@warmdrift/kgauto-compiler 2.0.0-alpha.85 → 2.0.0-alpha.87

package/dist/{ir-DfrM71pA.d.ts → ir-ZgN2KkuF.d.ts}

@@ -363,6 +363,26 @@ interface CompilePolicy {
      * advisor; otherwise unspecified.
      */
     posture?: 'locked' | 'preferred' | 'open';
+    /**
+     * alpha.87 — what compile() does with a model id no profile covers
+     * (gateway review §6; both cold readers hit this wall).
+     *
+     *   - `'refuse'` (default) — throw, naming `registerProfiles()` and this
+     *     option. The fleet's safety posture, unchanged: guards can only be
+     *     honest about models kgauto knows.
+     *   - `'best-effort'` — synthesize a conservative profile (provider
+     *     inferred from the id prefix; wire mechanics borrowed from a bundled
+     *     profile of that provider; NO cliffs, NO measured knowledge, cost
+     *     UNKNOWN and recorded as 0) and compile anyway, with a loud
+     *     `cliffWarnings` entry + `best_effort_profile` mutation naming
+     *     every guard that is skipped. If the provider cannot be inferred,
+     *     refusal still applies — best-effort cannot lower a wire format it
+     *     cannot name.
+     *
+     * The option converts "wait for the maintainer's weekend" into a
+     * consumer's informed choice; it does not make the unknown model safe.
+     */
+    onUnprofiledModel?: 'refuse' | 'best-effort';
 }
 /**
  * alpha.41 — entry in `PromptIR.models[]`. Either a literal model id (the
@@ -899,6 +919,69 @@ interface CompileResult {
          * number the cost-watcher can weigh against the lift.
          */
         disciplineGateTokens: number;
+        /**
+         * alpha.87 — which world answered THIS compile, per data source (the
+         * `per-source-freshness-marker-on-callresult` ask, PB s76). kgauto
+         * legitimately ranks on bundled profiles when cold and brain-re-ranked
+         * profiles when warm, and before this field the response said nothing
+         * about which had answered — PB drew three wrong conclusions across
+         * s74–s76 from observations taken at unknown temperature. A FLAT marker
+         * was rejected in the s76 consult: the app-scoped gates (measured
+         * failure, promotions, exclusions, promote-ready) carry cache state
+         * independent of the shared config snapshot, and a config-derived
+         * marker would read 'brain' in exactly the window where the
+         * measured-failure gate is still cold.
+         *
+         * Semantics: `config` is `'brain'` iff the shared `/v2/config` snapshot
+         * held fetched data at compile time (else `'bundled'`). The per-gate
+         * fields are `'warm'` iff a successful app-scoped load had completed in
+         * this process — `'cold'` covers never-loaded, load-failed, and
+         * feature-disabled alike, because in all three the gate read no data.
+         * With alpha.73/.86 prefetch + `gateWarmupMs`, a `call()` after warmup
+         * should read all-warm; a cold isolate's first compile may not.
+         */
+        freshness: {
+            config: 'bundled' | 'brain';
+            measuredFailure: 'cold' | 'warm';
+            promotions: 'cold' | 'warm';
+            exclusions: 'cold' | 'warm';
+            promoteReady: 'cold' | 'warm';
+        };
+        /**
+         * alpha.87 — per-section cost attribution (cc's
+         * `per-section-cost-attribution-in-diagnostics` ask, from Sacha's
+         * oracle read). `compile()` already receives sections IDed by the
+         * consumer, so this itemization is arithmetic on values the compiler
+         * computes anyway: zero model calls, no brain write, counts and IDs
+         * only — never section text (G3-compatible by construction). Always-on:
+         * an itemization nobody remembers to ask for is absent at the moment
+         * it is needed, and the payload is one small object per section.
+         *
+         * One row per INPUT section, in input order — dropped sections appear
+         * with `kept: false` and zero cost so a consumer can see what a pass
+         * removed and what it was worth.
+         *
+         * The cacheable interaction is deliberately two explicit columns
+         * rather than one blended number or a compile-time `cacheState` guess:
+         * whether a given call pays the first-call or the warm price is decided
+         * at the provider AFTER compile, so the honest compile-time statement
+         * is both prices. `estCostUsdFirstCall` is the section's tokens at the
+         * selected model's uncached input price (provider cache-WRITE
+         * surcharges, e.g. Anthropic's 1.25×, are not modeled — this
+         * understates first-call cost slightly). `estCostUsdWarm` applies the
+         * profile's cache-read discount when the section sits in an ACTIVE
+         * cacheable prefix (cacheable, and the prefix met the provider's
+         * minimum); otherwise it equals `estCostUsdFirstCall` — for a
+         * non-cached section the two worlds cost the same.
+         */
+        sections: Array<{
+            id: string;
+            tokens: number;
+            kept: boolean;
+            cacheable: boolean;
+            estCostUsdFirstCall: number;
+            estCostUsdWarm: number;
+        }>;
     };
     /**
      * alpha.33. Structured `system` for AI-SDK `streamText({ system })`
@@ -1084,6 +1167,13 @@ interface ShadowProbeConfig {
     skipSlowTierInSync?: boolean;
 }
 interface CallOptions {
+    /**
+     * alpha.86 — self-mark for synthetic callers (canaries, smokes, eval
+     * replays, probes). Passed through to every outcome row this call
+     * records, including failure rows. OMIT for real traffic. See
+     * {@link OutcomeSource}.
+     */
+    source?: OutcomeSource;
     /**
      * alpha.84 — fires once per FAILED attempt during the fallback walk, before
      * the walk continues. IC-Cairn's filing, 2026-07-29.
@@ -1458,9 +1548,34 @@ interface OracleScore {
     /** Free-form explanation for debugging. */
     rationale?: string;
 }
+/**
+ * alpha.86 — who originated an outcome row (migration 056).
+ *
+ * Absent/undefined means ORGANIC: a real consumer call on behalf of a real
+ * user. Consumers change nothing. Synthetic writers — canaries, smoke
+ * gates, eval replays, probe harnesses — self-mark, so liveness and volume
+ * rules can compute over consumer-originated rows only.
+ *
+ * The incident this closes (cost-watch 2026-08-06 → 08-08): playbacksam's
+ * daily canary wrote exactly 8 rows/day for 13 straight days while PB's
+ * organic traffic was near-zero, and the liveness rule scored PB the
+ * healthiest consumer in the portfolio. On 08-08 EVERY row the brain
+ * received in 24h was canary, and no rule could say so — distinguishing a
+ * replay from real traffic took a hand-reconstructed time-window +
+ * token-fingerprint argument, three mornings running. A consumer whose
+ * only rows are written by kgauto's own machinery is dark, not healthy;
+ * this column is what lets a rule print that sentence.
+ */
+type OutcomeSource = 'canary' | 'smoke' | 'eval' | 'probe' | 'synthetic';
 interface RecordInput {
     /** Handle from CompileResult. */
     handle: string;
+    /**
+     * alpha.86 — self-mark for synthetic writers (migration 056). OMIT for
+     * real traffic; never write an explicit "organic" value. See
+     * {@link OutcomeSource}.
+     */
+    source?: OutcomeSource;
     /** Actual tokens consumed (post-call). */
     tokensIn: number;
     tokensOut: number;
@@ -1753,4 +1868,4 @@ 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, 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 };
+export { type ApiKeys as A, type BestPracticeAdvisory as B, type CompilePolicy as C, type ToolDefinition as D, type EffortLevel as E, type FallbackReason as F, type Grounding as G, type HistoryCachePolicy as H, type IntentDeclaration as I, captureGoldenIr as J, hasMutation as K, mutationId as L, type Message as M, type NormalizedResponse as N, type OutcomeResult as O, type ProviderOverrides as P, parseGoldenCaptureRate as Q, type RecordInput as R, type SystemModelMessage as S, type ToolCall as T, resolveGoldenCaptureRate as U, shouldCaptureGolden as V, 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 OutcomeSource as h, type OracleScore as i, type Adapter as j, type PerAxisMetrics as k, type Provider as l, type ChainEntry as m, type CallAttempt as n, CallError as o, type ChainModelEntry as p, type ChainWithGrounding as q, type Constraints as r, type GoldenCaptureOptions as s, type MutationApplied as t, type NormalizedTokens as u, type OutcomeKind as v, type PerAxisMetricsByModel as w, type PromptSection as x, type SectionKind as y, type ShadowProbeConfig as z };

package/dist/key-health.js

@@ -28,7 +28,7 @@ __export(key_health_exports, {
 module.exports = __toCommonJS(key_health_exports);
 
 // src/version.ts
-var LIBRARY_VERSION = "2.0.0-alpha.85";
+var LIBRARY_VERSION = "2.0.0-alpha.87";
 
 // src/key-health.ts
 var JSON_HEADERS = { "Content-Type": "application/json" };

package/dist/key-health.mjs

@@ -3,7 +3,7 @@ import {
   KEY_FINGERPRINT_LENGTH,
   createKeyHealthRoute,
   keyFingerprint
-} from "./chunk-WXN7FNXP.mjs";
+} from "./chunk-GMZT4ZXD.mjs";
 export {
   KEY_FINGERPRINT_DOMAIN,
   KEY_FINGERPRINT_LENGTH,

package/dist/profiles.d.mts

@@ -1,4 +1,4 @@
-import { k as Provider } from './ir-JUoybVM-.mjs';
+import { l as Provider } from './ir-P3NEU8z4.mjs';
 import { IntentArchetypeName } from './dialect.mjs';
 
 /**
@@ -263,6 +263,15 @@ interface ModelProfile {
      * See `ArchetypeConvention` for field semantics.
      */
     archetypeConventions?: ArchetypeConvention[];
+    /**
+     * alpha.87 — true ONLY on profiles synthesized by
+     * `onUnprofiledModel: 'best-effort'`. Lets compile() (and any consumer)
+     * distinguish "kgauto knows this model" from "kgauto is winging it with
+     * borrowed wire mechanics" on every call, not just the one that
+     * triggered synthesis. Never set on bundled, brain, or
+     * `registerProfiles()` entries.
+     */
+    bestEffort?: true;
 }
 /**
  * Representative p50 latency (ms) per tier. Coarse on purpose — used only to
@@ -291,6 +300,38 @@ interface ProfileBrainHook {
 }
 /** @internal — called by models-brain.ts at module load. */
 declare function _setProfileBrainHook(hook: ProfileBrainHook): void;
+/**
+ * Register consumer-supplied profiles. They take precedence over both
+ * brain-curated and bundled profiles for the same id (explicit > learned >
+ * shipped — see block comment above for the tradeoff). Re-registering an
+ * id overwrites the previous consumer entry, including any best-effort
+ * synthesized profile.
+ */
+declare function registerProfiles(profiles: readonly ModelProfile[]): void;
+/** @internal — test hook. */
+declare function _testClearConsumerProfiles(): void;
+/** Provider inference for best-effort mode — id prefix conventions only. */
+declare function inferProviderFromId(id: string): Provider | undefined;
+/**
+ * alpha.87 — best-effort profile synthesis for unprofiled ids (opt-in via
+ * `CompilePolicy.onUnprofiledModel: 'best-effort'`; the default posture
+ * stays refuse). Returns undefined when the provider cannot be inferred
+ * from the id or kgauto ships no donor profile for that provider — in
+ * both cases the caller falls back to the standard refusal, which names
+ * `registerProfiles()` as the informed path.
+ *
+ * What a synthesized profile is: the inferred provider's wire mechanics
+ * (LoweringSpec / systemPromptMode / streaming are provider-wide, borrowed
+ * from a current bundled profile of that provider) around deliberately
+ * conservative capability claims — no cliffs, no recovery, no measured
+ * knowledge, `structuredOutput: 'none'`, sequential tools, modest output
+ * budget, and **cost 0 because cost is UNKNOWN** (cost gates and cost
+ * attribution are meaningless for this model; the compile carries a loud
+ * warning saying exactly that). It is registered into the consumer index
+ * on first synthesis so every later lookup in the process (fallback walk,
+ * passes, telemetry) sees one consistent object.
+ */
+declare function bestEffortProfile(id: string): ModelProfile | undefined;
 /**
  * Resolve a model id to its canonical form (alias → canonical, unknown ids
  * returned unchanged). This is the same resolution `getProfile` applies, so
@@ -308,4 +349,4 @@ declare function allProfiles(): readonly ModelProfile[];
 declare function allProfilesRaw(): readonly ModelProfile[];
 declare function profilesByProvider(provider: Provider): readonly ModelProfile[];
 
-export { ALIASES, type ArchetypeConvention, type CacheStrategy, type CliffRule, LATENCY_TIER_MS, type LatencyTier, type LoweringSpec, type ModelProfile, type RecoveryRule, type StructuredOutputCapability, type SystemPromptMode, _setProfileBrainHook, allProfiles, allProfilesRaw, getProfile, latencyTierOf, profilesByProvider, resolveModelAlias, tryGetProfile };
+export { ALIASES, type ArchetypeConvention, type CacheStrategy, type CliffRule, LATENCY_TIER_MS, type LatencyTier, type LoweringSpec, type ModelProfile, type RecoveryRule, type StructuredOutputCapability, type SystemPromptMode, _setProfileBrainHook, _testClearConsumerProfiles, allProfiles, allProfilesRaw, bestEffortProfile, getProfile, inferProviderFromId, latencyTierOf, profilesByProvider, registerProfiles, resolveModelAlias, tryGetProfile };

package/dist/profiles.d.ts

@@ -1,4 +1,4 @@
-import { k as Provider } from './ir-DfrM71pA.js';
+import { l as Provider } from './ir-ZgN2KkuF.js';
 import { IntentArchetypeName } from './dialect.js';
 
 /**
@@ -263,6 +263,15 @@ interface ModelProfile {
      * See `ArchetypeConvention` for field semantics.
      */
     archetypeConventions?: ArchetypeConvention[];
+    /**
+     * alpha.87 — true ONLY on profiles synthesized by
+     * `onUnprofiledModel: 'best-effort'`. Lets compile() (and any consumer)
+     * distinguish "kgauto knows this model" from "kgauto is winging it with
+     * borrowed wire mechanics" on every call, not just the one that
+     * triggered synthesis. Never set on bundled, brain, or
+     * `registerProfiles()` entries.
+     */
+    bestEffort?: true;
 }
 /**
  * Representative p50 latency (ms) per tier. Coarse on purpose — used only to
@@ -291,6 +300,38 @@ interface ProfileBrainHook {
 }
 /** @internal — called by models-brain.ts at module load. */
 declare function _setProfileBrainHook(hook: ProfileBrainHook): void;
+/**
+ * Register consumer-supplied profiles. They take precedence over both
+ * brain-curated and bundled profiles for the same id (explicit > learned >
+ * shipped — see block comment above for the tradeoff). Re-registering an
+ * id overwrites the previous consumer entry, including any best-effort
+ * synthesized profile.
+ */
+declare function registerProfiles(profiles: readonly ModelProfile[]): void;
+/** @internal — test hook. */
+declare function _testClearConsumerProfiles(): void;
+/** Provider inference for best-effort mode — id prefix conventions only. */
+declare function inferProviderFromId(id: string): Provider | undefined;
+/**
+ * alpha.87 — best-effort profile synthesis for unprofiled ids (opt-in via
+ * `CompilePolicy.onUnprofiledModel: 'best-effort'`; the default posture
+ * stays refuse). Returns undefined when the provider cannot be inferred
+ * from the id or kgauto ships no donor profile for that provider — in
+ * both cases the caller falls back to the standard refusal, which names
+ * `registerProfiles()` as the informed path.
+ *
+ * What a synthesized profile is: the inferred provider's wire mechanics
+ * (LoweringSpec / systemPromptMode / streaming are provider-wide, borrowed
+ * from a current bundled profile of that provider) around deliberately
+ * conservative capability claims — no cliffs, no recovery, no measured
+ * knowledge, `structuredOutput: 'none'`, sequential tools, modest output
+ * budget, and **cost 0 because cost is UNKNOWN** (cost gates and cost
+ * attribution are meaningless for this model; the compile carries a loud
+ * warning saying exactly that). It is registered into the consumer index
+ * on first synthesis so every later lookup in the process (fallback walk,
+ * passes, telemetry) sees one consistent object.
+ */
+declare function bestEffortProfile(id: string): ModelProfile | undefined;
 /**
  * Resolve a model id to its canonical form (alias → canonical, unknown ids
  * returned unchanged). This is the same resolution `getProfile` applies, so
@@ -308,4 +349,4 @@ declare function allProfiles(): readonly ModelProfile[];
 declare function allProfilesRaw(): readonly ModelProfile[];
 declare function profilesByProvider(provider: Provider): readonly ModelProfile[];
 
-export { ALIASES, type ArchetypeConvention, type CacheStrategy, type CliffRule, LATENCY_TIER_MS, type LatencyTier, type LoweringSpec, type ModelProfile, type RecoveryRule, type StructuredOutputCapability, type SystemPromptMode, _setProfileBrainHook, allProfiles, allProfilesRaw, getProfile, latencyTierOf, profilesByProvider, resolveModelAlias, tryGetProfile };
+export { ALIASES, type ArchetypeConvention, type CacheStrategy, type CliffRule, LATENCY_TIER_MS, type LatencyTier, type LoweringSpec, type ModelProfile, type RecoveryRule, type StructuredOutputCapability, type SystemPromptMode, _setProfileBrainHook, _testClearConsumerProfiles, allProfiles, allProfilesRaw, bestEffortProfile, getProfile, inferProviderFromId, latencyTierOf, profilesByProvider, registerProfiles, resolveModelAlias, tryGetProfile };