modelrig 0.11.1 → 0.12.0

package/dist/bakeoff/report.js

@@ -12,6 +12,7 @@ exports.formatDisagreementSummary = formatDisagreementSummary;
 exports.formatSwapDirection = formatSwapDirection;
 exports.printMatrix = printMatrix;
 exports.runEvalCaseBakeoffCli = runEvalCaseBakeoffCli;
+const turn_replay_1 = require("./turn-replay");
 const node_fs_1 = require("node:fs");
 const node_os_1 = require("node:os");
 const node_path_1 = require("node:path");
@@ -75,7 +76,7 @@ function bakeoffHeadline(artifact) {
         : "cost n/a";
     const ceiling = fmtCeiling(artifact);
     const r = p.regret;
-    return (`${p.variant} — regret ${fmtPts(r.regretMean)} pts ` +
+    return (`${p.variant}${arm?.armLabel !== undefined ? ` (${arm.armLabel})` : ""} — regret ${fmtPts(r.regretMean)} pts ` +
         `(CI ${fmtPts(r.regretCi95[0])}…${fmtPts(r.regretCi95[1])}, n=${r.nPaired}) vs reference ${ref.model}, ` +
         `${cost}${ceiling !== null ? ` vs ceiling ${ceiling}` : ""} — proposed`);
 }
@@ -178,7 +179,7 @@ function printMatrix(artifact, cacheAdjusted = []) {
         header.push("pairwise-win");
     const rows = artifact.variants.map((m) => {
         const base = [
-            m.isReference === true ? `${m.variant} [ref]` : m.variant,
+            m.isReference === true ? `${m.variant} [ref]` : m.armLabel !== undefined ? `${m.variant} (${m.armLabel})` : m.variant, // ARC-β R3-1
             String(m.samples),
             `${(m.conformRate * 100).toFixed(0)}%`,
             `[${(m.conformCi95[0] * 100).toFixed(0)},${(m.conformCi95[1] * 100).toFixed(0)}]`,
@@ -225,6 +226,12 @@ function printMatrix(artifact, cacheAdjusted = []) {
     // byte-identical to the pre-N6 matrix output.
     for (const l of formatDisagreementSummary(artifact))
         console.log(l);
+    for (const m of artifact.variants)
+        for (const l of (0, turn_replay_1.formatIncomparable)(m.incomparable ?? []))
+            console.log(`[${m.variant}] ${l}`); // ARC-β M3(c)
+    for (const m of artifact.variants)
+        for (const l of (0, turn_replay_1.formatFallbackServed)(m.servedByFallback ?? []))
+            console.log(`[${m.variant}] ${l}`); // ARC-β R3-1
     // Cache-adjusted section (spec §5.4) — printed only when the caller supplied
     // simulator rows; absent ⇒ the matrix output is byte-identical to pre-WS-B.
     for (const l of formatCacheAdjustedSection(cacheAdjusted))

package/dist/bakeoff/turn-replay.d.ts

@@ -0,0 +1,93 @@
+/**
+ * Turn replay — ARC-β (agent-archetypes spec §4, builder-prompt-02 §1.7).
+ *
+ * "Would a cheaper model have answered THIS turn as well, given the same
+ * conversation?" A messages route's capture row holds the history up to and
+ * including the turn's last user/tool message (run/messages.ts capture codec)
+ * and the served reply. The bake-off replays that history against each
+ * candidate, which answers the FINAL turn only, graded by the route's existing
+ * grader path with the regret bound unchanged. Replay NEVER continues a
+ * conversation past the captured turn — there is no simulated next user turn
+ * (spec §9 refuses user simulators).
+ *
+ * M15 GUARD (unruled → conservative, roadmap §2): turn replay re-sends real
+ * end-user text to candidate models and graders, so it runs only for routes
+ * whose capture is EXPLICITLY on under the existing capture posture
+ * (`effectiveCapture`: the control-plane flip wins, else the YAML flag).
+ * Otherwise the bake-off refuses with a teaching 422.
+ */
+import { RouteConfigError } from "../routes/loader";
+import type { TaskSettingsResolver } from "../settings/task-settings";
+import type { CaptureRow } from "../telemetry/capture";
+import type { RouteBundle } from "../types/route";
+import type { HistoryMessage } from "../run/messages";
+import type { CaptureReplayInput } from "../run/messages";
+import type { RigFailure } from "../types/failures";
+import type { ReplaySampleResult } from "./replay";
+/** The bake-off refused a messages route because capture is not explicitly on.
+ * A RouteConfigError (refused before any spend), carrying the HTTP status a
+ * server surface should answer with. */
+export declare class TurnReplayRefusedError extends RouteConfigError {
+    readonly httpStatus = 422;
+    constructor(route: string);
+}
+/** Throw {@link TurnReplayRefusedError} for a messages route whose capture is
+ * not explicitly on; a no-op for every variables route (byte-identical). */
+export declare function assertTurnReplayAllowed(route: Pick<RouteBundle, "route" | "inputKind" | "capture">, settings: TaskSettingsResolver | null): void;
+/** One replayable turn of a captured conversation. */
+export interface TurnReplayCase {
+    /** The route variables the system template renders from. */
+    readonly variables: Record<string, unknown>;
+    /** The conversation up to and including the turn's last user/tool message. */
+    readonly history: readonly HistoryMessage[];
+    /** The reply that was served for this turn (the capture's output). */
+    readonly reference: string;
+    /** The live call's tools (R2-1), when it declared any. */
+    readonly tools?: CaptureReplayInput["tools"];
+    readonly toolChoice?: CaptureReplayInput["toolChoice"];
+}
+/** The turn a capture row encodes, or null for a variables (non-messages) row. */
+export declare function turnReplayCase(capture: Pick<CaptureRow, "variablesJson" | "outputText">): TurnReplayCase | null;
+/** The incomparable marker for a replay failure, or null for an ordinary one. */
+export declare function incomparableFrom(failure: RigFailure): {
+    readonly model: string;
+    readonly reason: string;
+} | null;
+/** One "n cases incomparable for <model>: reason" line's data. */
+export interface IncomparableSummary {
+    readonly model: string;
+    readonly reason: string;
+    readonly n: number;
+}
+/**
+ * Split an arm's samples into the COMPARABLE ones (scored, paired for regret)
+ * and a per-(model, reason) count of the incomparable ones. An incomparable
+ * sample is neither a pass nor a failure: dropping it from both sides of the
+ * pairing keeps it from biasing the verdict.
+ */
+export declare function splitIncomparable(samples: readonly ReplaySampleResult[]): {
+    readonly comparable: readonly ReplaySampleResult[];
+    readonly incomparable: readonly IncomparableSummary[];
+};
+/** The report lines: "n cases incomparable for <model>: reason". */
+export declare function formatIncomparable(rows: readonly IncomparableSummary[]): string[];
+/** The `history_skipped` value on a replayed call's telemetry rows, if any. */
+export declare function skippedTagOf(rows: readonly {
+    readonly tagsJson: string;
+}[]): string | undefined;
+/** One "n cases served by <servedBy> (<skipped> skipped: <reason>)" row. */
+export interface FallbackServed {
+    readonly servedBy: string;
+    /** The skipped candidates, `provider/model=code` comma-joined (the tag value). */
+    readonly skipped: string;
+    readonly n: number;
+}
+/** Count an arm's comparable samples answered by a fallback, grouped by
+ * (served model, skipped set). Empty when nothing was skipped. */
+export declare function fallbackServed(samples: readonly ReplaySampleResult[]): FallbackServed[];
+/** "<first> → <fallback…> (fallback served n/N)" — the label an arm's numbers
+ * carry when any of its cases was served by a fallback. */
+export declare function fallbackArmLabel(first: string, rows: readonly FallbackServed[], total: number): string;
+/** The report lines for an arm's fallback-served cases. */
+export declare function formatFallbackServed(rows: readonly FallbackServed[]): string[];
+//# sourceMappingURL=turn-replay.d.ts.map

package/dist/bakeoff/turn-replay.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"turn-replay.d.ts","sourceRoot":"","sources":["../../src/bakeoff/turn-replay.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAEH,OAAO,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AAEpD,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,2BAA2B,CAAC;AACtE,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AACvD,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,gBAAgB,CAAC;AAClD,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AAEtD,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,iBAAiB,CAAC;AAC1D,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,UAAU,CAAC;AAEnD;;wCAEwC;AACxC,qBAAa,sBAAuB,SAAQ,gBAAgB;IAC1D,QAAQ,CAAC,UAAU,OAAO;gBACd,KAAK,EAAE,MAAM;CAS1B;AAED;4EAC4E;AAC5E,wBAAgB,uBAAuB,CACrC,KAAK,EAAE,IAAI,CAAC,WAAW,EAAE,OAAO,GAAG,WAAW,GAAG,SAAS,CAAC,EAC3D,QAAQ,EAAE,oBAAoB,GAAG,IAAI,GACpC,IAAI,CAGN;AAED,sDAAsD;AACtD,MAAM,WAAW,cAAc;IAC7B,4DAA4D;IAC5D,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAC5C,8EAA8E;IAC9E,QAAQ,CAAC,OAAO,EAAE,SAAS,cAAc,EAAE,CAAC;IAC5C,sEAAsE;IACtE,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,0DAA0D;IAC1D,QAAQ,CAAC,KAAK,CAAC,EAAE,kBAAkB,CAAC,OAAO,CAAC,CAAC;IAC7C,QAAQ,CAAC,UAAU,CAAC,EAAE,kBAAkB,CAAC,YAAY,CAAC,CAAC;CACxD;AAED,kFAAkF;AAClF,wBAAgB,cAAc,CAAC,OAAO,EAAE,IAAI,CAAC,UAAU,EAAE,eAAe,GAAG,YAAY,CAAC,GAAG,cAAc,GAAG,IAAI,CAU/G;AAID,iFAAiF;AACjF,wBAAgB,gBAAgB,CAAC,OAAO,EAAE,UAAU,GAAG;IAAE,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAA;CAAE,GAAG,IAAI,CAIhH;AAED,kEAAkE;AAClE,MAAM,WAAW,mBAAmB;IAClC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,CAAC,EAAE,MAAM,CAAC;CACpB;AAED;;;;;GAKG;AACH,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,SAAS,kBAAkB,EAAE,GAAG;IACzE,QAAQ,CAAC,UAAU,EAAE,SAAS,kBAAkB,EAAE,CAAC;IACnD,QAAQ,CAAC,YAAY,EAAE,SAAS,mBAAmB,EAAE,CAAC;CACvD,CAaA;AAED,oEAAoE;AACpE,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,SAAS,mBAAmB,EAAE,GAAG,MAAM,EAAE,CAEjF;AAOD,+EAA+E;AAC/E,wBAAgB,YAAY,CAAC,IAAI,EAAE,SAAS;IAAE,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAA;CAAE,EAAE,GAAG,MAAM,GAAG,SAAS,CAU/F;AAED,4EAA4E;AAC5E,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,kFAAkF;IAClF,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,CAAC,EAAE,MAAM,CAAC;CACpB;AAED;kEACkE;AAClE,wBAAgB,cAAc,CAAC,OAAO,EAAE,SAAS,kBAAkB,EAAE,GAAG,cAAc,EAAE,CAQvF;AAED;2DAC2D;AAC3D,wBAAgB,gBAAgB,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,SAAS,cAAc,EAAE,EAAE,KAAK,EAAE,MAAM,GAAG,MAAM,CAItG;AAED,2DAA2D;AAC3D,wBAAgB,oBAAoB,CAAC,IAAI,EAAE,SAAS,cAAc,EAAE,GAAG,MAAM,EAAE,CAW9E"}

package/dist/bakeoff/turn-replay.js

@@ -0,0 +1,150 @@
+"use strict";
+/**
+ * Turn replay — ARC-β (agent-archetypes spec §4, builder-prompt-02 §1.7).
+ *
+ * "Would a cheaper model have answered THIS turn as well, given the same
+ * conversation?" A messages route's capture row holds the history up to and
+ * including the turn's last user/tool message (run/messages.ts capture codec)
+ * and the served reply. The bake-off replays that history against each
+ * candidate, which answers the FINAL turn only, graded by the route's existing
+ * grader path with the regret bound unchanged. Replay NEVER continues a
+ * conversation past the captured turn — there is no simulated next user turn
+ * (spec §9 refuses user simulators).
+ *
+ * M15 GUARD (unruled → conservative, roadmap §2): turn replay re-sends real
+ * end-user text to candidate models and graders, so it runs only for routes
+ * whose capture is EXPLICITLY on under the existing capture posture
+ * (`effectiveCapture`: the control-plane flip wins, else the YAML flag).
+ * Otherwise the bake-off refuses with a teaching 422.
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.TurnReplayRefusedError = void 0;
+exports.assertTurnReplayAllowed = assertTurnReplayAllowed;
+exports.turnReplayCase = turnReplayCase;
+exports.incomparableFrom = incomparableFrom;
+exports.splitIncomparable = splitIncomparable;
+exports.formatIncomparable = formatIncomparable;
+exports.skippedTagOf = skippedTagOf;
+exports.fallbackServed = fallbackServed;
+exports.fallbackArmLabel = fallbackArmLabel;
+exports.formatFallbackServed = formatFallbackServed;
+const loader_1 = require("../routes/loader");
+const task_settings_1 = require("../settings/task-settings");
+const messages_1 = require("../run/messages");
+/** The bake-off refused a messages route because capture is not explicitly on.
+ * A RouteConfigError (refused before any spend), carrying the HTTP status a
+ * server surface should answer with. */
+class TurnReplayRefusedError extends loader_1.RouteConfigError {
+    constructor(route) {
+        super(`turn replay refused for route "${route}": it serves conversations (input: messages), and replaying them ` +
+            "re-sends end-user text to candidate models — allowed only where capture is explicitly on. Turn capture on " +
+            "for this route (`capture: true`, or the console capture flip), let it record real turns, then re-run the " +
+            "bake-off (see messages-lane.md#turn-replay)");
+        this.httpStatus = 422;
+        this.name = "TurnReplayRefusedError";
+    }
+}
+exports.TurnReplayRefusedError = TurnReplayRefusedError;
+/** Throw {@link TurnReplayRefusedError} for a messages route whose capture is
+ * not explicitly on; a no-op for every variables route (byte-identical). */
+function assertTurnReplayAllowed(route, settings) {
+    if (route.inputKind !== "messages")
+        return;
+    if (!(0, task_settings_1.effectiveCapture)(settings, route.route, route.capture))
+        throw new TurnReplayRefusedError(route.route);
+}
+/** The turn a capture row encodes, or null for a variables (non-messages) row. */
+function turnReplayCase(capture) {
+    const parsed = (0, messages_1.parseCaptureInput)(capture.variablesJson);
+    if (parsed.messages === undefined)
+        return null;
+    return {
+        variables: parsed.input,
+        history: parsed.messages,
+        reference: capture.outputText,
+        ...(parsed.tools !== undefined ? { tools: parsed.tools } : {}),
+        ...(parsed.toolChoice !== undefined ? { toolChoice: parsed.toolChoice } : {}),
+    };
+}
+// ── Incomparable cases (architect ruling M3(c), 2026-10-24) ──────────────────
+/** The incomparable marker for a replay failure, or null for an ordinary one. */
+function incomparableFrom(failure) {
+    if (!failure.message.startsWith(messages_1.HISTORY_INCOMPATIBLE_PREFIX))
+        return null;
+    const model = failure.provider !== undefined && failure.model !== undefined ? `${failure.provider}/${failure.model}` : "every candidate of the arm";
+    return { model, reason: failure.message.slice(messages_1.HISTORY_INCOMPATIBLE_PREFIX.length).replace(/^:\s*/, "") };
+}
+/**
+ * Split an arm's samples into the COMPARABLE ones (scored, paired for regret)
+ * and a per-(model, reason) count of the incomparable ones. An incomparable
+ * sample is neither a pass nor a failure: dropping it from both sides of the
+ * pairing keeps it from biasing the verdict.
+ */
+function splitIncomparable(samples) {
+    const counts = new Map();
+    const comparable = [];
+    for (const s of samples) {
+        if (s.incomparable === undefined) {
+            comparable.push(s);
+            continue;
+        }
+        const key = `${s.incomparable.model}\u0000${s.incomparable.reason}`;
+        const prev = counts.get(key);
+        counts.set(key, { model: s.incomparable.model, reason: s.incomparable.reason, n: (prev?.n ?? 0) + 1 });
+    }
+    return { comparable, incomparable: [...counts.values()] };
+}
+/** The report lines: "n cases incomparable for <model>: reason". */
+function formatIncomparable(rows) {
+    return rows.map((r) => `${r.n} case${r.n === 1 ? "" : "s"} incomparable for ${r.model}: ${r.reason}`);
+}
+// ── Fallback attribution (QA R3-1, 2026-10-24) ────────────────────────────────
+// Under ruling R2-2 a multi-candidate arm whose first candidate cannot take a
+// tool-loop history is served by a later one. The bake-off must say so — an
+// arm's numbers are never presented under a model that never saw the turn.
+/** The `history_skipped` value on a replayed call's telemetry rows, if any. */
+function skippedTagOf(rows) {
+    for (const r of rows) {
+        try {
+            const v = JSON.parse(r.tagsJson)[messages_1.HISTORY_SKIPPED_TAG];
+            if (typeof v === "string" && v !== "")
+                return v;
+        }
+        catch {
+            // a malformed tags blob carries no skip
+        }
+    }
+    return undefined;
+}
+/** Count an arm's comparable samples answered by a fallback, grouped by
+ * (served model, skipped set). Empty when nothing was skipped. */
+function fallbackServed(samples) {
+    const counts = new Map();
+    for (const s of samples) {
+        if (s.historySkipped === undefined || s.servedBy === undefined)
+            continue;
+        const key = `${s.servedBy}\u0000${s.historySkipped}`;
+        counts.set(key, { servedBy: s.servedBy, skipped: s.historySkipped, n: (counts.get(key)?.n ?? 0) + 1 });
+    }
+    return [...counts.values()];
+}
+/** "<first> → <fallback…> (fallback served n/N)" — the label an arm's numbers
+ * carry when any of its cases was served by a fallback. */
+function fallbackArmLabel(first, rows, total) {
+    const servedBy = [...new Set(rows.map((r) => r.servedBy))].join(", ");
+    const n = rows.reduce((a, r) => a + r.n, 0);
+    return `${first} → ${servedBy} (fallback served ${n}/${total})`;
+}
+/** The report lines for an arm's fallback-served cases. */
+function formatFallbackServed(rows) {
+    return rows.map((r) => {
+        const skipped = r.skipped
+            .split(",")
+            .map((entry) => {
+            const at = entry.lastIndexOf("=");
+            return at > 0 ? `${entry.slice(0, at)} skipped: ${entry.slice(at + 1)}` : `${entry} skipped`;
+        })
+            .join("; ");
+        return `${r.n} case${r.n === 1 ? "" : "s"} served by ${r.servedBy} (${skipped})`;
+    });
+}

package/dist/bakeoff/types.d.ts

@@ -4,6 +4,7 @@
  * effective $ per 1K conformant outputs (F1-corrected); RB-1 adds per-sample
  * value accuracy, paired regret vs a reference arm, and an objective-aware gate.
  */
+import type { FallbackServed, IncomparableSummary } from "./turn-replay";
 import type { ObjectiveMode } from "../types/route";
 import type { PairwiseStats } from "./pairwise";
 import type { DisagreementReport } from "./disagreement";
@@ -55,6 +56,17 @@ export interface VariantMetrics {
      * winning variant so the mirror/console read the artifact's decision rather
      * than recomputing a (mode-blind) winner client-side. */
     readonly proposed?: boolean;
+    /** ARC-β ruling M3(c): captured turns this arm's model could not accept
+     * natively — excluded from every number above and from regret; reported as
+     * "n cases incomparable for <model>: reason". Absent ⇒ none. */
+    readonly incomparable?: readonly IncomparableSummary[];
+    /** ARC-β R3-1: cases this arm's ladder answered with a FALLBACK because its
+     * earlier candidates were skipped for the history ("n cases served by
+     * <provider/model> (<skipped> skipped: <reason>)"). Absent ⇒ none. */
+    readonly servedByFallback?: readonly FallbackServed[];
+    /** ARC-β R3-1: present with `servedByFallback` — "<first> → <fallback>
+     * (fallback served n/N)", the label these numbers must be shown under. */
+    readonly armLabel?: string;
     /** RB-1: this arm's regret vs the reference; null on the reference arm and
      * when there is no reference. */
     readonly regret?: RegretStats | null;

package/dist/bakeoff/types.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/bakeoff/types.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,gBAAgB,CAAC;AACpD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAChD,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,gBAAgB,CAAC;AACzD,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,eAAe,CAAC;AAElD;sEACsE;AACtE,MAAM,MAAM,aAAa,GAAG,gBAAgB,GAAG,aAAa,CAAC;AAE7D;;8CAE8C;AAC9C,MAAM,WAAW,WAAW;IAC1B,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,UAAU,EAAE,SAAS,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC/C,qEAAqE;IACrE,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;CAC1B;AAED,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,WAAW,EAAE,SAAS,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAChD,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,iBAAiB,EAAE,MAAM,CAAC;IACnC,wFAAwF;IACxF,QAAQ,CAAC,iBAAiB,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1C,QAAQ,CAAC,aAAa,EAAE,MAAM,CAAC;IAC/B,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,+EAA+E;IAC/E,QAAQ,CAAC,2BAA2B,EAAE,MAAM,CAAC;IAC7C;;sFAEkF;IAClF,QAAQ,CAAC,eAAe,EAAE,SAAS,CAAC,MAAM,EAAE,MAAM,CAAC,GAAG,IAAI,CAAC;IAC3D;;sEAEkE;IAClE,QAAQ,CAAC,wBAAwB,EAAE,MAAM,GAAG,IAAI,CAAC;IACjD;6EACyE;IACzE,QAAQ,CAAC,cAAc,EAAE,OAAO,CAAC;IACjC;iFAC6E;IAC7E,QAAQ,CAAC,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IACpC,iEAAiE;IACjE,QAAQ,CAAC,aAAa,EAAE,aAAa,CAAC;IACtC,qDAAqD;IACrD,QAAQ,CAAC,WAAW,CAAC,EAAE,OAAO,CAAC;IAC/B;;6DAEyD;IACzD,QAAQ,CAAC,QAAQ,CAAC,EAAE,OAAO,CAAC;IAC5B;qCACiC;IACjC,QAAQ,CAAC,MAAM,CAAC,EAAE,WAAW,GAAG,IAAI,CAAC;IACrC,iFAAiF;IACjF,QAAQ,CAAC,gBAAgB,CAAC,EAAE,OAAO,CAAC;IACpC;;;oCAGgC;IAChC,QAAQ,CAAC,QAAQ,CAAC,EAAE,aAAa,GAAG,IAAI,CAAC;CAC1C;AAED,iFAAiF;AACjF,MAAM,WAAW,gBAAgB;IAC/B,QAAQ,CAAC,IAAI,EAAE,aAAa,CAAC;IAC7B,QAAQ,CAAC,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IAClC,QAAQ,CAAC,YAAY,EAAE;QAAE,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CAAC;IAC1F,QAAQ,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B;8EAC0E;IAC1E,QAAQ,CAAC,aAAa,CAAC,EAAE,OAAO,CAAC;CAClC;AAED,iEAAiE;AACjE,MAAM,WAAW,gBAAgB;IAC/B,mDAAmD;IACnD,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,wBAAwB;IACxB,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,mEAAmE;IACnE,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC;IACvB,2CAA2C;IAC3C,QAAQ,CAAC,IAAI,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,6DAA6D;IAC7D,QAAQ,CAAC,QAAQ,EAAE,OAAO,CAAC;IAC3B,kEAAkE;IAClE,QAAQ,CAAC,OAAO,EAAE,OAAO,CAAC;CAC3B;AAED,MAAM,WAAW,eAAe;IAC9B,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,QAAQ,EAAE,SAAS,cAAc,EAAE,CAAC;IAC7C;;8EAE0E;IAC1E,QAAQ,CAAC,QAAQ,EACb;QACE,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;QACzB,QAAQ,CAAC,qBAAqB,EAAE,MAAM,CAAC;QACvC,QAAQ,CAAC,MAAM,CAAC,EAAE,WAAW,GAAG,IAAI,CAAC;KACtC,GACD,IAAI,CAAC;IACT,sEAAsE;IACtE,QAAQ,CAAC,SAAS,CAAC,EAAE,gBAAgB,CAAC;IACtC,gFAAgF;IAChF,QAAQ,CAAC,SAAS,CAAC,EAAE,gBAAgB,GAAG,IAAI,CAAC;IAC7C,8DAA8D;IAC9D,QAAQ,CAAC,aAAa,CAAC,EAAE,aAAa,CAAC;IACvC;;;;oBAIgB;IAChB,QAAQ,CAAC,YAAY,CAAC,EAAE,kBAAkB,CAAC;IAC3C;;;;;qFAKiF;IACjF,QAAQ,CAAC,OAAO,CAAC,EAAE,YAAY,CAAC;CACjC"}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/bakeoff/types.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,KAAK,EAAE,cAAc,EAAE,mBAAmB,EAAE,MAAM,eAAe,CAAC;AACzE,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,gBAAgB,CAAC;AACpD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAChD,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,gBAAgB,CAAC;AACzD,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,eAAe,CAAC;AAElD;sEACsE;AACtE,MAAM,MAAM,aAAa,GAAG,gBAAgB,GAAG,aAAa,CAAC;AAE7D;;8CAE8C;AAC9C,MAAM,WAAW,WAAW;IAC1B,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,UAAU,EAAE,SAAS,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC/C,qEAAqE;IACrE,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;CAC1B;AAED,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,WAAW,EAAE,SAAS,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAChD,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,iBAAiB,EAAE,MAAM,CAAC;IACnC,wFAAwF;IACxF,QAAQ,CAAC,iBAAiB,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1C,QAAQ,CAAC,aAAa,EAAE,MAAM,CAAC;IAC/B,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,+EAA+E;IAC/E,QAAQ,CAAC,2BAA2B,EAAE,MAAM,CAAC;IAC7C;;sFAEkF;IAClF,QAAQ,CAAC,eAAe,EAAE,SAAS,CAAC,MAAM,EAAE,MAAM,CAAC,GAAG,IAAI,CAAC;IAC3D;;sEAEkE;IAClE,QAAQ,CAAC,wBAAwB,EAAE,MAAM,GAAG,IAAI,CAAC;IACjD;6EACyE;IACzE,QAAQ,CAAC,cAAc,EAAE,OAAO,CAAC;IACjC;iFAC6E;IAC7E,QAAQ,CAAC,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IACpC,iEAAiE;IACjE,QAAQ,CAAC,aAAa,EAAE,aAAa,CAAC;IACtC,qDAAqD;IACrD,QAAQ,CAAC,WAAW,CAAC,EAAE,OAAO,CAAC;IAC/B;;6DAEyD;IACzD,QAAQ,CAAC,QAAQ,CAAC,EAAE,OAAO,CAAC;IAC5B;;oEAEgE;IAChE,QAAQ,CAAC,YAAY,CAAC,EAAE,SAAS,mBAAmB,EAAE,CAAC;IACvD;;0EAEsE;IACtE,QAAQ,CAAC,gBAAgB,CAAC,EAAE,SAAS,cAAc,EAAE,CAAC;IACtD;8EAC0E;IAC1E,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAC3B;qCACiC;IACjC,QAAQ,CAAC,MAAM,CAAC,EAAE,WAAW,GAAG,IAAI,CAAC;IACrC,iFAAiF;IACjF,QAAQ,CAAC,gBAAgB,CAAC,EAAE,OAAO,CAAC;IACpC;;;oCAGgC;IAChC,QAAQ,CAAC,QAAQ,CAAC,EAAE,aAAa,GAAG,IAAI,CAAC;CAC1C;AAED,iFAAiF;AACjF,MAAM,WAAW,gBAAgB;IAC/B,QAAQ,CAAC,IAAI,EAAE,aAAa,CAAC;IAC7B,QAAQ,CAAC,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IAClC,QAAQ,CAAC,YAAY,EAAE;QAAE,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CAAC;IAC1F,QAAQ,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B;8EAC0E;IAC1E,QAAQ,CAAC,aAAa,CAAC,EAAE,OAAO,CAAC;CAClC;AAED,iEAAiE;AACjE,MAAM,WAAW,gBAAgB;IAC/B,mDAAmD;IACnD,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,wBAAwB;IACxB,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,mEAAmE;IACnE,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC;IACvB,2CAA2C;IAC3C,QAAQ,CAAC,IAAI,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,6DAA6D;IAC7D,QAAQ,CAAC,QAAQ,EAAE,OAAO,CAAC;IAC3B,kEAAkE;IAClE,QAAQ,CAAC,OAAO,EAAE,OAAO,CAAC;CAC3B;AAED,MAAM,WAAW,eAAe;IAC9B,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,QAAQ,EAAE,SAAS,cAAc,EAAE,CAAC;IAC7C;;8EAE0E;IAC1E,QAAQ,CAAC,QAAQ,EACb;QACE,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;QACzB,QAAQ,CAAC,qBAAqB,EAAE,MAAM,CAAC;QACvC,QAAQ,CAAC,MAAM,CAAC,EAAE,WAAW,GAAG,IAAI,CAAC;KACtC,GACD,IAAI,CAAC;IACT,sEAAsE;IACtE,QAAQ,CAAC,SAAS,CAAC,EAAE,gBAAgB,CAAC;IACtC,gFAAgF;IAChF,QAAQ,CAAC,SAAS,CAAC,EAAE,gBAAgB,GAAG,IAAI,CAAC;IAC7C,8DAA8D;IAC9D,QAAQ,CAAC,aAAa,CAAC,EAAE,aAAa,CAAC;IACvC;;;;oBAIgB;IAChB,QAAQ,CAAC,YAAY,CAAC,EAAE,kBAAkB,CAAC;IAC3C;;;;;qFAKiF;IACjF,QAAQ,CAAC,OAAO,CAAC,EAAE,YAAY,CAAC;CACjC"}

package/dist/feedback/grade.d.ts

@@ -11,7 +11,7 @@
  * (`rig.feedback` / `artifact.evaluate` / `run.evaluate`) and their adapters were
  * removed in GP-3 — every judgment now builds a grade directly.
  */
-import type { GradeCriterion, GradeRow, GradeSubject } from "../types/telemetry";
+import type { GradeCriterion, GradeKind, GradeRow, GradeSubject } from "../types/telemetry";
 /** Options for `rig.grade(subject, opts)`. `score` is the 0–1 primitive and is
  * required; everything else is optional display / provenance. */
 export interface GradeOpts {
@@ -26,7 +26,9 @@ export interface GradeOpts {
     };
     /** Written guidance saved for a later improvement path (was note/explanation). */
     readonly comment?: string;
-    readonly kind: "human" | "model-judge" | "deterministic";
+    /** `outcome` (ARC-3) is normally written through `rig.outcome`, which derives
+     * the subject, `grader = outcome:<name>` and the score for you. */
+    readonly kind: GradeKind;
     /** Optional specific grader name (e.g. "gpt-5-judge", "schema-check"). */
     readonly grader?: string;
     /** Provenance (e.g. "sdk", "console", a review-agent name). */
@@ -38,6 +40,9 @@ export interface GradeOpts {
      * the scrub + dedup/drop-empty-id/clamp guard at persist; this is carried
      * verbatim. */
     readonly criteria?: readonly GradeCriterion[];
+    /** ARC-3: the raw business value — allowed ONLY on `kind: "outcome"` (a value
+     * on any other kind throws: it would be refused server-side). */
+    readonly value?: number;
     /** The grade's own org — server-supplied ONLY, already authorized (T3). Omit
      * from application code (a rig has one org and the exporter knows it). */
     readonly orgId?: string;

package/dist/feedback/grade.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"grade.d.ts","sourceRoot":"","sources":["../../src/feedback/grade.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAGH,OAAO,KAAK,EAAE,cAAc,EAAE,QAAQ,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAEjF;iEACiE;AACjE,MAAM,WAAW,SAAS;IACxB,mFAAmF;IACnF,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB;6EACyE;IACzE,QAAQ,CAAC,KAAK,CAAC,EAAE;QAAE,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAA;KAAE,GAAG;QAAE,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;IACvE,kFAAkF;IAClF,QAAQ,CAAC,OAAO,CAAC,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,IAAI,EAAE,OAAO,GAAG,aAAa,GAAG,eAAe,CAAC;IACzD,0EAA0E;IAC1E,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,CAAC;IACzB,+DAA+D;IAC/D,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,CAAC;IACzB;;;;;mBAKe;IACf,QAAQ,CAAC,QAAQ,CAAC,EAAE,SAAS,cAAc,EAAE,CAAC;IAC9C;8EAC0E;IAC1E,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB,4EAA4E;IAC5E,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CACpC;AAQD;oFACoF;AACpF,wBAAgB,aAAa,CAAC,OAAO,EAAE,YAAY,EAAE,IAAI,EAAE,SAAS,GAAG,QAAQ,CA2B9E;AAED;;+DAE+D;AAC/D,wBAAgB,YAAY,CAAC,KAAK,EAAE;IAAE,WAAW,CAAC,GAAG,EAAE,QAAQ,GAAG,IAAI,CAAA;CAAE,GAAG;IACzE,KAAK,CAAC,OAAO,EAAE,YAAY,EAAE,IAAI,EAAE,SAAS,GAAG,QAAQ,CAAC;CACzD,CAQA"}
+{"version":3,"file":"grade.d.ts","sourceRoot":"","sources":["../../src/feedback/grade.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAGH,OAAO,KAAK,EAAE,cAAc,EAAE,SAAS,EAAE,QAAQ,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAE5F;iEACiE;AACjE,MAAM,WAAW,SAAS;IACxB,mFAAmF;IACnF,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB;6EACyE;IACzE,QAAQ,CAAC,KAAK,CAAC,EAAE;QAAE,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAA;KAAE,GAAG;QAAE,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;IACvE,kFAAkF;IAClF,QAAQ,CAAC,OAAO,CAAC,EAAE,MAAM,CAAC;IAC1B;uEACmE;IACnE,QAAQ,CAAC,IAAI,EAAE,SAAS,CAAC;IACzB,0EAA0E;IAC1E,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,CAAC;IACzB,+DAA+D;IAC/D,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,CAAC;IACzB;;;;;mBAKe;IACf,QAAQ,CAAC,QAAQ,CAAC,EAAE,SAAS,cAAc,EAAE,CAAC;IAC9C;qEACiE;IACjE,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB;8EAC0E;IAC1E,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB,4EAA4E;IAC5E,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CACpC;AAQD;oFACoF;AACpF,wBAAgB,aAAa,CAAC,OAAO,EAAE,YAAY,EAAE,IAAI,EAAE,SAAS,GAAG,QAAQ,CAoC9E;AAED;;+DAE+D;AAC/D,wBAAgB,YAAY,CAAC,KAAK,EAAE;IAAE,WAAW,CAAC,GAAG,EAAE,QAAQ,GAAG,IAAI,CAAA;CAAE,GAAG;IACzE,KAAK,CAAC,OAAO,EAAE,YAAY,EAAE,IAAI,EAAE,SAAS,GAAG,QAAQ,CAAC;CACzD,CAiBA"}

package/dist/feedback/grade.js

@@ -32,6 +32,11 @@ function buildGradeRow(subject, opts) {
     if (!Number.isFinite(opts.score) || opts.score < 0 || opts.score > 1) {
         throw new Error(`grade score must be in [0, 1], got ${opts.score}`);
     }
+    if (opts.value !== undefined && (opts.kind !== "outcome" || !Number.isFinite(opts.value))) {
+        throw new Error(opts.kind !== "outcome"
+            ? `grade value is only allowed on kind "outcome" (got kind "${opts.kind}") — put the number in score (0–1)`
+            : `outcome value must be a finite number, got ${String(opts.value)}`);
+    }
     const { scaleMax, scaleLabel } = scaleFields(opts.scale);
     return {
         id: (0, node_crypto_1.randomUUID)(),
@@ -51,6 +56,8 @@ function buildGradeRow(subject, opts) {
         // a single-score grade's row is byte-identical to pre-F4). Structural
         // normalization (dedup/clamp/scrub) is the server's job at persist.
         ...(opts.criteria !== undefined ? { criteria: opts.criteria } : {}),
+        // ARC-3: spread-iff so a non-outcome row is byte-identical to pre-ARC.
+        ...(opts.value !== undefined ? { value: opts.value } : {}),
     };
 }
 /** Build the rig's `grade` surface over a grade writer (the telemetry store).
@@ -59,6 +66,13 @@ function buildGradeRow(subject, opts) {
 function makeGradeApi(store) {
     return {
         grade(subject, opts) {
+            // L4: an outcome is written through rig.outcome (it derives the subject,
+            // the outcome:<name> grader, the score and the idempotent id) — a bare
+            // rig.grade({ kind: "outcome" }) would be dropped server-side, silently.
+            if (opts.kind === "outcome") {
+                throw new Error('rig.grade does not record outcomes — use rig.outcome({ key, name, score }) ' +
+                    "(it files the outcome under outcome:<name> with an idempotent id)");
+            }
             const row = buildGradeRow(subject, opts);
             store.recordGrade(row);
             return row;

package/dist/feedback/outcome.d.ts

@@ -0,0 +1,103 @@
+/**
+ * ARC-3 — outcomes: the delayed business ground truth, written as a GRADE.
+ *
+ * An outcome (resolved, reopened, booked, approved, CTR …) arrives hours to
+ * weeks after the call that earned it. It is stored as ONE `grades` row with
+ * `kind: "outcome"`, `grader: "outcome:<name>"`, `score` 0–1 (the primitive the
+ * verdict is derived from) and an optional raw `value` (00089). No new table,
+ * no pending store: a conversation-keyed outcome is stored
+ * `subject_kind = "run"`, `subject_id = <episode_key>` (the 00070 precedent) and
+ * the console joins it lazily at read time — so an outcome may arrive before,
+ * during or after its telemetry and nothing is ever dropped.
+ *
+ * This module is the ONE definition of the key mapping, the name rule, the
+ * score derivation and the idempotent id — shared by `rig.outcome` (SDK),
+ * `POST /v1/outcomes` (server) and the oracle's `record_outcome` (MCP), so the
+ * three surfaces cannot disagree about what an outcome is.
+ *
+ * An outcome is a RECEIPT, never a calibration gate (ARC-D3): outcome agreement
+ * never moves a grader to CALIBRATED on its own.
+ */
+import type { GradeRow, GradeSubject } from "../types/telemetry";
+/** What the outcome is ABOUT. A `conversation` key is the caller's own
+ * conversation id (= the run's `episode_key`); the others are ModelRig ids. */
+export type OutcomeKey = {
+    readonly run: string;
+} | {
+    readonly conversation: string;
+} | {
+    readonly inference: string;
+} | {
+    readonly artifact: string;
+};
+export interface OutcomeInput {
+    readonly key: OutcomeKey;
+    /** `^[a-z][a-z0-9_.-]{0,63}$` — e.g. `resolved`, `meeting_booked`, `ctr.7d`. */
+    readonly name: string;
+    /** 0–1, or a boolean (true ⇒ 1, false ⇒ 0). Required unless `value` and
+     * `scale.max` are both given, in which case score = clamp(value / max, 0, 1). */
+    readonly score?: number | boolean;
+    /** The raw business number (revenue, CTR, handle time). Stored as-is. */
+    readonly value?: number;
+    /** Display scale; `max` also drives the score derivation from `value`. */
+    readonly scale?: {
+        readonly max: number;
+    } | {
+        readonly label: string;
+    };
+    /** When the outcome happened (ISO-8601). Default: now. */
+    readonly at?: string;
+    /** Retry-safe id: the same (org, name, idempotencyKey) always yields the same
+     * grade id — UUIDv5(org, `${name}:${idempotencyKey}`), identical on the SDK and
+     * the API paths — so a retry never duplicates, and two DIFFERENT outcome names
+     * sent under one key (one webhook event) both land. One key = one outcome per name. */
+    readonly idempotencyKey?: string;
+    /** Optional free-text note (scrubbed server-side like every grade comment). */
+    readonly comment?: string;
+}
+/** The name rule — lowercase identifier, ≤64 chars. */
+export declare const OUTCOME_NAME_PATTERN: RegExp;
+export declare const OUTCOME_GRADER_PREFIX = "outcome:";
+/** A teaching validation failure (the server maps it to a 422). */
+export declare class OutcomeValidationError extends Error {
+    readonly fixHint: string;
+    constructor(message: string, fixHint: string);
+}
+/** The grade subject an outcome key maps to. `conversation` → a RUN subject whose
+ * id is the episode key (00070 precedent; resolved lazily at read time). */
+export declare function outcomeSubject(key: OutcomeKey): GradeSubject;
+/** Validate the outcome name; returns the grader name `outcome:<name>`. */
+export declare function outcomeGraderName(name: string): string;
+/** Derive the 0–1 score: an explicit score (or boolean) wins; else value/scale.max
+ * clamped to [0, 1]; else a teaching error. */
+export declare function deriveOutcomeScore(input: Pick<OutcomeInput, "score" | "value" | "scale">): number;
+/** RFC 4122 UUIDv5 (SHA-1, name-based) — deterministic, so an idempotency key
+ * maps to the same grade id on every retry. `namespace` must be a UUID. */
+export declare function uuidV5(name: string, namespace: string): string;
+/**
+ * THE id function for an idempotent outcome (fix round 1, L3) — shared by
+ * `rig.outcome` (SDK), `POST /v1/outcomes` and `/v1/ingest`:
+ * UUIDv5(namespace = org id, name = `${name}:${idempotencyKey}`). The SDK does
+ * not know its org locally, so it mints a PROVISIONAL id (null org → a fixed
+ * namespace) and ships the idempotency key on the wire; the server re-derives
+ * the id with the key's org at ingest, so both doors land on the same row.
+ */
+export declare function outcomeGradeId(orgId: string | null, name: string, idempotencyKey: string): string;
+/** Build the outcome's GradeRow. `orgId` keys the idempotent id (and names the
+ * grade's org on a server write); `source` is provenance ("sdk", "outcomes-api", …). */
+export declare function buildOutcomeGradeRow(input: OutcomeInput, ctx: {
+    readonly orgId?: string | null;
+    readonly source: string;
+}): GradeRow;
+/** `rig.outcome` — record one outcome locally; the existing grade exporter ships
+ * it (no new exporter). Returns the subject it was filed under. Throws an
+ * {@link OutcomeValidationError} on bad input (it is caller code, not telemetry). */
+export declare function makeOutcomeApi(store: {
+    recordGrade(row: GradeRow): void;
+}): {
+    outcome(input: OutcomeInput): {
+        readonly id: string;
+        readonly subject: GradeSubject;
+    };
+};
+//# sourceMappingURL=outcome.d.ts.map

package/dist/feedback/outcome.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"outcome.d.ts","sourceRoot":"","sources":["../../src/feedback/outcome.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAIH,OAAO,KAAK,EAAE,QAAQ,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAEjE;+EAC+E;AAC/E,MAAM,MAAM,UAAU,GAClB;IAAE,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAA;CAAE,GACxB;IAAE,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAA;CAAE,GACjC;IAAE,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAA;CAAE,GAC9B;IAAE,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAA;CAAE,CAAC;AAElC,MAAM,WAAW,YAAY;IAC3B,QAAQ,CAAC,GAAG,EAAE,UAAU,CAAC;IACzB,gFAAgF;IAChF,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB;qFACiF;IACjF,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC;IAClC,yEAAyE;IACzE,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB,0EAA0E;IAC1E,QAAQ,CAAC,KAAK,CAAC,EAAE;QAAE,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAA;KAAE,GAAG;QAAE,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;IACvE,0DAA0D;IAC1D,QAAQ,CAAC,EAAE,CAAC,EAAE,MAAM,CAAC;IACrB;;;2FAGuF;IACvF,QAAQ,CAAC,cAAc,CAAC,EAAE,MAAM,CAAC;IACjC,+EAA+E;IAC/E,QAAQ,CAAC,OAAO,CAAC,EAAE,MAAM,CAAC;CAC3B;AAED,uDAAuD;AACvD,eAAO,MAAM,oBAAoB,QAA6B,CAAC;AAC/D,eAAO,MAAM,qBAAqB,aAAa,CAAC;AAIhD,mEAAmE;AACnE,qBAAa,sBAAuB,SAAQ,KAAK;IAC/C,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;gBACb,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM;CAK7C;AAED;4EAC4E;AAC5E,wBAAgB,cAAc,CAAC,GAAG,EAAE,UAAU,GAAG,YAAY,CA6B5D;AAED,2EAA2E;AAC3E,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAQtD;AAED;+CAC+C;AAC/C,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,IAAI,CAAC,YAAY,EAAE,OAAO,GAAG,OAAO,GAAG,OAAO,CAAC,GAAG,MAAM,CAyBjG;AAED;2EAC2E;AAC3E,wBAAgB,MAAM,CAAC,IAAI,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,GAAG,MAAM,CAS9D;AAMD;;;;;;;GAOG;AACH,wBAAgB,cAAc,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI,EAAE,IAAI,EAAE,MAAM,EAAE,cAAc,EAAE,MAAM,GAAG,MAAM,CAGjG;AAcD;wFACwF;AACxF,wBAAgB,oBAAoB,CAClC,KAAK,EAAE,YAAY,EACnB,GAAG,EAAE;IAAE,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAAC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAA;CAAE,GAC/D,QAAQ,CA8BV;AAED;;qFAEqF;AACrF,wBAAgB,cAAc,CAAC,KAAK,EAAE;IAAE,WAAW,CAAC,GAAG,EAAE,QAAQ,GAAG,IAAI,CAAA;CAAE,GAAG;IAC3E,OAAO,CAAC,KAAK,EAAE,YAAY,GAAG;QAAE,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,OAAO,EAAE,YAAY,CAAA;KAAE,CAAC;CACvF,CAQA"}