@warmdrift/kgauto-compiler 2.0.0-alpha.77 → 2.0.0-alpha.78

package/dist/{ir-CRV1_taF.d.mts → ir-BFwWhj2s.d.mts}

@@ -442,6 +442,26 @@ type MutationApplied = {
     rankBefore?: number;
     rankAfter?: number;
 };
+/**
+ * alpha.78 — shape-proof accessor for a mutation entry (PB filing
+ * 2026-07-25 addendum #2). TWO public surfaces share the name
+ * `mutationsApplied` with DIFFERENT element types: `CompileResult` /
+ * `CallResult` carry `MutationApplied[]` (objects), while
+ * `compileForAISDKv6` carries `string[]` (ids) — each internally
+ * consistent, but a consumer coding to one surface's d.ts while reading the
+ * other's runtime gets a silent false negative (`typeof m === 'string'`
+ * compiles clean, runs clean, and is wrong on every row). Neither element
+ * type can change without breaking a live consumer, so the fix is an
+ * accessor that is correct on both.
+ */
+declare function mutationId(m: string | MutationApplied): string;
+/**
+ * alpha.78 — does any mutation in the list match `idOrPrefix` (exact id, or
+ * prefix when it ends with `*`)? Works on both `mutationsApplied` element
+ * shapes. `hasMutation(result.mutationsApplied, 'quality-gate-measured-*')`
+ * is the gate-detection idiom PB hand-rolled in their deep-smoke.
+ */
+declare function hasMutation(list: ReadonlyArray<string | MutationApplied> | undefined, idOrPrefix: string): boolean;
 /**
  * Target-specific wire request. Shape varies by provider — caller passes the
  * right field to the right SDK.
@@ -1090,6 +1110,36 @@ interface CallOptions {
      * later calls in the same isolate are gated either way).
      */
     gateWarmupMs?: number;
+    /**
+     * alpha.78 — per-ATTEMPT time bound on the provider leg, in ms
+     * (IC `per-call-timeout-on-call-options` + the portfolio-wide
+     * unbounded-LLM-call scan). Without it the only ceiling is the serverless
+     * route's `maxDuration`, and one hung socket eats the entire route budget
+     * before the fallback chain — the product's whole promise — can walk.
+     *
+     * Semantics per attempt shape:
+     *   - non-streaming: total bound via `AbortSignal.timeout()`; firing
+     *     classifies retryable `timeout` (L-061) so the chain WALKS.
+     *   - streaming: per-chunk STALL bound (max silence between bytes,
+     *     covering time-to-first-byte) — a total bound would cut legitimate
+     *     long streams. Also classifies `timeout` on fire.
+     *
+     * Each wire attempt (including the same-model retry) gets a fresh budget.
+     * Default: none — today's unbounded behavior, additive and opt-in; a
+     * consumer-side bounds scan (tt-intel s114 shape) can then require it
+     * mechanically. A hang recorded before this existed reads as model
+     * latency in brain telemetry — transport stalls were indistinguishable
+     * from slow models.
+     */
+    attemptTimeoutMs?: number;
+    /**
+     * alpha.78 — caller-owned cancellation (route-level deadline propagation).
+     * Composed per-attempt with `attemptTimeoutMs` via `AbortSignal.any`.
+     * Aborting classifies terminal `aborted` and stops the WHOLE chain walk —
+     * the caller cancelled, so walking would spend money on an answer nobody
+     * is waiting for. Distinct from a timeout, which walks.
+     */
+    abortSignal?: AbortSignal;
     toolRelevanceThreshold?: number;
     compressHistoryAfter?: number;
     /** Override API keys (defaults: process.env). */
@@ -1654,4 +1704,4 @@ interface PerAxisMetrics {
 /** Per-axis metrics keyed by model — used for chain-comparison views. */
 type PerAxisMetricsByModel = Record<string, PerAxisMetrics>;
 
-export { type ApiKeys as A, type BestPracticeAdvisory as B, type CompilePolicy as C, captureGoldenIr as D, type EffortLevel as E, type FallbackReason as F, type Grounding as G, type HistoryCachePolicy as H, type IntentDeclaration as I, parseGoldenCaptureRate as J, resolveGoldenCaptureRate as K, shouldCaptureGolden as L, type Message as M, type NormalizedResponse as N, type OutcomeResult as O, type ProviderOverrides as P, type RecordInput as R, type SystemModelMessage as S, type ToolCall as T, type CompiledRequest as a, type PromptIR as b, type CallOptions as c, type CallResult as d, type CompileResult as e, type SectionRewrite as f, type RecordOutcomeInput as g, type OracleScore as h, type Adapter as i, type PerAxisMetrics as j, type Provider as k, type ChainEntry as l, type CallAttempt as m, CallError as n, type ChainModelEntry as o, type ChainWithGrounding as p, type Constraints as q, type GoldenCaptureOptions as r, type MutationApplied as s, type NormalizedTokens as t, type OutcomeKind as u, type PerAxisMetricsByModel as v, type PromptSection as w, type SectionKind as x, type ShadowProbeConfig as y, type ToolDefinition as z };
+export { type ApiKeys as A, type BestPracticeAdvisory as B, type CompilePolicy as C, captureGoldenIr as D, type EffortLevel as E, type FallbackReason as F, type Grounding as G, type HistoryCachePolicy as H, type IntentDeclaration as I, hasMutation as J, mutationId as K, parseGoldenCaptureRate as L, type Message as M, type NormalizedResponse as N, type OutcomeResult as O, type ProviderOverrides as P, resolveGoldenCaptureRate as Q, type RecordInput as R, type SystemModelMessage as S, type ToolCall as T, shouldCaptureGolden as U, type CompiledRequest as a, type PromptIR as b, type CallOptions as c, type CallResult as d, type CompileResult as e, type SectionRewrite as f, type RecordOutcomeInput as g, type OracleScore as h, type Adapter as i, type PerAxisMetrics as j, type Provider as k, type ChainEntry as l, type CallAttempt as m, CallError as n, type ChainModelEntry as o, type ChainWithGrounding as p, type Constraints as q, type GoldenCaptureOptions as r, type MutationApplied as s, type NormalizedTokens as t, type OutcomeKind as u, type PerAxisMetricsByModel as v, type PromptSection as w, type SectionKind as x, type ShadowProbeConfig as y, type ToolDefinition as z };

package/dist/{ir-C02EIElQ.d.ts → ir-DZKS1tI7.d.ts}

@@ -442,6 +442,26 @@ type MutationApplied = {
     rankBefore?: number;
     rankAfter?: number;
 };
+/**
+ * alpha.78 — shape-proof accessor for a mutation entry (PB filing
+ * 2026-07-25 addendum #2). TWO public surfaces share the name
+ * `mutationsApplied` with DIFFERENT element types: `CompileResult` /
+ * `CallResult` carry `MutationApplied[]` (objects), while
+ * `compileForAISDKv6` carries `string[]` (ids) — each internally
+ * consistent, but a consumer coding to one surface's d.ts while reading the
+ * other's runtime gets a silent false negative (`typeof m === 'string'`
+ * compiles clean, runs clean, and is wrong on every row). Neither element
+ * type can change without breaking a live consumer, so the fix is an
+ * accessor that is correct on both.
+ */
+declare function mutationId(m: string | MutationApplied): string;
+/**
+ * alpha.78 — does any mutation in the list match `idOrPrefix` (exact id, or
+ * prefix when it ends with `*`)? Works on both `mutationsApplied` element
+ * shapes. `hasMutation(result.mutationsApplied, 'quality-gate-measured-*')`
+ * is the gate-detection idiom PB hand-rolled in their deep-smoke.
+ */
+declare function hasMutation(list: ReadonlyArray<string | MutationApplied> | undefined, idOrPrefix: string): boolean;
 /**
  * Target-specific wire request. Shape varies by provider — caller passes the
  * right field to the right SDK.
@@ -1090,6 +1110,36 @@ interface CallOptions {
      * later calls in the same isolate are gated either way).
      */
     gateWarmupMs?: number;
+    /**
+     * alpha.78 — per-ATTEMPT time bound on the provider leg, in ms
+     * (IC `per-call-timeout-on-call-options` + the portfolio-wide
+     * unbounded-LLM-call scan). Without it the only ceiling is the serverless
+     * route's `maxDuration`, and one hung socket eats the entire route budget
+     * before the fallback chain — the product's whole promise — can walk.
+     *
+     * Semantics per attempt shape:
+     *   - non-streaming: total bound via `AbortSignal.timeout()`; firing
+     *     classifies retryable `timeout` (L-061) so the chain WALKS.
+     *   - streaming: per-chunk STALL bound (max silence between bytes,
+     *     covering time-to-first-byte) — a total bound would cut legitimate
+     *     long streams. Also classifies `timeout` on fire.
+     *
+     * Each wire attempt (including the same-model retry) gets a fresh budget.
+     * Default: none — today's unbounded behavior, additive and opt-in; a
+     * consumer-side bounds scan (tt-intel s114 shape) can then require it
+     * mechanically. A hang recorded before this existed reads as model
+     * latency in brain telemetry — transport stalls were indistinguishable
+     * from slow models.
+     */
+    attemptTimeoutMs?: number;
+    /**
+     * alpha.78 — caller-owned cancellation (route-level deadline propagation).
+     * Composed per-attempt with `attemptTimeoutMs` via `AbortSignal.any`.
+     * Aborting classifies terminal `aborted` and stops the WHOLE chain walk —
+     * the caller cancelled, so walking would spend money on an answer nobody
+     * is waiting for. Distinct from a timeout, which walks.
+     */
+    abortSignal?: AbortSignal;
     toolRelevanceThreshold?: number;
     compressHistoryAfter?: number;
     /** Override API keys (defaults: process.env). */
@@ -1654,4 +1704,4 @@ interface PerAxisMetrics {
 /** Per-axis metrics keyed by model — used for chain-comparison views. */
 type PerAxisMetricsByModel = Record<string, PerAxisMetrics>;
 
-export { type ApiKeys as A, type BestPracticeAdvisory as B, type CompilePolicy as C, captureGoldenIr as D, type EffortLevel as E, type FallbackReason as F, type Grounding as G, type HistoryCachePolicy as H, type IntentDeclaration as I, parseGoldenCaptureRate as J, resolveGoldenCaptureRate as K, shouldCaptureGolden as L, type Message as M, type NormalizedResponse as N, type OutcomeResult as O, type ProviderOverrides as P, type RecordInput as R, type SystemModelMessage as S, type ToolCall as T, type CompiledRequest as a, type PromptIR as b, type CallOptions as c, type CallResult as d, type CompileResult as e, type SectionRewrite as f, type RecordOutcomeInput as g, type OracleScore as h, type Adapter as i, type PerAxisMetrics as j, type Provider as k, type ChainEntry as l, type CallAttempt as m, CallError as n, type ChainModelEntry as o, type ChainWithGrounding as p, type Constraints as q, type GoldenCaptureOptions as r, type MutationApplied as s, type NormalizedTokens as t, type OutcomeKind as u, type PerAxisMetricsByModel as v, type PromptSection as w, type SectionKind as x, type ShadowProbeConfig as y, type ToolDefinition as z };
+export { type ApiKeys as A, type BestPracticeAdvisory as B, type CompilePolicy as C, captureGoldenIr as D, type EffortLevel as E, type FallbackReason as F, type Grounding as G, type HistoryCachePolicy as H, type IntentDeclaration as I, hasMutation as J, mutationId as K, parseGoldenCaptureRate as L, type Message as M, type NormalizedResponse as N, type OutcomeResult as O, type ProviderOverrides as P, resolveGoldenCaptureRate as Q, type RecordInput as R, type SystemModelMessage as S, type ToolCall as T, shouldCaptureGolden as U, type CompiledRequest as a, type PromptIR as b, type CallOptions as c, type CallResult as d, type CompileResult as e, type SectionRewrite as f, type RecordOutcomeInput as g, type OracleScore as h, type Adapter as i, type PerAxisMetrics as j, type Provider as k, type ChainEntry as l, type CallAttempt as m, CallError as n, type ChainModelEntry as o, type ChainWithGrounding as p, type Constraints as q, type GoldenCaptureOptions as r, type MutationApplied as s, type NormalizedTokens as t, type OutcomeKind as u, type PerAxisMetricsByModel as v, type PromptSection as w, type SectionKind as x, type ShadowProbeConfig as y, type ToolDefinition as z };

package/dist/key-health.js

@@ -25,7 +25,7 @@ __export(key_health_exports, {
 module.exports = __toCommonJS(key_health_exports);
 
 // src/version.ts
-var LIBRARY_VERSION = "2.0.0-alpha.77";
+var LIBRARY_VERSION = "2.0.0-alpha.78";
 
 // src/key-health.ts
 var JSON_HEADERS = { "Content-Type": "application/json" };

package/dist/key-health.mjs

@@ -1,6 +1,6 @@
 import {
   createKeyHealthRoute
-} from "./chunk-GB7VJQ6C.mjs";
+} from "./chunk-QVD2QWST.mjs";
 export {
   createKeyHealthRoute
 };

package/dist/profiles.d.mts

@@ -1,4 +1,4 @@
-import { k as Provider } from './ir-CRV1_taF.mjs';
+import { k as Provider } from './ir-BFwWhj2s.mjs';
 import { IntentArchetypeName } from './dialect.mjs';
 
 /**
@@ -22,7 +22,17 @@ interface CliffRule {
     /** Threshold — meaning depends on metric. */
     threshold: number;
     /** What action to take when triggered. */
-    action: 'downgrade_quality_warning' | 'drop_to_top_relevant' | 'force_thinking_budget_zero' | 'force_terse_output' | 'escalate_target' | 'strip_tools';
+    action: 'downgrade_quality_warning' | 'drop_to_top_relevant' | 'force_thinking_budget_zero' | 'force_terse_output' | 'escalate_target' | 'strip_tools'
+    /**
+     * alpha.78 — apply the band-dominating QUALITY_GATE_PENALTY (same lever
+     * as the alpha.49 schema-weak gate) when the IR declares
+     * structuredOutput and the metric crosses threshold. For models whose
+     * declared structured-output support is MEASURED not to hold above an
+     * input size on an archetype — evidence-shaped, unlike the archetype-
+     * wide `structuredOutputHint: 'avoid'`. Only meaningful with
+     * `metric: 'input_tokens'` (+ usually `whenIntent`).
+     */
+     | 'quality_gate_structured';
     /**
      * Optional: only fire this cliff when the IR's intent.archetype matches.
      * Used for archetype-specific failure modes (e.g. Gemini Flash returns

package/dist/profiles.d.ts

@@ -1,4 +1,4 @@
-import { k as Provider } from './ir-C02EIElQ.js';
+import { k as Provider } from './ir-DZKS1tI7.js';
 import { IntentArchetypeName } from './dialect.js';
 
 /**
@@ -22,7 +22,17 @@ interface CliffRule {
     /** Threshold — meaning depends on metric. */
     threshold: number;
     /** What action to take when triggered. */
-    action: 'downgrade_quality_warning' | 'drop_to_top_relevant' | 'force_thinking_budget_zero' | 'force_terse_output' | 'escalate_target' | 'strip_tools';
+    action: 'downgrade_quality_warning' | 'drop_to_top_relevant' | 'force_thinking_budget_zero' | 'force_terse_output' | 'escalate_target' | 'strip_tools'
+    /**
+     * alpha.78 — apply the band-dominating QUALITY_GATE_PENALTY (same lever
+     * as the alpha.49 schema-weak gate) when the IR declares
+     * structuredOutput and the metric crosses threshold. For models whose
+     * declared structured-output support is MEASURED not to hold above an
+     * input size on an archetype — evidence-shaped, unlike the archetype-
+     * wide `structuredOutputHint: 'avoid'`. Only meaningful with
+     * `metric: 'input_tokens'` (+ usually `whenIntent`).
+     */
+     | 'quality_gate_structured';
     /**
      * Optional: only fire this cliff when the IR's intent.archetype matches.
      * Used for archetype-specific failure modes (e.g. Gemini Flash returns

package/dist/profiles.js

@@ -395,6 +395,22 @@ var PROFILES_RAW = [
         threshold: 16,
         action: "drop_to_top_relevant",
         reason: "Haiku reliability degrades above ~16 tools"
+      },
+      {
+        // alpha.78 — the declared `structuredOutput: 'grammar'` does NOT
+        // hold on long-input summarize. MEASURED (brain, playbacksam):
+        // 21 disambiguated `structured_output_parse_failed` fallover rows
+        // 2026-07-22..27, tokens_in 12,280–31,450; PB's gate counted 20/20
+        // in-window failures. Clean traffic p50 sits at ~9K tokens_in, so
+        // 12K gates the failing band without touching the working one.
+        // Short-input summarize carries no failure evidence and stays
+        // ungated — this is why it's a cliff, not an archetype-wide
+        // `structuredOutputHint: 'avoid'`.
+        metric: "input_tokens",
+        threshold: 12e3,
+        action: "quality_gate_structured",
+        whenIntent: "summarize",
+        reason: "Structured-output parse failures at 100% in-window on long-input summarize (measured on playbacksam, 2026-07-25..27; haiku only led when input size made price dominate, then failed every time)."
       }
     ],
     costInputPer1m: 1,

package/dist/profiles.mjs

@@ -8,7 +8,7 @@ import {
   latencyTierOf,
   profilesByProvider,
   tryGetProfile
-} from "./chunk-N36LE3MK.mjs";
+} from "./chunk-VVRDFE6T.mjs";
 export {
   ALIASES,
   LATENCY_TIER_MS,

package/dist/{types-C5BX9bzi.d.ts → types-B4kz3Vs0.d.ts}

@@ -1,4 +1,4 @@
-import { s as MutationApplied, B as BestPracticeAdvisory, F as FallbackReason, m as CallAttempt } from './ir-C02EIElQ.js';
+import { s as MutationApplied, B as BestPracticeAdvisory, F as FallbackReason, m as CallAttempt } from './ir-DZKS1tI7.js';
 
 /**
  * Glass-Box observability types (alpha.17).

package/dist/{types-CkRvjxrI.d.ts → types-D_fLt_Xv.d.ts}

@@ -1,4 +1,4 @@
-import { i as Adapter, F as FallbackReason, x as SectionKind } from './ir-C02EIElQ.js';
+import { i as Adapter, F as FallbackReason, x as SectionKind } from './ir-DZKS1tI7.js';
 
 /**
  * Wire contract for the Glass-Box Chrome extension's brain-poll endpoint.

package/dist/{types-Caq-hOIj.d.mts → types-DpcAMmk-.d.mts}

@@ -1,4 +1,4 @@
-import { s as MutationApplied, B as BestPracticeAdvisory, F as FallbackReason, m as CallAttempt } from './ir-CRV1_taF.mjs';
+import { s as MutationApplied, B as BestPracticeAdvisory, F as FallbackReason, m as CallAttempt } from './ir-BFwWhj2s.mjs';
 
 /**
  * Glass-Box observability types (alpha.17).

package/dist/{types-3WdGMqin.d.mts → types-hjzSWxtv.d.mts}

@@ -1,4 +1,4 @@
-import { i as Adapter, F as FallbackReason, x as SectionKind } from './ir-CRV1_taF.mjs';
+import { i as Adapter, F as FallbackReason, x as SectionKind } from './ir-BFwWhj2s.mjs';
 
 /**
  * Wire contract for the Glass-Box Chrome extension's brain-poll endpoint.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@warmdrift/kgauto-compiler",
-  "version": "2.0.0-alpha.77",
+  "version": "2.0.0-alpha.78",
   "description": "Prompt compiler with executable provider knowledge for multi-model AI apps: normalized multi-provider transport with fallback chains, compile-time cliff guards, a curated model registry, and a telemetry flight recorder. Swap models without rewriting prompts.",
   "main": "./dist/index.js",
   "module": "./dist/index.mjs",