@sanity/ailf 2.0.0 → 2.0.1

package/dist/lib/discovery-report.js

@@ -0,0 +1,57 @@
+/**
+ * lib/discovery-report.ts — DEPRECATED re-export shim.
+ *
+ * The real implementation has moved to pipeline/discovery-report.ts.
+ *
+ * @deprecated Import from ../pipeline/discovery-report.js instead.
+ */
+import { existsSync, readFileSync, writeFileSync } from "node:fs";
+import { dirname, join, resolve } from "node:path";
+import { fileURLToPath } from "node:url";
+export { formatDiscoveryMarkdown, generateDiscoveryReport, } from "../pipeline/discovery-report.js";
+import { formatDiscoveryMarkdown, generateDiscoveryReport, } from "../pipeline/discovery-report.js";
+const __dirname = dirname(fileURLToPath(import.meta.url));
+const ROOT = resolve(__dirname, "..", "..");
+/**
+ * Legacy main() entry point.
+ * @deprecated Use generateDiscoveryReport() + formatDiscoveryMarkdown() directly.
+ */
+export function main() {
+    const args = process.argv.slice(2);
+    let output;
+    const areaFilter = [];
+    let summaryPath = join(ROOT, "results", "latest", "score-summary.json");
+    for (let i = 0; i < args.length; i++) {
+        if (args[i] === "--output" && args[i + 1]) {
+            output = args[++i];
+        }
+        else if (args[i] === "--area" && args[i + 1]) {
+            areaFilter.push(...args[++i].split(","));
+        }
+        else if (args[i] === "--input" && args[i + 1]) {
+            summaryPath = args[++i];
+        }
+        else if (!args[i].startsWith("-")) {
+            summaryPath = args[i];
+        }
+    }
+    if (!existsSync(summaryPath)) {
+        console.error(`❌ Score summary not found: ${summaryPath}`);
+        console.error("Run an agentic evaluation first: pnpm pipeline -- --mode agentic");
+        process.exit(1);
+    }
+    const summary = JSON.parse(readFileSync(summaryPath, "utf-8"));
+    const report = generateDiscoveryReport(summary, areaFilter.length > 0 ? areaFilter : undefined);
+    const markdown = formatDiscoveryMarkdown(report);
+    if (output) {
+        writeFileSync(output, markdown, "utf-8");
+        console.log(`✅ Discovery report written to ${output}`);
+    }
+    else {
+        console.log(markdown);
+    }
+}
+if (process.argv[1]?.endsWith("discovery-report.ts") ||
+    process.argv[1]?.endsWith("discovery-report.js")) {
+    main();
+}

package/dist/lib/fetch-docs.d.ts

@@ -0,0 +1,30 @@
+/**
+ * Fetch-docs.ts — CLI entry point for documentation fetching.
+ *
+ * This is a standalone script for direct CLI invocation:
+ *   npx tsx src/lib/fetch-docs.ts [--source name] [--include-feature-areas] [--include-corpus]
+ *
+ * The pipeline uses ctx.docFetcher (SanityDocFetcher adapter) directly via
+ * FetchDocsStep — this file is NOT called by the pipeline.
+ *
+ * Capabilities:
+ *   1. Canonical contexts     — delegates to SanityDocFetcher adapter (always)
+ *   2. Feature-area contexts  — one file per GROQ feature area query
+ *      (opt-in via --include-feature-areas)
+ *   3. Full corpus            — all articles in one file
+ *      (opt-in via --include-corpus)
+ */
+import "dotenv/config";
+import { type ResolvedSourceConfig } from "../sources.js";
+/** Options for the fetch-docs main() function. */
+export interface FetchDocsOptions {
+    /** Fetch full corpus (for retrieval experiments) */
+    includeCorpus?: boolean;
+    /** Fetch feature-area contexts in addition to canonical contexts */
+    includeFeatureAreas?: boolean;
+    /** Pre-resolved source config (skips loadSource() call) */
+    resolvedSource?: ResolvedSourceConfig;
+    /** Documentation source name (e.g., "branch", "local") */
+    source?: string;
+}
+export declare function main(options?: FetchDocsOptions): Promise<void>;

package/dist/lib/fetch-docs.js

@@ -0,0 +1,171 @@
+/**
+ * Fetch-docs.ts — CLI entry point for documentation fetching.
+ *
+ * This is a standalone script for direct CLI invocation:
+ *   npx tsx src/lib/fetch-docs.ts [--source name] [--include-feature-areas] [--include-corpus]
+ *
+ * The pipeline uses ctx.docFetcher (SanityDocFetcher adapter) directly via
+ * FetchDocsStep — this file is NOT called by the pipeline.
+ *
+ * Capabilities:
+ *   1. Canonical contexts     — delegates to SanityDocFetcher adapter (always)
+ *   2. Feature-area contexts  — one file per GROQ feature area query
+ *      (opt-in via --include-feature-areas)
+ *   3. Full corpus            — all articles in one file
+ *      (opt-in via --include-corpus)
+ */
+// oxlint-disable-next-line import/no-unassigned-import -- side-effect: loads .env into process.env
+import "dotenv/config";
+import { mkdirSync, writeFileSync } from "fs";
+import { dirname, join } from "path";
+import { SanityDocFetcher } from "../adapters/doc-fetchers/sanity-doc-fetcher.js";
+import { YamlTaskSource } from "../adapters/task-sources/index.js";
+import { getSanityClient } from "../sanity/client.js";
+import { toMarkdown } from "../sanity/portable-text.js";
+import { ALL_ARTICLES_QUERY, ALL_FEATURE_AREAS, FEATURE_AREA_QUERIES, } from "../sanity/queries.js";
+import { loadSource } from "../sources.js";
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+const ROOT = join(dirname(new URL(import.meta.url).pathname), "..", "..");
+function escapeNunjucks(text) {
+    return text.replace(/\{\{|\}\}/g, (match) => match === "{{" ? '{{ "{{" }}' : '{{ "}}" }}');
+}
+function estimateTokens(text) {
+    return Math.ceil(text.length / 4);
+}
+function formatArticle(doc) {
+    const sectionLabel = doc.section ? `Section: ${doc.section.title}\n` : "";
+    const desc = doc.description ? `${doc.description}\n\n` : "";
+    const markdown = toMarkdown(doc.content ?? []);
+    return `## ${doc.title}\n\n${sectionLabel}${desc}${markdown}`;
+}
+// ---------------------------------------------------------------------------
+// Feature-area contexts (opt-in, not part of the port)
+// ---------------------------------------------------------------------------
+async function generateFeatureAreaContexts(source) {
+    const client = getSanityClient(undefined, source);
+    const contextsDir = join(ROOT, "contexts");
+    mkdirSync(contextsDir, { recursive: true });
+    console.log("Generating feature-area contexts...\n");
+    for (const feature of ALL_FEATURE_AREAS) {
+        const query = FEATURE_AREA_QUERIES[feature];
+        const docs = await client.fetch(query);
+        if (docs.length === 0) {
+            console.warn(`  [warn] No articles found for "${feature}"`);
+            continue;
+        }
+        const combined = docs.map(formatArticle).join("\n\n---\n\n");
+        const outPath = join(contextsDir, `${feature}.md`);
+        writeFileSync(outPath, escapeNunjucks(combined));
+        console.log(`  ${feature}: ${docs.length} articles, ~${estimateTokens(combined)} tokens`);
+    }
+}
+// ---------------------------------------------------------------------------
+// Full corpus (opt-in, not part of the port)
+// ---------------------------------------------------------------------------
+async function generateFullCorpus(source) {
+    const client = getSanityClient(undefined, source);
+    console.log("\nGenerating full corpus...");
+    const docs = await client.fetch(ALL_ARTICLES_QUERY);
+    const corpus = docs
+        .map((d) => {
+        const ptBlocks = d.content ?? [];
+        const markdown = toMarkdown(ptBlocks);
+        return (`## ${d.title}\n\n` +
+            // oxlint-disable-next-line @typescript-eslint/prefer-nullish-coalescing -- empty string title should fall back to "General"
+            `Section: ${d.section?.title || "General"}\n` +
+            `URL: ${source.baseUrl}/${d.slug}\n\n` +
+            markdown);
+    })
+        .join("\n\n---\n\n");
+    const outDir = join(ROOT, "contexts");
+    mkdirSync(outDir, { recursive: true });
+    writeFileSync(join(outDir, "full-corpus.md"), escapeNunjucks(corpus));
+    console.log(`  full-corpus.md: ${docs.length} articles, ~${estimateTokens(corpus)} tokens`);
+}
+export async function main(options) {
+    console.log("=== ai-literacy-framework — Documentation Fetcher ===\n");
+    const args = process.argv.slice(2);
+    // Options win over argv (backward compatible)
+    const includeFeatureAreas = options?.includeFeatureAreas ?? args.includes("--include-feature-areas");
+    const includeCorpus = options?.includeCorpus ?? args.includes("--include-corpus");
+    // Pre-resolved source wins over name-based lookup (backward compatible)
+    const sourceIdx = args.indexOf("--source");
+    const sourceName = options?.source ?? (sourceIdx !== -1 ? args[sourceIdx + 1] : undefined);
+    const source = options?.resolvedSource ?? loadSource(sourceName);
+    console.log(`  Source: ${sourceName ?? "default (production)"}`);
+    console.log(`  Base URL: ${source.baseUrl}`);
+    if (source.dataset) {
+        console.log(`  Dataset: ${source.dataset}`);
+    }
+    if (source.perspective) {
+        console.log(`  Perspective: ${source.perspective}`);
+    }
+    if (source.documentIds && source.documentIds.length > 0) {
+        console.log(`  Documents: ${source.documentIds.length} document ID(s)`);
+        for (const id of source.documentIds) {
+            console.log(`             ${id}`);
+        }
+    }
+    if (source.urls.length > 0) {
+        console.log(`  URLs:      ${source.urls.length} direct URL(s)`);
+        for (const u of source.urls) {
+            console.log(`             ${u}`);
+        }
+    }
+    console.log();
+    if (includeFeatureAreas) {
+        await generateFeatureAreaContexts(source);
+    }
+    // Canonical contexts — delegate to the SanityDocFetcher adapter
+    const taskSource = new YamlTaskSource(ROOT);
+    const tasks = await taskSource.loadTasks();
+    const tasksWithDocs = tasks.filter((t) => t.canonicalDocs.length > 0);
+    if (tasksWithDocs.length > 0) {
+        console.log("\nGenerating canonical (gold-retrieval) contexts...\n");
+        const fetcher = new SanityDocFetcher(ROOT);
+        const result = await fetcher.fetch(tasksWithDocs, source);
+        // Write metadata files inline (CLI mode)
+        const contextsDir = join(ROOT, "contexts");
+        mkdirSync(contextsDir, { recursive: true });
+        if (result.metadata?.manifest) {
+            const path = join(contextsDir, "document-manifest.json");
+            writeFileSync(path, JSON.stringify(result.metadata.manifest, null, 2));
+            console.log(`\n  📋 Document manifest: ${result.metadata.manifest.length} docs → contexts/document-manifest.json`);
+        }
+        if (result.metadata?.releaseImpact) {
+            const path = join(contextsDir, "release-impact.json");
+            writeFileSync(path, JSON.stringify(result.metadata.releaseImpact, null, 2));
+            console.log("  📄 Release impact written to contexts/release-impact.json");
+        }
+        if (result.metadata?.documentOverlay) {
+            const path = join(contextsDir, "document-overlay.json");
+            writeFileSync(path, JSON.stringify(result.metadata.documentOverlay, null, 2));
+            console.log("  📄 Document overlay written to contexts/document-overlay.json");
+        }
+        if (result.metadata?.urlFetch) {
+            const path = join(contextsDir, "url-fetch.json");
+            writeFileSync(path, JSON.stringify(result.metadata.urlFetch, null, 2));
+            console.log("  📄 URL fetch metadata written to contexts/url-fetch.json");
+        }
+        console.log(`\n  Canonical contexts: ${result.contexts.length} tasks`);
+        for (const ctx of result.contexts) {
+            console.log(`    ${ctx.taskId}: ${ctx.slugs.length} doc(s), ~${ctx.tokenCount ?? 0} tokens`);
+        }
+    }
+    if (includeCorpus) {
+        await generateFullCorpus(source);
+    }
+    console.log("\nDone!");
+}
+// ---------------------------------------------------------------------------
+// Main — only run when invoked directly
+// ---------------------------------------------------------------------------
+if (process.argv[1]?.endsWith("fetch-docs.ts") ||
+    process.argv[1]?.endsWith("fetch-docs.js")) {
+    main().catch((err) => {
+        console.error("Fatal error:", err);
+        process.exit(1);
+    });
+}

package/dist/lib/generate-configs.d.ts

@@ -0,0 +1,25 @@
+/**
+ * lib/generate-configs.ts — DEPRECATED re-export shim.
+ *
+ * The real implementation has moved to pipeline/generate-configs.ts.
+ * This shim preserves backward compatibility for:
+ *   - Direct CLI invocation: `tsx src/lib/generate-configs.ts`
+ *   - Test imports that haven't been updated yet
+ *
+ * TODO: Update all importers to use pipeline/generate-configs.ts, then delete this file.
+ *
+ * @deprecated Import from ../pipeline/generate-configs.js instead.
+ */
+export { discoverTaskFiles, extractModelName, extractProvider, generateConfigs, loadPrompts, mergeConfig, modelMatchesMode, type GenerateConfigsOptions, } from "../pipeline/generate-configs.js";
+import type { GenerateConfigsOptions } from "../pipeline/generate-configs.js";
+/**
+ * Legacy main() entry point — wraps generateConfigs() with env var fallbacks.
+ *
+ * Only used for direct CLI invocation (`tsx src/lib/generate-configs.ts`).
+ * Command handlers and orchestration steps should call generateConfigs() directly.
+ *
+ * @deprecated Use generateConfigs() from pipeline/generate-configs.ts instead.
+ */
+export declare function main(options?: Omit<GenerateConfigsOptions, "rootDir"> & {
+    rootDir?: string;
+}): void;

package/dist/lib/generate-configs.js

@@ -0,0 +1,42 @@
+/**
+ * lib/generate-configs.ts — DEPRECATED re-export shim.
+ *
+ * The real implementation has moved to pipeline/generate-configs.ts.
+ * This shim preserves backward compatibility for:
+ *   - Direct CLI invocation: `tsx src/lib/generate-configs.ts`
+ *   - Test imports that haven't been updated yet
+ *
+ * TODO: Update all importers to use pipeline/generate-configs.ts, then delete this file.
+ *
+ * @deprecated Import from ../pipeline/generate-configs.js instead.
+ */
+import { dirname, resolve } from "path";
+import { fileURLToPath } from "url";
+// Re-export everything from the real implementation
+export { discoverTaskFiles, extractModelName, extractProvider, generateConfigs, loadPrompts, mergeConfig, modelMatchesMode, } from "../pipeline/generate-configs.js";
+import { generateConfigs } from "../pipeline/generate-configs.js";
+const __dirname = dirname(fileURLToPath(import.meta.url));
+const ROOT = resolve(__dirname, "../..");
+/**
+ * Legacy main() entry point — wraps generateConfigs() with env var fallbacks.
+ *
+ * Only used for direct CLI invocation (`tsx src/lib/generate-configs.ts`).
+ * Command handlers and orchestration steps should call generateConfigs() directly.
+ *
+ * @deprecated Use generateConfigs() from pipeline/generate-configs.ts instead.
+ */
+export function main(options) {
+    generateConfigs({
+        rootDir: options?.rootDir ?? ROOT,
+        allowedOrigins: options?.allowedOrigins,
+        filter: options?.filter,
+        resolvedSource: options?.resolvedSource,
+        searchMode: options?.searchMode,
+        source: options?.source ?? process.argv[2] ?? process.env.DOC_SOURCE,
+    });
+}
+// Only run when invoked directly (not when imported for testing)
+if (process.argv[1]?.endsWith("generate-configs.ts") ||
+    process.argv[1]?.endsWith("generate-configs.js")) {
+    main();
+}

package/dist/lib/grader-api.d.ts

@@ -0,0 +1,21 @@
+/**
+ * lib/grader-api.ts — DEPRECATED re-export shim.
+ *
+ * The real implementation has moved to pipeline/grader-api.ts.
+ * This shim preserves backward compatibility for:
+ *   - Direct CLI invocation and other lib/ files that import from here
+ *   - Test imports that haven't been updated yet
+ *
+ * TODO: Update all importers to use pipeline/grader-api.ts, then delete this file.
+ *
+ * @deprecated Import from ../pipeline/grader-api.js instead.
+ */
+export { extractScore, gradeOnce, resolveProvider, } from "../pipeline/grader-api.js";
+/**
+ * Backward-compatible loadGraderModel that uses EVAL_ROOT.
+ * @deprecated Use loadGraderModel(rootDir) from pipeline/grader-api.js instead.
+ */
+export declare function loadGraderModel(): {
+    id: string;
+    label: string;
+};

package/dist/lib/grader-api.js

@@ -0,0 +1,34 @@
+/**
+ * lib/grader-api.ts — DEPRECATED re-export shim.
+ *
+ * The real implementation has moved to pipeline/grader-api.ts.
+ * This shim preserves backward compatibility for:
+ *   - Direct CLI invocation and other lib/ files that import from here
+ *   - Test imports that haven't been updated yet
+ *
+ * TODO: Update all importers to use pipeline/grader-api.ts, then delete this file.
+ *
+ * @deprecated Import from ../pipeline/grader-api.js instead.
+ */
+import { config as dotenvConfig } from "dotenv";
+import { existsSync } from "fs";
+import { dirname, resolve } from "path";
+import { fileURLToPath } from "url";
+// Re-export everything from the real implementation
+export { extractScore, gradeOnce, resolveProvider, } from "../pipeline/grader-api.js";
+import { loadGraderModel as _loadGraderModel } from "../pipeline/grader-api.js";
+const __dirname = dirname(fileURLToPath(import.meta.url));
+const EVAL_ROOT = resolve(__dirname, "..", "..");
+// Load root .env (two levels above packages/eval/) so API keys are available
+// even when this module is invoked via `tsx` from packages/eval/.
+const rootEnvPath = resolve(EVAL_ROOT, "..", "..", ".env");
+if (existsSync(rootEnvPath)) {
+    dotenvConfig({ override: true, path: rootEnvPath });
+}
+/**
+ * Backward-compatible loadGraderModel that uses EVAL_ROOT.
+ * @deprecated Use loadGraderModel(rootDir) from pipeline/grader-api.js instead.
+ */
+export function loadGraderModel() {
+    return _loadGraderModel(EVAL_ROOT);
+}

package/dist/lib/grader-compare.d.ts

@@ -0,0 +1,19 @@
+/**
+ * lib/grader-compare.ts — DEPRECATED re-export shim.
+ *
+ * The real implementation has moved to pipeline/grader-compare-runner.ts.
+ * The pure analysis functions live in pipeline/grader-comparison.ts.
+ * This shim preserves backward compatibility for direct CLI invocation.
+ *
+ * TODO: Update all importers to use pipeline/ modules, then delete this file.
+ *
+ * @deprecated Import from ../pipeline/grader-compare-runner.js instead.
+ */
+import "dotenv/config";
+export { formatComparisonReport, runGraderCompare, type GraderCompareRunnerOptions, } from "../pipeline/grader-compare-runner.js";
+export { compareGraders, type GraderComparison, type GraderScore, type GraderScoreSet, } from "../pipeline/grader-comparison.js";
+/**
+ * Legacy main() entry point.
+ * @deprecated Use runGraderCompare() from pipeline/grader-compare-runner.js instead.
+ */
+export declare function main(): Promise<void>;

package/dist/lib/grader-compare.js

@@ -0,0 +1,91 @@
+/**
+ * lib/grader-compare.ts — DEPRECATED re-export shim.
+ *
+ * The real implementation has moved to pipeline/grader-compare-runner.ts.
+ * The pure analysis functions live in pipeline/grader-comparison.ts.
+ * This shim preserves backward compatibility for direct CLI invocation.
+ *
+ * TODO: Update all importers to use pipeline/ modules, then delete this file.
+ *
+ * @deprecated Import from ../pipeline/grader-compare-runner.js instead.
+ */
+// oxlint-disable-next-line import/no-unassigned-import -- side-effect: loads .env into process.env
+import "dotenv/config";
+import { dirname, resolve } from "path";
+import { fileURLToPath } from "url";
+import { load } from "js-yaml";
+import { existsSync, readFileSync } from "fs";
+// Re-export from pipeline modules
+export { formatComparisonReport, runGraderCompare, } from "../pipeline/grader-compare-runner.js";
+export { compareGraders, } from "../pipeline/grader-comparison.js";
+import { runGraderCompare } from "../pipeline/grader-compare-runner.js";
+const __dirname = dirname(fileURLToPath(import.meta.url));
+const ROOT = resolve(__dirname, "..", "..");
+function parseCliArgs() {
+    const args = process.argv.slice(2);
+    function getAllOptions(name) {
+        const results = [];
+        const flag = `--${name}`;
+        for (let i = 0; i < args.length; i++) {
+            if (args[i] === flag && i + 1 < args.length) {
+                results.push(args[i + 1]);
+            }
+        }
+        return results;
+    }
+    function getOption(name) {
+        const idx = args.indexOf(`--${name}`);
+        return idx !== -1 && idx + 1 < args.length ? args[idx + 1] : undefined;
+    }
+    const candidateArgs = getAllOptions("candidate");
+    let candidates;
+    if (candidateArgs.length > 0) {
+        candidates = candidateArgs.map((id) => ({
+            id,
+            label: id.split(":").pop() ?? id,
+        }));
+    }
+    else {
+        const modelsPath = resolve(ROOT, "config", "models.yaml");
+        if (existsSync(modelsPath)) {
+            const raw = readFileSync(modelsPath, "utf-8");
+            const data = load(raw);
+            const configCandidates = data?.["grader-candidates"] ?? [];
+            candidates = configCandidates.map((c) => ({
+                id: c.id,
+                label: c.label ?? c.id.split(":").pop() ?? c.id,
+            }));
+        }
+        else {
+            candidates = [];
+        }
+    }
+    return {
+        candidates,
+        format: getOption("format") ?? "table",
+        outputPath: getOption("output"),
+        resultsPath: getOption("results") ?? "results/latest/eval-results.json",
+    };
+}
+/**
+ * Legacy main() entry point.
+ * @deprecated Use runGraderCompare() from pipeline/grader-compare-runner.js instead.
+ */
+export async function main() {
+    const { candidates, format, outputPath, resultsPath } = parseCliArgs();
+    await runGraderCompare({
+        candidates,
+        format: format,
+        outputPath,
+        resultsPath,
+        rootDir: ROOT,
+    });
+}
+// Only run when invoked directly
+if (process.argv[1]?.endsWith("grader-compare.ts") ||
+    process.argv[1]?.endsWith("grader-compare.js")) {
+    main().catch((err) => {
+        console.error("❌ Fatal error:", err);
+        process.exit(1);
+    });
+}

package/dist/lib/grader-consistency.d.ts

@@ -0,0 +1,27 @@
+/**
+ * lib/grader-consistency.ts — DEPRECATED re-export shim.
+ *
+ * The real implementation has moved to pipeline/grader-consistency-runner.ts.
+ * The pure analysis functions live in pipeline/grader-consistency.ts.
+ * This shim preserves backward compatibility for direct CLI invocation.
+ *
+ * TODO: Update all importers to use pipeline/ modules, then delete this file.
+ *
+ * @deprecated Import from ../pipeline/grader-consistency-runner.js instead.
+ */
+import "dotenv/config";
+export { extractGradingJudgments, formatConsistencyReport, runGraderConsistency, type GraderConsistencyRunnerOptions, } from "../pipeline/grader-consistency-runner.js";
+export { analyzeConsistency, type GraderConsistency, type ReplicatedGrading, } from "../pipeline/grader-consistency.js";
+/** Options for the legacy main() function. */
+export interface GraderConsistencyOptions {
+    /** Number of additional grading replications (default: 5) */
+    replications?: number;
+    /** Path to eval-results.json */
+    resultsPath?: string;
+}
+/**
+ * Legacy main() entry point — wraps runGraderConsistency() with CLI arg parsing.
+ *
+ * @deprecated Use runGraderConsistency() from pipeline/grader-consistency-runner.js instead.
+ */
+export declare function main(options?: GraderConsistencyOptions): Promise<void>;

package/dist/lib/grader-consistency.js

@@ -0,0 +1,79 @@
+/**
+ * lib/grader-consistency.ts — DEPRECATED re-export shim.
+ *
+ * The real implementation has moved to pipeline/grader-consistency-runner.ts.
+ * The pure analysis functions live in pipeline/grader-consistency.ts.
+ * This shim preserves backward compatibility for direct CLI invocation.
+ *
+ * TODO: Update all importers to use pipeline/ modules, then delete this file.
+ *
+ * @deprecated Import from ../pipeline/grader-consistency-runner.js instead.
+ */
+// oxlint-disable-next-line import/no-unassigned-import -- side-effect: loads .env into process.env
+import "dotenv/config";
+import { dirname, join, resolve } from "path";
+import { fileURLToPath } from "url";
+// Re-export from pipeline modules
+export { extractGradingJudgments, formatConsistencyReport, runGraderConsistency, } from "../pipeline/grader-consistency-runner.js";
+export { analyzeConsistency, } from "../pipeline/grader-consistency.js";
+import { runGraderConsistency } from "../pipeline/grader-consistency-runner.js";
+const __dirname = dirname(fileURLToPath(import.meta.url));
+const ROOT = resolve(__dirname, "..", "..");
+// ---------------------------------------------------------------------------
+// CLI argument parsing (used when called from CLI)
+// ---------------------------------------------------------------------------
+function parseCliArgs() {
+    const args = process.argv.slice(2);
+    function getFlag(name) {
+        return args.includes(`--${name}`);
+    }
+    function getOption(name) {
+        const idx = args.indexOf(`--${name}`);
+        return idx !== -1 && idx + 1 < args.length ? args[idx + 1] : undefined;
+    }
+    const showHelp = getFlag("help") || getFlag("h");
+    if (showHelp) {
+        console.log(`
+Usage: pnpm grader-consistency [options]
+
+Measure grader consistency by re-grading existing eval responses N times.
+
+Options:
+  --replications <n>   Number of additional grading replications (default: 5)
+  --results <path>     Path to eval-results.json (default: results/latest/eval-results.json)
+  --help, -h           Show this help
+`);
+        process.exit(0);
+    }
+    const replicationsStr = getOption("replications") ?? "5";
+    return {
+        replications: parseInt(replicationsStr, 10),
+        resultsPath: getOption("results") ??
+            join(ROOT, "results", "latest", "eval-results.json"),
+    };
+}
+/**
+ * Legacy main() entry point — wraps runGraderConsistency() with CLI arg parsing.
+ *
+ * @deprecated Use runGraderConsistency() from pipeline/grader-consistency-runner.js instead.
+ */
+export async function main(options) {
+    const cliArgs = options ? undefined : parseCliArgs();
+    const replications = options?.replications ?? cliArgs?.replications ?? 5;
+    const resultsPath = options?.resultsPath ??
+        cliArgs?.resultsPath ??
+        join(ROOT, "results", "latest", "eval-results.json");
+    await runGraderConsistency({
+        replications,
+        resultsPath,
+        rootDir: ROOT,
+    });
+}
+// Only run when invoked directly
+if (process.argv[1]?.endsWith("grader-consistency.ts") ||
+    process.argv[1]?.endsWith("grader-consistency.js")) {
+    main().catch((err) => {
+        console.error("❌ Fatal error:", err);
+        process.exit(1);
+    });
+}

package/dist/lib/grader-sensitivity.d.ts

@@ -0,0 +1,19 @@
+/**
+ * lib/grader-sensitivity.ts — DEPRECATED re-export shim.
+ *
+ * The real implementation has moved to pipeline/grader-sensitivity-runner.ts.
+ * The pure analysis functions live in pipeline/grader-sensitivity.ts.
+ * This shim preserves backward compatibility for direct CLI invocation.
+ *
+ * TODO: Update all importers to use pipeline/ modules, then delete this file.
+ *
+ * @deprecated Import from ../pipeline/grader-sensitivity-runner.js instead.
+ */
+import "dotenv/config";
+export { formatSensitivityReport, runGraderSensitivity, type GraderSensitivityRunnerOptions, } from "../pipeline/grader-sensitivity-runner.js";
+export { analyzeSensitivity, type GraderSensitivityResult, type SensitivityPair, } from "../pipeline/grader-sensitivity.js";
+/**
+ * Legacy main() entry point.
+ * @deprecated Use runGraderSensitivity() from pipeline/grader-sensitivity-runner.js instead.
+ */
+export declare function main(): Promise<void>;