@cleocode/core 2026.9.22 → 2026.9.24

package/dist/decide/bench/spot-check.js

@@ -0,0 +1,221 @@
+/**
+ * Owner spot-check for the System One benchmark dataset (T12495).
+ *
+ * Before any spend, a small random sample stratified by site and label is
+ * written to a file. The orchestrator shows each item to the owner through
+ * the ask tool (the item carries the proposed label and the allowed labels
+ * as ready-made options) and writes the answers to a corrections file.
+ * {@link applyBenchCorrections} then relabels (or drops) those rows and marks
+ * them owner-verified. Both steps are offline: no provider is contacted.
+ *
+ * @task T12495
+ * @epic T12486
+ */
+import { readFile } from 'node:fs/promises';
+import { z } from 'zod';
+import { createSeededRandom, DEFAULT_BENCH_SEED, shuffled } from './stats.js';
+import { labelsForSite } from './types.js';
+/** Default spot-check sample size. */
+export const DEFAULT_SPOT_CHECK_SIZE = 30;
+/** Characters of each text shown in a spot-check item. */
+const SPOT_CHECK_TEXT_CHARS = 280;
+/** Clip a text for display. */
+function brief(text) {
+    const flat = text.replace(/\s+/g, ' ').trim();
+    return flat.length <= SPOT_CHECK_TEXT_CHARS
+        ? flat
+        : `${flat.slice(0, SPOT_CHECK_TEXT_CHARS - 1)}…`;
+}
+/** One-paragraph summary of a row. */
+function summarize(row) {
+    if (row.site === 'duplicateDetection') {
+        const { a, b } = row.input;
+        return `Are these the same work? [${a.id}] ${brief(a.title)} — ${brief(a.description)} || [${b.id}] ${brief(b.title)} — ${brief(b.description)}`;
+    }
+    if (row.site === 'observationType') {
+        return `Which type is this observation? [${row.input.id}] ${brief(row.input.title)} — ${brief(row.input.text)}`;
+    }
+    const { newer, older } = row.input;
+    return `Does the newer decision conflict with or replace the older one? NEWER [${newer.id}] ${brief(newer.decision)} (${brief(newer.rationale)}) || OLDER [${older.id}] ${brief(older.decision)} (${brief(older.rationale)})`;
+}
+/**
+ * Largest-remainder allocation of `size` slots over strata, at least one
+ * slot per stratum while slots last.
+ */
+function allocate(counts, size) {
+    const total = counts.reduce((s, c) => s + c, 0);
+    if (total === 0)
+        return counts.map(() => 0);
+    const target = Math.min(size, total);
+    const alloc = counts.map((c) => (c > 0 && target >= counts.length ? 1 : 0));
+    let left = target - alloc.reduce((s, a) => s + a, 0);
+    const exact = counts.map((c) => (c / total) * left);
+    for (const [i, e] of exact.entries()) {
+        const add = Math.min(Math.floor(e), (counts[i] ?? 0) - (alloc[i] ?? 0));
+        alloc[i] = (alloc[i] ?? 0) + add;
+        left -= add;
+    }
+    const order = exact
+        .map((e, i) => ({ i, r: e - Math.floor(e) }))
+        .sort((x, y) => y.r - x.r || x.i - y.i);
+    while (left > 0) {
+        let moved = false;
+        for (const { i } of order) {
+            if (left === 0)
+                break;
+            if ((alloc[i] ?? 0) < (counts[i] ?? 0)) {
+                alloc[i] = (alloc[i] ?? 0) + 1;
+                left -= 1;
+                moved = true;
+            }
+        }
+        if (!moved)
+            break;
+    }
+    return alloc;
+}
+/**
+ * Draw a random sample stratified by site and label.
+ *
+ * @param rows - Dataset rows.
+ * @param opts - Sample size (default {@link DEFAULT_SPOT_CHECK_SIZE}) and seed.
+ * @returns The spot-check file contents.
+ */
+export function sampleBenchSpotCheck(rows, opts = {}) {
+    const seed = opts.seed ?? DEFAULT_BENCH_SEED;
+    const random = createSeededRandom(seed ^ 0x5eed);
+    const strata = new Map();
+    for (const r of rows) {
+        const key = `${r.site}\u0000${r.label}`;
+        strata.set(key, [...(strata.get(key) ?? []), r]);
+    }
+    const keys = [...strata.keys()].sort();
+    const alloc = allocate(keys.map((k) => strata.get(k)?.length ?? 0), opts.size ?? DEFAULT_SPOT_CHECK_SIZE);
+    const items = [];
+    for (const [i, key] of keys.entries()) {
+        for (const row of shuffled(strata.get(key) ?? [], random).slice(0, alloc[i] ?? 0)) {
+            const allowed = labelsForSite(row.site);
+            items.push({
+                id: row.id,
+                site: row.site,
+                proposedLabel: row.label,
+                options: [row.label, ...allowed.filter((l) => l !== row.label)],
+                summary: summarize(row),
+                provenance: row.provenance,
+                ownerVerified: row.ownerVerified === true,
+            });
+        }
+    }
+    return {
+        schemaVersion: 1,
+        generatedAt: (opts.now ?? new Date()).toISOString(),
+        seed,
+        instructions: 'For each item, ask the owner (ask tool) to pick one of `options` (the proposed label is first) or "drop" when the pair is not a fair test. Write the answers as {"corrections":[{"id","label"} | {"id","drop":true}]} and pass the file to `cleo decide bench --corrections <file>`. Confirming the proposed label also marks the row owner-verified.',
+        items: shuffled(items, random),
+    };
+}
+/** Zod schema of one correction. */
+const correctionSchema = z.union([
+    z.object({
+        id: z.string().min(1),
+        label: z.string().min(1),
+        note: z.string().optional(),
+    }),
+    z.object({ id: z.string().min(1), drop: z.literal(true), note: z.string().optional() }),
+]);
+/** Zod schema of a corrections file: `{ corrections: [...] }` or a bare array. */
+const correctionsFileSchema = z.union([
+    z.object({ corrections: z.array(correctionSchema) }),
+    z.array(correctionSchema),
+]);
+/**
+ * Parse a corrections file.
+ *
+ * @param text - File contents (JSON).
+ * @returns The corrections.
+ * @throws Error when the JSON or its shape is invalid.
+ */
+export function parseBenchCorrections(text) {
+    let json;
+    try {
+        json = JSON.parse(text);
+    }
+    catch {
+        throw new Error('corrections file is not JSON');
+    }
+    const parsed = correctionsFileSchema.safeParse(json);
+    if (!parsed.success) {
+        throw new Error('corrections file must be {"corrections":[{"id","label"}|{"id","drop":true}]}');
+    }
+    return Array.isArray(parsed.data) ? parsed.data : parsed.data.corrections;
+}
+/**
+ * Read and parse a corrections file.
+ *
+ * @param path - Corrections file.
+ * @returns The corrections.
+ */
+export async function readBenchCorrections(path) {
+    return parseBenchCorrections(await readFile(path, 'utf-8'));
+}
+/**
+ * Apply owner corrections: relabel or drop the named rows and mark every
+ * corrected (or confirmed) row owner-verified. A relabelled row keeps its
+ * original label in `correctedFrom`.
+ *
+ * @param rows - Dataset rows.
+ * @param corrections - Owner answers.
+ * @returns The new rows and a receipt.
+ */
+export function applyBenchCorrections(rows, corrections) {
+    const byId = new Map(corrections.map((c) => [c.id, c]));
+    const seen = new Set();
+    const rejected = [];
+    let confirmed = 0;
+    let relabelled = 0;
+    let dropped = 0;
+    const out = [];
+    for (const row of rows) {
+        const c = byId.get(row.id);
+        if (!c) {
+            out.push(row);
+            continue;
+        }
+        seen.add(row.id);
+        const note = c.note ? { ownerNote: c.note } : {};
+        if ('drop' in c) {
+            dropped++;
+            continue;
+        }
+        if (!labelsForSite(row.site).includes(c.label)) {
+            rejected.push({ id: row.id, label: c.label });
+            out.push(row);
+            continue;
+        }
+        if (c.label === row.label) {
+            confirmed++;
+            out.push({ ...row, ...note, ownerVerified: true });
+        }
+        else {
+            relabelled++;
+            out.push({
+                ...row,
+                ...note,
+                label: c.label,
+                correctedFrom: row.correctedFrom ?? row.label,
+                ownerVerified: true,
+            });
+        }
+    }
+    return {
+        rows: out,
+        receipt: {
+            confirmed,
+            relabelled,
+            dropped,
+            unmatched: corrections.filter((c) => !seen.has(c.id)).map((c) => c.id),
+            rejected,
+        },
+    };
+}
+//# sourceMappingURL=spot-check.js.map

package/dist/decide/bench/spot-check.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"spot-check.js","sourceRoot":"","sources":["../../../src/decide/bench/spot-check.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAC5C,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,QAAQ,EAAE,MAAM,YAAY,CAAC;AAC9E,OAAO,EAAiB,aAAa,EAAE,MAAM,YAAY,CAAC;AAE1D,sCAAsC;AACtC,MAAM,CAAC,MAAM,uBAAuB,GAAG,EAAE,CAAC;AAE1C,0DAA0D;AAC1D,MAAM,qBAAqB,GAAG,GAAG,CAAC;AAkClC,+BAA+B;AAC/B,SAAS,KAAK,CAAC,IAAY;IACzB,MAAM,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC;IAC9C,OAAO,IAAI,CAAC,MAAM,IAAI,qBAAqB;QACzC,CAAC,CAAC,IAAI;QACN,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,qBAAqB,GAAG,CAAC,CAAC,GAAG,CAAC;AACrD,CAAC;AAED,sCAAsC;AACtC,SAAS,SAAS,CAAC,GAAa;IAC9B,IAAI,GAAG,CAAC,IAAI,KAAK,oBAAoB,EAAE,CAAC;QACtC,MAAM,EAAE,CAAC,EAAE,CAAC,EAAE,GAAG,GAAG,CAAC,KAAK,CAAC;QAC3B,OAAO,6BAA6B,CAAC,CAAC,EAAE,KAAK,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,WAAW,CAAC,QAAQ,CAAC,CAAC,EAAE,KAAK,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,WAAW,CAAC,EAAE,CAAC;IACnJ,CAAC;IACD,IAAI,GAAG,CAAC,IAAI,KAAK,iBAAiB,EAAE,CAAC;QACnC,OAAO,oCAAoC,GAAG,CAAC,KAAK,CAAC,EAAE,KAAK,KAAK,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC;IAClH,CAAC;IACD,MAAM,EAAE,KAAK,EAAE,KAAK,EAAE,GAAG,GAAG,CAAC,KAAK,CAAC;IACnC,OAAO,0EAA0E,KAAK,CAAC,EAAE,KAAK,KAAK,CAAC,KAAK,CAAC,QAAQ,CAAC,KAAK,KAAK,CAAC,KAAK,CAAC,SAAS,CAAC,eAAe,KAAK,CAAC,EAAE,KAAK,KAAK,CAAC,KAAK,CAAC,QAAQ,CAAC,KAAK,KAAK,CAAC,KAAK,CAAC,SAAS,CAAC,GAAG,CAAC;AAChO,CAAC;AAED;;;GAGG;AACH,SAAS,QAAQ,CAAC,MAAyB,EAAE,IAAY;IACvD,MAAM,KAAK,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC;IAChD,IAAI,KAAK,KAAK,CAAC;QAAE,OAAO,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC;IAC5C,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;IACrC,MAAM,KAAK,GAAa,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,MAAM,IAAI,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACtF,IAAI,IAAI,GAAG,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC;IACrD,MAAM,KAAK,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,GAAG,IAAI,CAAC,CAAC;IACpD,KAAK,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,KAAK,CAAC,OAAO,EAAE,EAAE,CAAC;QACrC,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACxE,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,GAAG,CAAC;QACjC,IAAI,IAAI,GAAG,CAAC;IACd,CAAC;IACD,MAAM,KAAK,GAAG,KAAK;SAChB,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;SAC5C,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;IAC1C,OAAO,IAAI,GAAG,CAAC,EAAE,CAAC;QAChB,IAAI,KAAK,GAAG,KAAK,CAAC;QAClB,KAAK,MAAM,EAAE,CAAC,EAAE,IAAI,KAAK,EAAE,CAAC;YAC1B,IAAI,IAAI,KAAK,CAAC;gBAAE,MAAM;YACtB,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC;gBACvC,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;gBAC/B,IAAI,IAAI,CAAC,CAAC;gBACV,KAAK,GAAG,IAAI,CAAC;YACf,CAAC;QACH,CAAC;QACD,IAAI,CAAC,KAAK;YAAE,MAAM;IACpB,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,oBAAoB,CAClC,IAAyB,EACzB,OAAgF,EAAE;IAElF,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,IAAI,kBAAkB,CAAC;IAC7C,MAAM,MAAM,GAAG,kBAAkB,CAAC,IAAI,GAAG,MAAM,CAAC,CAAC;IACjD,MAAM,MAAM,GAAG,IAAI,GAAG,EAAsB,CAAC;IAC7C,KAAK,MAAM,CAAC,IAAI,IAAI,EAAE,CAAC;QACrB,MAAM,GAAG,GAAG,GAAG,CAAC,CAAC,IAAI,SAAS,CAAC,CAAC,KAAK,EAAE,CAAC;QACxC,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;IACnD,CAAC;IACD,MAAM,IAAI,GAAG,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IACvC,MAAM,KAAK,GAAG,QAAQ,CACpB,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,MAAM,IAAI,CAAC,CAAC,EAC3C,IAAI,CAAC,IAAI,IAAI,uBAAuB,CACrC,CAAC;IACF,MAAM,KAAK,GAAyB,EAAE,CAAC;IACvC,KAAK,MAAM,CAAC,CAAC,EAAE,GAAG,CAAC,IAAI,IAAI,CAAC,OAAO,EAAE,EAAE,CAAC;QACtC,KAAK,MAAM,GAAG,IAAI,QAAQ,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,EAAE,EAAE,MAAM,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC;YAClF,MAAM,OAAO,GAAG,aAAa,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;YACxC,KAAK,CAAC,IAAI,CAAC;gBACT,EAAE,EAAE,GAAG,CAAC,EAAE;gBACV,IAAI,EAAE,GAAG,CAAC,IAAI;gBACd,aAAa,EAAE,GAAG,CAAC,KAAK;gBACxB,OAAO,EAAE,CAAC,GAAG,CAAC,KAAK,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,KAAK,GAAG,CAAC,KAAK,CAAC,CAAC;gBAC/D,OAAO,EAAE,SAAS,CAAC,GAAG,CAAC;gBACvB,UAAU,EAAE,GAAG,CAAC,UAAU;gBAC1B,aAAa,EAAE,GAAG,CAAC,aAAa,KAAK,IAAI;aAC1C,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IACD,OAAO;QACL,aAAa,EAAE,CAAC;QAChB,WAAW,EAAE,CAAC,IAAI,CAAC,GAAG,IAAI,IAAI,IAAI,EAAE,CAAC,CAAC,WAAW,EAAE;QACnD,IAAI;QACJ,YAAY,EACV,uVAAuV;QACzV,KAAK,EAAE,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;KAC/B,CAAC;AACJ,CAAC;AAED,oCAAoC;AACpC,MAAM,gBAAgB,GAAG,CAAC,CAAC,KAAK,CAAC;IAC/B,CAAC,CAAC,MAAM,CAAC;QACP,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QACrB,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QACxB,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;KAC5B,CAAC;IACF,CAAC,CAAC,MAAM,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,EAAE,CAAC;CACxF,CAAC,CAAC;AAEH,kFAAkF;AAClF,MAAM,qBAAqB,GAAG,CAAC,CAAC,KAAK,CAAC;IACpC,CAAC,CAAC,MAAM,CAAC,EAAE,WAAW,EAAE,CAAC,CAAC,KAAK,CAAC,gBAAgB,CAAC,EAAE,CAAC;IACpD,CAAC,CAAC,KAAK,CAAC,gBAAgB,CAAC;CAC1B,CAAC,CAAC;AAmBH;;;;;;GAMG;AACH,MAAM,UAAU,qBAAqB,CAAC,IAAY;IAChD,IAAI,IAAa,CAAC;IAClB,IAAI,CAAC;QACH,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC1B,CAAC;IAAC,MAAM,CAAC;QACP,MAAM,IAAI,KAAK,CAAC,8BAA8B,CAAC,CAAC;IAClD,CAAC;IACD,MAAM,MAAM,GAAG,qBAAqB,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC;IACrD,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;QACpB,MAAM,IAAI,KAAK,CAAC,8EAA8E,CAAC,CAAC;IAClG,CAAC;IACD,OAAO,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC;AAC5E,CAAC;AAED;;;;;GAKG;AACH,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,IAAY;IACrD,OAAO,qBAAqB,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,CAAC;AAC9D,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,qBAAqB,CACnC,IAAyB,EACzB,WAAuC;IAEvC,MAAM,IAAI,GAAG,IAAI,GAAG,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IACxD,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAC;IAC/B,MAAM,QAAQ,GAAoC,EAAE,CAAC;IACrD,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,MAAM,GAAG,GAAe,EAAE,CAAC;IAC3B,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACvB,MAAM,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QAC3B,IAAI,CAAC,CAAC,EAAE,CAAC;YACP,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YACd,SAAS;QACX,CAAC;QACD,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QACjB,MAAM,IAAI,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACjD,IAAI,MAAM,IAAI,CAAC,EAAE,CAAC;YAChB,OAAO,EAAE,CAAC;YACV,SAAS;QACX,CAAC;QACD,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC;YAC/C,QAAQ,CAAC,IAAI,CAAC,EAAE,EAAE,EAAE,GAAG,CAAC,EAAE,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC;YAC9C,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YACd,SAAS;QACX,CAAC;QACD,IAAI,CAAC,CAAC,KAAK,KAAK,GAAG,CAAC,KAAK,EAAE,CAAC;YAC1B,SAAS,EAAE,CAAC;YACZ,GAAG,CAAC,IAAI,CAAC,EAAE,GAAG,GAAG,EAAE,GAAG,IAAI,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;QACrD,CAAC;aAAM,CAAC;YACN,UAAU,EAAE,CAAC;YACb,GAAG,CAAC,IAAI,CAAC;gBACP,GAAG,GAAG;gBACN,GAAG,IAAI;gBACP,KAAK,EAAE,CAAC,CAAC,KAAK;gBACd,aAAa,EAAE,GAAG,CAAC,aAAa,IAAI,GAAG,CAAC,KAAK;gBAC7C,aAAa,EAAE,IAAI;aACpB,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IACD,OAAO;QACL,IAAI,EAAE,GAAG;QACT,OAAO,EAAE;YACP,SAAS;YACT,UAAU;YACV,OAAO;YACP,SAAS,EAAE,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACtE,QAAQ;SACT;KACF,CAAC;AACJ,CAAC"}

package/dist/decide/bench/stats.d.ts

@@ -0,0 +1,57 @@
+/**
+ * Deterministic randomness and summary statistics for the System One
+ * benchmark (T12495). Every random choice (negative sampling, the owner
+ * spot-check sample) goes through a seeded generator so a dataset can be
+ * rebuilt byte for byte.
+ *
+ * @task T12495
+ * @epic T12486
+ */
+/** Default seed: the task number, so an unseeded run is still reproducible. */
+export declare const DEFAULT_BENCH_SEED = 12495;
+/** A seeded generator of floats in [0, 1). */
+export type BenchRandom = () => number;
+/**
+ * Mulberry32: a small, fast, seeded PRNG (not cryptographic).
+ *
+ * @param seed - 32-bit seed.
+ * @returns A generator of floats in [0, 1).
+ */
+export declare function createSeededRandom(seed: number): BenchRandom;
+/**
+ * Fisher–Yates shuffle into a new array.
+ *
+ * @param items - Items to shuffle.
+ * @param random - Seeded generator.
+ * @returns A shuffled copy.
+ */
+export declare function shuffled<T>(items: readonly T[], random: BenchRandom): T[];
+/**
+ * The `p`-th percentile (nearest-rank) of `values`.
+ *
+ * @param values - Samples.
+ * @param p - Percentile in [0, 100].
+ * @returns The percentile, or `null` when there are no samples.
+ */
+export declare function percentile(values: readonly number[], p: number): number | null;
+/** Mean and spread of one metric across repeated runs. */
+export interface BenchSpread {
+    /** Runs that produced a value. */
+    readonly n: number;
+    /** Arithmetic mean. */
+    readonly mean: number | null;
+    /** Sample standard deviation (0 for a single run). */
+    readonly stddev: number | null;
+    /** Smallest value. */
+    readonly min: number | null;
+    /** Largest value. */
+    readonly max: number | null;
+}
+/**
+ * Mean, sample standard deviation, min and max of the non-null values.
+ *
+ * @param values - One value per run; `null` means the run produced none.
+ * @returns The spread.
+ */
+export declare function spreadOf(values: readonly (number | null)[]): BenchSpread;
+//# sourceMappingURL=stats.d.ts.map

package/dist/decide/bench/stats.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"stats.d.ts","sourceRoot":"","sources":["../../../src/decide/bench/stats.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,+EAA+E;AAC/E,eAAO,MAAM,kBAAkB,QAAQ,CAAC;AAExC,8CAA8C;AAC9C,MAAM,MAAM,WAAW,GAAG,MAAM,MAAM,CAAC;AAEvC;;;;;GAKG;AACH,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,MAAM,GAAG,WAAW,CAS5D;AAED;;;;;;GAMG;AACH,wBAAgB,QAAQ,CAAC,CAAC,EAAE,KAAK,EAAE,SAAS,CAAC,EAAE,EAAE,MAAM,EAAE,WAAW,GAAG,CAAC,EAAE,CAWzE;AAED;;;;;;GAMG;AACH,wBAAgB,UAAU,CAAC,MAAM,EAAE,SAAS,MAAM,EAAE,EAAE,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAK9E;AAED,0DAA0D;AAC1D,MAAM,WAAW,WAAW;IAC1B,kCAAkC;IAClC,QAAQ,CAAC,CAAC,EAAE,MAAM,CAAC;IACnB,uBAAuB;IACvB,QAAQ,CAAC,IAAI,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,sDAAsD;IACtD,QAAQ,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,sBAAsB;IACtB,QAAQ,CAAC,GAAG,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,qBAAqB;IACrB,QAAQ,CAAC,GAAG,EAAE,MAAM,GAAG,IAAI,CAAC;CAC7B;AAED;;;;;GAKG;AACH,wBAAgB,QAAQ,CAAC,MAAM,EAAE,SAAS,CAAC,MAAM,GAAG,IAAI,CAAC,EAAE,GAAG,WAAW,CAaxE"}

package/dist/decide/bench/stats.js

@@ -0,0 +1,82 @@
+/**
+ * Deterministic randomness and summary statistics for the System One
+ * benchmark (T12495). Every random choice (negative sampling, the owner
+ * spot-check sample) goes through a seeded generator so a dataset can be
+ * rebuilt byte for byte.
+ *
+ * @task T12495
+ * @epic T12486
+ */
+/** Default seed: the task number, so an unseeded run is still reproducible. */
+export const DEFAULT_BENCH_SEED = 12495;
+/**
+ * Mulberry32: a small, fast, seeded PRNG (not cryptographic).
+ *
+ * @param seed - 32-bit seed.
+ * @returns A generator of floats in [0, 1).
+ */
+export function createSeededRandom(seed) {
+    let a = seed >>> 0;
+    return () => {
+        a = (a + 0x6d2b79f5) >>> 0;
+        let t = a;
+        t = Math.imul(t ^ (t >>> 15), t | 1);
+        t ^= t + Math.imul(t ^ (t >>> 7), t | 61);
+        return ((t ^ (t >>> 14)) >>> 0) / 4294967296;
+    };
+}
+/**
+ * Fisher–Yates shuffle into a new array.
+ *
+ * @param items - Items to shuffle.
+ * @param random - Seeded generator.
+ * @returns A shuffled copy.
+ */
+export function shuffled(items, random) {
+    const out = [...items];
+    for (let i = out.length - 1; i > 0; i--) {
+        const j = Math.floor(random() * (i + 1));
+        const tmp = out[i];
+        const other = out[j];
+        if (tmp === undefined || other === undefined)
+            continue;
+        out[i] = other;
+        out[j] = tmp;
+    }
+    return out;
+}
+/**
+ * The `p`-th percentile (nearest-rank) of `values`.
+ *
+ * @param values - Samples.
+ * @param p - Percentile in [0, 100].
+ * @returns The percentile, or `null` when there are no samples.
+ */
+export function percentile(values, p) {
+    if (values.length === 0)
+        return null;
+    const sorted = [...values].sort((x, y) => x - y);
+    const rank = Math.min(sorted.length, Math.max(1, Math.ceil((p / 100) * sorted.length)));
+    return sorted[rank - 1] ?? null;
+}
+/**
+ * Mean, sample standard deviation, min and max of the non-null values.
+ *
+ * @param values - One value per run; `null` means the run produced none.
+ * @returns The spread.
+ */
+export function spreadOf(values) {
+    const xs = values.filter((v) => v !== null && Number.isFinite(v));
+    if (xs.length === 0)
+        return { n: 0, mean: null, stddev: null, min: null, max: null };
+    const mean = xs.reduce((s, x) => s + x, 0) / xs.length;
+    const variance = xs.length > 1 ? xs.reduce((s, x) => s + (x - mean) ** 2, 0) / (xs.length - 1) : 0;
+    return {
+        n: xs.length,
+        mean,
+        stddev: Math.sqrt(variance),
+        min: Math.min(...xs),
+        max: Math.max(...xs),
+    };
+}
+//# sourceMappingURL=stats.js.map

package/dist/decide/bench/stats.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"stats.js","sourceRoot":"","sources":["../../../src/decide/bench/stats.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,+EAA+E;AAC/E,MAAM,CAAC,MAAM,kBAAkB,GAAG,KAAK,CAAC;AAKxC;;;;;GAKG;AACH,MAAM,UAAU,kBAAkB,CAAC,IAAY;IAC7C,IAAI,CAAC,GAAG,IAAI,KAAK,CAAC,CAAC;IACnB,OAAO,GAAG,EAAE;QACV,CAAC,GAAG,CAAC,CAAC,GAAG,UAAU,CAAC,KAAK,CAAC,CAAC;QAC3B,IAAI,CAAC,GAAG,CAAC,CAAC;QACV,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QACrC,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC;QAC1C,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,GAAG,UAAU,CAAC;IAC/C,CAAC,CAAC;AACJ,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,QAAQ,CAAI,KAAmB,EAAE,MAAmB;IAClE,MAAM,GAAG,GAAG,CAAC,GAAG,KAAK,CAAC,CAAC;IACvB,KAAK,IAAI,CAAC,GAAG,GAAG,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QACxC,MAAM,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QACzC,MAAM,GAAG,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC;QACnB,MAAM,KAAK,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC;QACrB,IAAI,GAAG,KAAK,SAAS,IAAI,KAAK,KAAK,SAAS;YAAE,SAAS;QACvD,GAAG,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC;QACf,GAAG,CAAC,CAAC,CAAC,GAAG,GAAG,CAAC;IACf,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,UAAU,CAAC,MAAyB,EAAE,CAAS;IAC7D,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,IAAI,CAAC;IACrC,MAAM,MAAM,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACjD,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,GAAG,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;IACxF,OAAO,MAAM,CAAC,IAAI,GAAG,CAAC,CAAC,IAAI,IAAI,CAAC;AAClC,CAAC;AAgBD;;;;;GAKG;AACH,MAAM,UAAU,QAAQ,CAAC,MAAkC;IACzD,MAAM,EAAE,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,CAAC,KAAK,IAAI,IAAI,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;IAC/E,IAAI,EAAE,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC,EAAE,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,IAAI,EAAE,GAAG,EAAE,IAAI,EAAE,CAAC;IACrF,MAAM,IAAI,GAAG,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,MAAM,CAAC;IACvD,MAAM,QAAQ,GACZ,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACpF,OAAO;QACL,CAAC,EAAE,EAAE,CAAC,MAAM;QACZ,IAAI;QACJ,MAAM,EAAE,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC;QAC3B,GAAG,EAAE,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC;QACpB,GAAG,EAAE,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC;KACrB,CAAC;AACJ,CAAC"}

package/dist/decide/bench/types.d.ts

@@ -0,0 +1,216 @@
+/**
+ * Types and row schemas for the System One accuracy benchmark (T12495).
+ *
+ * A benchmark row is ONE labelled question built from CLEO's own history
+ * (duplicate task pairs, explicitly typed observations, superseding decision
+ * pairs), already redacted. Rows are stored as JSONL; every row carries its
+ * provenance (source record ids and the rule that labelled it) so the owner
+ * spot-check and any later audit can trace the label back to the store.
+ *
+ * @task T12495
+ * @epic T12486
+ */
+import { z } from 'zod';
+/** The benchmarked decision sites, by their `decide.sites.*` key name. */
+export declare const BENCH_SITES: readonly ["duplicateDetection", "observationType", "decisionContradiction"];
+/** One of {@link BENCH_SITES}. */
+export type BenchSite = (typeof BENCH_SITES)[number];
+/** Labels of the duplicate-detection site. `duplicate` is the positive class. */
+export declare const DUPLICATE_LABELS: readonly ["duplicate", "distinct"];
+/** Labels of the decision-contradiction site. `conflict` is the positive class. */
+export declare const CONTRADICTION_LABELS: readonly ["conflict", "compatible"];
+/**
+ * The labels a row of `site` may carry.
+ *
+ * @param site - Benchmarked site.
+ * @returns The allowed labels, positive class first for the two-label sites.
+ */
+export declare function labelsForSite(site: BenchSite): readonly string[];
+/**
+ * The positive class of a two-label site, or `null` for a multi-class site
+ * (whose precision, recall, F1 and false-positive rate are macro averages).
+ *
+ * @param site - Benchmarked site.
+ * @returns The positive label, or `null`.
+ */
+export declare function positiveLabel(site: BenchSite): string | null;
+/** Why a row carries its label. */
+export declare const BENCH_LABEL_RULES: readonly ["duplicates-relation", "cancelled-duplicate-reason", "note-duplicate-reference", "same-parent-unrelated", "explicit-type", "supersedes-edge", "random-unlinked-pair"];
+/** One of {@link BENCH_LABEL_RULES}. */
+export type BenchLabelRule = (typeof BENCH_LABEL_RULES)[number];
+/** Zod schema of one dataset row (one JSONL line). */
+export declare const benchRowSchema: z.ZodDiscriminatedUnion<[z.ZodObject<{
+    site: z.ZodLiteral<"duplicateDetection">;
+    input: z.ZodObject<{
+        a: z.ZodObject<{
+            id: z.ZodString;
+            title: z.ZodString;
+            description: z.ZodString;
+        }, z.core.$strip>;
+        b: z.ZodObject<{
+            id: z.ZodString;
+            title: z.ZodString;
+            description: z.ZodString;
+        }, z.core.$strip>;
+    }, z.core.$strip>;
+    id: z.ZodString;
+    label: z.ZodString;
+    provenance: z.ZodObject<{
+        sourceIds: z.ZodArray<z.ZodString>;
+        rule: z.ZodEnum<{
+            "duplicates-relation": "duplicates-relation";
+            "cancelled-duplicate-reason": "cancelled-duplicate-reason";
+            "note-duplicate-reference": "note-duplicate-reference";
+            "same-parent-unrelated": "same-parent-unrelated";
+            "explicit-type": "explicit-type";
+            "supersedes-edge": "supersedes-edge";
+            "random-unlinked-pair": "random-unlinked-pair";
+        }>;
+        reason: z.ZodString;
+    }, z.core.$strip>;
+    ownerVerified: z.ZodOptional<z.ZodBoolean>;
+    correctedFrom: z.ZodOptional<z.ZodString>;
+    ownerNote: z.ZodOptional<z.ZodString>;
+}, z.core.$strip>, z.ZodObject<{
+    site: z.ZodLiteral<"observationType">;
+    input: z.ZodObject<{
+        id: z.ZodString;
+        title: z.ZodString;
+        text: z.ZodString;
+    }, z.core.$strip>;
+    id: z.ZodString;
+    label: z.ZodString;
+    provenance: z.ZodObject<{
+        sourceIds: z.ZodArray<z.ZodString>;
+        rule: z.ZodEnum<{
+            "duplicates-relation": "duplicates-relation";
+            "cancelled-duplicate-reason": "cancelled-duplicate-reason";
+            "note-duplicate-reference": "note-duplicate-reference";
+            "same-parent-unrelated": "same-parent-unrelated";
+            "explicit-type": "explicit-type";
+            "supersedes-edge": "supersedes-edge";
+            "random-unlinked-pair": "random-unlinked-pair";
+        }>;
+        reason: z.ZodString;
+    }, z.core.$strip>;
+    ownerVerified: z.ZodOptional<z.ZodBoolean>;
+    correctedFrom: z.ZodOptional<z.ZodString>;
+    ownerNote: z.ZodOptional<z.ZodString>;
+}, z.core.$strip>, z.ZodObject<{
+    site: z.ZodLiteral<"decisionContradiction">;
+    input: z.ZodObject<{
+        newer: z.ZodObject<{
+            id: z.ZodString;
+            type: z.ZodOptional<z.ZodString>;
+            decision: z.ZodString;
+            rationale: z.ZodString;
+        }, z.core.$strip>;
+        older: z.ZodObject<{
+            id: z.ZodString;
+            type: z.ZodOptional<z.ZodString>;
+            decision: z.ZodString;
+            rationale: z.ZodString;
+        }, z.core.$strip>;
+    }, z.core.$strip>;
+    id: z.ZodString;
+    label: z.ZodString;
+    provenance: z.ZodObject<{
+        sourceIds: z.ZodArray<z.ZodString>;
+        rule: z.ZodEnum<{
+            "duplicates-relation": "duplicates-relation";
+            "cancelled-duplicate-reason": "cancelled-duplicate-reason";
+            "note-duplicate-reference": "note-duplicate-reference";
+            "same-parent-unrelated": "same-parent-unrelated";
+            "explicit-type": "explicit-type";
+            "supersedes-edge": "supersedes-edge";
+            "random-unlinked-pair": "random-unlinked-pair";
+        }>;
+        reason: z.ZodString;
+    }, z.core.$strip>;
+    ownerVerified: z.ZodOptional<z.ZodBoolean>;
+    correctedFrom: z.ZodOptional<z.ZodString>;
+    ownerNote: z.ZodOptional<z.ZodString>;
+}, z.core.$strip>], "site">;
+/** One labelled, redacted benchmark row. */
+export type BenchRow = z.infer<typeof benchRowSchema>;
+/** Where a row came from and why it carries its label. */
+export type BenchProvenance = BenchRow['provenance'];
+/** A task as the dataset builder reads it (from the task accessor). */
+export interface BenchTaskRecord {
+    /** Task id. */
+    readonly id: string;
+    /** Title. */
+    readonly title: string;
+    /** Description (empty when absent). */
+    readonly description: string;
+    /** Status. */
+    readonly status: string;
+    /** Parent id, when any. */
+    readonly parentId: string | null;
+    /** Cancellation reason, when cancelled with one. */
+    readonly cancellationReason?: string;
+    /** Free-text notes. */
+    readonly notes: readonly string[];
+    /** Typed relations to other tasks. */
+    readonly relates: readonly {
+        readonly taskId: string;
+        readonly type: string;
+        readonly reason?: string;
+    }[];
+}
+/** An observation as the dataset builder reads it (from the brain accessor). */
+export interface BenchObservationRecord {
+    /** Observation id. */
+    readonly id: string;
+    /** Stored type. */
+    readonly type: string;
+    /** Title. */
+    readonly title: string;
+    /** Narrative (the text the keyword heuristic classifies). */
+    readonly narrative: string;
+}
+/** A decision as the dataset builder reads it (from the brain accessor). */
+export interface BenchDecisionRecord {
+    /** Decision id. */
+    readonly id: string;
+    /** Decision type. */
+    readonly type: string;
+    /** Decision text. */
+    readonly decision: string;
+    /** Rationale. */
+    readonly rationale: string;
+    /** Id of the decision this one supersedes, when any. */
+    readonly supersedes?: string | null;
+    /** Id of the decision that superseded this one, when any. */
+    readonly supersededBy?: string | null;
+}
+/**
+ * The store port the dataset builder reads through. The default
+ * implementation (`./source.ts`) uses the core task and brain accessors;
+ * tests inject records directly.
+ */
+export interface BenchSource {
+    /** Every task, including cancelled and archived ones. */
+    tasks(): Promise<readonly BenchTaskRecord[]>;
+    /** Every observation, including invalidated ones. */
+    observations(): Promise<readonly BenchObservationRecord[]>;
+    /** Every decision, including superseded ones. */
+    decisions(): Promise<readonly BenchDecisionRecord[]>;
+}
+/**
+ * One provider connection the benchmark compares. The key is sent only as a
+ * bearer token and is never written to the results, the report or the log.
+ */
+export interface BenchConnection {
+    /** Profile name shown in the results (e.g. `layahost`, `jev`). */
+    readonly name: string;
+    /** Provider kind (`layahost`, `jev`, …); informational. */
+    readonly provider: string;
+    /** Jev-compatible base URL. */
+    readonly baseUrl: string;
+    /** API key. */
+    readonly apiKey: string;
+    /** Model; absent → the provider's default. */
+    readonly model?: string;
+}
+//# sourceMappingURL=types.d.ts.map

package/dist/decide/bench/types.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../src/decide/bench/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAGxB,0EAA0E;AAC1E,eAAO,MAAM,WAAW,6EAId,CAAC;AAEX,kCAAkC;AAClC,MAAM,MAAM,SAAS,GAAG,CAAC,OAAO,WAAW,CAAC,CAAC,MAAM,CAAC,CAAC;AAErD,iFAAiF;AACjF,eAAO,MAAM,gBAAgB,oCAAqC,CAAC;AAEnE,mFAAmF;AACnF,eAAO,MAAM,oBAAoB,qCAAsC,CAAC;AAExE;;;;;GAKG;AACH,wBAAgB,aAAa,CAAC,IAAI,EAAE,SAAS,GAAG,SAAS,MAAM,EAAE,CAIhE;AAED;;;;;;GAMG;AACH,wBAAgB,aAAa,CAAC,IAAI,EAAE,SAAS,GAAG,MAAM,GAAG,IAAI,CAI5D;AAED,mCAAmC;AACnC,eAAO,MAAM,iBAAiB,iLAQpB,CAAC;AAEX,wCAAwC;AACxC,MAAM,MAAM,cAAc,GAAG,CAAC,OAAO,iBAAiB,CAAC,CAAC,MAAM,CAAC,CAAC;AA8BhE,sDAAsD;AACtD,eAAO,MAAM,cAAc;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;2BAgBzB,CAAC;AAEH,4CAA4C;AAC5C,MAAM,MAAM,QAAQ,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,cAAc,CAAC,CAAC;AAEtD,0DAA0D;AAC1D,MAAM,MAAM,eAAe,GAAG,QAAQ,CAAC,YAAY,CAAC,CAAC;AAErD,uEAAuE;AACvE,MAAM,WAAW,eAAe;IAC9B,eAAe;IACf,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,aAAa;IACb,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,uCAAuC;IACvC,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,cAAc;IACd,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,2BAA2B;IAC3B,QAAQ,CAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACjC,oDAAoD;IACpD,QAAQ,CAAC,kBAAkB,CAAC,EAAE,MAAM,CAAC;IACrC,uBAAuB;IACvB,QAAQ,CAAC,KAAK,EAAE,SAAS,MAAM,EAAE,CAAC;IAClC,sCAAsC;IACtC,QAAQ,CAAC,OAAO,EAAE,SAAS;QACzB,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;QACxB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;QACtB,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,CAAC;KAC1B,EAAE,CAAC;CACL;AAED,gFAAgF;AAChF,MAAM,WAAW,sBAAsB;IACrC,sBAAsB;IACtB,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,mBAAmB;IACnB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,aAAa;IACb,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,6DAA6D;IAC7D,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;CAC5B;AAED,4EAA4E;AAC5E,MAAM,WAAW,mBAAmB;IAClC,mBAAmB;IACnB,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,qBAAqB;IACrB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,qBAAqB;IACrB,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,iBAAiB;IACjB,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,wDAAwD;IACxD,QAAQ,CAAC,UAAU,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACpC,6DAA6D;IAC7D,QAAQ,CAAC,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CACvC;AAED;;;;GAIG;AACH,MAAM,WAAW,WAAW;IAC1B,yDAAyD;IACzD,KAAK,IAAI,OAAO,CAAC,SAAS,eAAe,EAAE,CAAC,CAAC;IAC7C,qDAAqD;IACrD,YAAY,IAAI,OAAO,CAAC,SAAS,sBAAsB,EAAE,CAAC,CAAC;IAC3D,iDAAiD;IACjD,SAAS,IAAI,OAAO,CAAC,SAAS,mBAAmB,EAAE,CAAC,CAAC;CACtD;AAED;;;GAGG;AACH,MAAM,WAAW,eAAe;IAC9B,kEAAkE;IAClE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,2DAA2D;IAC3D,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,+BAA+B;IAC/B,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,eAAe;IACf,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,8CAA8C;IAC9C,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;CACzB"}