@tangle-network/agent-eval 0.216.0 → 0.216.1

package/dist/matrix/index.d.ts

@@ -1,4 +1,4 @@
-import { p as CostProvenance } from "../cost-ledger-CO7UyEcQ.js";
+import { l as CostProvenance } from "../cost-ledger-Jk62OBfF.js";
 import { t as DefaultVerdict } from "../verdict-E4eRNf7-.js";
-import { a as MatrixCell, c as CellSpend, i as MatrixAxis, l as readCellSpend, n as AxisSummary, o as MatrixResult, r as CellResult, s as RunAgentMatrixOptions, t as runAgentMatrix, u as withCellSpend } from "../index-Ch9VxCpK.js";
+import { a as MatrixCell, c as CellSpend, i as MatrixAxis, l as readCellSpend, n as AxisSummary, o as MatrixResult, r as CellResult, s as RunAgentMatrixOptions, t as runAgentMatrix, u as withCellSpend } from "../index-BO41eGj5.js";
 export { type AxisSummary, type CellResult, type CellSpend, type CostProvenance, type DefaultVerdict, type MatrixAxis, type MatrixCell, type MatrixResult, type RunAgentMatrixOptions, readCellSpend, runAgentMatrix, withCellSpend };

package/dist/matrix/index.js

@@ -1,2 +1,2 @@
-import { n as readCellSpend, r as withCellSpend, t as runAgentMatrix } from "../matrix-_BV9-Lk5.js";
+import { n as readCellSpend, r as withCellSpend, t as runAgentMatrix } from "../matrix-Dm0BQW7U.js";
 export { readCellSpend, runAgentMatrix, withCellSpend };

package/dist/{matrix-CHMiebFx.d.ts → matrix-DSrSxjjy.d.ts}

@@ -1,6 +1,6 @@
-import { p as CostProvenance } from "./cost-ledger-CO7UyEcQ.js";
-import { w as JudgeScore } from "./types-B4C5nFw6.js";
-import { o as MatrixResult } from "./index-Ch9VxCpK.js";
+import { l as CostProvenance } from "./cost-ledger-Jk62OBfF.js";
+import { w as JudgeScore } from "./types-Cq8mSbq8.js";
+import { o as MatrixResult } from "./index-BO41eGj5.js";
 import { AgentProfile } from "@tangle-network/agent-interface";
 //#region src/multishot/types.d.ts
 interface MultishotMessage {
@@ -398,4 +398,4 @@ interface RunMultishotMatrixResult {
 declare function runMultishotMatrix<TPersona extends MultishotPersona>(opts: RunMultishotMatrixOptions<TPersona>): Promise<RunMultishotMatrixResult>;
 //#endregion
 export { MultishotToolExecutor as A, MultishotFatalToolError as C, MultishotShape as D, MultishotResult as E, assertMultishotShotResult as F, MultishotTransportRequest as M, MultishotTransportResponse as N, MultishotShotResultError as O, MultishotTransportToolCall as P, MultishotDriverEmptyError as S, MultishotPersona as T, JudgeRunResult as _, MultishotCellOutput as a, runJudge as b, RunMultishotMatrixResult as c, MultishotShot as d, RunMultishotOptions as f, JudgeDimension as g, JudgeConfig as h, ConversationJudgeInput as i, MultishotTransport as j, MultishotToolDefinition as k, computeCellComposite as l, DEFAULT_JUDGE_MODEL as m, CellCompositeInput as n, MultishotJudges as o, runMultishot as p, CellCompositeScore as r, RunMultishotMatrixOptions as s, ArtifactJudgeInput as t, runMultishotMatrix as u, renderDimensions as v, MultishotMessage as w, MultishotArtifact as x, renderJsonFooter as y };
-//# sourceMappingURL=matrix-CHMiebFx.d.ts.map
+//# sourceMappingURL=matrix-DSrSxjjy.d.ts.map

package/dist/{matrix-CHMiebFx.d.ts.map → matrix-DSrSxjjy.d.ts.map}

@@ -1 +1 @@
-{"version":3,"file":"matrix-CHMiebFx.d.ts","names":[],"sources":["../src/multishot/types.ts","../src/multishot/judges.ts","../src/multishot/multishot.ts","../src/multishot/matrix.ts"],"mappings":";;;;;UAIiB;EACf;EACA;EACA;EACA,YAAY;IAAQ;IAAY;IAAc,MAAM;;;UAGrC;EACf;EACA;EACA;IAAc;IAAc,MAAM;;EAClC;;UAGe;EACf,YAAY;EACZ,WAAW;EACX;EACA;;;EAGA;;;;;;;;EAQA,iBAAiB;;UAGF;EACf;EACA;IACE;IACA;IACA,YAAY;;;;;;UAOC;EACf;EACA,UAAU,MAAM;EAChB,QAAQ;EACR;EACA;EACA,SAAS;;UAGM;EACf;EACA;EACA;IAAY;IAAc;;;UAGX;EACf;IAAW;IAAyB,aAAa;;EACjD;IAAU;IAAwB;;;;EAGlC;;;;EAIA;;;;;;;;KASU,sBACV,KAAK,8BACF,QAAQ;KAED,yBACV,MAAM,yBACN;;EAEE,WAAW;EACX,SAAS;MAER;EAAU;EAAiB;;UAEf;;EAEf;;GAEC;;;;;;;;UASc,eAAe,iBAAiB;;EAE/C,eAAe,SAAS;;;EAGxB,2BAA2B,SAAS;;cAGzB,kCAAkC;WACjB;EAA5B,YAA4B;;cAMjB,gCAAgC;EAC3C,YAAY;;cAMD,iCAAiC;EAC5C,YAAY;;;;;;;;;;;;;;;;;;;iBAyBE,0BAA0B,yBAAyB,SAAS;;;cCvI/D;UAEI;;EAEf;;EAEA;;UAGe,YAAY;;EAE3B;;;EAGA,WAAW;;EAEX;;EAEA,YAAY;;EAEZ;;;EAGA,cAAc,OAAO;;EAErB;;UAGe;;EAEf,OAAO;;EAEP,MAAM;;iBAGc,SAAS,QAC7B,OAAO,YAAY,SACnB,OAAO,SACN,QAAQ;;;iBAsIK,iBAAiB,eAAe;;iBAKhC,iBAAiB,eAAe;;;UCxK/B,oBAAoB,iBAAiB;EACpD,SAAS;EACT,SAAS;;;EAGT,QAAQ,eAAe;;EAEvB,QAAQ;;EAER,gBAAgB,eAAe;;;;EAI/B,mBAAmB;EACnB;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA,gBAAgB;;;;EAIhB,iBAAiB;;;EAGjB,gBAAgB;EAChB,SAAS;;;;;;;;;;;KAYC,cAAc,iBAAiB,qBACzC,MAAM,oBAAoB,cACvB,QAAQ;;;;;;;;;;;;;;;;;iBA2BS,aAAa,iBAAiB,kBAClD,MAAM,oBAAoB,YACzB,QAAQ;;;UClFM,uBAAuB,iBAAiB;EACvD,YAAY;EACZ,SAAS;;UAGM,mBAAmB,iBAAiB;EACnD,UAAU;EACV,SAAS;;UAGM,gBAAgB,iBAAiB;;EAEhD,cAAc,YAAY,uBAAuB;;EAEjD,aAAa,YAAY,mBAAmB;;EAE5C,iBAAiB,YAAY,mBAAmB;;EAEhD;;EAEA;;UAGe;EACf;EACA,cAAc;EACd;IACE,aAAa,MAAM;MAAe;MAAc;;IAChD;;EAEF;IACE,aAAa,MAAM;MAAe;MAAc;;IAChD;;;UAIa,0BAA0B,iBAAiB;;EAE1D,UAAU;IAAQ;IAAY,OAAO;;;EAErC,UAAU;;;EAGV,QAAQ,eAAe;;EAEvB,QAAQ,gBAAgB;;EAExB,QAAQ;;EAER,gBAAgB,eAAe;;EAE/B,mBAAmB;;EAEnB;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;EAIA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;EAIA,gBAAgB;;EAEhB,iBAAiB;;;EAGjB,gBAAgB;;;;;;;;;;;;;;;EAehB,UAAU,cAAc;;;;;UAMT;EACf;EACA;EACA;;UAoBe;EACf,cAAc;;EAEd,cAAc,cAAc;;EAE5B,iBAAiB,cAAc;;;;;;;iBAQjB,qBAAqB,OAAO;EAC1C;EACA;EACA;EACA;;UAuBe;EACf,QAAQ,aAAa;;iBAGD,mBAAmB,iBAAiB,kBACxD,MAAM,0BAA0B,YAC/B,QAAQ"}
+{"version":3,"file":"matrix-DSrSxjjy.d.ts","names":[],"sources":["../src/multishot/types.ts","../src/multishot/judges.ts","../src/multishot/multishot.ts","../src/multishot/matrix.ts"],"mappings":";;;;;UAIiB;EACf;EACA;EACA;EACA,YAAY;IAAQ;IAAY;IAAc,MAAM;;;UAGrC;EACf;EACA;EACA;IAAc;IAAc,MAAM;;EAClC;;UAGe;EACf,YAAY;EACZ,WAAW;EACX;EACA;;;EAGA;;;;;;;;EAQA,iBAAiB;;UAGF;EACf;EACA;IACE;IACA;IACA,YAAY;;;;;;UAOC;EACf;EACA,UAAU,MAAM;EAChB,QAAQ;EACR;EACA;EACA,SAAS;;UAGM;EACf;EACA;EACA;IAAY;IAAc;;;UAGX;EACf;IAAW;IAAyB,aAAa;;EACjD;IAAU;IAAwB;;;;EAGlC;;;;EAIA;;;;;;;;KASU,sBACV,KAAK,8BACF,QAAQ;KAED,yBACV,MAAM,yBACN;;EAEE,WAAW;EACX,SAAS;MAER;EAAU;EAAiB;;UAEf;;EAEf;;GAEC;;;;;;;;UASc,eAAe,iBAAiB;;EAE/C,eAAe,SAAS;;;EAGxB,2BAA2B,SAAS;;cAGzB,kCAAkC;WACjB;EAA5B,YAA4B;;cAMjB,gCAAgC;EAC3C,YAAY;;cAMD,iCAAiC;EAC5C,YAAY;;;;;;;;;;;;;;;;;;;iBAyBE,0BAA0B,yBAAyB,SAAS;;;cCvI/D;UAEI;;EAEf;;EAEA;;UAGe,YAAY;;EAE3B;;;EAGA,WAAW;;EAEX;;EAEA,YAAY;;EAEZ;;;EAGA,cAAc,OAAO;;EAErB;;UAGe;;EAEf,OAAO;;EAEP,MAAM;;iBAGc,SAAS,QAC7B,OAAO,YAAY,SACnB,OAAO,SACN,QAAQ;;;iBAsIK,iBAAiB,eAAe;;iBAKhC,iBAAiB,eAAe;;;UCxK/B,oBAAoB,iBAAiB;EACpD,SAAS;EACT,SAAS;;;EAGT,QAAQ,eAAe;;EAEvB,QAAQ;;EAER,gBAAgB,eAAe;;;;EAI/B,mBAAmB;EACnB;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA,gBAAgB;;;;EAIhB,iBAAiB;;;EAGjB,gBAAgB;EAChB,SAAS;;;;;;;;;;;KAYC,cAAc,iBAAiB,qBACzC,MAAM,oBAAoB,cACvB,QAAQ;;;;;;;;;;;;;;;;;iBA2BS,aAAa,iBAAiB,kBAClD,MAAM,oBAAoB,YACzB,QAAQ;;;UClFM,uBAAuB,iBAAiB;EACvD,YAAY;EACZ,SAAS;;UAGM,mBAAmB,iBAAiB;EACnD,UAAU;EACV,SAAS;;UAGM,gBAAgB,iBAAiB;;EAEhD,cAAc,YAAY,uBAAuB;;EAEjD,aAAa,YAAY,mBAAmB;;EAE5C,iBAAiB,YAAY,mBAAmB;;EAEhD;;EAEA;;UAGe;EACf;EACA,cAAc;EACd;IACE,aAAa,MAAM;MAAe;MAAc;;IAChD;;EAEF;IACE,aAAa,MAAM;MAAe;MAAc;;IAChD;;;UAIa,0BAA0B,iBAAiB;;EAE1D,UAAU;IAAQ;IAAY,OAAO;;;EAErC,UAAU;;;EAGV,QAAQ,eAAe;;EAEvB,QAAQ,gBAAgB;;EAExB,QAAQ;;EAER,gBAAgB,eAAe;;EAE/B,mBAAmB;;EAEnB;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;EAIA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;EAIA,gBAAgB;;EAEhB,iBAAiB;;;EAGjB,gBAAgB;;;;;;;;;;;;;;;EAehB,UAAU,cAAc;;;;;UAMT;EACf;EACA;EACA;;UAoBe;EACf,cAAc;;EAEd,cAAc,cAAc;;EAE5B,iBAAiB,cAAc;;;;;;;iBAQjB,qBAAqB,OAAO;EAC1C;EACA;EACA;EACA;;UAuBe;EACf,QAAQ,aAAa;;iBAGD,mBAAmB,iBAAiB,kBACxD,MAAM,0BAA0B,YAC/B,QAAQ"}

package/dist/{matrix-_BV9-Lk5.js → matrix-Dm0BQW7U.js}

@@ -1,4 +1,5 @@
 import { t as interpolatedQuantile } from "./quantile--0yb07pD.js";
+import { t as isNonnegativeFiniteNumber } from "./nonnegative-number-DvdMxg_f.js";
 //#region src/matrix/cell-spend.ts
 /**
 * Spend carrier for a cell that fails after it has already spent money.
@@ -72,7 +73,7 @@ function readCellSpend(error) {
 	const raw = error[CELL_SPEND_KEY];
 	if (typeof raw !== "object" || raw === null) return void 0;
 	const candidate = raw;
-	if (!isAmount(candidate.costUsd) || !isAmount(candidate.durationMs)) return void 0;
+	if (!isNonnegativeFiniteNumber(candidate.costUsd) || !isNonnegativeFiniteNumber(candidate.durationMs)) return void 0;
 	const kind = candidate.kind;
 	if (kind !== "observed" && kind !== "estimated" && kind !== "uncaptured") return void 0;
 	return {
@@ -81,13 +82,8 @@ function readCellSpend(error) {
 		kind
 	};
 }
-/** A number the matrix can add to a running cost or duration: finite and not
-*  negative. Anything else would poison a sum the cost ceiling reads. */
-function isAmount(value) {
-	return typeof value === "number" && Number.isFinite(value) && value >= 0;
-}
 function assertAmount(value, field) {
-	if (!isAmount(value)) throw new TypeError(`withCellSpend: ${field} must be a finite number >= 0, received ${String(value)}`);
+	if (!isNonnegativeFiniteNumber(value)) throw new TypeError(`withCellSpend: ${field} must be a finite number >= 0, received ${String(value)}`);
 }
 //#endregion
 //#region src/matrix/aggregation.ts
@@ -278,8 +274,8 @@ function makeErrorResult(err) {
 *  steps back mid-cell makes `Date.now() - startedAt` negative, so this is a
 *  reachable state, not only a caller bug. */
 function billableResult(result, onBadDuration) {
-	if (!isAmount(result.costUsd)) return makeErrorResult(/* @__PURE__ */ new RangeError(`runCell reported costUsd=${String(result.costUsd)}; a cell cost must be a finite number >= 0 so the cost ceiling stays enforceable`));
-	if (isAmount(result.durationMs)) return result;
+	if (!isNonnegativeFiniteNumber(result.costUsd)) return makeErrorResult(/* @__PURE__ */ new RangeError(`runCell reported costUsd=${String(result.costUsd)}; a cell cost must be a finite number >= 0 so the cost ceiling stays enforceable`));
+	if (isNonnegativeFiniteNumber(result.durationMs)) return result;
 	onBadDuration(result.durationMs);
 	return {
 		...result,
@@ -292,7 +288,7 @@ async function runAgentMatrix(opts) {
 	const maxConcurrency = Math.max(1, opts.maxConcurrency ?? 4);
 	const costCeiling = opts.costCeiling ?? Number.POSITIVE_INFINITY;
 	const aggregateBy = opts.aggregateBy ?? opts.axes.map((a) => a.name);
-	if (opts.maxCellCostUsd !== void 0 && !isAmount(opts.maxCellCostUsd)) throw new RangeError(`runAgentMatrix: maxCellCostUsd must be a finite number >= 0, received ${String(opts.maxCellCostUsd)}`);
+	if (opts.maxCellCostUsd !== void 0 && !isNonnegativeFiniteNumber(opts.maxCellCostUsd)) throw new RangeError(`runAgentMatrix: maxCellCostUsd must be a finite number >= 0, received ${String(opts.maxCellCostUsd)}`);
 	const base = cartesian(opts.axes);
 	const filtered = opts.filter ? base.filter((c) => opts.filter(c)) : base;
 	const filteredOut = base.length - filtered.length;
@@ -414,4 +410,4 @@ async function runAgentMatrix(opts) {
 //#endregion
 export { readCellSpend as n, withCellSpend as r, runAgentMatrix as t };
 
-//# sourceMappingURL=matrix-_BV9-Lk5.js.map
+//# sourceMappingURL=matrix-Dm0BQW7U.js.map

package/dist/matrix-Dm0BQW7U.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"matrix-Dm0BQW7U.js","names":["isAmount","isAmount"],"sources":["../src/matrix/cell-spend.ts","../src/matrix/aggregation.ts","../src/matrix/runner.ts"],"sourcesContent":["/**\n * Spend carrier for a cell that fails after it has already spent money.\n *\n * `runCell` reports what a cell cost by returning a `CellResult`. A throw\n * carries no result, so the runner cannot know what the cell spent before it\n * failed: the cumulative sum the cost ceiling reads, and `totalCostUsd`, both\n * miss that money. A throw that carries a `CellSpend` closes the gap — the\n * runner bills the failed cell for the spend the carrier declares, and records\n * a cell that declared nothing as `uncaptured` instead of a zero that reads\n * like \"this cell spent nothing\".\n *\n * The carrier is read structurally, never by `instanceof`, so a throw that\n * crosses a package or realm boundary still bills.\n */\n\nimport { isNonnegativeFiniteNumber as isAmount } from '../math/nonnegative-number'\n\n/** Property key the carrier occupies on a thrown object. Stable across\n *  versions — two copies of this package must agree on it. */\nconst CELL_SPEND_KEY = '__agentEvalCellSpend'\n\n/** Money and wall time a cell consumed before it threw. */\nexport interface CellSpend {\n  /** KNOWN subtotal in USD. When `kind` is `uncaptured` the cell spent this\n   *  much AND an unknown amount more, so the value must not be read as the\n   *  cell's total. */\n  costUsd: number\n  /** Wall time the cell consumed before the throw, in milliseconds. */\n  durationMs: number\n  /** `observed` = provider-reported amounts, `estimated` = computed from token\n   *  prices, `uncaptured` = `costUsd` is a subtotal and part of the spend\n   *  could not be measured. */\n  kind: 'observed' | 'estimated' | 'uncaptured'\n}\n\n/**\n * Return `error` carrying `spend`, for a throw site inside `runCell`:\n * `throw withCellSpend(err, { costUsd, durationMs, kind })`.\n *\n * A value that cannot hold the carrier — a thrown string or number, or a\n * frozen, sealed, or otherwise non-extensible object — is wrapped in an\n * `Error` that keeps the original as `cause`, so the spend is never dropped.\n * A later call overwrites an earlier carrier — the outer frame knows more\n * than the inner one.\n *\n * Throws `TypeError` on a non-finite or negative amount. A poisoned amount\n * would reach `cumulativeCost` and disable the cost ceiling for the whole run,\n * because `NaN >= ceiling` is false.\n */\nexport function withCellSpend(error: unknown, spend: CellSpend): unknown {\n  assertAmount(spend.costUsd, 'costUsd')\n  assertAmount(spend.durationMs, 'durationMs')\n  if (spend.kind !== 'observed' && spend.kind !== 'estimated' && spend.kind !== 'uncaptured') {\n    throw new TypeError(\n      `withCellSpend: kind must be observed, estimated or uncaptured, received ${String(spend.kind)}`,\n    )\n  }\n  const carrier: CellSpend = {\n    costUsd: spend.costUsd,\n    durationMs: spend.durationMs,\n    kind: spend.kind,\n  }\n  if (typeof error === 'object' && error !== null && attach(error, carrier)) return error\n  // A primitive cannot hold a property, and a frozen or sealed object refuses\n  // one. Carry the spend on a fresh Error that keeps the original as `cause`\n  // rather than lose it — a dropped carrier bills the cell as uncaptured,\n  // which is the under-billing this module exists to prevent.\n  const wrapper = new Error(`cell failed: ${messageOf(error)}`, { cause: error })\n  attach(wrapper, carrier)\n  return wrapper\n}\n\n/** True when the carrier landed. `defineProperty` throws on a non-extensible\n *  target, so the caller must be able to fall back. */\nfunction attach(target: object, carrier: CellSpend): boolean {\n  try {\n    Object.defineProperty(target, CELL_SPEND_KEY, {\n      value: carrier,\n      enumerable: false,\n      configurable: true,\n      writable: true,\n    })\n    return true\n  } catch {\n    return false\n  }\n}\n\nfunction messageOf(error: unknown): string {\n  if (error instanceof Error) return error.message\n  return String(error)\n}\n\n/** Read the spend a thrown value carries. `undefined` when the value carries\n *  none, or carries one that does not satisfy `CellSpend` — the runner records\n *  both as `uncaptured`. */\nexport function readCellSpend(error: unknown): CellSpend | undefined {\n  if (typeof error !== 'object' || error === null) return undefined\n  const raw = (error as Record<string, unknown>)[CELL_SPEND_KEY]\n  if (typeof raw !== 'object' || raw === null) return undefined\n  const candidate = raw as Record<string, unknown>\n  if (!isAmount(candidate.costUsd) || !isAmount(candidate.durationMs)) return undefined\n  const kind = candidate.kind\n  if (kind !== 'observed' && kind !== 'estimated' && kind !== 'uncaptured') return undefined\n  return { costUsd: candidate.costUsd, durationMs: candidate.durationMs, kind }\n}\n\n/** A number the matrix can add to a running cost or duration: finite and not\n *  negative. Anything else would poison a sum the cost ceiling reads. */\nexport { isNonnegativeFiniteNumber as isAmount } from '../math/nonnegative-number'\n\nfunction assertAmount(value: number, field: string): void {\n  if (!isAmount(value)) {\n    throw new TypeError(\n      `withCellSpend: ${field} must be a finite number >= 0, received ${String(value)}`,\n    )\n  }\n}\n","/**\n * Per-axis aggregation of cell runs into `AxisSummary` rows.\n *\n * Pure: consumes the final `cells: [{cell, runs}]` array and returns the\n * `byAxis` table. Error runs contribute 0 to passRate and meanScore. Cost and\n * duration always count — the budget was spent regardless. A run whose cost is\n * a subtotal rather than a total raises `costUncapturedCells` for its bucket,\n * so a per-axis cost row says when it under-counts.\n */\n\nimport { interpolatedQuantile } from '../math/quantile'\nimport type { AxisSummary, CellResult, MatrixAxis, MatrixCell, MatrixResult } from './types'\n\ninterface Row<Output> {\n  cell: MatrixCell\n  result: CellResult<Output>\n}\n\nfunction flattenRuns<Output>(cells: MatrixResult<Output>['cells']): Row<Output>[] {\n  const rows: Row<Output>[] = []\n  for (const { cell, runs } of cells) {\n    for (const result of runs) rows.push({ cell, result })\n  }\n  return rows\n}\n\nfunction quantile(sorted: number[], q: number): number {\n  return sorted.length ? interpolatedQuantile(sorted, q) : 0\n}\n\nexport function summariseRows<Output>(\n  rows: Row<Output>[],\n  axisName: string,\n  axisValue: string,\n): AxisSummary {\n  if (rows.length === 0) {\n    return {\n      axisName,\n      axisValue,\n      cells: 0,\n      passRate: 0,\n      meanScore: 0,\n      p50Score: 0,\n      p90Score: 0,\n      totalCostUsd: 0,\n      costUncapturedCells: 0,\n      meanDurationMs: 0,\n    }\n  }\n  let pass = 0\n  let scoreSum = 0\n  let costSum = 0\n  let durSum = 0\n  let costUncapturedCells = 0\n  const scores: number[] = []\n  for (const { result } of rows) {\n    const errored = result.error !== undefined\n    const score = errored ? 0 : result.verdict.score\n    const valid = !errored && result.verdict.valid\n    if (valid) pass++\n    scoreSum += score\n    scores.push(score)\n    costSum += result.costUsd\n    durSum += result.durationMs\n    if (result.costProvenance?.kind === 'uncaptured') costUncapturedCells++\n  }\n  scores.sort((a, b) => a - b)\n  return {\n    axisName,\n    axisValue,\n    cells: rows.length,\n    passRate: pass / rows.length,\n    meanScore: scoreSum / rows.length,\n    p50Score: quantile(scores, 0.5),\n    p90Score: quantile(scores, 0.9),\n    totalCostUsd: costSum,\n    costUncapturedCells,\n    meanDurationMs: durSum / rows.length,\n  }\n}\n\nfunction bucketBy<Output>(\n  rows: Row<Output>[],\n  axisName: string,\n  labelFor: (id: string) => string,\n): Record<string, AxisSummary> {\n  const buckets = new Map<string, Row<Output>[]>()\n  for (const row of rows) {\n    const slot = row.cell.axes[axisName]\n    if (!slot) continue\n    const id = slot.id\n    let arr = buckets.get(id)\n    if (!arr) {\n      arr = []\n      buckets.set(id, arr)\n    }\n    arr.push(row)\n  }\n  const out: Record<string, AxisSummary> = {}\n  // Sorted keys for deterministic JSON serialisation.\n  for (const id of [...buckets.keys()].sort()) {\n    out[id] = summariseRows(buckets.get(id) as Row<Output>[], axisName, labelFor(id))\n  }\n  return out\n}\n\nexport function buildByAxis<Output>(\n  cells: MatrixResult<Output>['cells'],\n  axes: MatrixAxis<unknown>[],\n  aggregateBy: string[],\n): Record<string, Record<string, AxisSummary>> {\n  const rows = flattenRuns(cells)\n  const byName = new Map(axes.map((a) => [a.name, a]))\n  const byAxis: Record<string, Record<string, AxisSummary>> = {}\n  for (const name of aggregateBy) {\n    const axis = byName.get(name)\n    const labelFor = (id: string): string => {\n      if (!axis?.label) return id\n      const found = axis.values.find((v) => v.id === id)\n      if (!found) return id\n      return axis.label(found.value, id)\n    }\n    byAxis[name] = bucketBy(rows, name, labelFor)\n  }\n  return byAxis\n}\n","/**\n * N-axis cartesian runner.\n *\n * Expansion order: cartesian over `axes` in declared order, then `reps` as the\n * inner-most dim → `ordinal = (cartIdx * reps) + rep`. The returned\n * `cells[]` is sorted by `ordinal` so concurrent execution does not reorder\n * the output.\n *\n * Scheduling is a sliding window of in-flight promises capped at\n * `maxConcurrency`. The window stops admitting new cells when the cost\n * ceiling trips or the abort signal fires; in-flight cells finish.\n */\n\nimport { buildByAxis } from './aggregation'\nimport { isAmount, readCellSpend } from './cell-spend'\nimport type {\n  CellResult,\n  CostProvenance,\n  MatrixAxis,\n  MatrixCell,\n  MatrixResult,\n  RunAgentMatrixOptions,\n} from './types'\n\ninterface BaseCell {\n  axes: Record<string, { id: string; value: unknown }>\n}\n\nfunction cartesian(axes: MatrixAxis<unknown>[]): BaseCell[] {\n  // Empty axes (`values=[]`) collapse the whole product to zero cells. An\n  // empty `axes` array yields a single empty-axes cell — degenerate but\n  // valid (caller is iterating only reps).\n  if (axes.length === 0) return [{ axes: {} }]\n  for (const a of axes) if (a.values.length === 0) return []\n  const out: BaseCell[] = []\n  const idx = new Array(axes.length).fill(0)\n  while (true) {\n    const slot: Record<string, { id: string; value: unknown }> = {}\n    for (let i = 0; i < axes.length; i++) {\n      const axis = axes[i] as MatrixAxis<unknown>\n      const v = axis.values[idx[i] as number] as { id: string; value: unknown }\n      slot[axis.name] = { id: v.id, value: v.value }\n    }\n    out.push({ axes: slot })\n    // Increment like an odometer, left-most axis is fastest.\n    let i = 0\n    while (i < axes.length) {\n      const next = (idx[i] as number) + 1\n      const axis = axes[i] as MatrixAxis<unknown>\n      if (next < axis.values.length) {\n        idx[i] = next\n        break\n      }\n      idx[i] = 0\n      i++\n    }\n    if (i === axes.length) break\n  }\n  return out\n}\n\nfunction makeMatrixId(): string {\n  // Stable id-like string: time + 8 random hex chars. Avoids node:crypto\n  // import to keep the matrix dep-free.\n  const t = Date.now().toString(36)\n  let r = ''\n  for (let i = 0; i < 8; i++) r += Math.floor(Math.random() * 16).toString(16)\n  return `mtx_${t}_${r}`\n}\n\n/** Fresh per result — a shared object would let one consumer's mutation reach\n *  every other cell in the run. */\nfunction uncaptured(): CostProvenance {\n  return { kind: 'uncaptured', usd: null }\n}\n\nfunction makeErrorResult<Output>(err: unknown): CellResult<Output> {\n  const e = err as { message?: string; name?: string }\n  const spend = readCellSpend(err)\n  return {\n    output: undefined as unknown as Output,\n    verdict: { valid: false, score: 0 },\n    costUsd: spend?.costUsd ?? 0,\n    durationMs: spend?.durationMs ?? 0,\n    costProvenance:\n      spend === undefined || spend.kind === 'uncaptured'\n        ? uncaptured()\n        : { kind: spend.kind, usd: spend.costUsd },\n    error: {\n      message: typeof e?.message === 'string' ? e.message : String(err),\n      kind: typeof e?.name === 'string' ? e.name : 'Error',\n    },\n  }\n}\n\n/** Make a result safe to sum.\n *\n *  A `costUsd` that cannot be summed fails the cell: `NaN` in `cumulativeCost`\n *  makes `>= costCeiling` false forever, so one bad cell would disable the\n *  ceiling for every cell after it.\n *\n *  A bad `durationMs` reaches only `meanDurationMs`, never the ceiling, so it\n *  cannot justify throwing away a real verdict and a real cost. It is reported\n *  as 0 and warned about, and the rest of the result stands. A wall clock that\n *  steps back mid-cell makes `Date.now() - startedAt` negative, so this is a\n *  reachable state, not only a caller bug. */\nfunction billableResult<Output>(\n  result: CellResult<Output>,\n  onBadDuration: (reported: unknown) => void,\n): CellResult<Output> {\n  if (!isAmount(result.costUsd)) {\n    return makeErrorResult<Output>(\n      new RangeError(\n        `runCell reported costUsd=${String(result.costUsd)}; a cell cost must be a finite number >= 0 so the cost ceiling stays enforceable`,\n      ),\n    )\n  }\n  if (isAmount(result.durationMs)) return result\n  onBadDuration(result.durationMs)\n  return { ...result, durationMs: 0 }\n}\n\nexport async function runAgentMatrix<Output>(\n  opts: RunAgentMatrixOptions<Output>,\n): Promise<MatrixResult<Output>> {\n  const startedAt = Date.now()\n  const reps = Math.max(1, opts.reps ?? 1)\n  const maxConcurrency = Math.max(1, opts.maxConcurrency ?? 4)\n  const costCeiling = opts.costCeiling ?? Number.POSITIVE_INFINITY\n  const aggregateBy = opts.aggregateBy ?? opts.axes.map((a) => a.name)\n  if (opts.maxCellCostUsd !== undefined && !isAmount(opts.maxCellCostUsd)) {\n    throw new RangeError(\n      `runAgentMatrix: maxCellCostUsd must be a finite number >= 0, received ${String(opts.maxCellCostUsd)}`,\n    )\n  }\n\n  const base = cartesian(opts.axes)\n  const filtered = opts.filter\n    ? base.filter((c) => (opts.filter as (b: BaseCell) => boolean)(c))\n    : base\n  const filteredOut = base.length - filtered.length\n\n  const planned: MatrixCell[] = []\n  for (let i = 0; i < filtered.length; i++) {\n    for (let r = 0; r < reps; r++) {\n      planned.push({\n        axes: (filtered[i] as BaseCell).axes,\n        rep: r,\n        ordinal: i * reps + r,\n      })\n    }\n  }\n\n  const cellRecords: Array<{ cell: MatrixCell; runs: CellResult<Output>[] }> = []\n  // What the ceiling reads. It runs ahead of the reported total whenever a\n  // cell's cost is a subtotal and `maxCellCostUsd` bounds what it could have\n  // been — a budget must assume the worst about spend it cannot see.\n  let cumulativeCost = 0\n  let costCeilingReached = false\n  let runsExecuted = 0\n  let cellsUnscheduled = 0\n  let costUncapturedCells = 0\n  let badDurationCells = 0\n\n  const aborted = (): boolean => opts.signal?.aborted === true\n\n  // Per-run abort controller forwards the external signal so cell executors\n  // see cancellation. We don't expose it on `MatrixCell` — the signature on\n  // `runCell` per the public API is `(cell) => Promise<...>`. Executors that\n  // need cancellation use the external signal directly via closure.\n\n  let inFlight = 0\n  let cursor = 0\n  let resolveAll: (() => void) | undefined\n  const done = new Promise<void>((res) => {\n    resolveAll = res\n  })\n\n  const pump = (): void => {\n    while (inFlight < maxConcurrency && cursor < planned.length) {\n      if (aborted() || costCeilingReached) {\n        // Drain remaining as unscheduled.\n        const left = planned.length - cursor\n        cellsUnscheduled += left\n        cursor = planned.length\n        break\n      }\n      const cell = planned[cursor++] as MatrixCell\n      inFlight++\n      // Lazily allocate the record so cells appear in `cells[]` in any\n      // arrival order; we sort by ordinal at the end.\n      const record = { cell, runs: [] as CellResult<Output>[] }\n      cellRecords.push(record)\n      const promise: Promise<CellResult<Output>> = (async () => {\n        try {\n          return await opts.runCell(cell)\n        } catch (err) {\n          return makeErrorResult<Output>(err)\n        }\n      })()\n      promise.then((settled) => {\n        const result = billableResult(settled, (reported) => {\n          badDurationCells++\n          if (badDurationCells === 1) {\n            // eslint-disable-next-line no-console\n            console.warn(\n              `[matrix] a cell reported durationMs=${String(reported)}; recorded as 0 — meanDurationMs under-reports this run`,\n            )\n          }\n        })\n        record.runs.push(result)\n        runsExecuted++\n        const uncapturedCost = result.costProvenance?.kind === 'uncaptured'\n        cumulativeCost +=\n          uncapturedCost && opts.maxCellCostUsd !== undefined\n            ? Math.max(result.costUsd, opts.maxCellCostUsd)\n            : result.costUsd\n        if (uncapturedCost) {\n          costUncapturedCells++\n          if (costUncapturedCells === 1) {\n            // eslint-disable-next-line no-console\n            console.warn(\n              \"[matrix] a cell's cost is a subtotal, not a total — totalCostUsd and the cost ceiling under-count this run\",\n            )\n          }\n        }\n        if (cumulativeCost >= costCeiling && !costCeilingReached) {\n          costCeilingReached = true\n          // eslint-disable-next-line no-console\n          console.warn('[matrix] cost ceiling reached')\n        }\n        try {\n          opts.onCellComplete?.(cell, result)\n        } catch {\n          // onCellComplete is observational — swallow throws so a noisy\n          // callback can't tank the run.\n        }\n        inFlight--\n        if (cursor < planned.length) {\n          pump()\n        } else if (inFlight === 0) {\n          resolveAll?.()\n        }\n      })\n    }\n    if (cursor >= planned.length && inFlight === 0) resolveAll?.()\n  }\n\n  const onAbort = (): void => {\n    // External abort: stop scheduling. In-flight cells finish; their\n    // executors observe `opts.signal.aborted` directly via closure.\n    if (cursor < planned.length) {\n      cellsUnscheduled += planned.length - cursor\n      cursor = planned.length\n    }\n    if (inFlight === 0) resolveAll?.()\n  }\n  if (opts.signal) {\n    if (opts.signal.aborted) {\n      cellsUnscheduled = planned.length\n      cursor = planned.length\n      resolveAll?.()\n    } else {\n      opts.signal.addEventListener('abort', onAbort, { once: true })\n    }\n  }\n\n  if (planned.length === 0) {\n    resolveAll?.()\n  } else {\n    pump()\n  }\n\n  await done\n  if (opts.signal) opts.signal.removeEventListener('abort', onAbort)\n\n  cellRecords.sort((a, b) => a.cell.ordinal - b.cell.ordinal)\n\n  let pass = 0\n  let scoreSum = 0\n  let totalCost = 0\n  let runCount = 0\n  for (const { runs } of cellRecords) {\n    for (const r of runs) {\n      runCount++\n      const errored = r.error !== undefined\n      if (!errored && r.verdict.valid) pass++\n      scoreSum += errored ? 0 : r.verdict.score\n      totalCost += r.costUsd\n    }\n  }\n\n  const byAxis = buildByAxis(cellRecords, opts.axes, aggregateBy)\n\n  return {\n    cells: cellRecords,\n    byAxis,\n    summary: {\n      totalCells: planned.length,\n      runsExecuted,\n      cellsSkipped: cellsUnscheduled + filteredOut * reps,\n      overallPassRate: runCount === 0 ? 0 : pass / runCount,\n      overallMeanScore: runCount === 0 ? 0 : scoreSum / runCount,\n      totalCostUsd: totalCost,\n      costUncapturedCells,\n      ceilingChargedUsd: cumulativeCost,\n      durationMs: Date.now() - startedAt,\n    },\n    matrixId: makeMatrixId(),\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;AAmBA,MAAM,iBAAiB;;;;;;;;;;;;;;;AA8BvB,SAAgB,cAAc,OAAgB,OAA2B;CACvE,aAAa,MAAM,SAAS,SAAS;CACrC,aAAa,MAAM,YAAY,YAAY;CAC3C,IAAI,MAAM,SAAS,cAAc,MAAM,SAAS,eAAe,MAAM,SAAS,cAC5E,MAAM,IAAI,UACR,2EAA2E,OAAO,MAAM,IAAI,GAC9F;CAEF,MAAM,UAAqB;EACzB,SAAS,MAAM;EACf,YAAY,MAAM;EAClB,MAAM,MAAM;CACd;CACA,IAAI,OAAO,UAAU,YAAY,UAAU,QAAQ,OAAO,OAAO,OAAO,GAAG,OAAO;CAKlF,MAAM,UAAU,IAAI,MAAM,gBAAgB,UAAU,KAAK,KAAK,EAAE,OAAO,MAAM,CAAC;CAC9E,OAAO,SAAS,OAAO;CACvB,OAAO;AACT;;;AAIA,SAAS,OAAO,QAAgB,SAA6B;CAC3D,IAAI;EACF,OAAO,eAAe,QAAQ,gBAAgB;GAC5C,OAAO;GACP,YAAY;GACZ,cAAc;GACd,UAAU;EACZ,CAAC;EACD,OAAO;CACT,QAAQ;EACN,OAAO;CACT;AACF;AAEA,SAAS,UAAU,OAAwB;CACzC,IAAI,iBAAiB,OAAO,OAAO,MAAM;CACzC,OAAO,OAAO,KAAK;AACrB;;;;AAKA,SAAgB,cAAc,OAAuC;CACnE,IAAI,OAAO,UAAU,YAAY,UAAU,MAAM,OAAO,KAAA;CACxD,MAAM,MAAO,MAAkC;CAC/C,IAAI,OAAO,QAAQ,YAAY,QAAQ,MAAM,OAAO,KAAA;CACpD,MAAM,YAAY;CAClB,IAAI,CAACA,0BAAS,UAAU,OAAO,KAAK,CAACA,0BAAS,UAAU,UAAU,GAAG,OAAO,KAAA;CAC5E,MAAM,OAAO,UAAU;CACvB,IAAI,SAAS,cAAc,SAAS,eAAe,SAAS,cAAc,OAAO,KAAA;CACjF,OAAO;EAAE,SAAS,UAAU;EAAS,YAAY,UAAU;EAAY;CAAK;AAC9E;AAMA,SAAS,aAAa,OAAe,OAAqB;CACxD,IAAI,CAACA,0BAAS,KAAK,GACjB,MAAM,IAAI,UACR,kBAAkB,MAAM,0CAA0C,OAAO,KAAK,GAChF;AAEJ;;;;;;;;;;;;ACnGA,SAAS,YAAoB,OAAqD;CAChF,MAAM,OAAsB,CAAC;CAC7B,KAAK,MAAM,EAAE,MAAM,UAAU,OAC3B,KAAK,MAAM,UAAU,MAAM,KAAK,KAAK;EAAE;EAAM;CAAO,CAAC;CAEvD,OAAO;AACT;AAEA,SAAS,SAAS,QAAkB,GAAmB;CACrD,OAAO,OAAO,SAAS,qBAAqB,QAAQ,CAAC,IAAI;AAC3D;AAEA,SAAgB,cACd,MACA,UACA,WACa;CACb,IAAI,KAAK,WAAW,GAClB,OAAO;EACL;EACA;EACA,OAAO;EACP,UAAU;EACV,WAAW;EACX,UAAU;EACV,UAAU;EACV,cAAc;EACd,qBAAqB;EACrB,gBAAgB;CAClB;CAEF,IAAI,OAAO;CACX,IAAI,WAAW;CACf,IAAI,UAAU;CACd,IAAI,SAAS;CACb,IAAI,sBAAsB;CAC1B,MAAM,SAAmB,CAAC;CAC1B,KAAK,MAAM,EAAE,YAAY,MAAM;EAC7B,MAAM,UAAU,OAAO,UAAU,KAAA;EACjC,MAAM,QAAQ,UAAU,IAAI,OAAO,QAAQ;EAE3C,IADc,CAAC,WAAW,OAAO,QAAQ,OAC9B;EACX,YAAY;EACZ,OAAO,KAAK,KAAK;EACjB,WAAW,OAAO;EAClB,UAAU,OAAO;EACjB,IAAI,OAAO,gBAAgB,SAAS,cAAc;CACpD;CACA,OAAO,MAAM,GAAG,MAAM,IAAI,CAAC;CAC3B,OAAO;EACL;EACA;EACA,OAAO,KAAK;EACZ,UAAU,OAAO,KAAK;EACtB,WAAW,WAAW,KAAK;EAC3B,UAAU,SAAS,QAAQ,EAAG;EAC9B,UAAU,SAAS,QAAQ,EAAG;EAC9B,cAAc;EACd;EACA,gBAAgB,SAAS,KAAK;CAChC;AACF;AAEA,SAAS,SACP,MACA,UACA,UAC6B;CAC7B,MAAM,0BAAU,IAAI,IAA2B;CAC/C,KAAK,MAAM,OAAO,MAAM;EACtB,MAAM,OAAO,IAAI,KAAK,KAAK;EAC3B,IAAI,CAAC,MAAM;EACX,MAAM,KAAK,KAAK;EAChB,IAAI,MAAM,QAAQ,IAAI,EAAE;EACxB,IAAI,CAAC,KAAK;GACR,MAAM,CAAC;GACP,QAAQ,IAAI,IAAI,GAAG;EACrB;EACA,IAAI,KAAK,GAAG;CACd;CACA,MAAM,MAAmC,CAAC;CAE1C,KAAK,MAAM,MAAM,CAAC,GAAG,QAAQ,KAAK,CAAC,CAAC,CAAC,KAAK,GACxC,IAAI,MAAM,cAAc,QAAQ,IAAI,EAAE,GAAoB,UAAU,SAAS,EAAE,CAAC;CAElF,OAAO;AACT;AAEA,SAAgB,YACd,OACA,MACA,aAC6C;CAC7C,MAAM,OAAO,YAAY,KAAK;CAC9B,MAAM,SAAS,IAAI,IAAI,KAAK,KAAK,MAAM,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC;CACnD,MAAM,SAAsD,CAAC;CAC7D,KAAK,MAAM,QAAQ,aAAa;EAC9B,MAAM,OAAO,OAAO,IAAI,IAAI;EAC5B,MAAM,YAAY,OAAuB;GACvC,IAAI,CAAC,MAAM,OAAO,OAAO;GACzB,MAAM,QAAQ,KAAK,OAAO,MAAM,MAAM,EAAE,OAAO,EAAE;GACjD,IAAI,CAAC,OAAO,OAAO;GACnB,OAAO,KAAK,MAAM,MAAM,OAAO,EAAE;EACnC;EACA,OAAO,QAAQ,SAAS,MAAM,MAAM,QAAQ;CAC9C;CACA,OAAO;AACT;;;;;;;;;;;;;;;ACjGA,SAAS,UAAU,MAAyC;CAI1D,IAAI,KAAK,WAAW,GAAG,OAAO,CAAC,EAAE,MAAM,CAAC,EAAE,CAAC;CAC3C,KAAK,MAAM,KAAK,MAAM,IAAI,EAAE,OAAO,WAAW,GAAG,OAAO,CAAC;CACzD,MAAM,MAAkB,CAAC;CACzB,MAAM,MAAM,IAAI,MAAM,KAAK,MAAM,CAAC,CAAC,KAAK,CAAC;CACzC,OAAO,MAAM;EACX,MAAM,OAAuD,CAAC;EAC9D,KAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;GACpC,MAAM,OAAO,KAAK;GAClB,MAAM,IAAI,KAAK,OAAO,IAAI;GAC1B,KAAK,KAAK,QAAQ;IAAE,IAAI,EAAE;IAAI,OAAO,EAAE;GAAM;EAC/C;EACA,IAAI,KAAK,EAAE,MAAM,KAAK,CAAC;EAEvB,IAAI,IAAI;EACR,OAAO,IAAI,KAAK,QAAQ;GACtB,MAAM,OAAQ,IAAI,KAAgB;GAElC,IAAI,OADS,KAAK,EACH,CAAC,OAAO,QAAQ;IAC7B,IAAI,KAAK;IACT;GACF;GACA,IAAI,KAAK;GACT;EACF;EACA,IAAI,MAAM,KAAK,QAAQ;CACzB;CACA,OAAO;AACT;AAEA,SAAS,eAAuB;CAG9B,MAAM,IAAI,KAAK,IAAI,CAAC,CAAC,SAAS,EAAE;CAChC,IAAI,IAAI;CACR,KAAK,IAAI,IAAI,GAAG,IAAI,GAAG,KAAK,KAAK,KAAK,MAAM,KAAK,OAAO,IAAI,EAAE,CAAC,CAAC,SAAS,EAAE;CAC3E,OAAO,OAAO,EAAE,GAAG;AACrB;;;AAIA,SAAS,aAA6B;CACpC,OAAO;EAAE,MAAM;EAAc,KAAK;CAAK;AACzC;AAEA,SAAS,gBAAwB,KAAkC;CACjE,MAAM,IAAI;CACV,MAAM,QAAQ,cAAc,GAAG;CAC/B,OAAO;EACL,QAAQ,KAAA;EACR,SAAS;GAAE,OAAO;GAAO,OAAO;EAAE;EAClC,SAAS,OAAO,WAAW;EAC3B,YAAY,OAAO,cAAc;EACjC,gBACE,UAAU,KAAA,KAAa,MAAM,SAAS,eAClC,WAAW,IACX;GAAE,MAAM,MAAM;GAAM,KAAK,MAAM;EAAQ;EAC7C,OAAO;GACL,SAAS,OAAO,GAAG,YAAY,WAAW,EAAE,UAAU,OAAO,GAAG;GAChE,MAAM,OAAO,GAAG,SAAS,WAAW,EAAE,OAAO;EAC/C;CACF;AACF;;;;;;;;;;;;AAaA,SAAS,eACP,QACA,eACoB;CACpB,IAAI,CAACC,0BAAS,OAAO,OAAO,GAC1B,OAAO,gCACL,IAAI,WACF,4BAA4B,OAAO,OAAO,OAAO,EAAE,iFACrD,CACF;CAEF,IAAIA,0BAAS,OAAO,UAAU,GAAG,OAAO;CACxC,cAAc,OAAO,UAAU;CAC/B,OAAO;EAAE,GAAG;EAAQ,YAAY;CAAE;AACpC;AAEA,eAAsB,eACpB,MAC+B;CAC/B,MAAM,YAAY,KAAK,IAAI;CAC3B,MAAM,OAAO,KAAK,IAAI,GAAG,KAAK,QAAQ,CAAC;CACvC,MAAM,iBAAiB,KAAK,IAAI,GAAG,KAAK,kBAAkB,CAAC;CAC3D,MAAM,cAAc,KAAK,eAAe,OAAO;CAC/C,MAAM,cAAc,KAAK,eAAe,KAAK,KAAK,KAAK,MAAM,EAAE,IAAI;CACnE,IAAI,KAAK,mBAAmB,KAAA,KAAa,CAACA,0BAAS,KAAK,cAAc,GACpE,MAAM,IAAI,WACR,yEAAyE,OAAO,KAAK,cAAc,GACrG;CAGF,MAAM,OAAO,UAAU,KAAK,IAAI;CAChC,MAAM,WAAW,KAAK,SAClB,KAAK,QAAQ,MAAO,KAAK,OAAoC,CAAC,CAAC,IAC/D;CACJ,MAAM,cAAc,KAAK,SAAS,SAAS;CAE3C,MAAM,UAAwB,CAAC;CAC/B,KAAK,IAAI,IAAI,GAAG,IAAI,SAAS,QAAQ,KACnC,KAAK,IAAI,IAAI,GAAG,IAAI,MAAM,KACxB,QAAQ,KAAK;EACX,MAAO,SAAS,EAAE,CAAc;EAChC,KAAK;EACL,SAAS,IAAI,OAAO;CACtB,CAAC;CAIL,MAAM,cAAuE,CAAC;CAI9E,IAAI,iBAAiB;CACrB,IAAI,qBAAqB;CACzB,IAAI,eAAe;CACnB,IAAI,mBAAmB;CACvB,IAAI,sBAAsB;CAC1B,IAAI,mBAAmB;CAEvB,MAAM,gBAAyB,KAAK,QAAQ,YAAY;CAOxD,IAAI,WAAW;CACf,IAAI,SAAS;CACb,IAAI;CACJ,MAAM,OAAO,IAAI,SAAe,QAAQ;EACtC,aAAa;CACf,CAAC;CAED,MAAM,aAAmB;EACvB,OAAO,WAAW,kBAAkB,SAAS,QAAQ,QAAQ;GAC3D,IAAI,QAAQ,KAAK,oBAAoB;IAEnC,MAAM,OAAO,QAAQ,SAAS;IAC9B,oBAAoB;IACpB,SAAS,QAAQ;IACjB;GACF;GACA,MAAM,OAAO,QAAQ;GACrB;GAGA,MAAM,SAAS;IAAE;IAAM,MAAM,CAAC;GAA0B;GACxD,YAAY,KAAK,MAAM;GAQvB,CAP8C,YAAY;IACxD,IAAI;KACF,OAAO,MAAM,KAAK,QAAQ,IAAI;IAChC,SAAS,KAAK;KACZ,OAAO,gBAAwB,GAAG;IACpC;GACF,EAAA,CACM,CAAC,CAAC,MAAM,YAAY;IACxB,MAAM,SAAS,eAAe,UAAU,aAAa;KACnD;KACA,IAAI,qBAAqB,GAEvB,QAAQ,KACN,uCAAuC,OAAO,QAAQ,EAAE,wDAC1D;IAEJ,CAAC;IACD,OAAO,KAAK,KAAK,MAAM;IACvB;IACA,MAAM,iBAAiB,OAAO,gBAAgB,SAAS;IACvD,kBACE,kBAAkB,KAAK,mBAAmB,KAAA,IACtC,KAAK,IAAI,OAAO,SAAS,KAAK,cAAc,IAC5C,OAAO;IACb,IAAI,gBAAgB;KAClB;KACA,IAAI,wBAAwB,GAE1B,QAAQ,KACN,4GACF;IAEJ;IACA,IAAI,kBAAkB,eAAe,CAAC,oBAAoB;KACxD,qBAAqB;KAErB,QAAQ,KAAK,+BAA+B;IAC9C;IACA,IAAI;KACF,KAAK,iBAAiB,MAAM,MAAM;IACpC,QAAQ,CAGR;IACA;IACA,IAAI,SAAS,QAAQ,QACnB,KAAK;SACA,IAAI,aAAa,GACtB,aAAa;GAEjB,CAAC;EACH;EACA,IAAI,UAAU,QAAQ,UAAU,aAAa,GAAG,aAAa;CAC/D;CAEA,MAAM,gBAAsB;EAG1B,IAAI,SAAS,QAAQ,QAAQ;GAC3B,oBAAoB,QAAQ,SAAS;GACrC,SAAS,QAAQ;EACnB;EACA,IAAI,aAAa,GAAG,aAAa;CACnC;CACA,IAAI,KAAK,QAAQ;EACf,IAAI,KAAK,OAAO,SAAS;GACvB,mBAAmB,QAAQ;GAC3B,SAAS,QAAQ;GACjB,aAAa;EACf,OACE,KAAK,OAAO,iBAAiB,SAAS,SAAS,EAAE,MAAM,KAAK,CAAC;CAEjE;CAEA,IAAI,QAAQ,WAAW,GACrB,aAAa;MAEb,KAAK;CAGP,MAAM;CACN,IAAI,KAAK,QAAQ,KAAK,OAAO,oBAAoB,SAAS,OAAO;CAEjE,YAAY,MAAM,GAAG,MAAM,EAAE,KAAK,UAAU,EAAE,KAAK,OAAO;CAE1D,IAAI,OAAO;CACX,IAAI,WAAW;CACf,IAAI,YAAY;CAChB,IAAI,WAAW;CACf,KAAK,MAAM,EAAE,UAAU,aACrB,KAAK,MAAM,KAAK,MAAM;EACpB;EACA,MAAM,UAAU,EAAE,UAAU,KAAA;EAC5B,IAAI,CAAC,WAAW,EAAE,QAAQ,OAAO;EACjC,YAAY,UAAU,IAAI,EAAE,QAAQ;EACpC,aAAa,EAAE;CACjB;CAKF,OAAO;EACL,OAAO;EACP,QAJa,YAAY,aAAa,KAAK,MAAM,WAI5C;EACL,SAAS;GACP,YAAY,QAAQ;GACpB;GACA,cAAc,mBAAmB,cAAc;GAC/C,iBAAiB,aAAa,IAAI,IAAI,OAAO;GAC7C,kBAAkB,aAAa,IAAI,IAAI,WAAW;GAClD,cAAc;GACd;GACA,mBAAmB;GACnB,YAAY,KAAK,IAAI,IAAI;EAC3B;EACA,UAAU,aAAa;CACzB;AACF"}

package/dist/{paired-tests-jbwPdSz5.js → mean-DHoMiudD.js}

@@ -1,6 +1,6 @@
-import { s as ValidationError } from "./errors-Dngq5h35.js";
-import { a as medianInPlace, i as makeRng, n as binomialHalfUpperTail, t as assertFiniteSample } from "./internal-CRHwS9iz.js";
-import { t as studentTCdf } from "./student-t-BQeKJx2N.js";
+import { s as ValidationError } from "./errors-DVZiqS7K.js";
+import { a as medianInPlace, i as makeRng, n as binomialHalfUpperTail, t as assertFiniteSample } from "./internal-Epl-YBZW.js";
+import { t as studentTCdf } from "./student-t-DVPW7CdL.js";
 //#region src/statistics/paired-tests.ts
 /**
 * Paired significance tests on continuous scores: the paired t-test, the
@@ -208,6 +208,12 @@ function pairedDeltaTieFraction(before, after) {
 */
 const DECISION_PAIRED_DELTA_STATISTIC = "mean";
 //#endregion
-export { pairedSignTest as a, pairedDeltaTieFraction as i, DECISION_PAIRED_DELTA_STATISTIC as n, pairedTTest as o, pairedBootstrap as r, BOOTSTRAP_GATE_MIN_N as t };
+//#region src/statistics/mean.ts
+/** Arithmetic mean. An empty input has no mean and returns NaN. */
+function mean(values) {
+	return values.reduce((sum, value) => sum + value, 0) / values.length;
+}
+//#endregion
+export { pairedDeltaTieFraction as a, pairedBootstrap as i, BOOTSTRAP_GATE_MIN_N as n, pairedSignTest as o, DECISION_PAIRED_DELTA_STATISTIC as r, pairedTTest as s, mean as t };
 
-//# sourceMappingURL=paired-tests-jbwPdSz5.js.map
+//# sourceMappingURL=mean-DHoMiudD.js.map

package/dist/mean-DHoMiudD.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"mean-DHoMiudD.js","names":[],"sources":["../src/statistics/paired-tests.ts","../src/statistics/mean.ts"],"sourcesContent":["/**\n * Paired significance tests on continuous scores: the paired t-test, the\n * promotion-gate paired bootstrap, the exact sign test, and the package-wide\n * paired-delta decision statistic.\n */\n\nimport { ValidationError } from '../errors'\nimport { studentTCdf } from '../math/student-t'\nimport { assertFiniteSample, binomialHalfUpperTail, makeRng, medianInPlace } from './internal'\n\nexport interface PairedTTestResult {\n  /** Null when the statistic is undefined — see {@link pairedTTest}. */\n  t: number | null\n  df: number\n  /** Null exactly when `t` is null. */\n  p: number | null\n}\n\n/**\n * Paired t-test — before/after measurements on the SAME items.\n * Pairing removes inter-item variance, giving tighter significance than\n * an unpaired test when comparing prompt v1 vs prompt v2 on identical\n * scenarios.\n *\n * Returns `t = p = null` where the statistic is undefined: fewer than two\n * pairs, or a non-zero constant delta whose observed variance is zero. A\n * constant shift carries no information about the variance it would have to\n * be compared against, so the honest answer is \"undefined\", not `p = 0` —\n * three observations cannot buy absolute certainty. This is the same contract\n * {@link pairedCohensDz} states for the same condition. An all-zero delta is\n * different: it is a measured null, and returns `t = 0, p = 1`.\n */\nexport function pairedTTest(before: number[], after: number[]): PairedTTestResult {\n  if (before.length !== after.length) {\n    throw new ValidationError(\n      `pairedTTest: unequal sample sizes (${before.length} vs ${after.length})`,\n    )\n  }\n  assertFiniteSample('pairedTTest', 'before', before)\n  assertFiniteSample('pairedTTest', 'after', after)\n  const n = before.length\n  if (n < 2) return { t: null, df: 0, p: null }\n\n  const diffs = before.map((b, i) => after[i]! - b)\n  const mean = diffs.reduce((a, b) => a + b, 0) / n\n  const variance = diffs.reduce((acc, d) => acc + (d - mean) ** 2, 0) / (n - 1)\n  const se = Math.sqrt(variance / n)\n  if (se === 0) {\n    return mean === 0 ? { t: 0, df: n - 1, p: 1 } : { t: null, df: n - 1, p: null }\n  }\n\n  const t = mean / se\n  const df = n - 1\n  const p = 2 * (1 - studentTCdf(Math.abs(t), df))\n  return { t, df, p }\n}\n\n// ── Paired bootstrap (promotion-gate effect size) ────────────────────\n\nexport interface PairedBootstrapResult {\n  /** Number of paired observations. */\n  n: number\n  /** Median of paired deltas (after − before). */\n  median: number\n  /** Mean of paired deltas. */\n  mean: number\n  /** Lower bound of the bootstrap CI on the chosen statistic. */\n  low: number\n  /** Upper bound of the bootstrap CI on the chosen statistic. */\n  high: number\n  /** Confidence level used (e.g. 0.95). */\n  confidence: number\n  /** Number of bootstrap resamples used. */\n  resamples: number\n  /** False below {@link BOOTSTRAP_GATE_MIN_N}. See {@link pairedBootstrap}. */\n  gateEligible: boolean\n}\n\n/**\n * Pairs below which a percentile bootstrap interval is descriptive spread only.\n *\n * `P(low > 0)` under a true null, against a nominal 2.5 %, measured over 4000\n * seeded trials: 13.53 % at n = 3, 3.52 % at n = 10, 3.10 % at n = 20 on the\n * median; 13.85 %, 4.90 %, 3.80 % on the mean. This is intrinsic to resampling\n * three points, not an implementation error — scipy's BCa gives 16.0 % on the\n * same n = 3 data — so no change to the estimator moves it. Below this floor\n * the decision belongs to the exact sign test or exact signed-rank test.\n */\nexport const BOOTSTRAP_GATE_MIN_N = 20\n\nexport interface PairedBootstrapOptions {\n  /** Confidence level. Default 0.95. */\n  confidence?: number\n  /** Bootstrap resample count. Default 2000. */\n  resamples?: number\n  /** Statistic to bootstrap. Default 'median'. */\n  statistic?: 'median' | 'mean'\n  /** Deterministic seed. If omitted, derived from the deltas so the interval\n   *  is reproducible regardless. */\n  seed?: number\n}\n\n/**\n * Paired bootstrap on (after − before) deltas. Returns a CI on the chosen\n * statistic (median by default); pairs are resampled with replacement. Throws\n * on unequal sample sizes.\n *\n * `low > threshold` carries the stated confidence ONLY at `n ≥\n * {@link BOOTSTRAP_GATE_MIN_N}`, which `gateEligible` reports. Below it the\n * check fires under a true null several times more often than nominal, so the\n * interval is descriptive spread and a promotion must not turn on it.\n */\nexport function pairedBootstrap(\n  before: number[],\n  after: number[],\n  opts: PairedBootstrapOptions = {},\n): PairedBootstrapResult {\n  if (before.length !== after.length) {\n    throw new Error(`pairedBootstrap: unequal sample sizes (${before.length} vs ${after.length})`)\n  }\n  const confidence = opts.confidence ?? 0.95\n  const resamples = opts.resamples ?? 2000\n  const statistic = opts.statistic ?? 'median'\n  if (confidence <= 0 || confidence >= 1) {\n    throw new Error(`pairedBootstrap: confidence must be in (0,1), got ${confidence}`)\n  }\n\n  const n = before.length\n  const deltas = before.map((b, i) => after[i]! - b)\n  const gateEligible = n >= BOOTSTRAP_GATE_MIN_N\n  if (n === 0) {\n    return { n: 0, median: 0, mean: 0, low: 0, high: 0, confidence, resamples, gateEligible }\n  }\n  if (n === 1) {\n    const d = deltas[0]!\n    return { n: 1, median: d, mean: d, low: d, high: d, confidence, resamples, gateEligible }\n  }\n\n  const rng = makeRng(opts.seed, deltas)\n  const samples = new Array<number>(resamples)\n  for (let b = 0; b < resamples; b++) {\n    if (statistic === 'mean') {\n      let sum = 0\n      for (let k = 0; k < n; k++) {\n        sum += deltas[Math.floor(rng() * n)]!\n      }\n      samples[b] = sum / n\n    } else {\n      const acc = new Array<number>(n)\n      for (let k = 0; k < n; k++) {\n        acc[k] = deltas[Math.floor(rng() * n)]!\n      }\n      samples[b] = medianInPlace(acc)\n    }\n  }\n  samples.sort((a, b) => a - b)\n\n  const alpha = 1 - confidence\n  const lowIdx = Math.floor((alpha / 2) * resamples)\n  const highIdx = Math.min(resamples - 1, Math.ceil((1 - alpha / 2) * resamples) - 1)\n\n  return {\n    n,\n    median: medianInPlace([...deltas]),\n    mean: deltas.reduce((s, x) => s + x, 0) / n,\n    low: samples[lowIdx]!,\n    high: samples[Math.max(highIdx, lowIdx)]!,\n    confidence,\n    resamples,\n    gateEligible,\n  }\n}\n\n/** Pre-registered direction for a one-sided paired sign test. */\nexport type SignTestAlternative = 'greater' | 'less'\n\n/** Exact one-sided sign-test result for paired numeric differences. */\nexport interface PairedSignTestResult {\n  /** Total supplied differences, including zero ties. */\n  n: number\n  /** Strictly positive differences. */\n  positive: number\n  /** Strictly negative differences. */\n  negative: number\n  /** Zero differences excluded from the binomial test. */\n  ties: number\n  /** Non-zero differences used by the binomial test. */\n  nNonTies: number\n  /** Direction of the pre-registered alternative hypothesis. */\n  alternative: SignTestAlternative\n  /** Exact one-sided p-value under P(positive) = P(negative) = 0.5. */\n  pValue: number\n}\n\n/**\n * Exact one-sided sign test over paired differences.\n *\n * Pass `after[i] - before[i]` for each matched item. `alternative = 'greater'`\n * tests whether positive signs are more likely than negative signs and returns\n * `P(Binomial(nNonTies, 0.5) >= positive)`. `alternative = 'less'` treats\n * negative signs as successes instead. With a continuous difference\n * distribution this is the usual directional median test. Exact zero\n * differences are ties and do not enter the binomial denominator. All-tie and\n * empty inputs return p = 1. Every input difference must be finite, and the\n * direction must be chosen explicitly so a caller cannot select it after\n * seeing the signs.\n */\nexport function pairedSignTest(\n  differences: readonly number[],\n  alternative: SignTestAlternative,\n): PairedSignTestResult {\n  if (alternative !== 'greater' && alternative !== 'less') {\n    throw new ValidationError(\n      `pairedSignTest: alternative must be 'greater' or 'less', got ${alternative}`,\n    )\n  }\n\n  let positive = 0\n  let negative = 0\n  let ties = 0\n  for (let i = 0; i < differences.length; i++) {\n    const difference = differences[i]!\n    if (!Number.isFinite(difference)) {\n      throw new ValidationError(\n        `pairedSignTest: difference at index ${i} must be finite, got ${difference}`,\n      )\n    }\n    if (difference > 0) positive++\n    else if (difference < 0) negative++\n    else ties++\n  }\n\n  const nNonTies = positive + negative\n  const successes = alternative === 'greater' ? positive : negative\n  return {\n    n: differences.length,\n    positive,\n    negative,\n    ties,\n    nNonTies,\n    alternative,\n    pValue: binomialHalfUpperTail(successes, nNonTies),\n  }\n}\n\n/** Fraction of paired observations whose delta is an exact tie (|after − before|\n *  < 1e-9). Throws on unequal sample sizes; 0 pairs ⇒ 0. */\nexport function pairedDeltaTieFraction(\n  before: ArrayLike<number>,\n  after: ArrayLike<number>,\n): number {\n  if (before.length !== after.length) {\n    throw new Error(\n      `pairedDeltaTieFraction: unequal sample sizes (${before.length} vs ${after.length})`,\n    )\n  }\n  const n = before.length\n  if (n === 0) return 0\n  let ties = 0\n  for (let i = 0; i < n; i++) {\n    if (Math.abs(after[i]! - before[i]!) < 1e-9) ties++\n  }\n  return ties / n\n}\n\n/**\n * The paired-delta statistic a DECISION is computed on, package-wide.\n *\n * The mean paired delta is the estimator that answers the question a promotion\n * gate asks — \"by how much did the candidate move the score\" — in the caller's\n * own units, and it equals the aggregate lift everyone quotes. The MEDIAN\n * answers a different question and loses the answer to this one in every regime\n * eval data actually lands in:\n *   - TWO-POINT (pass/fail) outcomes on any encoding: the delta vector lives in\n *     {−s, 0, +s} dominated by zeros, so the median and its whole bootstrap CI\n *     are pinned at exactly 0 however large the shift. (Decide these on\n *     {@link pairedRiskDifferenceExact} instead — same estimand, exact interval.)\n *   - TIE-DOMINATED outcomes: at half the pairs tied the sample median is 0 by\n *     construction, and `ci.low > threshold` then answers \"no\" forever at a\n *     non-negative threshold and \"yes\" forever at a negative one.\n *   - LOW-CARDINALITY outcomes, even well below half ties: judge dimensions on\n *     integer 0-100, and block scores like {⅔, 1} from averaging pass/fail\n *     leaves, put the median on a coarse lattice whose bootstrap percentiles\n *     land on atoms. Measured: 26 blocks of 3 pass/fail leaves carrying a real\n *     +12.8pp lift, only 23% of pairs tied, gives a median CI of [0, 0.333] —\n *     lower bound exactly 0, so a gate at threshold 0 refuses a real lift.\n * That last case is why there is no tie-fraction threshold here: any cutoff on\n * ties leaves the lattice case open on the other side of it.\n *\n * `heldoutSignificance` has defaulted to the mean since #316 for the same\n * reason. The median remains available per call site for callers who\n * specifically want outlier robustness and accept the blindness.\n */\nexport const DECISION_PAIRED_DELTA_STATISTIC: 'mean' = 'mean'\n","/** Arithmetic mean. An empty input has no mean and returns NaN. */\nexport function mean(values: readonly number[]): number {\n  return values.reduce((sum, value) => sum + value, 0) / values.length\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;AAgCA,SAAgB,YAAY,QAAkB,OAAoC;CAChF,IAAI,OAAO,WAAW,MAAM,QAC1B,MAAM,IAAI,gBACR,sCAAsC,OAAO,OAAO,MAAM,MAAM,OAAO,EACzE;CAEF,mBAAmB,eAAe,UAAU,MAAM;CAClD,mBAAmB,eAAe,SAAS,KAAK;CAChD,MAAM,IAAI,OAAO;CACjB,IAAI,IAAI,GAAG,OAAO;EAAE,GAAG;EAAM,IAAI;EAAG,GAAG;CAAK;CAE5C,MAAM,QAAQ,OAAO,KAAK,GAAG,MAAM,MAAM,KAAM,CAAC;CAChD,MAAM,OAAO,MAAM,QAAQ,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI;CAChD,MAAM,WAAW,MAAM,QAAQ,KAAK,MAAM,OAAO,IAAI,SAAS,GAAG,CAAC,KAAK,IAAI;CAC3E,MAAM,KAAK,KAAK,KAAK,WAAW,CAAC;CACjC,IAAI,OAAO,GACT,OAAO,SAAS,IAAI;EAAE,GAAG;EAAG,IAAI,IAAI;EAAG,GAAG;CAAE,IAAI;EAAE,GAAG;EAAM,IAAI,IAAI;EAAG,GAAG;CAAK;CAGhF,MAAM,IAAI,OAAO;CACjB,MAAM,KAAK,IAAI;CAEf,OAAO;EAAE;EAAG;EAAI,GADN,KAAK,IAAI,YAAY,KAAK,IAAI,CAAC,GAAG,EAAE;CAC5B;AACpB;;;;;;;;;;;AAiCA,MAAa,uBAAuB;;;;;;;;;;;AAwBpC,SAAgB,gBACd,QACA,OACA,OAA+B,CAAC,GACT;CACvB,IAAI,OAAO,WAAW,MAAM,QAC1B,MAAM,IAAI,MAAM,0CAA0C,OAAO,OAAO,MAAM,MAAM,OAAO,EAAE;CAE/F,MAAM,aAAa,KAAK,cAAc;CACtC,MAAM,YAAY,KAAK,aAAa;CACpC,MAAM,YAAY,KAAK,aAAa;CACpC,IAAI,cAAc,KAAK,cAAc,GACnC,MAAM,IAAI,MAAM,qDAAqD,YAAY;CAGnF,MAAM,IAAI,OAAO;CACjB,MAAM,SAAS,OAAO,KAAK,GAAG,MAAM,MAAM,KAAM,CAAC;CACjD,MAAM,eAAe,KAAA;CACrB,IAAI,MAAM,GACR,OAAO;EAAE,GAAG;EAAG,QAAQ;EAAG,MAAM;EAAG,KAAK;EAAG,MAAM;EAAG;EAAY;EAAW;CAAa;CAE1F,IAAI,MAAM,GAAG;EACX,MAAM,IAAI,OAAO;EACjB,OAAO;GAAE,GAAG;GAAG,QAAQ;GAAG,MAAM;GAAG,KAAK;GAAG,MAAM;GAAG;GAAY;GAAW;EAAa;CAC1F;CAEA,MAAM,MAAM,QAAQ,KAAK,MAAM,MAAM;CACrC,MAAM,UAAU,IAAI,MAAc,SAAS;CAC3C,KAAK,IAAI,IAAI,GAAG,IAAI,WAAW,KAC7B,IAAI,cAAc,QAAQ;EACxB,IAAI,MAAM;EACV,KAAK,IAAI,IAAI,GAAG,IAAI,GAAG,KACrB,OAAO,OAAO,KAAK,MAAM,IAAI,IAAI,CAAC;EAEpC,QAAQ,KAAK,MAAM;CACrB,OAAO;EACL,MAAM,MAAM,IAAI,MAAc,CAAC;EAC/B,KAAK,IAAI,IAAI,GAAG,IAAI,GAAG,KACrB,IAAI,KAAK,OAAO,KAAK,MAAM,IAAI,IAAI,CAAC;EAEtC,QAAQ,KAAK,cAAc,GAAG;CAChC;CAEF,QAAQ,MAAM,GAAG,MAAM,IAAI,CAAC;CAE5B,MAAM,QAAQ,IAAI;CAClB,MAAM,SAAS,KAAK,MAAO,QAAQ,IAAK,SAAS;CACjD,MAAM,UAAU,KAAK,IAAI,YAAY,GAAG,KAAK,MAAM,IAAI,QAAQ,KAAK,SAAS,IAAI,CAAC;CAElF,OAAO;EACL;EACA,QAAQ,cAAc,CAAC,GAAG,MAAM,CAAC;EACjC,MAAM,OAAO,QAAQ,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI;EAC1C,KAAK,QAAQ;EACb,MAAM,QAAQ,KAAK,IAAI,SAAS,MAAM;EACtC;EACA;EACA;CACF;AACF;;;;;;;;;;;;;;AAoCA,SAAgB,eACd,aACA,aACsB;CACtB,IAAI,gBAAgB,aAAa,gBAAgB,QAC/C,MAAM,IAAI,gBACR,gEAAgE,aAClE;CAGF,IAAI,WAAW;CACf,IAAI,WAAW;CACf,IAAI,OAAO;CACX,KAAK,IAAI,IAAI,GAAG,IAAI,YAAY,QAAQ,KAAK;EAC3C,MAAM,aAAa,YAAY;EAC/B,IAAI,CAAC,OAAO,SAAS,UAAU,GAC7B,MAAM,IAAI,gBACR,uCAAuC,EAAE,uBAAuB,YAClE;EAEF,IAAI,aAAa,GAAG;OACf,IAAI,aAAa,GAAG;OACpB;CACP;CAEA,MAAM,WAAW,WAAW;CAC5B,MAAM,YAAY,gBAAgB,YAAY,WAAW;CACzD,OAAO;EACL,GAAG,YAAY;EACf;EACA;EACA;EACA;EACA;EACA,QAAQ,sBAAsB,WAAW,QAAQ;CACnD;AACF;;;AAIA,SAAgB,uBACd,QACA,OACQ;CACR,IAAI,OAAO,WAAW,MAAM,QAC1B,MAAM,IAAI,MACR,iDAAiD,OAAO,OAAO,MAAM,MAAM,OAAO,EACpF;CAEF,MAAM,IAAI,OAAO;CACjB,IAAI,MAAM,GAAG,OAAO;CACpB,IAAI,OAAO;CACX,KAAK,IAAI,IAAI,GAAG,IAAI,GAAG,KACrB,IAAI,KAAK,IAAI,MAAM,KAAM,OAAO,EAAG,IAAI,MAAM;CAE/C,OAAO,OAAO;AAChB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA8BA,MAAa,kCAA0C;;;;ACpSvD,SAAgB,KAAK,QAAmC;CACtD,OAAO,OAAO,QAAQ,KAAK,UAAU,MAAM,OAAO,CAAC,IAAI,OAAO;AAChE"}

package/dist/meta-eval/index.d.ts

@@ -1,11 +1,11 @@
 import { f as Run } from "../schema-2g7trUwH.js";
 import { a as ContinuousCalibrationResult, c as SelfPreferenceResult, d as calibrateJudgeContinuous, f as continuousAgreement, h as verbosityBias, i as ContinuousAgreementOptions, l as VerbosityBiasResult, m as selfPreference, n as CandidateScore, o as GoldenItem, p as positionalBias, r as ContinuousAgreement, s as PositionalBiasResult, t as CalibrationResult, u as calibrateJudge } from "../judge-calibration-DFtEMlde.js";
-import { A as SentinelThresholds, B as validateSentinelSnapshot, C as JudgeSentinelOptions, D as SentinelSetOptions, E as SentinelReport, F as inMemorySentinelStore, I as judgeSentinelReport, L as snapshotFromAgreement, M as SnapshotMeta, N as evalHealthStamp, O as SentinelSnapshot, P as fileSentinelStore, R as snapshotFromCalibration, S as EvalHealthStamp, T as SentinelMetrics, _ as assertJudgeMayGate, a as JudgeCalibrationSet, b as registerCalibrationSet, c as JudgeGateDecision, d as JudgeGateRefusedError, f as JudgeGateStatus, g as addCalibrationVerdicts, h as OwnerVerdict, i as JudgeAgreement, j as SentinelTrend, k as SentinelStore, l as JudgeGateInput, m as JudgeRunMeta, n as CalibrationMerge, o as JudgeConfusion, p as JudgeRun, r as DEFAULT_JUDGE_GATE_POLICY, s as JudgeExampleAgreement, t as CalibrationExample, u as JudgeGatePolicy, v as judgeGateDecision, w as SentinelMetricName, x as snapshotFromJudgeAgreement, y as measureJudgeAgreement, z as snapshotFromSentinelSet } from "../judge-gate-DNcUrRFA.js";
+import { A as SentinelThresholds, B as validateSentinelSnapshot, C as JudgeSentinelOptions, D as SentinelSetOptions, E as SentinelReport, F as inMemorySentinelStore, I as judgeSentinelReport, L as snapshotFromAgreement, M as SnapshotMeta, N as evalHealthStamp, O as SentinelSnapshot, P as fileSentinelStore, R as snapshotFromCalibration, S as EvalHealthStamp, T as SentinelMetrics, _ as assertJudgeMayGate, a as JudgeCalibrationSet, b as registerCalibrationSet, c as JudgeGateDecision, d as JudgeGateRefusedError, f as JudgeGateStatus, g as addCalibrationVerdicts, h as OwnerVerdict, i as JudgeAgreement, j as SentinelTrend, k as SentinelStore, l as JudgeGateInput, m as JudgeRunMeta, n as CalibrationMerge, o as JudgeConfusion, p as JudgeRun, r as DEFAULT_JUDGE_GATE_POLICY, s as JudgeExampleAgreement, t as CalibrationExample, u as JudgeGatePolicy, v as judgeGateDecision, w as SentinelMetricName, x as snapshotFromJudgeAgreement, y as measureJudgeAgreement, z as snapshotFromSentinelSet } from "../judge-gate-CE7CQAgt.js";
 import { r as LedgerHash } from "../canonical-CFpojCN5.js";
 import { s as TraceStore } from "../store-C4avFjIi.js";
 import { a as OutcomeFilter, i as InMemoryOutcomeStore, n as FileSystemOutcomeStore, o as OutcomeStore, r as FileSystemOutcomeStoreOptions, s as OutcomeStoreError, t as DeploymentOutcome } from "../outcome-store-CNt4iZ67.js";
-import { a as ComputedInterval } from "../ast-hI-vjW6J.js";
-import { a as RubricPredictiveValidityReport, c as CorrelationInterval, i as RubricPredictiveValidityInput, l as OutcomeReduction, n as RubricOutcomeExclusion, o as RubricRanking, r as RubricOutcomePair, s as rubricPredictiveValidity, t as OutcomeMetricSpec } from "../rubric-predictive-validity-2SgHxoa5.js";
+import { t as ComputedInterval } from "../interval-ohuOJurX.js";
+import { a as RubricPredictiveValidityReport, c as CorrelationInterval, i as RubricPredictiveValidityInput, l as OutcomeReduction, n as RubricOutcomeExclusion, o as RubricRanking, r as RubricOutcomePair, s as rubricPredictiveValidity, t as OutcomeMetricSpec } from "../rubric-predictive-validity-C03EC7ZE.js";
 import { z } from "zod";
 //#region src/meta-eval/correlation-study.d.ts
 export interface EvalMetricSpec {

package/dist/meta-eval/index.js

@@ -1,12 +1,12 @@
-import { n as CaptureIntegrityError, s as ValidationError } from "../errors-Dngq5h35.js";
-import { a as hashCanonical, i as compareCodeUnits } from "../canonical-BsZVgBhW.js";
-import { t as mulberry32 } from "../random-Dn5fPWkt.js";
-import { a as selfPreference, i as positionalBias, n as calibrateJudgeContinuous, o as verbosityBias, r as continuousAgreement, t as calibrateJudge } from "../judge-calibration-N44V04Se.js";
-import { l as runMetricExtractor } from "../query-B1D1qp2Z.js";
-import { r as computeInterval } from "../ast-DJ6Bc7ab.js";
-import { a as judgeGateDecision, c as snapshotFromJudgeAgreement, d as inMemorySentinelStore, f as judgeSentinelReport, g as validateSentinelSnapshot, h as snapshotFromSentinelSet, i as assertJudgeMayGate, l as evalHealthStamp, m as snapshotFromCalibration, n as JudgeGateRefusedError, o as measureJudgeAgreement, p as snapshotFromAgreement, r as addCalibrationVerdicts, s as registerCalibrationSet, t as DEFAULT_JUDGE_GATE_POLICY, u as fileSentinelStore } from "../judge-gate-B1qYnkpi.js";
-import { t as clusterBootstrapMean } from "../cluster-bootstrap-B3JywQQn.js";
-import { a as reduceOutcomeMetric, i as hasVariation, n as assertUniqueObservationIds, o as validateObservationOptions, r as correlationSummary, t as rubricPredictiveValidity } from "../rubric-predictive-validity-BsgkBfT0.js";
+import { n as CaptureIntegrityError, s as ValidationError } from "../errors-DVZiqS7K.js";
+import { a as hashCanonical, i as compareCodeUnits } from "../canonical-B2RqLofv.js";
+import { t as mulberry32 } from "../random-ByUSnV9w.js";
+import { a as selfPreference, i as positionalBias, n as calibrateJudgeContinuous, o as verbosityBias, r as continuousAgreement, t as calibrateJudge } from "../judge-calibration-BpCy5XQl.js";
+import { l as runMetricExtractor } from "../query-BUCZCK7F.js";
+import { a as judgeGateDecision, c as snapshotFromJudgeAgreement, d as inMemorySentinelStore, f as judgeSentinelReport, g as validateSentinelSnapshot, h as snapshotFromSentinelSet, i as assertJudgeMayGate, l as evalHealthStamp, m as snapshotFromCalibration, n as JudgeGateRefusedError, o as measureJudgeAgreement, p as snapshotFromAgreement, r as addCalibrationVerdicts, s as registerCalibrationSet, t as DEFAULT_JUDGE_GATE_POLICY, u as fileSentinelStore } from "../judge-gate-iY9s4L44.js";
+import { t as computeInterval } from "../interval-CELpye1n.js";
+import { t as clusterBootstrapMean } from "../cluster-bootstrap-qcUWYFl-.js";
+import { a as reduceOutcomeMetric, i as hasVariation, n as assertUniqueObservationIds, o as validateObservationOptions, r as correlationSummary, t as rubricPredictiveValidity } from "../rubric-predictive-validity-DaFW70y9.js";
 import { n as InMemoryOutcomeStore, r as OutcomeStoreError, t as FileSystemOutcomeStore } from "../outcome-store-DUcMj-3D.js";
 import { z } from "zod";
 //#region src/meta-eval/calibration.ts