@bastani/atomic 0.9.40 → 0.9.41-alpha.2

package/dist/core/structured-output/deadline.js

@@ -0,0 +1,110 @@
+import { SETTINGS_DEFAULTS } from "../settings-defaults.js";
+/** The largest delay `setTimeout` honors; a larger one fires immediately. */
+const MAX_DEADLINE_MS = 2_147_483_647;
+const CANCELLED_MESSAGE = "Structured output cancelled; no decision was accepted.";
+export class StructuredOutputDeadlineError extends Error {
+    constructor(timeoutMs, attempts) {
+        super(`Structured output decision exceeded its ${timeoutMs}ms deadline; no decision was accepted.${attempts ? ` ${attempts}` : ""}`);
+        this.name = "StructuredOutputDeadlineError";
+    }
+}
+/** The failure for an aborted decision: the deadline error when the deadline fired, otherwise a plain cancellation. */
+export function abortFailure(signal) {
+    return signal.reason instanceof StructuredOutputDeadlineError ? signal.reason : new Error(CANCELLED_MESSAGE);
+}
+export function validateTimeoutOverride(value) {
+    if (value === undefined)
+        return;
+    if (!Number.isSafeInteger(value) || value <= 0) {
+        throw new Error("Structured output timeoutMs must be a positive integer.");
+    }
+}
+/**
+ * The wall-clock limit for one decision: an explicit `timeoutMs`, else the session's `httpIdleTimeoutMs`, else that
+ * setting's default. Like the provider clients, an idle timeout of `0` (disabled) becomes the largest timer delay.
+ */
+export function resolveDecisionDeadlineMs(timeoutMs, httpIdleTimeoutMs) {
+    validateTimeoutOverride(timeoutMs);
+    const requested = timeoutMs ?? httpIdleTimeoutMs ?? SETTINGS_DEFAULTS.httpIdleTimeoutMs;
+    return requested === 0 ? MAX_DEADLINE_MS : Math.min(requested, MAX_DEADLINE_MS);
+}
+function formatDuration(milliseconds) {
+    const ms = Math.round(milliseconds);
+    if (ms < 1_000)
+        return `${ms}ms`;
+    if (ms < 60_000)
+        return `${(ms / 1_000).toFixed(1)}s`;
+    const totalSeconds = Math.round(ms / 1_000);
+    const hours = Math.floor(totalSeconds / 3_600);
+    const minutes = Math.floor((totalSeconds % 3_600) / 60);
+    const pad = (value) => String(value).padStart(2, "0");
+    const clock = hours > 0 ? `${hours}h${pad(minutes)}m` : `${minutes}m`;
+    return `${clock}${pad(totalSeconds % 60)}s`;
+}
+/** One model request, from send to response; transport retries and their waits are not part of it. Records only the stop reason and timing, never content. */
+export class DecisionAttempt {
+    startedAt = performance.now();
+    endedAt;
+    stopReason;
+    settle(stopReason) {
+        this.endedAt ??= performance.now();
+        if (stopReason !== undefined)
+            this.stopReason = stopReason;
+    }
+    describe(now) {
+        const outcome = this.endedAt === undefined
+            ? "still generating"
+            : this.stopReason === undefined
+                ? "no response"
+                : `stopReason=${this.stopReason}`;
+        return `${outcome} ${formatDuration((this.endedAt ?? now) - this.startedAt)}`;
+    }
+}
+/** How every generation of one decision ended and how long it took, across repairs, retries and candidates. */
+export class DecisionTrace {
+    attempts = [];
+    get size() {
+        return this.attempts.length;
+    }
+    begin() {
+        const attempt = new DecisionAttempt();
+        this.attempts.push(attempt);
+        return attempt;
+    }
+    /** Mark every generation still in flight as ended without a response. */
+    settleOpen() {
+        for (const attempt of this.attempts)
+            attempt.settle();
+    }
+    /** The attempts from `first` onward, numbered from 1, or an empty string when there are none. */
+    describe(first = 0) {
+        const now = performance.now();
+        const shown = this.attempts.slice(first);
+        if (shown.length === 0)
+            return "";
+        return `Attempts: ${shown.map((attempt, index) => `${index + 1}) ${attempt.describe(now)}`).join("; ")}.`;
+    }
+}
+/**
+ * Run a whole decision under one wall-clock limit. `run` receives a signal that aborts with the caller's reason when
+ * the caller aborts and with a {@link StructuredOutputDeadlineError} when the limit passes, so the existing abort
+ * plumbing cancels the in-flight generation. The timer is always released.
+ */
+export async function withinDecisionDeadline(callerSignal, timeoutMs, run) {
+    const controller = new AbortController();
+    const trace = new DecisionTrace();
+    const forwardCallerAbort = () => controller.abort(callerSignal?.reason);
+    if (callerSignal?.aborted)
+        forwardCallerAbort();
+    else
+        callerSignal?.addEventListener("abort", forwardCallerAbort, { once: true });
+    const timer = setTimeout(() => controller.abort(new StructuredOutputDeadlineError(timeoutMs, trace.describe())), timeoutMs);
+    try {
+        return await run({ signal: controller.signal, trace });
+    }
+    finally {
+        clearTimeout(timer);
+        callerSignal?.removeEventListener("abort", forwardCallerAbort);
+    }
+}
+//# sourceMappingURL=deadline.js.map

package/dist/core/structured-output/deadline.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"deadline.js","sourceRoot":"","sources":["../../../src/core/structured-output/deadline.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,iBAAiB,EAAE,MAAM,yBAAyB,CAAC;AAE5D,6EAA6E;AAC7E,MAAM,eAAe,GAAG,aAAa,CAAC;AAEtC,MAAM,iBAAiB,GAAG,wDAAwD,CAAC;AAEnF,MAAM,OAAO,6BAA8B,SAAQ,KAAK;IACvD,YAAY,SAAiB,EAAE,QAAgB;QAC9C,KAAK,CACJ,2CAA2C,SAAS,yCAAyC,QAAQ,CAAC,CAAC,CAAC,IAAI,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,CAC7H,CAAC;QACF,IAAI,CAAC,IAAI,GAAG,+BAA+B,CAAC;IAC7C,CAAC;CACD;AAED,uHAAuH;AACvH,MAAM,UAAU,YAAY,CAAC,MAAmB;IAC/C,OAAO,MAAM,CAAC,MAAM,YAAY,6BAA6B,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,iBAAiB,CAAC,CAAC;AAC9G,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,KAAyB;IAChE,IAAI,KAAK,KAAK,SAAS;QAAE,OAAO;IAChC,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,KAAK,IAAI,CAAC,EAAE,CAAC;QAChD,MAAM,IAAI,KAAK,CAAC,yDAAyD,CAAC,CAAC;IAC5E,CAAC;AACF,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,yBAAyB,CAAC,SAA6B,EAAE,iBAA0B;IAClG,uBAAuB,CAAC,SAAS,CAAC,CAAC;IACnC,MAAM,SAAS,GAAG,SAAS,IAAI,iBAAiB,IAAI,iBAAiB,CAAC,iBAAiB,CAAC;IACxF,OAAO,SAAS,KAAK,CAAC,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,SAAS,EAAE,eAAe,CAAC,CAAC;AACjF,CAAC;AAED,SAAS,cAAc,CAAC,YAAoB;IAC3C,MAAM,EAAE,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC;IACpC,IAAI,EAAE,GAAG,KAAK;QAAE,OAAO,GAAG,EAAE,IAAI,CAAC;IACjC,IAAI,EAAE,GAAG,MAAM;QAAE,OAAO,GAAG,CAAC,EAAE,GAAG,KAAK,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;IACtD,MAAM,YAAY,GAAG,IAAI,CAAC,KAAK,CAAC,EAAE,GAAG,KAAK,CAAC,CAAC;IAC5C,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,GAAG,KAAK,CAAC,CAAC;IAC/C,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,YAAY,GAAG,KAAK,CAAC,GAAG,EAAE,CAAC,CAAC;IACxD,MAAM,GAAG,GAAG,CAAC,KAAa,EAAE,EAAE,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;IAC9D,MAAM,KAAK,GAAG,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,KAAK,IAAI,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,OAAO,GAAG,CAAC;IACtE,OAAO,GAAG,KAAK,GAAG,GAAG,CAAC,YAAY,GAAG,EAAE,CAAC,GAAG,CAAC;AAC7C,CAAC;AAED,8JAA8J;AAC9J,MAAM,OAAO,eAAe;IAClB,SAAS,GAAG,WAAW,CAAC,GAAG,EAAE,CAAC;IACvC,OAAO,CAAqB;IAC5B,UAAU,CAAqB;IAE/B,MAAM,CAAC,UAAmB;QACzB,IAAI,CAAC,OAAO,KAAK,WAAW,CAAC,GAAG,EAAE,CAAC;QACnC,IAAI,UAAU,KAAK,SAAS;YAAE,IAAI,CAAC,UAAU,GAAG,UAAU,CAAC;IAC5D,CAAC;IAED,QAAQ,CAAC,GAAW;QACnB,MAAM,OAAO,GACZ,IAAI,CAAC,OAAO,KAAK,SAAS;YACzB,CAAC,CAAC,kBAAkB;YACpB,CAAC,CAAC,IAAI,CAAC,UAAU,KAAK,SAAS;gBAC9B,CAAC,CAAC,aAAa;gBACf,CAAC,CAAC,cAAc,IAAI,CAAC,UAAU,EAAE,CAAC;QACrC,OAAO,GAAG,OAAO,IAAI,cAAc,CAAC,CAAC,IAAI,CAAC,OAAO,IAAI,GAAG,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC;IAC/E,CAAC;CACD;AAED,+GAA+G;AAC/G,MAAM,OAAO,aAAa;IACR,QAAQ,GAAsB,EAAE,CAAC;IAElD,IAAI,IAAI;QACP,OAAO,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC;IAC7B,CAAC;IAED,KAAK;QACJ,MAAM,OAAO,GAAG,IAAI,eAAe,EAAE,CAAC;QACtC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAC5B,OAAO,OAAO,CAAC;IAChB,CAAC;IAED,yEAAyE;IACzE,UAAU;QACT,KAAK,MAAM,OAAO,IAAI,IAAI,CAAC,QAAQ;YAAE,OAAO,CAAC,MAAM,EAAE,CAAC;IACvD,CAAC;IAED,iGAAiG;IACjG,QAAQ,CAAC,KAAK,GAAG,CAAC;QACjB,MAAM,GAAG,GAAG,WAAW,CAAC,GAAG,EAAE,CAAC;QAC9B,MAAM,KAAK,GAAG,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;QACzC,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,EAAE,CAAC;QAClC,OAAO,aAAa,KAAK,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,KAAK,GAAG,CAAC,KAAK,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC;IAC3G,CAAC;CACD;AAOD;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,sBAAsB,CAC3C,YAAqC,EACrC,SAAiB,EACjB,GAA0C;IAE1C,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;IACzC,MAAM,KAAK,GAAG,IAAI,aAAa,EAAE,CAAC;IAClC,MAAM,kBAAkB,GAAG,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,CAAC,YAAY,EAAE,MAAM,CAAC,CAAC;IACxE,IAAI,YAAY,EAAE,OAAO;QAAE,kBAAkB,EAAE,CAAC;;QAC3C,YAAY,EAAE,gBAAgB,CAAC,OAAO,EAAE,kBAAkB,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;IACjF,MAAM,KAAK,GAAG,UAAU,CACvB,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,CAAC,IAAI,6BAA6B,CAAC,SAAS,EAAE,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC,EACtF,SAAS,CACT,CAAC;IACF,IAAI,CAAC;QACJ,OAAO,MAAM,GAAG,CAAC,EAAE,MAAM,EAAE,UAAU,CAAC,MAAM,EAAE,KAAK,EAAE,CAAC,CAAC;IACxD,CAAC;YAAS,CAAC;QACV,YAAY,CAAC,KAAK,CAAC,CAAC;QACpB,YAAY,EAAE,mBAAmB,CAAC,OAAO,EAAE,kBAAkB,CAAC,CAAC;IAChE,CAAC;AACF,CAAC","sourcesContent":["import { SETTINGS_DEFAULTS } from \"../settings-defaults.ts\";\n\n/** The largest delay `setTimeout` honors; a larger one fires immediately. */\nconst MAX_DEADLINE_MS = 2_147_483_647;\n\nconst CANCELLED_MESSAGE = \"Structured output cancelled; no decision was accepted.\";\n\nexport class StructuredOutputDeadlineError extends Error {\n\tconstructor(timeoutMs: number, attempts: string) {\n\t\tsuper(\n\t\t\t`Structured output decision exceeded its ${timeoutMs}ms deadline; no decision was accepted.${attempts ? ` ${attempts}` : \"\"}`,\n\t\t);\n\t\tthis.name = \"StructuredOutputDeadlineError\";\n\t}\n}\n\n/** The failure for an aborted decision: the deadline error when the deadline fired, otherwise a plain cancellation. */\nexport function abortFailure(signal: AbortSignal): Error {\n\treturn signal.reason instanceof StructuredOutputDeadlineError ? signal.reason : new Error(CANCELLED_MESSAGE);\n}\n\nexport function validateTimeoutOverride(value: number | undefined): void {\n\tif (value === undefined) return;\n\tif (!Number.isSafeInteger(value) || value <= 0) {\n\t\tthrow new Error(\"Structured output timeoutMs must be a positive integer.\");\n\t}\n}\n\n/**\n * The wall-clock limit for one decision: an explicit `timeoutMs`, else the session's `httpIdleTimeoutMs`, else that\n * setting's default. Like the provider clients, an idle timeout of `0` (disabled) becomes the largest timer delay.\n */\nexport function resolveDecisionDeadlineMs(timeoutMs: number | undefined, httpIdleTimeoutMs?: number): number {\n\tvalidateTimeoutOverride(timeoutMs);\n\tconst requested = timeoutMs ?? httpIdleTimeoutMs ?? SETTINGS_DEFAULTS.httpIdleTimeoutMs;\n\treturn requested === 0 ? MAX_DEADLINE_MS : Math.min(requested, MAX_DEADLINE_MS);\n}\n\nfunction formatDuration(milliseconds: number): string {\n\tconst ms = Math.round(milliseconds);\n\tif (ms < 1_000) return `${ms}ms`;\n\tif (ms < 60_000) return `${(ms / 1_000).toFixed(1)}s`;\n\tconst totalSeconds = Math.round(ms / 1_000);\n\tconst hours = Math.floor(totalSeconds / 3_600);\n\tconst minutes = Math.floor((totalSeconds % 3_600) / 60);\n\tconst pad = (value: number) => String(value).padStart(2, \"0\");\n\tconst clock = hours > 0 ? `${hours}h${pad(minutes)}m` : `${minutes}m`;\n\treturn `${clock}${pad(totalSeconds % 60)}s`;\n}\n\n/** One model request, from send to response; transport retries and their waits are not part of it. Records only the stop reason and timing, never content. */\nexport class DecisionAttempt {\n\treadonly startedAt = performance.now();\n\tendedAt: number | undefined;\n\tstopReason: string | undefined;\n\n\tsettle(stopReason?: string): void {\n\t\tthis.endedAt ??= performance.now();\n\t\tif (stopReason !== undefined) this.stopReason = stopReason;\n\t}\n\n\tdescribe(now: number): string {\n\t\tconst outcome =\n\t\t\tthis.endedAt === undefined\n\t\t\t\t? \"still generating\"\n\t\t\t\t: this.stopReason === undefined\n\t\t\t\t\t? \"no response\"\n\t\t\t\t\t: `stopReason=${this.stopReason}`;\n\t\treturn `${outcome} ${formatDuration((this.endedAt ?? now) - this.startedAt)}`;\n\t}\n}\n\n/** How every generation of one decision ended and how long it took, across repairs, retries and candidates. */\nexport class DecisionTrace {\n\tprivate readonly attempts: DecisionAttempt[] = [];\n\n\tget size(): number {\n\t\treturn this.attempts.length;\n\t}\n\n\tbegin(): DecisionAttempt {\n\t\tconst attempt = new DecisionAttempt();\n\t\tthis.attempts.push(attempt);\n\t\treturn attempt;\n\t}\n\n\t/** Mark every generation still in flight as ended without a response. */\n\tsettleOpen(): void {\n\t\tfor (const attempt of this.attempts) attempt.settle();\n\t}\n\n\t/** The attempts from `first` onward, numbered from 1, or an empty string when there are none. */\n\tdescribe(first = 0): string {\n\t\tconst now = performance.now();\n\t\tconst shown = this.attempts.slice(first);\n\t\tif (shown.length === 0) return \"\";\n\t\treturn `Attempts: ${shown.map((attempt, index) => `${index + 1}) ${attempt.describe(now)}`).join(\"; \")}.`;\n\t}\n}\n\nexport interface DecisionRun {\n\treadonly signal: AbortSignal;\n\treadonly trace: DecisionTrace;\n}\n\n/**\n * Run a whole decision under one wall-clock limit. `run` receives a signal that aborts with the caller's reason when\n * the caller aborts and with a {@link StructuredOutputDeadlineError} when the limit passes, so the existing abort\n * plumbing cancels the in-flight generation. The timer is always released.\n */\nexport async function withinDecisionDeadline<T>(\n\tcallerSignal: AbortSignal | undefined,\n\ttimeoutMs: number,\n\trun: (decision: DecisionRun) => Promise<T>,\n): Promise<T> {\n\tconst controller = new AbortController();\n\tconst trace = new DecisionTrace();\n\tconst forwardCallerAbort = () => controller.abort(callerSignal?.reason);\n\tif (callerSignal?.aborted) forwardCallerAbort();\n\telse callerSignal?.addEventListener(\"abort\", forwardCallerAbort, { once: true });\n\tconst timer = setTimeout(\n\t\t() => controller.abort(new StructuredOutputDeadlineError(timeoutMs, trace.describe())),\n\t\ttimeoutMs,\n\t);\n\ttry {\n\t\treturn await run({ signal: controller.signal, trace });\n\t} finally {\n\t\tclearTimeout(timer);\n\t\tcallerSignal?.removeEventListener(\"abort\", forwardCallerAbort);\n\t}\n}\n"]}

package/dist/core/structured-output/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/core/structured-output/index.ts"],"names":[],"mappings":"AAUA,OAAO,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAQ/C,OAAO,KAAK,EAGX,qBAAqB,EAErB,uBAAuB,EACvB,sBAAsB,EACtB,MAAM,YAAY,CAAC;AAEpB,OAAO,EAAE,kBAAkB,EAAE,MAAM,eAAe,CAAC;AACnD,YAAY,EACX,qBAAqB,EACrB,2BAA2B,EAC3B,wBAAwB,EACxB,qBAAqB,EACrB,uBAAuB,EACvB,sBAAsB,GACtB,MAAM,YAAY,CAAC;AA8OpB,wBAAsB,wBAAwB,CAAC,CAAC,SAAS,OAAO,EAC/D,OAAO,EAAE,uBAAuB,CAAC,CAAC,CAAC,GACjC,OAAO,CAAC,sBAAsB,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAyE5C;AA8JD,6GAA6G;AAC7G,wBAAsB,UAAU,CAAC,CAAC,SAAS,OAAO,EACjD,OAAO,EAAE,qBAAqB,CAAC,CAAC,CAAC;AACjC,iGAAiG;AACjG,gBAAgB,CAAC,EAAE,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC,KAAK,OAAO,GAC9C,OAAO,CAAC,sBAAsB,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAU5C"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/core/structured-output/index.ts"],"names":[],"mappings":"AAUA,OAAO,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAU/C,OAAO,KAAK,EAGX,qBAAqB,EAErB,uBAAuB,EACvB,sBAAsB,EACtB,MAAM,YAAY,CAAC;AAEpB,OAAO,EAAE,kBAAkB,EAAE,MAAM,eAAe,CAAC;AACnD,YAAY,EACX,qBAAqB,EACrB,2BAA2B,EAC3B,wBAAwB,EACxB,qBAAqB,EACrB,uBAAuB,EACvB,sBAAsB,GACtB,MAAM,YAAY,CAAC;AAkRpB,wBAAsB,wBAAwB,CAAC,CAAC,SAAS,OAAO,EAC/D,OAAO,EAAE,uBAAuB,CAAC,CAAC,CAAC,GACjC,OAAO,CAAC,sBAAsB,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAM5C;AAkQD,6GAA6G;AAC7G,wBAAsB,UAAU,CAAC,CAAC,SAAS,OAAO,EACjD,OAAO,EAAE,qBAAqB,CAAC,CAAC,CAAC;AACjC,iGAAiG;AACjG,gBAAgB,CAAC,EAAE,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC,KAAK,OAAO,GAC9C,OAAO,CAAC,sBAAsB,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAa5C"}

package/dist/core/structured-output/index.js

@@ -4,7 +4,9 @@ import { raceWithAbortSignal } from "../../utils/abort.js";
 import { reportModelRoutingDebug } from "../model-routing-debug.js";
 import { STRUCTURED_OUTPUT_TOOL_NAME } from "../tools/structured-output.js";
 import { compileChoiceSchema } from "./choice-schema.js";
+import { abortFailure, resolveDecisionDeadlineMs, withinDecisionDeadline } from "./deadline.js";
 import { InvalidDecisionOutputError } from "./invalid-output.js";
+import { describeSchemaErrors, schemaMismatchMessage, unparseableArgumentsMessage } from "./rejection-detail.js";
 import { resolveRouterModel } from "./resolver.js";
 export { resolveRouterModel } from "./resolver.js";
 const DEFAULT_DECISION_RETRY = Object.freeze({ enabled: true, maxRetries: 3, baseDelayMs: 2000 });
@@ -45,7 +47,7 @@ function validateState(state) {
         throw new Error("Structured output requires a nonempty named state object containing the task and relevant context text.");
     }
 }
-async function inferChat(request, model, signal, assertActive) {
+async function inferChat(request, model, signal, assertActive, trace) {
     // Anthropic catalog configuration may opt into server-side fallback. A decision must not.
     const decisionModel = model.api === "anthropic-messages"
         ? { ...model, compat: { ...model.compat, allowedFallbackModels: [] } }
@@ -58,34 +60,44 @@ async function inferChat(request, model, signal, assertActive) {
     let response;
     assertActive();
     try {
-        response = await retryAssistantCall(() => request.modelRegistry
-            .streamSimple(decisionModel, {
-            systemPrompt: `${STRUCTURED_DECISION_POLICY}\n\n${request.instructions}\n\nCall ${STRUCTURED_OUTPUT_TOOL_NAME} exactly once with the decision. Do not use prose or other tools.`,
-            messages: [
-                {
-                    role: "user",
-                    content: JSON.stringify(Object.keys(request.classifier.questions).length
-                        ? { state: request.state, questions: request.classifier.questions }
-                        : { state: request.state }),
-                    timestamp: Date.now(),
-                },
-            ],
-            tools: [
-                {
-                    name: tool.name,
-                    description: tool.description,
-                    parameters: tool.parameters,
-                    constrainedSampling: { type: "json_schema", strict: "prefer" },
-                },
-            ],
-        }, {
-            signal,
-            maxRetries: 0,
-            transport: "sse",
-            toolChoice: "auto",
-            ...(request.maxTokens === undefined ? {} : { maxTokens: request.maxTokens }),
-        })
-            .result(), request.retry ?? DEFAULT_DECISION_RETRY, signal);
+        response = await retryAssistantCall(() => {
+            const attempt = trace.begin();
+            return request.modelRegistry
+                .streamSimple(decisionModel, {
+                systemPrompt: `${STRUCTURED_DECISION_POLICY}\n\n${request.instructions}\n\nCall ${STRUCTURED_OUTPUT_TOOL_NAME} exactly once with the decision. Do not use prose or other tools.`,
+                messages: [
+                    {
+                        role: "user",
+                        content: JSON.stringify(Object.keys(request.classifier.questions).length
+                            ? { state: request.state, questions: request.classifier.questions }
+                            : { state: request.state }),
+                        timestamp: Date.now(),
+                    },
+                ],
+                tools: [
+                    {
+                        name: tool.name,
+                        description: tool.description,
+                        parameters: tool.parameters,
+                        constrainedSampling: { type: "json_schema", strict: "prefer" },
+                    },
+                ],
+            }, {
+                signal,
+                maxRetries: 0,
+                transport: "sse",
+                toolChoice: "auto",
+                ...(request.maxTokens === undefined ? {} : { maxTokens: request.maxTokens }),
+            })
+                .result()
+                .then((message) => {
+                attempt.settle(message.stopReason);
+                return message;
+            }, (error) => {
+                attempt.settle();
+                throw error;
+            });
+        }, request.retry ?? DEFAULT_DECISION_RETRY, signal);
     }
     catch (error) {
         signal.throwIfAborted();
@@ -116,6 +128,11 @@ async function inferChat(request, model, signal, assertActive) {
         error.usage = { inputTokens: response.usage.input, outputTokens: response.usage.output };
         throw error;
     }
+    if (calls[0].argumentsError) {
+        const error = new InvalidDecisionOutputError(unparseableArgumentsMessage(calls[0].argumentsError));
+        error.usage = { inputTokens: response.usage.input, outputTokens: response.usage.output };
+        throw error;
+    }
     // These are result arguments, never executable tool calls. Strict validation happens below.
     return {
         value: calls[0].arguments,
@@ -126,7 +143,7 @@ async function inferChat(request, model, signal, assertActive) {
 }
 function assertNotCancelled(signal) {
     if (signal?.aborted)
-        throw new Error("Structured output cancelled; no decision was accepted.");
+        throw abortFailure(signal);
 }
 function resolveCandidate(id, request) {
     if (typeof id !== "string" || !id.trim() || id.trim() !== id || id === "auto")
@@ -145,6 +162,14 @@ function resolveCandidate(id, request) {
         return { kind: "classifier", fullId: id, model: classifier };
     throw new Error(`Structured output model is unavailable or not a chat or classifier model: ${id}`);
 }
+function resolveCandidateOutcome(id, request) {
+    try {
+        return { candidate: resolveCandidate(id, request) };
+    }
+    catch (error) {
+        return { id, failure: error instanceof Error ? error : new Error(String(error)) };
+    }
+}
 class ClassifierDecisionError extends Error {
     /** Status and error type only; provider bodies can echo private state. */
     detail;
@@ -156,7 +181,7 @@ class ClassifierDecisionError extends Error {
     }
 }
 const CLASSIFIER_ABORTED = "Structured output classifier request was aborted; no fallback was attempted.";
-async function inferClassifier(request, selected, signal) {
+async function inferClassifier(request, selected, signal, trace) {
     const classify = request.modelRegistry.classify;
     if (!classify)
         throw new ClassifierDecisionError();
@@ -169,6 +194,7 @@ async function inferClassifier(request, selected, signal) {
         },
     ]));
     let result;
+    const attempt = trace.begin();
     try {
         result = await classify.call(request.modelRegistry, selected.model, { state: request.state, questions }, {
             signal,
@@ -184,6 +210,7 @@ async function inferClassifier(request, selected, signal) {
     signal.throwIfAborted();
     if (!result || typeof result !== "object")
         throw new ClassifierDecisionError();
+    attempt.settle(result.stopReason);
     if (result.stopReason === "aborted")
         throw new TerminalDecisionError(CLASSIFIER_ABORTED);
     if (result.stopReason !== "stop")
@@ -209,6 +236,11 @@ async function inferClassifier(request, selected, signal) {
     return { value, model: selected.fullId, responseModel: result.model, usage: { inputTokens: 0, outputTokens: 0 } };
 }
 export async function generateStructuredOutput(request) {
+    assertNotCancelled(request.signal);
+    const timeoutMs = resolveDecisionDeadlineMs(request.timeoutMs);
+    return withinDecisionDeadline(request.signal, timeoutMs, ({ signal, trace }) => generateWithinDeadline({ ...request, signal }, trace));
+}
+async function generateWithinDeadline(request, trace) {
     assertNotCancelled(request.signal);
     const current = request.currentModel;
     if (current && (current.id === "auto" || !isModelType(current, "chat")))
@@ -219,6 +251,7 @@ export async function generateStructuredOutput(request) {
         throw new Error("Structured output requires complete judgment instructions.");
     const snapshot = {
         ...request,
+        trace,
         state: jsonSnapshot(request.state),
         schema: jsonSnapshot(request.schema),
         instructions: request.instructions,
@@ -230,12 +263,20 @@ export async function generateStructuredOutput(request) {
     ];
     if (!ids[0])
         throw new Error("Structured output requires model or currentModel.");
-    const selected = [...new Set(ids)].map((id) => resolveCandidate(id, request));
+    const resolutions = [...new Set(ids)].map((id) => resolveCandidateOutcome(id, request));
+    const resolutionFailures = resolutions.filter((resolution) => "failure" in resolution);
+    if (resolutionFailures.length === resolutions.length)
+        throw resolutionFailures[0].failure;
     const choices = compileChoiceSchema(snapshot.schema);
     const modelAttempts = [];
     let lastDiagnostic;
-    for (const candidate of selected) {
+    for (const resolution of resolutions) {
         assertNotCancelled(request.signal);
+        if ("failure" in resolution) {
+            modelAttempts.push({ model: resolution.id, error: resolution.failure.message });
+            continue;
+        }
+        const candidate = resolution.candidate;
         if (candidate.kind === "classifier" && !choices) {
             modelAttempts.push({
                 model: candidate.fullId,
@@ -274,7 +315,10 @@ export async function generateStructuredOutput(request) {
                     : "Structured output inference failed; no fallback was attempted.");
             }
             lastDiagnostic = error instanceof OutputRepairExhaustedError ? error : undefined;
-            modelAttempts.push({ model: candidate.fullId, error: "Structured output inference failed." });
+            modelAttempts.push({
+                model: candidate.fullId,
+                error: error instanceof OutputRepairExhaustedError ? error.message : "Structured output inference failed.",
+            });
         }
     }
     if (lastDiagnostic)
@@ -301,7 +345,7 @@ class TerminalDecisionError extends Error {
 class CandidateProviderError extends Error {
 }
 async function inferDecision(request, repairs, validateDecision, fallbackModel) {
-    request.signal?.throwIfAborted();
+    request.signal.throwIfAborted();
     validateMaxTokensOverride(request.maxTokens);
     validateState(request.state);
     if (!request.instructions?.trim())
@@ -340,13 +384,15 @@ async function inferDecision(request, repairs, validateDecision, fallbackModel)
     }
     let fallback;
     const controller = new AbortController();
-    const abort = () => controller.abort(new Error("Structured output cancelled; no decision was accepted."));
-    request.signal?.addEventListener("abort", abort, { once: true });
+    const abort = () => controller.abort(abortFailure(request.signal));
+    request.signal.addEventListener("abort", abort, { once: true });
     const assertActive = () => controller.signal.throwIfAborted();
     try {
-        if (request.signal?.aborted)
+        if (request.signal.aborted)
             abort();
         const usage = { inputTokens: 0, outputTokens: 0 };
+        let rejection;
+        let firstAttempt = request.trace.size;
         for (let attempt = 0;;) {
             assertActive();
             const current = {
@@ -354,12 +400,12 @@ async function inferDecision(request, repairs, validateDecision, fallbackModel)
                 model: selected,
                 instructions: attempt === 0
                     ? snapshot.instructions
-                    : `${snapshot.instructions}\n\nThe previous response failed output validation. Return a complete valid decision satisfying the original schema, candidates and constraints. Do not change the task or invent values.`,
+                    : `${snapshot.instructions}\n\nThe previous response failed output validation${rejection ? `: ${rejection}` : "."} Return a complete valid decision satisfying the original schema, candidates and constraints. Do not change the task or invent values.`,
             };
             try {
                 const result = await raceWithAbortSignal(selected.kind === "classifier"
-                    ? inferClassifier(current, selected, controller.signal)
-                    : inferChat(current, selected.model, controller.signal, assertActive), controller.signal);
+                    ? inferClassifier(current, selected, controller.signal, request.trace)
+                    : inferChat(current, selected.model, controller.signal, assertActive, request.trace), controller.signal);
                 assertActive();
                 usage.inputTokens += result.usage.inputTokens;
                 usage.outputTokens += result.usage.outputTokens;
@@ -372,12 +418,14 @@ async function inferDecision(request, repairs, validateDecision, fallbackModel)
                 }
                 if (request.candidateFallback && selected.kind === "chat")
                     omitOptionalNulls(value, snapshot.schema);
-                if (!Check(snapshot.schema, value) || (validateDecision && !validateDecision(value)))
-                    throw new InvalidDecisionOutputError("Invalid structured output: response does not match the decision schema.");
+                const matchesSchema = Check(snapshot.schema, value);
+                if (!matchesSchema || (validateDecision && !validateDecision(value)))
+                    throw new InvalidDecisionOutputError(schemaMismatchMessage(matchesSchema ? [] : describeSchemaErrors(snapshot.schema, value)));
                 assertActive();
                 return { ...result, value, usage, ...(fallback ? { fallback } : {}) };
             }
             catch (error) {
+                request.trace.settleOpen();
                 assertActive();
                 if (selected.kind === "classifier" &&
                     fallbackChat &&
@@ -394,7 +442,9 @@ async function inferDecision(request, repairs, validateDecision, fallbackModel)
                     };
                     reportModelRoutingDebug(`Classifier routing failed${error instanceof ClassifierDecisionError && error.detail ? ` (${error.detail})` : ""}; falling back to current chat model ${fallback.to} for this routing decision.`);
                     selected = { kind: "chat", fullId: fallback.to, model: fallbackChat };
+                    rejection = undefined;
                     attempt = 0;
+                    firstAttempt = request.trace.size;
                     continue;
                 }
                 if (!(error instanceof InvalidDecisionOutputError))
@@ -404,17 +454,22 @@ async function inferDecision(request, repairs, validateDecision, fallbackModel)
                     usage.outputTokens += error.usage.outputTokens;
                 }
                 if (attempt < repairs) {
+                    rejection = error.message;
                     attempt++;
                     continue;
                 }
                 const scope = request.candidateFallback ? "Structured" : fallback ? "Chat fallback" : "Routing";
-                const message = `${error.message} ${repairs ? `${scope} output repair exhausted after ${repairs + 1} attempts.` : "No repair request was made."}`;
+                const exhaustion = repairs
+                    ? `${scope} output repair exhausted after ${repairs + 1} attempts.`
+                    : "No repair request was made.";
+                const attempts = request.trace.describe(firstAttempt);
+                const message = `${error.message} ${exhaustion}${attempts ? ` ${attempts}` : ""}`;
                 throw request.candidateFallback ? new OutputRepairExhaustedError(message) : new Error(message);
             }
         }
     }
     finally {
-        request.signal?.removeEventListener("abort", abort);
+        request.signal.removeEventListener("abort", abort);
     }
 }
 /** Resolve only prerequisite routing inference. Does not execute the selected action or alter chat/tools. */
@@ -422,12 +477,13 @@ export async function routeModel(request,
 /** Pure correlated-field validation against original candidates, never live admission checks. */
 validateDecision) {
     request.signal?.throwIfAborted();
-    const { settings, currentModel, ...inference } = request;
+    const { settings, currentModel, timeoutMs, ...inference } = request;
     const model = resolveRouterModel({ settings, currentModel, modelRegistry: request.modelRegistry });
     const fallback = model.kind !== "chat" && currentModel && currentModel.id !== "auto" && isModelType(currentModel, "chat")
         ? currentModel
         : undefined;
     const retry = inference.retry ?? settings.getRetrySettings?.() ?? DEFAULT_DECISION_RETRY;
-    return inferDecision({ ...inference, model, retry }, 3, validateDecision, fallback);
+    const deadlineMs = resolveDecisionDeadlineMs(timeoutMs, settings.getHttpIdleTimeoutMs?.());
+    return withinDecisionDeadline(request.signal, deadlineMs, ({ signal, trace }) => inferDecision({ ...inference, model, retry, signal, trace }, 3, validateDecision, fallback));
 }
 //# sourceMappingURL=index.js.map