI",
+        long: "CI-triggered — the evaluation ran automatically as part of a pull request or merge pipeline.",
+    },
+    triggerSchedule: {
+        label: "Scheduled",
+        long: "Scheduled — the evaluation ran on a recurring schedule (e.g. nightly or weekly) to track score trends over time.",
+    },
+    triggerWebhook: {
+        label: "Webhook",
+        long: "Webhook-triggered — a content change in Sanity triggered the evaluation automatically. Helps catch doc regressions early.",
+    },
+    triggerCrossRepo: {
+        label: "Cross-Repo",
+        long: "Cross-repo — triggered from another repository via the dispatch API. Used when external repos want to validate their docs against AILF tasks.",
+    },
+};

package/dist/_vendor/ailf-shared/help-content.d.ts

@@ -0,0 +1,10 @@
+/**
+ * Re-export of the build-generated help-topic table.
+ *
+ * The underlying file `src/generated/help-content.ts` is emitted by
+ * `scripts/extract-help.ts` and is gitignored. Run `pnpm extract-help`
+ * (invoked automatically by this package's `prebuild`) to (re)generate it.
+ *
+ * @see scripts/extract-help.ts
+ */
+export { HELP_TOPICS } from "./generated/help-content.js";

package/dist/_vendor/ailf-shared/help-content.js

@@ -0,0 +1,10 @@
+/**
+ * Re-export of the build-generated help-topic table.
+ *
+ * The underlying file `src/generated/help-content.ts` is emitted by
+ * `scripts/extract-help.ts` and is gitignored. Run `pnpm extract-help`
+ * (invoked automatically by this package's `prebuild`) to (re)generate it.
+ *
+ * @see scripts/extract-help.ts
+ */
+export { HELP_TOPICS } from "./generated/help-content.js";

package/dist/_vendor/ailf-shared/help-topics.d.ts

@@ -0,0 +1,26 @@
+/**
+ * Help topic type — extracted from `:::help` remark directives in
+ * `docs/**\/*.md` by `scripts/extract-help.ts`. Consumed by both the
+ * Studio plugin and the App SDK dashboard.
+ *
+ * The extraction script emits `src/generated/help-content.ts` (gitignored)
+ * within this package. Consumers import `HELP_TOPICS` from the package
+ * barrel rather than reaching into the generated path directly.
+ *
+ * @see scripts/extract-help.ts
+ * @see docs/design-docs/contextual-help-sidebar.md
+ */
+export interface HelpTopic {
+    /** URL-safe identifier — matches the #id in the :::help directive */
+    id: string;
+    /** Display title shown in the drawer header */
+    title: string;
+    /** Markdown body content (rendered in the drawer) */
+    body: string;
+    /** Source file path (for debugging / "Edit this page" links) */
+    source: string;
+    /** Related topic IDs — rendered as "See also" links */
+    related?: string[];
+    /** Tags for search/filtering */
+    tags?: string[];
+}

package/dist/_vendor/ailf-shared/help-topics.js

@@ -0,0 +1 @@
+export {};

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

@@ -21,6 +21,10 @@ export { computeCanaryDrift, type CanaryDriftReport, type CanaryReportSlim, type
 export { type DocumentRef } from "./document-ref.js";
 export { makeEditorialReference, type EditorialReference, type MakeEditorialReferenceArgs, } from "./editorial-reference.js";
 export { FEATURE_FLAGS, type FeatureFlag, type FeatureFlagKey, } from "./feature-flags.js";
+export { DEFAULT_GCS_ARTIFACT_BUCKET } from "./gcs-defaults.js";
+export { GLOSSARY, type GlossaryEntry, type GlossarySlug } from "./glossary.js";
+export { HELP_TOPICS } from "./help-content.js";
+export { type HelpTopic } from "./help-topics.js";
 export { GRADE_BOUNDARIES, scoreGrade, type ScoreGrade, } from "./score-grades.js";
 export { NOISE_THRESHOLD } from "./noise-threshold.js";
 export { CANONICAL_EVAL_MODES, LEGACY_EVAL_MODE_ALIASES, LITERACY_VARIANTS, RAW_EVAL_MODES, type EvalMode, type LiteracyVariant, type RawEvalMode, } from "./eval-modes.js";

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

@@ -20,6 +20,9 @@
 export { computeCanaryDrift, } from "./canary-drift.js";
 export { makeEditorialReference, } from "./editorial-reference.js";
 export { FEATURE_FLAGS, } from "./feature-flags.js";
+export { DEFAULT_GCS_ARTIFACT_BUCKET } from "./gcs-defaults.js";
+export { GLOSSARY } from "./glossary.js";
+export { HELP_TOPICS } from "./help-content.js";
 export { GRADE_BOUNDARIES, scoreGrade, } from "./score-grades.js";
 export { NOISE_THRESHOLD } from "./noise-threshold.js";
 export { CANONICAL_EVAL_MODES, LEGACY_EVAL_MODE_ALIASES, LITERACY_VARIANTS, RAW_EVAL_MODES, } from "./eval-modes.js";

package/dist/composition-root.js

@@ -18,6 +18,7 @@
 import { promises as fs } from "node:fs";
 import path from "node:path";
 import { ARTIFACT_EXPORT_PHASE_ID, DIAGNOSIS_CARD_GENERATORS, InMemoryPluginRegistry, NoOpArtifactWriter, NoOpProgressReporter, createDiagnosisRunner, createLLMClient, generateRunId, isArtifactType, modelId, } from "./_vendor/ailf-core/index.js";
+import { DEFAULT_GCS_ARTIFACT_BUCKET } from "./_vendor/ailf-shared/index.js";
 import { JudgmentAttributionSchema } from "./adapters/attribution/per-entry-attribution-writer.js";
 import { AccumulatingArtifactWriter } from "./artifact-capture/accumulating-artifact-writer.js";
 import { ApiGatewayArtifactWriter } from "./artifact-capture/api-gateway-artifact-writer.js";
@@ -162,12 +163,13 @@ function createProgressReporter() {
     });
 }
 /**
- * Shared GCS bucket for report artifacts. Matches the gateway default at
- * packages/api/src/routes/artifacts.ts — both sides assume ailf-artifacts
- * unless explicitly overridden. The gateway's signing credentials are scoped
- * to this bucket, so alternate names require reconfiguring the gateway.
+ * Shared GCS bucket for report artifacts. Canonical default lives in
+ * `@sanity/ailf-shared` so the gateway (`packages/api/src/routes/runs.ts`)
+ * and the dashboard read the same value. The gateway's signing credentials
+ * are scoped to this bucket, so alternate names require reconfiguring the
+ * gateway in addition to setting `AILF_GCS_ARTIFACT_BUCKET`.
  */
-const DEFAULT_ARTIFACT_BUCKET = "ailf-artifacts";
+const DEFAULT_ARTIFACT_BUCKET = DEFAULT_GCS_ARTIFACT_BUCKET;
 /**
  * D0033 M4 default root for local artifacts when `--artifacts-dir` is unset.
  * Mirrors the pre-W0050 capture root so existing dev tooling (Studio

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

@@ -10,21 +10,25 @@
  * Designed to run in any HTTP environment: Cloudflare Workers, Vercel
  * functions, Express, Hono, etc.
  *
- * Supports two scoping modes:
- * - **Release-scoped** — requires `perspective` field
- * - **Task-scoped** — requires `tasks` array (optionally with `areas`)
- *
- * At least one of `perspective` or `tasks` must be present.
+ * The eval-request document carries a canonical `PipelineRequest` JSON
+ * blob in its `pipelineRequest` field (see W0239). The handler parses it
+ * via `PipelineRequestSchema` from `@sanity/ailf-core` and forwards it
+ * to the dispatcher as-is. Scoping (release-scoped via `perspective`,
+ * task-scoped via `tasks`) is asserted on the parsed `PipelineRequest`
+ * — at least one must be present.
  *
  * Flow:
  * 1. Receive eval request payload (from Sanity webhook projection)
- * 2. Validate: must be `ailf.evalRequest` type, `pending` status,
- *    with either `perspective` or `tasks`
- * 3. Dispatch evaluation to GitHub Actions via `repository_dispatch`
- *    with `external-eval` event type and scoped client payload
- * 4. On success: PATCH the eval request document → `status: "dispatched"`
- * 5. On failure: PATCH the eval request document → `status: "failed"` + error
- * 6. Return a structured result
+ * 2. Validate envelope: must be `ailf.evalRequest` type, `pending` status,
+ *    `pipelineRequest` present
+ * 3. Parse + Zod-validate `pipelineRequest` against `PipelineRequestSchema`
+ * 4. Assert scoping: parsed request must have `perspective` or `tasks`
+ * 5. Dispatch evaluation to GitHub Actions via `repository_dispatch`
+ *    with `external-eval` event type — the parsed `PipelineRequest`
+ *    rides as `client_payload.request` unchanged
+ * 6. On success: PATCH the eval request document → `status: "dispatched"`
+ * 7. On failure: PATCH the eval request document → `status: "failed"` + error
+ * 8. Return a structured result
  *
  * ## Sanity Manage Webhook Configuration
  *
@@ -44,38 +48,37 @@
  * @see .github/workflows/external-eval.yml — receiving workflow
  * @see docs/design-docs/report-store/visibility-workflows.md
  */
-/** Projected shape of an `ailf.evalRequest` document from a Sanity webhook. */
+/**
+ * Projected shape of an `ailf.evalRequest` document from a Sanity webhook.
+ *
+ * Per the W0239 schema redesign, request-scope fields (mode, perspective,
+ * tasks, areas, debug, tag, etc.) ride inside the `pipelineRequest` JSON
+ * blob — the canonical `PipelineRequest` serialization. The handler parses
+ * it via `PipelineRequestSchema` from `@sanity/ailf-core` and forwards it
+ * to the dispatcher as-is.
+ */
 export interface EvalRequestPayload {
     /** The Sanity document _id */
     _id: string;
     /** The Sanity document _type (should be "ailf.evalRequest") */
     _type: string;
-    /** Feature areas to scope the evaluation (task-scoped evals) */
-    areas?: string[];
-    /** Sanity dataset */
+    /** Sanity dataset hosting the eval-request document itself */
     dataset: string;
-    /** Run in debug mode */
-    debug?: boolean;
     /** Error message (only if status is "failed") */
     error?: string;
-    /** Evaluation mode */
-    mode: string;
-    /** Content release perspective ID (release-scoped evals) */
-    perspective?: string;
-    /** Sanity project ID */
+    /**
+     * Canonical `PipelineRequest` JSON. Source of truth for the dispatch
+     * payload. Parses against `PipelineRequestSchema` from `@sanity/ailf-core`.
+     */
+    pipelineRequest: string;
+    /** Sanity project ID hosting the eval-request document itself */
     projectId: string;
     /** ISO datetime of when the request was created */
     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 */
-    tag?: string;
-    /** Specific task IDs to evaluate (task-scoped evals) */
-    tasks?: string[];
 }
 /** Configuration for the eval request handler. */
 export interface EvalRequestHandlerConfig {