pi-smart-router 0.1.0

package/dist/config/pi-model-mapper.js

@@ -0,0 +1,216 @@
+/**
+ * Pi model registry → ModelProfile mapper.
+ *
+ * Maps pi `Model` objects (provider + id) to router fleet entries using
+ * pattern-based lookup for known families. Unknown models receive conservative
+ * economical-cloud defaults.
+ */
+const LOCAL_PROVIDERS = new Set(['lmstudio', 'ollama']);
+/** Conservative context limits when YAML and LiteLLM registry have no entry (SP-092). */
+export const DEFAULT_MODEL_LIMITS = {
+    'zero-tier': { max_input_tokens: 32_768, max_output_tokens: 4_096 },
+    'economical-cloud': { max_input_tokens: 128_000, max_output_tokens: 8_192 },
+    'frontier-cloud': { max_input_tokens: 200_000, max_output_tokens: 16_384 },
+};
+export function getDefaultLimitsForTier(tier) {
+    return { ...DEFAULT_MODEL_LIMITS[tier] };
+}
+const LOCAL_DEFAULTS = {
+    tier: 'zero-tier',
+    capabilities: { reasoning: 0.3, code_gen: 0.6, tool_use: 0.1 },
+    performance: {
+        latency_p50_ms: 120,
+        verbosity_factor: 0.9,
+        cache_friendly: true,
+    },
+    pricing: {
+        registry_key: 'local/free',
+        fallback_cost_per_1m: 0.0,
+    },
+    endpoint: 'http://localhost:1234/v1',
+};
+const FRONTIER_DEFAULTS = {
+    tier: 'frontier-cloud',
+    capabilities: { reasoning: 0.95, code_gen: 0.95, tool_use: 0.95 },
+    performance: {
+        latency_p50_ms: 450,
+        verbosity_factor: 1.2,
+        cache_friendly: true,
+    },
+    pricing: {
+        fallback_cost_per_1m: 3.0,
+    },
+};
+const ECONOMICAL_DEFAULTS = {
+    tier: 'economical-cloud',
+    capabilities: { reasoning: 0.7, code_gen: 0.75, tool_use: 0.7 },
+    performance: {
+        latency_p50_ms: 280,
+        verbosity_factor: 0.95,
+        cache_friendly: true,
+    },
+    pricing: {
+        fallback_cost_per_1m: 0.8,
+    },
+};
+const UNKNOWN_DEFAULTS = {
+    tier: 'economical-cloud',
+    capabilities: { reasoning: 0.6, code_gen: 0.65, tool_use: 0.6 },
+    performance: {
+        latency_p50_ms: 350,
+        verbosity_factor: 1.0,
+        cache_friendly: true,
+    },
+    pricing: {
+        fallback_cost_per_1m: 1.0,
+    },
+};
+/**
+ * Default virtual subscription-quota cost for Cursor frontier models (SP-096 / #70).
+ * `fallback_cost_per_1m` stays 0 (no per-token API billing); `quota_cost_per_1m`
+ * is used only for frugality scoring and telemetry so economical API models are
+ * not dominated solely because subscription registry cost is zero.
+ */
+export const DEFAULT_CURSOR_QUOTA_COST_PER_1M = 3.0;
+/**
+ * Opaque pi/Cursor fleet placeholder id `default` — SP-098 / #70.
+ * Without an explicit rule this id fell through to UNKNOWN_DEFAULTS (economical-cloud)
+ * and won turn_envelope lowest-cost selection for tool_result turns.
+ */
+const OPAQUE_FLEET_DEFAULT_ID = 'default';
+/**
+ * Cursor opaque-auto models (`cursor/auto`, etc.) — SP-086 / #40.
+ * Frontier tier: Cursor picks the underlying model; HyDRA needs high capability
+ * scores so these are not dominated by mapped economical Gemini/OpenAI models.
+ * Registry `fallback_cost_per_1m` is zero (subscription billing); virtual quota
+ * cost (SP-096) drives frugality scoring and telemetry.
+ */
+const CURSOR_AUTO_DEFAULTS = {
+    tier: 'frontier-cloud',
+    capabilities: { reasoning: 0.9, code_gen: 0.9, tool_use: 0.95 },
+    performance: {
+        latency_p50_ms: 350,
+        verbosity_factor: 1.0,
+        cache_friendly: false,
+    },
+    pricing: {
+        fallback_cost_per_1m: 0.0,
+        quota_cost_per_1m: DEFAULT_CURSOR_QUOTA_COST_PER_1M,
+    },
+};
+/**
+ * Cursor Composer coding models (`composer-latest`, `composer-*`) — SP-086 / #40.
+ * Frontier tier with strong code_gen; zero API fallback with virtual quota cost (SP-096).
+ */
+const COMPOSER_DEFAULTS = {
+    tier: 'frontier-cloud',
+    capabilities: { reasoning: 0.85, code_gen: 0.95, tool_use: 0.9 },
+    performance: {
+        latency_p50_ms: 400,
+        verbosity_factor: 1.05,
+        cache_friendly: false,
+    },
+    pricing: {
+        fallback_cost_per_1m: 0.0,
+        quota_cost_per_1m: DEFAULT_CURSOR_QUOTA_COST_PER_1M,
+    },
+};
+/** Ordered rules — first match wins. */
+const MODEL_PATTERN_RULES = [
+    { pattern: /claude[-_.]?opus|claude[-_.]?sonnet|opus|sonnet/i, defaults: FRONTIER_DEFAULTS },
+    { pattern: /claude[-_.]?haiku|haiku/i, defaults: ECONOMICAL_DEFAULTS },
+    { pattern: /gpt[-_.]?5\.5|gpt-5-5/i, defaults: FRONTIER_DEFAULTS },
+    { pattern: /gpt[-_.]?5\.1|gpt[-_.]?5[-_.]?mini|gpt[-_.]?mini/i, defaults: ECONOMICAL_DEFAULTS },
+    { pattern: /gemini[-_.]?2\.5[-_.]?pro|gemini-2-5-pro/i, defaults: FRONTIER_DEFAULTS },
+    { pattern: /gemini[-_.]?3.*pro/i, defaults: FRONTIER_DEFAULTS },
+    { pattern: /gemini.*pro/i, defaults: FRONTIER_DEFAULTS },
+    { pattern: /gemini.*flash|gemini-flash/i, defaults: ECONOMICAL_DEFAULTS },
+    { pattern: /^composer[-_]/i, defaults: COMPOSER_DEFAULTS },
+    { pattern: /^cursor\//i, defaults: CURSOR_AUTO_DEFAULTS },
+];
+function normalizeProvider(provider) {
+    return provider.trim().toLowerCase();
+}
+function matchPatternRules(id) {
+    for (const rule of MODEL_PATTERN_RULES) {
+        if (rule.pattern.test(id)) {
+            return rule.defaults;
+        }
+    }
+    return undefined;
+}
+/**
+ * Registry cost → USD per 1M tokens: average of input and output rates scaled to 1M.
+ * Matches `computeCostPer1MTokens` in litellm-fetch (SP-045/SP-046).
+ * Returns undefined when both input and output rates are zero (use pattern default).
+ */
+function deriveFallbackCostPer1M(cost) {
+    const { input, output } = cost;
+    if (input === 0 && output === 0) {
+        return undefined;
+    }
+    const inputPer1M = input * 1_000_000;
+    const outputPer1M = output * 1_000_000;
+    return (inputPer1M + outputPer1M) / 2;
+}
+function resolveFallbackCost(input, defaults) {
+    if (input.cost === undefined) {
+        return defaults.pricing.fallback_cost_per_1m;
+    }
+    const fromRegistry = deriveFallbackCostPer1M(input.cost);
+    return fromRegistry ?? defaults.pricing.fallback_cost_per_1m;
+}
+function buildProfile(input, defaults) {
+    const provider = input.provider.trim();
+    const registryKey = `${normalizeProvider(provider)}/${input.id}`;
+    const profile = {
+        id: input.id,
+        tier: defaults.tier,
+        provider,
+        capabilities: { ...defaults.capabilities },
+        pricing: {
+            registry_key: defaults.pricing.registry_key ?? registryKey,
+            fallback_cost_per_1m: resolveFallbackCost(input, defaults),
+            ...(defaults.pricing.quota_cost_per_1m !== undefined
+                ? { quota_cost_per_1m: defaults.pricing.quota_cost_per_1m }
+                : {}),
+        },
+    };
+    const withPerformance = defaults.performance !== undefined
+        ? { ...profile, performance: { ...defaults.performance } }
+        : profile;
+    if (defaults.endpoint !== undefined) {
+        return { ...withPerformance, endpoint: defaults.endpoint };
+    }
+    return withPerformance;
+}
+/**
+ * Map a pi model registry entry to a router ModelProfile.
+ */
+export function mapPiModelToProfile(input) {
+    const provider = normalizeProvider(input.provider);
+    if (LOCAL_PROVIDERS.has(provider)) {
+        // Local models stay free — registry cost must not override zero-tier pricing.
+        const localInput = {
+            provider: input.provider,
+            id: input.id,
+            ...(input.name !== undefined ? { name: input.name } : {}),
+        };
+        return buildProfile(localInput, LOCAL_DEFAULTS);
+    }
+    if (input.id === OPAQUE_FLEET_DEFAULT_ID) {
+        return buildProfile(input, CURSOR_AUTO_DEFAULTS);
+    }
+    const matched = matchPatternRules(input.id);
+    if (matched) {
+        return buildProfile(input, matched);
+    }
+    return buildProfile(input, UNKNOWN_DEFAULTS);
+}
+/**
+ * Map an array of pi registry models to a router fleet catalog.
+ */
+export function mapFleetFromRegistry(models) {
+    return models.map(mapPiModelToProfile);
+}
+//# sourceMappingURL=pi-model-mapper.js.map

package/dist/config/pi-model-mapper.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"pi-model-mapper.js","sourceRoot":"","sources":["../../src/config/pi-model-mapper.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAuCH,MAAM,eAAe,GAAG,IAAI,GAAG,CAAC,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAC,CAAC;AAExD,yFAAyF;AACzF,MAAM,CAAC,MAAM,oBAAoB,GAAwC;IACvE,WAAW,EAAE,EAAE,gBAAgB,EAAE,MAAM,EAAE,iBAAiB,EAAE,KAAK,EAAE;IACnE,kBAAkB,EAAE,EAAE,gBAAgB,EAAE,OAAO,EAAE,iBAAiB,EAAE,KAAK,EAAE;IAC3E,gBAAgB,EAAE,EAAE,gBAAgB,EAAE,OAAO,EAAE,iBAAiB,EAAE,MAAM,EAAE;CAC3E,CAAC;AAEF,MAAM,UAAU,uBAAuB,CAAC,IAAU;IAChD,OAAO,EAAE,GAAG,oBAAoB,CAAC,IAAI,CAAC,EAAE,CAAC;AAC3C,CAAC;AAED,MAAM,cAAc,GAAwB;IAC1C,IAAI,EAAE,WAAW;IACjB,YAAY,EAAE,EAAE,SAAS,EAAE,GAAG,EAAE,QAAQ,EAAE,GAAG,EAAE,QAAQ,EAAE,GAAG,EAAE;IAC9D,WAAW,EAAE;QACX,cAAc,EAAE,GAAG;QACnB,gBAAgB,EAAE,GAAG;QACrB,cAAc,EAAE,IAAI;KACrB;IACD,OAAO,EAAE;QACP,YAAY,EAAE,YAAY;QAC1B,oBAAoB,EAAE,GAAG;KAC1B;IACD,QAAQ,EAAE,0BAA0B;CACrC,CAAC;AAEF,MAAM,iBAAiB,GAAwB;IAC7C,IAAI,EAAE,gBAAgB;IACtB,YAAY,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE;IACjE,WAAW,EAAE;QACX,cAAc,EAAE,GAAG;QACnB,gBAAgB,EAAE,GAAG;QACrB,cAAc,EAAE,IAAI;KACrB;IACD,OAAO,EAAE;QACP,oBAAoB,EAAE,GAAG;KAC1B;CACF,CAAC;AAEF,MAAM,mBAAmB,GAAwB;IAC/C,IAAI,EAAE,kBAAkB;IACxB,YAAY,EAAE,EAAE,SAAS,EAAE,GAAG,EAAE,QAAQ,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE;IAC/D,WAAW,EAAE;QACX,cAAc,EAAE,GAAG;QACnB,gBAAgB,EAAE,IAAI;QACtB,cAAc,EAAE,IAAI;KACrB;IACD,OAAO,EAAE;QACP,oBAAoB,EAAE,GAAG;KAC1B;CACF,CAAC;AAEF,MAAM,gBAAgB,GAAwB;IAC5C,IAAI,EAAE,kBAAkB;IACxB,YAAY,EAAE,EAAE,SAAS,EAAE,GAAG,EAAE,QAAQ,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE;IAC/D,WAAW,EAAE;QACX,cAAc,EAAE,GAAG;QACnB,gBAAgB,EAAE,GAAG;QACrB,cAAc,EAAE,IAAI;KACrB;IACD,OAAO,EAAE;QACP,oBAAoB,EAAE,GAAG;KAC1B;CACF,CAAC;AAEF;;;;;GAKG;AACH,MAAM,CAAC,MAAM,gCAAgC,GAAG,GAAG,CAAC;AAEpD;;;;GAIG;AACH,MAAM,uBAAuB,GAAG,SAAS,CAAC;AAE1C;;;;;;GAMG;AACH,MAAM,oBAAoB,GAAwB;IAChD,IAAI,EAAE,gBAAgB;IACtB,YAAY,EAAE,EAAE,SAAS,EAAE,GAAG,EAAE,QAAQ,EAAE,GAAG,EAAE,QAAQ,EAAE,IAAI,EAAE;IAC/D,WAAW,EAAE;QACX,cAAc,EAAE,GAAG;QACnB,gBAAgB,EAAE,GAAG;QACrB,cAAc,EAAE,KAAK;KACtB;IACD,OAAO,EAAE;QACP,oBAAoB,EAAE,GAAG;QACzB,iBAAiB,EAAE,gCAAgC;KACpD;CACF,CAAC;AAEF;;;GAGG;AACH,MAAM,iBAAiB,GAAwB;IAC7C,IAAI,EAAE,gBAAgB;IACtB,YAAY,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE;IAChE,WAAW,EAAE;QACX,cAAc,EAAE,GAAG;QACnB,gBAAgB,EAAE,IAAI;QACtB,cAAc,EAAE,KAAK;KACtB;IACD,OAAO,EAAE;QACP,oBAAoB,EAAE,GAAG;QACzB,iBAAiB,EAAE,gCAAgC;KACpD;CACF,CAAC;AAEF,wCAAwC;AACxC,MAAM,mBAAmB,GAAgC;IACvD,EAAE,OAAO,EAAE,kDAAkD,EAAE,QAAQ,EAAE,iBAAiB,EAAE;IAC5F,EAAE,OAAO,EAAE,0BAA0B,EAAE,QAAQ,EAAE,mBAAmB,EAAE;IACtE,EAAE,OAAO,EAAE,wBAAwB,EAAE,QAAQ,EAAE,iBAAiB,EAAE;IAClE,EAAE,OAAO,EAAE,mDAAmD,EAAE,QAAQ,EAAE,mBAAmB,EAAE;IAC/F,EAAE,OAAO,EAAE,2CAA2C,EAAE,QAAQ,EAAE,iBAAiB,EAAE;IACrF,EAAE,OAAO,EAAE,qBAAqB,EAAE,QAAQ,EAAE,iBAAiB,EAAE;IAC/D,EAAE,OAAO,EAAE,cAAc,EAAE,QAAQ,EAAE,iBAAiB,EAAE;IACxD,EAAE,OAAO,EAAE,6BAA6B,EAAE,QAAQ,EAAE,mBAAmB,EAAE;IACzE,EAAE,OAAO,EAAE,gBAAgB,EAAE,QAAQ,EAAE,iBAAiB,EAAE;IAC1D,EAAE,OAAO,EAAE,YAAY,EAAE,QAAQ,EAAE,oBAAoB,EAAE;CAC1D,CAAC;AAEF,SAAS,iBAAiB,CAAC,QAAgB;IACzC,OAAO,QAAQ,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;AACvC,CAAC;AAED,SAAS,iBAAiB,CAAC,EAAU;IACnC,KAAK,MAAM,IAAI,IAAI,mBAAmB,EAAE,CAAC;QACvC,IAAI,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,CAAC;YAC1B,OAAO,IAAI,CAAC,QAAQ,CAAC;QACvB,CAAC;IACH,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED;;;;GAIG;AACH,SAAS,uBAAuB,CAAC,IAAoB;IACnD,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,GAAG,IAAI,CAAC;IAC/B,IAAI,KAAK,KAAK,CAAC,IAAI,MAAM,KAAK,CAAC,EAAE,CAAC;QAChC,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,MAAM,UAAU,GAAG,KAAK,GAAG,SAAS,CAAC;IACrC,MAAM,WAAW,GAAG,MAAM,GAAG,SAAS,CAAC;IACvC,OAAO,CAAC,UAAU,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;AACxC,CAAC;AAED,SAAS,mBAAmB,CAC1B,KAAmB,EACnB,QAA6B;IAE7B,IAAI,KAAK,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;QAC7B,OAAO,QAAQ,CAAC,OAAO,CAAC,oBAAoB,CAAC;IAC/C,CAAC;IAED,MAAM,YAAY,GAAG,uBAAuB,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IACzD,OAAO,YAAY,IAAI,QAAQ,CAAC,OAAO,CAAC,oBAAoB,CAAC;AAC/D,CAAC;AAED,SAAS,YAAY,CAAC,KAAmB,EAAE,QAA6B;IACtE,MAAM,QAAQ,GAAG,KAAK,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC;IACvC,MAAM,WAAW,GAAG,GAAG,iBAAiB,CAAC,QAAQ,CAAC,IAAI,KAAK,CAAC,EAAE,EAAE,CAAC;IAEjE,MAAM,OAAO,GAAiB;QAC5B,EAAE,EAAE,KAAK,CAAC,EAAE;QACZ,IAAI,EAAE,QAAQ,CAAC,IAAI;QACnB,QAAQ;QACR,YAAY,EAAE,EAAE,GAAG,QAAQ,CAAC,YAAY,EAAE;QAC1C,OAAO,EAAE;YACP,YAAY,EAAE,QAAQ,CAAC,OAAO,CAAC,YAAY,IAAI,WAAW;YAC1D,oBAAoB,EAAE,mBAAmB,CAAC,KAAK,EAAE,QAAQ,CAAC;YAC1D,GAAG,CAAC,QAAQ,CAAC,OAAO,CAAC,iBAAiB,KAAK,SAAS;gBAClD,CAAC,CAAC,EAAE,iBAAiB,EAAE,QAAQ,CAAC,OAAO,CAAC,iBAAiB,EAAE;gBAC3D,CAAC,CAAC,EAAE,CAAC;SACR;KACF,CAAC;IAEF,MAAM,eAAe,GACnB,QAAQ,CAAC,WAAW,KAAK,SAAS;QAChC,CAAC,CAAC,EAAE,GAAG,OAAO,EAAE,WAAW,EAAE,EAAE,GAAG,QAAQ,CAAC,WAAW,EAAE,EAAE;QAC1D,CAAC,CAAC,OAAO,CAAC;IAEd,IAAI,QAAQ,CAAC,QAAQ,KAAK,SAAS,EAAE,CAAC;QACpC,OAAO,EAAE,GAAG,eAAe,EAAE,QAAQ,EAAE,QAAQ,CAAC,QAAQ,EAAE,CAAC;IAC7D,CAAC;IAED,OAAO,eAAe,CAAC;AACzB,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,mBAAmB,CAAC,KAAmB;IACrD,MAAM,QAAQ,GAAG,iBAAiB,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;IAEnD,IAAI,eAAe,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;QAClC,8EAA8E;QAC9E,MAAM,UAAU,GAAiB;YAC/B,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,EAAE,EAAE,KAAK,CAAC,EAAE;YACZ,GAAG,CAAC,KAAK,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAC1D,CAAC;QACF,OAAO,YAAY,CAAC,UAAU,EAAE,cAAc,CAAC,CAAC;IAClD,CAAC;IAED,IAAI,KAAK,CAAC,EAAE,KAAK,uBAAuB,EAAE,CAAC;QACzC,OAAO,YAAY,CAAC,KAAK,EAAE,oBAAoB,CAAC,CAAC;IACnD,CAAC;IAED,MAAM,OAAO,GAAG,iBAAiB,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;IAC5C,IAAI,OAAO,EAAE,CAAC;QACZ,OAAO,YAAY,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;IACtC,CAAC;IAED,OAAO,YAAY,CAAC,KAAK,EAAE,gBAAgB,CAAC,CAAC;AAC/C,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,oBAAoB,CAAC,MAA+B;IAClE,OAAO,MAAM,CAAC,GAAG,CAAC,mBAAmB,CAAC,CAAC;AACzC,CAAC"}

package/dist/config/routing-clusters-loader.d.ts

@@ -0,0 +1,35 @@
+/**
+ * Routing cluster catalog loader — reads routing-clusters.yaml, validates via Zod,
+ * and precomputes centroid embeddings from reference prompts at load time.
+ *
+ * Maps to GitHub issue #55 / SP-099.
+ */
+import type { RoutingCluster, RoutingClusterCatalog } from '../domain/types/index.js';
+/** Prefix for cluster match reason codes (`cluster_low_stakes_general`, etc.). */
+export declare const CLUSTER_REASON_CODE_PREFIX = "cluster_";
+export interface TextEmbedder {
+    embed(text: string): Promise<Float32Array>;
+}
+export interface LoadRoutingClustersOptions {
+    readonly filePath?: string;
+    readonly embedder: TextEmbedder;
+}
+export declare class RoutingClustersLoaderError extends Error {
+    readonly name = "RoutingClustersLoaderError";
+    constructor(message: string, options?: ErrorOptions);
+}
+/** Stable reason-code prefix for a cluster id. */
+export declare function clusterReasonCode(clusterId: string): string;
+/**
+ * Parse and validate routing cluster YAML without computing centroids.
+ *
+ * @throws {RoutingClustersLoaderError} when validation fails.
+ */
+export declare function parseRoutingClustersYaml(raw: string): readonly RoutingCluster[];
+/**
+ * Load, validate, and precompute centroid embeddings for the routing cluster catalog.
+ *
+ * @throws {RoutingClustersLoaderError} when the file cannot be read or validation fails.
+ */
+export declare function loadRoutingClusters(options: LoadRoutingClustersOptions): Promise<RoutingClusterCatalog>;
+//# sourceMappingURL=routing-clusters-loader.d.ts.map

package/dist/config/routing-clusters-loader.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"routing-clusters-loader.d.ts","sourceRoot":"","sources":["../../src/config/routing-clusters-loader.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAQH,OAAO,KAAK,EAEV,cAAc,EACd,qBAAqB,EACtB,MAAM,0BAA0B,CAAC;AAElC,kFAAkF;AAClF,eAAO,MAAM,0BAA0B,aAAa,CAAC;AAErD,MAAM,WAAW,YAAY;IAC3B,KAAK,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,YAAY,CAAC,CAAC;CAC5C;AAED,MAAM,WAAW,0BAA0B;IACzC,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,QAAQ,EAAE,YAAY,CAAC;CACjC;AAED,qBAAa,0BAA2B,SAAQ,KAAK;IACnD,SAAkB,IAAI,gCAAgC;gBAE1C,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,YAAY;CAGpD;AAED,kDAAkD;AAClD,wBAAgB,iBAAiB,CAAC,SAAS,EAAE,MAAM,GAAG,MAAM,CAE3D;AAQD;;;;GAIG;AACH,wBAAgB,wBAAwB,CAAC,GAAG,EAAE,MAAM,GAAG,SAAS,cAAc,EAAE,CAkB/E;AAiDD;;;;GAIG;AACH,wBAAsB,mBAAmB,CACvC,OAAO,EAAE,0BAA0B,GAClC,OAAO,CAAC,qBAAqB,CAAC,CAkBhC"}

package/dist/config/routing-clusters-loader.js

@@ -0,0 +1,100 @@
+/**
+ * Routing cluster catalog loader — reads routing-clusters.yaml, validates via Zod,
+ * and precomputes centroid embeddings from reference prompts at load time.
+ *
+ * Maps to GitHub issue #55 / SP-099.
+ */
+import { readFileSync } from 'node:fs';
+import { resolve } from 'node:path';
+import { parse as parseYaml } from 'yaml';
+import { RoutingClustersFileSchema } from '../domain/types/schemas.js';
+/** Prefix for cluster match reason codes (`cluster_low_stakes_general`, etc.). */
+export const CLUSTER_REASON_CODE_PREFIX = 'cluster_';
+export class RoutingClustersLoaderError extends Error {
+    name = 'RoutingClustersLoaderError';
+    constructor(message, options) {
+        super(message, options);
+    }
+}
+/** Stable reason-code prefix for a cluster id. */
+export function clusterReasonCode(clusterId) {
+    return `${CLUSTER_REASON_CODE_PREFIX}${clusterId}`;
+}
+function formatZodIssues(error) {
+    return error.issues
+        .map((issue) => `  - ${issue.path.join('.')}: ${issue.message}`)
+        .join('\n');
+}
+/**
+ * Parse and validate routing cluster YAML without computing centroids.
+ *
+ * @throws {RoutingClustersLoaderError} when validation fails.
+ */
+export function parseRoutingClustersYaml(raw) {
+    let parsed;
+    try {
+        parsed = parseYaml(raw);
+    }
+    catch (err) {
+        const message = err instanceof Error ? err.message : String(err);
+        throw new RoutingClustersLoaderError(`Failed to parse YAML: ${message}`, { cause: err });
+    }
+    const result = RoutingClustersFileSchema.safeParse(parsed);
+    if (!result.success) {
+        throw new RoutingClustersLoaderError(`Invalid routing cluster catalog:\n${formatZodIssues(result.error)}`, { cause: result.error });
+    }
+    return result.data.clusters;
+}
+function computeCentroid(embeddings) {
+    if (embeddings.length === 0) {
+        throw new RoutingClustersLoaderError('Cannot compute centroid from zero embeddings');
+    }
+    const dim = embeddings[0].length;
+    const centroid = new Float32Array(dim);
+    for (const embedding of embeddings) {
+        if (embedding.length !== dim) {
+            throw new RoutingClustersLoaderError(`Embedding shape mismatch: expected ${dim}, got ${embedding.length}`);
+        }
+        for (let i = 0; i < dim; i++) {
+            centroid[i] = (centroid[i] ?? 0) + (embedding[i] ?? 0);
+        }
+    }
+    for (let i = 0; i < dim; i++) {
+        centroid[i] = (centroid[i] ?? 0) / embeddings.length;
+    }
+    return centroid;
+}
+async function loadClusterCentroids(clusters, embedder) {
+    const loaded = [];
+    for (const cluster of clusters) {
+        const embeddings = [];
+        for (const prompt of cluster.reference_prompts) {
+            embeddings.push(await embedder.embed(prompt));
+        }
+        loaded.push({
+            ...cluster,
+            centroid: computeCentroid(embeddings),
+        });
+    }
+    return loaded;
+}
+/**
+ * Load, validate, and precompute centroid embeddings for the routing cluster catalog.
+ *
+ * @throws {RoutingClustersLoaderError} when the file cannot be read or validation fails.
+ */
+export async function loadRoutingClusters(options) {
+    const filePath = options.filePath ?? resolve('config', 'routing-clusters.yaml');
+    let raw;
+    try {
+        raw = readFileSync(filePath, 'utf8');
+    }
+    catch (err) {
+        const message = err instanceof Error ? err.message : String(err);
+        throw new RoutingClustersLoaderError(`Failed to read routing clusters file: ${message}`, { cause: err });
+    }
+    const clusters = parseRoutingClustersYaml(raw);
+    const loaded = await loadClusterCentroids(clusters, options.embedder);
+    return { clusters: loaded };
+}
+//# sourceMappingURL=routing-clusters-loader.js.map

package/dist/config/routing-clusters-loader.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"routing-clusters-loader.js","sourceRoot":"","sources":["../../src/config/routing-clusters-loader.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AACvC,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAEpC,OAAO,EAAE,KAAK,IAAI,SAAS,EAAE,MAAM,MAAM,CAAC;AAE1C,OAAO,EAAE,yBAAyB,EAAE,MAAM,4BAA4B,CAAC;AAOvE,kFAAkF;AAClF,MAAM,CAAC,MAAM,0BAA0B,GAAG,UAAU,CAAC;AAWrD,MAAM,OAAO,0BAA2B,SAAQ,KAAK;IACjC,IAAI,GAAG,4BAA4B,CAAC;IAEtD,YAAY,OAAe,EAAE,OAAsB;QACjD,KAAK,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;IAC1B,CAAC;CACF;AAED,kDAAkD;AAClD,MAAM,UAAU,iBAAiB,CAAC,SAAiB;IACjD,OAAO,GAAG,0BAA0B,GAAG,SAAS,EAAE,CAAC;AACrD,CAAC;AAED,SAAS,eAAe,CAAC,KAA+E;IACtG,OAAO,KAAK,CAAC,MAAM;SAChB,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,KAAK,CAAC,OAAO,EAAE,CAAC;SAC/D,IAAI,CAAC,IAAI,CAAC,CAAC;AAChB,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,wBAAwB,CAAC,GAAW;IAClD,IAAI,MAAe,CAAC;IACpB,IAAI,CAAC;QACH,MAAM,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC;IAC1B,CAAC;IAAC,OAAO,GAAY,EAAE,CAAC;QACtB,MAAM,OAAO,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QACjE,MAAM,IAAI,0BAA0B,CAAC,yBAAyB,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,GAAG,EAAE,CAAC,CAAC;IAC3F,CAAC;IAED,MAAM,MAAM,GAAG,yBAAyB,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;IAC3D,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;QACpB,MAAM,IAAI,0BAA0B,CAClC,qCAAqC,eAAe,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,EACpE,EAAE,KAAK,EAAE,MAAM,CAAC,KAAK,EAAE,CACxB,CAAC;IACJ,CAAC;IAED,OAAO,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC;AAC9B,CAAC;AAED,SAAS,eAAe,CAAC,UAAmC;IAC1D,IAAI,UAAU,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC5B,MAAM,IAAI,0BAA0B,CAAC,8CAA8C,CAAC,CAAC;IACvF,CAAC;IAED,MAAM,GAAG,GAAG,UAAU,CAAC,CAAC,CAAE,CAAC,MAAM,CAAC;IAClC,MAAM,QAAQ,GAAG,IAAI,YAAY,CAAC,GAAG,CAAC,CAAC;IAEvC,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;QACnC,IAAI,SAAS,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;YAC7B,MAAM,IAAI,0BAA0B,CAClC,sCAAsC,GAAG,SAAS,SAAS,CAAC,MAAM,EAAE,CACrE,CAAC;QACJ,CAAC;QACD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC;YAC7B,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC;QACzD,CAAC;IACH,CAAC;IAED,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC;QAC7B,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,UAAU,CAAC,MAAM,CAAC;IACvD,CAAC;IAED,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,KAAK,UAAU,oBAAoB,CACjC,QAAmC,EACnC,QAAsB;IAEtB,MAAM,MAAM,GAA2B,EAAE,CAAC;IAE1C,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,MAAM,UAAU,GAAmB,EAAE,CAAC;QACtC,KAAK,MAAM,MAAM,IAAI,OAAO,CAAC,iBAAiB,EAAE,CAAC;YAC/C,UAAU,CAAC,IAAI,CAAC,MAAM,QAAQ,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC;QAChD,CAAC;QAED,MAAM,CAAC,IAAI,CAAC;YACV,GAAG,OAAO;YACV,QAAQ,EAAE,eAAe,CAAC,UAAU,CAAC;SACtC,CAAC,CAAC;IACL,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,mBAAmB,CACvC,OAAmC;IAEnC,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,IAAI,OAAO,CAAC,QAAQ,EAAE,uBAAuB,CAAC,CAAC;IAEhF,IAAI,GAAW,CAAC;IAChB,IAAI,CAAC;QACH,GAAG,GAAG,YAAY,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;IACvC,CAAC;IAAC,OAAO,GAAY,EAAE,CAAC;QACtB,MAAM,OAAO,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QACjE,MAAM,IAAI,0BAA0B,CAClC,yCAAyC,OAAO,EAAE,EAClD,EAAE,KAAK,EAAE,GAAG,EAAE,CACf,CAAC;IACJ,CAAC;IAED,MAAM,QAAQ,GAAG,wBAAwB,CAAC,GAAG,CAAC,CAAC;IAC/C,MAAM,MAAM,GAAG,MAAM,oBAAoB,CAAC,QAAQ,EAAE,OAAO,CAAC,QAAQ,CAAC,CAAC;IAEtE,OAAO,EAAE,QAAQ,EAAE,MAAM,EAAE,CAAC;AAC9B,CAAC"}

package/dist/domain/delegation/delegation-context.d.ts

@@ -0,0 +1,23 @@
+/**
+ * Delegation context normalization — provider-agnostic replay identity fix.
+ *
+ * pi-ai transformMessages compares assistant message provider/api/model to the
+ * target model. Virtual smart-router tags break isSameModel and strip replay
+ * state (thoughtSignature, thinkingSignature, etc.).
+ */
+import type { Api, Context, Message, Model } from '@earendil-works/pi-ai/compat';
+import type { ExecutionModel } from './execution-ledger.js';
+export declare const VIRTUAL_ROUTER_PROVIDER: "smart-router";
+export declare const VIRTUAL_ROUTER_MODEL_ID: "auto";
+export declare function isVirtualRouterIdentity(provider: string, modelId: string): boolean;
+export declare function hasReplaySensitiveState(messages: readonly Message[]): boolean;
+export interface NormalizeDelegationContextOptions {
+    readonly virtualProvider?: typeof VIRTUAL_ROUTER_PROVIDER;
+    readonly sessionExecution?: ExecutionModel | null;
+}
+/**
+ * Rewrite assistant messages tagged with the virtual router to the executing
+ * model identity pi-ai expects for same-model replay.
+ */
+export declare function normalizeDelegationContext<TApi extends Api>(context: Context, targetModel: Model<TApi>, options?: NormalizeDelegationContextOptions): Context;
+//# sourceMappingURL=delegation-context.d.ts.map

package/dist/domain/delegation/delegation-context.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"delegation-context.d.ts","sourceRoot":"","sources":["../../../src/domain/delegation/delegation-context.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,KAAK,EACV,GAAG,EAEH,OAAO,EACP,OAAO,EACP,KAAK,EACN,MAAM,8BAA8B,CAAC;AAEtC,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AAE5D,eAAO,MAAM,uBAAuB,EAAG,cAAuB,CAAC;AAC/D,eAAO,MAAM,uBAAuB,EAAG,MAAe,CAAC;AAEvD,wBAAgB,uBAAuB,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,OAAO,CAElF;AAED,wBAAgB,uBAAuB,CAAC,QAAQ,EAAE,SAAS,OAAO,EAAE,GAAG,OAAO,CA+B7E;AAcD,MAAM,WAAW,iCAAiC;IAChD,QAAQ,CAAC,eAAe,CAAC,EAAE,OAAO,uBAAuB,CAAC;IAC1D,QAAQ,CAAC,gBAAgB,CAAC,EAAE,cAAc,GAAG,IAAI,CAAC;CACnD;AAED;;;GAGG;AACH,wBAAgB,0BAA0B,CAAC,IAAI,SAAS,GAAG,EACzD,OAAO,EAAE,OAAO,EAChB,WAAW,EAAE,KAAK,CAAC,IAAI,CAAC,EACxB,OAAO,CAAC,EAAE,iCAAiC,GAC1C,OAAO,CA2BT"}

package/dist/domain/delegation/delegation-context.js

@@ -0,0 +1,73 @@
+/**
+ * Delegation context normalization — provider-agnostic replay identity fix.
+ *
+ * pi-ai transformMessages compares assistant message provider/api/model to the
+ * target model. Virtual smart-router tags break isSameModel and strip replay
+ * state (thoughtSignature, thinkingSignature, etc.).
+ */
+export const VIRTUAL_ROUTER_PROVIDER = 'smart-router';
+export const VIRTUAL_ROUTER_MODEL_ID = 'auto';
+export function isVirtualRouterIdentity(provider, modelId) {
+    return provider === VIRTUAL_ROUTER_PROVIDER && modelId === VIRTUAL_ROUTER_MODEL_ID;
+}
+export function hasReplaySensitiveState(messages) {
+    for (const message of messages) {
+        if (message.role !== 'assistant') {
+            continue;
+        }
+        for (const block of message.content) {
+            if (block.type === 'thinking') {
+                if (block.redacted) {
+                    return true;
+                }
+                if (block.thinkingSignature && block.thinkingSignature.length > 0) {
+                    return true;
+                }
+            }
+            if (block.type === 'text' && block.textSignature && block.textSignature.length > 0) {
+                return true;
+            }
+            if (block.type === 'toolCall' &&
+                block.thoughtSignature &&
+                block.thoughtSignature.length > 0) {
+                return true;
+            }
+        }
+    }
+    return false;
+}
+function rewriteAssistantIdentity(message, executionModel) {
+    return {
+        ...message,
+        provider: executionModel.provider,
+        api: executionModel.api,
+        model: executionModel.id,
+    };
+}
+/**
+ * Rewrite assistant messages tagged with the virtual router to the executing
+ * model identity pi-ai expects for same-model replay.
+ */
+export function normalizeDelegationContext(context, targetModel, options) {
+    const virtualProvider = options?.virtualProvider ?? VIRTUAL_ROUTER_PROVIDER;
+    const fallbackExecution = {
+        provider: targetModel.provider,
+        api: targetModel.api,
+        id: targetModel.id,
+    };
+    const sessionExecution = options?.sessionExecution ?? null;
+    const executionForVirtual = sessionExecution ?? fallbackExecution;
+    const messages = context.messages.map((message) => {
+        if (message.role !== 'assistant') {
+            return message;
+        }
+        const assistant = message;
+        if (assistant.provider === virtualProvider &&
+            assistant.model === VIRTUAL_ROUTER_MODEL_ID) {
+            return rewriteAssistantIdentity(assistant, executionForVirtual);
+        }
+        return message;
+    });
+    return { ...context, messages };
+}
+//# sourceMappingURL=delegation-context.js.map

package/dist/domain/delegation/delegation-context.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"delegation-context.js","sourceRoot":"","sources":["../../../src/domain/delegation/delegation-context.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAYH,MAAM,CAAC,MAAM,uBAAuB,GAAG,cAAuB,CAAC;AAC/D,MAAM,CAAC,MAAM,uBAAuB,GAAG,MAAe,CAAC;AAEvD,MAAM,UAAU,uBAAuB,CAAC,QAAgB,EAAE,OAAe;IACvE,OAAO,QAAQ,KAAK,uBAAuB,IAAI,OAAO,KAAK,uBAAuB,CAAC;AACrF,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,QAA4B;IAClE,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,IAAI,OAAO,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;YACjC,SAAS;QACX,CAAC;QAED,KAAK,MAAM,KAAK,IAAI,OAAO,CAAC,OAAO,EAAE,CAAC;YACpC,IAAI,KAAK,CAAC,IAAI,KAAK,UAAU,EAAE,CAAC;gBAC9B,IAAI,KAAK,CAAC,QAAQ,EAAE,CAAC;oBACnB,OAAO,IAAI,CAAC;gBACd,CAAC;gBACD,IAAI,KAAK,CAAC,iBAAiB,IAAI,KAAK,CAAC,iBAAiB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;oBAClE,OAAO,IAAI,CAAC;gBACd,CAAC;YACH,CAAC;YAED,IAAI,KAAK,CAAC,IAAI,KAAK,MAAM,IAAI,KAAK,CAAC,aAAa,IAAI,KAAK,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBACnF,OAAO,IAAI,CAAC;YACd,CAAC;YAED,IACE,KAAK,CAAC,IAAI,KAAK,UAAU;gBACzB,KAAK,CAAC,gBAAgB;gBACtB,KAAK,CAAC,gBAAgB,CAAC,MAAM,GAAG,CAAC,EACjC,CAAC;gBACD,OAAO,IAAI,CAAC;YACd,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,wBAAwB,CAC/B,OAAyB,EACzB,cAA8B;IAE9B,OAAO;QACL,GAAG,OAAO;QACV,QAAQ,EAAE,cAAc,CAAC,QAAwC;QACjE,GAAG,EAAE,cAAc,CAAC,GAAG;QACvB,KAAK,EAAE,cAAc,CAAC,EAAE;KACzB,CAAC;AACJ,CAAC;AAOD;;;GAGG;AACH,MAAM,UAAU,0BAA0B,CACxC,OAAgB,EAChB,WAAwB,EACxB,OAA2C;IAE3C,MAAM,eAAe,GAAG,OAAO,EAAE,eAAe,IAAI,uBAAuB,CAAC;IAC5E,MAAM,iBAAiB,GAAmB;QACxC,QAAQ,EAAE,WAAW,CAAC,QAAQ;QAC9B,GAAG,EAAE,WAAW,CAAC,GAAG;QACpB,EAAE,EAAE,WAAW,CAAC,EAAE;KACnB,CAAC;IACF,MAAM,gBAAgB,GAAG,OAAO,EAAE,gBAAgB,IAAI,IAAI,CAAC;IAC3D,MAAM,mBAAmB,GAAG,gBAAgB,IAAI,iBAAiB,CAAC;IAElE,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE;QAChD,IAAI,OAAO,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;YACjC,OAAO,OAAO,CAAC;QACjB,CAAC;QAED,MAAM,SAAS,GAAG,OAAO,CAAC;QAC1B,IACE,SAAS,CAAC,QAAQ,KAAK,eAAe;YACtC,SAAS,CAAC,KAAK,KAAK,uBAAuB,EAC3C,CAAC;YACD,OAAO,wBAAwB,CAAC,SAAS,EAAE,mBAAmB,CAAC,CAAC;QAClE,CAAC;QAED,OAAO,OAAO,CAAC;IACjB,CAAC,CAAC,CAAC;IAEH,OAAO,EAAE,GAAG,OAAO,EAAE,QAAQ,EAAE,CAAC;AAClC,CAAC"}

package/dist/domain/delegation/execution-ledger.d.ts

@@ -0,0 +1,19 @@
+/**
+ * Tracks the last successfully executing model per session.
+ *
+ * Used to rewrite virtual smart-router assistant identity on replay so pi-ai
+ * transformMessages sees the model that actually produced prior turns.
+ */
+import type { Api } from '@earendil-works/pi-ai/compat';
+export interface ExecutionModel {
+    readonly provider: string;
+    readonly api: Api;
+    readonly id: string;
+}
+export declare class ExecutionLedger {
+    private readonly bySession;
+    recordSuccess(sessionId: string, model: ExecutionModel): void;
+    getLastExecution(sessionId: string): ExecutionModel | null;
+    clear(sessionId: string): void;
+}
+//# sourceMappingURL=execution-ledger.d.ts.map

package/dist/domain/delegation/execution-ledger.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"execution-ledger.d.ts","sourceRoot":"","sources":["../../../src/domain/delegation/execution-ledger.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,KAAK,EAAE,GAAG,EAAE,MAAM,8BAA8B,CAAC;AAExD,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,GAAG,EAAE,GAAG,CAAC;IAClB,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;CACrB;AAED,qBAAa,eAAe;IAC1B,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAqC;IAE/D,aAAa,CAAC,SAAS,EAAE,MAAM,EAAE,KAAK,EAAE,cAAc,GAAG,IAAI;IAQ7D,gBAAgB,CAAC,SAAS,EAAE,MAAM,GAAG,cAAc,GAAG,IAAI;IAI1D,KAAK,CAAC,SAAS,EAAE,MAAM,GAAG,IAAI;CAG/B"}

package/dist/domain/delegation/execution-ledger.js

@@ -0,0 +1,23 @@
+/**
+ * Tracks the last successfully executing model per session.
+ *
+ * Used to rewrite virtual smart-router assistant identity on replay so pi-ai
+ * transformMessages sees the model that actually produced prior turns.
+ */
+export class ExecutionLedger {
+    bySession = new Map();
+    recordSuccess(sessionId, model) {
+        this.bySession.set(sessionId, {
+            provider: model.provider,
+            api: model.api,
+            id: model.id,
+        });
+    }
+    getLastExecution(sessionId) {
+        return this.bySession.get(sessionId) ?? null;
+    }
+    clear(sessionId) {
+        this.bySession.delete(sessionId);
+    }
+}
+//# sourceMappingURL=execution-ledger.js.map

package/dist/domain/delegation/execution-ledger.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"execution-ledger.js","sourceRoot":"","sources":["../../../src/domain/delegation/execution-ledger.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAUH,MAAM,OAAO,eAAe;IACT,SAAS,GAAG,IAAI,GAAG,EAA0B,CAAC;IAE/D,aAAa,CAAC,SAAiB,EAAE,KAAqB;QACpD,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,SAAS,EAAE;YAC5B,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,GAAG,EAAE,KAAK,CAAC,GAAG;YACd,EAAE,EAAE,KAAK,CAAC,EAAE;SACb,CAAC,CAAC;IACL,CAAC;IAED,gBAAgB,CAAC,SAAiB;QAChC,OAAO,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,SAAS,CAAC,IAAI,IAAI,CAAC;IAC/C,CAAC;IAED,KAAK,CAAC,SAAiB;QACrB,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;IACnC,CAAC;CACF"}

package/dist/domain/delegation/output-headroom.d.ts

@@ -0,0 +1,39 @@
+/**
+ * Delegation output headroom — SP-108.
+ *
+ * Reserves output budget from the model context window before provider dispatch.
+ * Prevents stopReason "length" with zero output when input consumes the window.
+ */
+import type { Api, Model } from '@earendil-works/pi-ai/compat';
+import type { ModelProfile } from '../types/index.js';
+export declare const MIN_OUTPUT_TOKEN_FLOOR = 256;
+export declare const DEFAULT_OUTPUT_HEADROOM_BUFFER = 64;
+export declare const MIN_OUTPUT_TOKEN_FLOOR_ENV = "MIN_OUTPUT_TOKEN_FLOOR";
+export declare const OUTPUT_HEADROOM_BUFFER_ENV = "OUTPUT_HEADROOM_BUFFER";
+export interface OutputHeadroomConfig {
+    readonly minOutputFloor?: number;
+    readonly buffer?: number;
+}
+export interface OutputHeadroomFit {
+    readonly kind: 'fit';
+    readonly maxTokens: number;
+    readonly contextWindow: number;
+}
+export interface OutputHeadroomNoFit {
+    readonly kind: 'no_fit';
+    readonly contextWindow: number;
+    readonly availableOutputTokens: number;
+}
+export type OutputHeadroomResult = OutputHeadroomFit | OutputHeadroomNoFit;
+export declare function resolveMinOutputFloor(config?: OutputHeadroomConfig): number;
+export declare function resolveOutputHeadroomBuffer(config?: OutputHeadroomConfig): number;
+export declare function resolveContextWindow(profile: ModelProfile, registryModel?: Model<Api>): number;
+export declare function resolveMaxOutputCap(profile: ModelProfile, registryModel?: Model<Api>): number;
+/**
+ * Compute delegatable maxTokens and whether output headroom meets the floor.
+ *
+ * Validates `estimatedInputTokens + minOutputFloor <= contextWindow` and returns
+ * `maxTokens = min(maxOutputCap, contextWindow - estimatedInput - buffer)`.
+ */
+export declare function computeOutputHeadroom(profile: ModelProfile, estimatedInputTokens: number, config?: OutputHeadroomConfig, registryModel?: Model<Api>): OutputHeadroomResult;
+//# sourceMappingURL=output-headroom.d.ts.map

package/dist/domain/delegation/output-headroom.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"output-headroom.d.ts","sourceRoot":"","sources":["../../../src/domain/delegation/output-headroom.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,KAAK,EAAE,GAAG,EAAE,KAAK,EAAE,MAAM,8BAA8B,CAAC;AAE/D,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,mBAAmB,CAAC;AAEtD,eAAO,MAAM,sBAAsB,MAAM,CAAC;AAC1C,eAAO,MAAM,8BAA8B,KAAK,CAAC;AAEjD,eAAO,MAAM,0BAA0B,2BAA2B,CAAC;AACnE,eAAO,MAAM,0BAA0B,2BAA2B,CAAC;AAEnE,MAAM,WAAW,oBAAoB;IACnC,QAAQ,CAAC,cAAc,CAAC,EAAE,MAAM,CAAC;IACjC,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,CAAC;CAC1B;AAED,MAAM,WAAW,iBAAiB;IAChC,QAAQ,CAAC,IAAI,EAAE,KAAK,CAAC;IACrB,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,aAAa,EAAE,MAAM,CAAC;CAChC;AAED,MAAM,WAAW,mBAAmB;IAClC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC;IACxB,QAAQ,CAAC,aAAa,EAAE,MAAM,CAAC;IAC/B,QAAQ,CAAC,qBAAqB,EAAE,MAAM,CAAC;CACxC;AAED,MAAM,MAAM,oBAAoB,GAAG,iBAAiB,GAAG,mBAAmB,CAAC;AAgB3E,wBAAgB,qBAAqB,CAAC,MAAM,CAAC,EAAE,oBAAoB,GAAG,MAAM,CAK3E;AAED,wBAAgB,2BAA2B,CAAC,MAAM,CAAC,EAAE,oBAAoB,GAAG,MAAM,CAKjF;AAED,wBAAgB,oBAAoB,CAClC,OAAO,EAAE,YAAY,EACrB,aAAa,CAAC,EAAE,KAAK,CAAC,GAAG,CAAC,GACzB,MAAM,CAMR;AAED,wBAAgB,mBAAmB,CACjC,OAAO,EAAE,YAAY,EACrB,aAAa,CAAC,EAAE,KAAK,CAAC,GAAG,CAAC,GACzB,MAAM,CAMR;AAED;;;;;GAKG;AACH,wBAAgB,qBAAqB,CACnC,OAAO,EAAE,YAAY,EACrB,oBAAoB,EAAE,MAAM,EAC5B,MAAM,CAAC,EAAE,oBAAoB,EAC7B,aAAa,CAAC,EAAE,KAAK,CAAC,GAAG,CAAC,GACzB,oBAAoB,CAiCtB"}

package/dist/domain/delegation/output-headroom.js

@@ -0,0 +1,78 @@
+/**
+ * Delegation output headroom — SP-108.
+ *
+ * Reserves output budget from the model context window before provider dispatch.
+ * Prevents stopReason "length" with zero output when input consumes the window.
+ */
+export const MIN_OUTPUT_TOKEN_FLOOR = 256;
+export const DEFAULT_OUTPUT_HEADROOM_BUFFER = 64;
+export const MIN_OUTPUT_TOKEN_FLOOR_ENV = 'MIN_OUTPUT_TOKEN_FLOOR';
+export const OUTPUT_HEADROOM_BUFFER_ENV = 'OUTPUT_HEADROOM_BUFFER';
+function readPositiveIntEnv(name, fallback) {
+    const raw = process.env[name];
+    if (raw === undefined || raw.trim() === '') {
+        return fallback;
+    }
+    const parsed = Number(raw);
+    if (!Number.isFinite(parsed) || parsed <= 0) {
+        return fallback;
+    }
+    return Math.floor(parsed);
+}
+export function resolveMinOutputFloor(config) {
+    if (config?.minOutputFloor !== undefined) {
+        return config.minOutputFloor;
+    }
+    return readPositiveIntEnv(MIN_OUTPUT_TOKEN_FLOOR_ENV, MIN_OUTPUT_TOKEN_FLOOR);
+}
+export function resolveOutputHeadroomBuffer(config) {
+    if (config?.buffer !== undefined) {
+        return config.buffer;
+    }
+    return readPositiveIntEnv(OUTPUT_HEADROOM_BUFFER_ENV, DEFAULT_OUTPUT_HEADROOM_BUFFER);
+}
+export function resolveContextWindow(profile, registryModel) {
+    return (profile.limits?.max_input_tokens ??
+        registryModel?.contextWindow ??
+        Number.MAX_SAFE_INTEGER);
+}
+export function resolveMaxOutputCap(profile, registryModel) {
+    return (profile.limits?.max_output_tokens ??
+        registryModel?.maxTokens ??
+        Number.MAX_SAFE_INTEGER);
+}
+/**
+ * Compute delegatable maxTokens and whether output headroom meets the floor.
+ *
+ * Validates `estimatedInputTokens + minOutputFloor <= contextWindow` and returns
+ * `maxTokens = min(maxOutputCap, contextWindow - estimatedInput - buffer)`.
+ */
+export function computeOutputHeadroom(profile, estimatedInputTokens, config, registryModel) {
+    const contextWindow = resolveContextWindow(profile, registryModel);
+    const minFloor = resolveMinOutputFloor(config);
+    const buffer = resolveOutputHeadroomBuffer(config);
+    const maxOutputCap = resolveMaxOutputCap(profile, registryModel);
+    if (contextWindow === Number.MAX_SAFE_INTEGER) {
+        const maxTokens = Math.min(maxOutputCap, Number.MAX_SAFE_INTEGER);
+        return { kind: 'fit', maxTokens, contextWindow };
+    }
+    const availableOutputTokens = contextWindow - estimatedInputTokens;
+    if (estimatedInputTokens + minFloor > contextWindow) {
+        return {
+            kind: 'no_fit',
+            contextWindow,
+            availableOutputTokens: Math.max(0, availableOutputTokens),
+        };
+    }
+    const rawMaxTokens = contextWindow - estimatedInputTokens - buffer;
+    const maxTokens = Math.min(maxOutputCap, rawMaxTokens);
+    if (maxTokens < minFloor) {
+        return {
+            kind: 'no_fit',
+            contextWindow,
+            availableOutputTokens: rawMaxTokens,
+        };
+    }
+    return { kind: 'fit', maxTokens, contextWindow };
+}
+//# sourceMappingURL=output-headroom.js.map