@beingmartinbmc/ojas 0.2.0

package/dist/agni/index.js

@@ -0,0 +1,328 @@
+"use strict";
+/**
+ * Ojas Agni (ओजस अग्नि) — AI Cognitive Metabolism System
+ *
+ * Regulates token budgets, tool economy, latency, and reasoning effort.
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.Agni = exports.charBasedTokenEstimator = void 0;
+const types_1 = require("../types");
+const model_router_1 = require("./model-router");
+const response_distiller_1 = require("./response-distiller");
+/**
+ * Default token estimator: `Math.ceil(text.length / 4)`. Conservative,
+ * platform-stable, no external dependency. Off by up to ~25% from real
+ * tokenizers (`cl100k_base`, `o200k_base`) on natural-language English,
+ * and worse on code, JSON, and non-English text. The number is
+ * *directionally* correct but not a model-billing prediction.
+ *
+ * Plug in a real tokenizer via `new Agni({}, { tokenEstimator: ... })`.
+ * The optional `createTiktokenEstimator()` adapter in
+ * `src/agni/tiktoken-adapter.ts` wraps OpenAI's `tiktoken` if installed.
+ */
+exports.charBasedTokenEstimator = {
+    name: 'char/4',
+    estimate(text) {
+        if (typeof text !== 'string' || text.length === 0)
+            return 0;
+        return Math.ceil(text.length / 4);
+    },
+};
+function now() {
+    return new Date().toISOString();
+}
+function clamp(v, min = 0, max = 1) {
+    return Math.max(min, Math.min(max, v));
+}
+function healthScore(value, source) {
+    return { value: clamp(value), timestamp: now(), source };
+}
+/**
+ * Serialize a value to a string for size estimation, gracefully handling
+ * circular references. A single bad trace must not crash health assessment.
+ */
+function safeStringify(value) {
+    if (value === null || value === undefined)
+        return '';
+    try {
+        return JSON.stringify(value) ?? '';
+    }
+    catch {
+        try {
+            return String(value);
+        }
+        catch {
+            return '[unserializable]';
+        }
+    }
+}
+class Agni {
+    policy;
+    traces = [];
+    tokenEstimator;
+    modelRouter;
+    responseDistiller;
+    /**
+     * Running sum of estimated tokens removed by `distillResponse`.
+     * Used by the token-negative learning-budget invariant in
+     * `lifetimeOutputTokensSaved()`.
+     */
+    outputTokensSaved = 0;
+    constructor(policy = {}, options = {}) {
+        this.policy = this.validatePolicy({ ...types_1.DEFAULT_METABOLISM_POLICY, ...policy });
+        this.tokenEstimator = options.tokenEstimator ?? exports.charBasedTokenEstimator;
+        this.modelRouter = options.modelRouter ?? new model_router_1.ConfidenceRoutingTable();
+        this.responseDistiller = options.responseDistiller ?? response_distiller_1.defaultResponseDistiller;
+    }
+    /**
+     * Identifier of the token estimator Agni is using. Useful for stamping on
+     * reports / benchmarks so the reader knows whether numbers are char/4 or
+     * a real tokenizer. Stable across calls.
+     */
+    getTokenEstimatorName() {
+        return this.tokenEstimator.name;
+    }
+    /** Identifier of the model router currently wired in. */
+    getModelRouterName() {
+        return this.modelRouter.name;
+    }
+    /** Identifier of the response distiller currently wired in. */
+    getResponseDistillerName() {
+        return this.responseDistiller.name;
+    }
+    /**
+     * Record an observed pass/fail outcome for a task class so the model
+     * router can build a confidence interval. Delegates to the configured
+     * `ModelRouter` — see `ConfidenceRoutingTable` for the default.
+     */
+    recordTaskOutcome(taskClass, succeeded) {
+        this.modelRouter.recordOutcome(taskClass, succeeded);
+    }
+    /**
+     * Recommend a routing decision (`'cheap'` vs `'flagship'`) for a task
+     * class. The router never calls an LLM itself — the caller wires the
+     * decision to their own provider gateway. Fail-closed: on insufficient
+     * data, low CI lower bound, or a configured safety class, returns
+     * `flagship`.
+     */
+    recommendRoute(taskClass) {
+        return this.modelRouter.recommend(taskClass);
+    }
+    /**
+     * Snapshot of what the router has learned, sorted by observation
+     * count (most-observed task classes first). Useful for debugging or
+     * for an operator dashboard.
+     */
+    getRouterReport() {
+        return this.modelRouter.report();
+    }
+    /**
+     * Distill an agent **output** to remove non-load-bearing filler
+     * (preserves code blocks). Updates the lifetime
+     * `outputTokensSaved` counter so callers can audit the
+     * token-negative invariant.
+     */
+    distillResponse(text) {
+        const result = this.responseDistiller.distill(text);
+        this.outputTokensSaved += result.tokensRemoved;
+        return result;
+    }
+    /**
+     * Lifetime estimated tokens saved by `distillResponse` since this
+     * Agni was constructed. Used by callers (or downstream tooling) to
+     * enforce a token-negative learning-budget invariant — i.e. a
+     * self-improvement loop should never spend more than a small fraction
+     * of lifetime savings on its own learning.
+     */
+    lifetimeOutputTokensSaved() {
+        return this.outputTokensSaved;
+    }
+    validatePolicy(policy) {
+        const finite = (v) => typeof v === 'number' && Number.isFinite(v);
+        if (!finite(policy.targetTokensPerTask) || policy.targetTokensPerTask <= 0) {
+            throw new Error('Agni: targetTokensPerTask must be a positive finite number');
+        }
+        if (!finite(policy.maxLatencyMs) || policy.maxLatencyMs <= 0) {
+            throw new Error('Agni: maxLatencyMs must be a positive finite number');
+        }
+        if (!Number.isInteger(policy.maxToolCallsPerTask) || policy.maxToolCallsPerTask <= 0) {
+            throw new Error('Agni: maxToolCallsPerTask must be a positive integer');
+        }
+        if (!finite(policy.highRiskReasoningBoost) || policy.highRiskReasoningBoost < 0 || policy.highRiskReasoningBoost > 1) {
+            throw new Error('Agni: highRiskReasoningBoost must be a finite number in [0,1]');
+        }
+        if (policy.maxTraces !== undefined && (!Number.isInteger(policy.maxTraces) || policy.maxTraces < 0)) {
+            throw new Error('Agni: maxTraces must be a non-negative integer if set');
+        }
+        if (policy.targetCostUsdPerTask !== undefined && (!finite(policy.targetCostUsdPerTask) || policy.targetCostUsdPerTask <= 0)) {
+            throw new Error('Agni: targetCostUsdPerTask must be a positive finite number if set');
+        }
+        return policy;
+    }
+    enforceTraceLimit() {
+        const cap = this.policy.maxTraces ?? 0;
+        if (cap > 0 && this.traces.length > cap) {
+            this.traces.splice(0, this.traces.length - cap);
+        }
+    }
+    recordTrace(trace) {
+        this.traces.push(trace);
+        this.enforceTraceLimit();
+    }
+    estimateReasoningEffort(riskScore, ambiguityScore) {
+        const pressure = clamp((riskScore * 0.6) + (ambiguityScore * 0.4) + this.policy.highRiskReasoningBoost);
+        if (pressure >= 0.75)
+            return 'deep';
+        if (pressure >= 0.4)
+            return 'standard';
+        return 'light';
+    }
+    assess(traces = this.traces) {
+        if (traces.length === 0) {
+            return {
+                tokenEfficiency: healthScore(0.5, 'agni.tokenEfficiency'),
+                toolEconomy: healthScore(0.5, 'agni.toolEconomy'),
+                reasoningEfficiency: healthScore(0.5, 'agni.reasoningEfficiency'),
+                latencyEfficiency: healthScore(0.5, 'agni.latencyEfficiency'),
+                costPressure: 0,
+            };
+        }
+        const avgLatency = traces.reduce((sum, trace) => sum + trace.durationMs, 0) / traces.length;
+        const failureRate = traces.filter((trace) => !trace.success).length / traces.length;
+        const reasoningWaste = traces.filter((trace) => (trace.reasoning?.length ?? 0) > 4000 && !trace.success).length / traces.length;
+        // Prefer explicit telemetry. Fall back to input+output token components,
+        // then finally to the configured `tokenEstimator` over the serialized
+        // payload + reasoning text. The estimator is `charBasedTokenEstimator`
+        // by default; pass a real tokenizer via `new Agni({}, {tokenEstimator})`
+        // (e.g. `createTiktokenEstimator('cl100k_base')`) for production-grade
+        // cost claims.
+        const tokensPerTrace = traces.map((trace) => {
+            if (typeof trace.tokensUsed === 'number' && Number.isFinite(trace.tokensUsed) && trace.tokensUsed >= 0) {
+                return trace.tokensUsed;
+            }
+            const componentTokens = (trace.inputTokens ?? 0) + (trace.outputTokens ?? 0);
+            if (componentTokens > 0 && Number.isFinite(componentTokens))
+                return componentTokens;
+            const inputStr = typeof trace.input === 'string' ? trace.input : safeStringify(trace.input);
+            const outputStr = typeof trace.output === 'string' ? trace.output : safeStringify(trace.output);
+            const reasoningStr = trace.reasoning ?? '';
+            // Use the configured estimator. We sum independently so the estimator
+            // sees natural sub-strings rather than one concatenated blob — most
+            // tokenizers are sensitive to whitespace and joining can over-count.
+            let est = 0;
+            try {
+                est = this.tokenEstimator.estimate(inputStr)
+                    + this.tokenEstimator.estimate(outputStr)
+                    + this.tokenEstimator.estimate(reasoningStr);
+            }
+            catch {
+                // Contract says estimators must not throw; if one does, fall back to
+                // char/4 so a single bad estimator can't crash health assessment.
+                est = exports.charBasedTokenEstimator.estimate(inputStr + outputStr + reasoningStr);
+            }
+            return est;
+        });
+        const avgTokens = tokensPerTrace.reduce((s, n) => s + n, 0) / tokensPerTrace.length;
+        const avgToolCalls = traces.reduce((sum, trace) => {
+            const n = trace.toolCalls;
+            return sum + (typeof n === 'number' && Number.isFinite(n) && n >= 0 ? n : 0);
+        }, 0) / traces.length;
+        const avgCostUsd = traces.reduce((sum, trace) => {
+            const n = trace.costUsd;
+            return sum + (typeof n === 'number' && Number.isFinite(n) && n >= 0 ? n : 0);
+        }, 0) / traces.length;
+        const tokenEfficiency = 1 - Math.max(0, avgTokens - this.policy.targetTokensPerTask) / this.policy.targetTokensPerTask;
+        const latencyEfficiency = 1 - Math.max(0, avgLatency - this.policy.maxLatencyMs) / this.policy.maxLatencyMs;
+        const toolCallEfficiency = 1 - Math.max(0, avgToolCalls - this.policy.maxToolCallsPerTask) / this.policy.maxToolCallsPerTask;
+        const toolEconomy = Math.min(1 - failureRate, toolCallEfficiency);
+        const reasoningEfficiency = 1 - reasoningWaste;
+        const costComponent = this.policy.targetCostUsdPerTask
+            ? Math.min(1, avgCostUsd / this.policy.targetCostUsdPerTask)
+            : 0;
+        const costPressure = clamp((avgTokens / this.policy.targetTokensPerTask) * 0.4 +
+            (avgLatency / this.policy.maxLatencyMs) * 0.25 +
+            (avgToolCalls / this.policy.maxToolCallsPerTask) * 0.15 +
+            failureRate * 0.15 +
+            costComponent * 0.05);
+        return {
+            tokenEfficiency: healthScore(tokenEfficiency, 'agni.tokenEfficiency'),
+            toolEconomy: healthScore(toolEconomy, 'agni.toolEconomy'),
+            reasoningEfficiency: healthScore(reasoningEfficiency, 'agni.reasoningEfficiency'),
+            latencyEfficiency: healthScore(latencyEfficiency, 'agni.latencyEfficiency'),
+            costPressure,
+        };
+    }
+    recommend(traces = this.traces) {
+        const health = this.assess(traces);
+        const recs = [];
+        if (health.tokenEfficiency.value < 0.6) {
+            recs.push({
+                module: 'agni',
+                severity: 'warning',
+                message: 'Token metabolism is inefficient.',
+                action: 'Compress context, reduce retrieval volume, and summarize verbose tool outputs.',
+            });
+        }
+        if (health.latencyEfficiency.value < 0.6) {
+            recs.push({
+                module: 'agni',
+                severity: 'warning',
+                message: 'Latency metabolism is degraded.',
+                action: 'Reduce tool fan-out and prefer cached or lower-latency sources.',
+            });
+        }
+        if (health.costPressure > 0.85) {
+            recs.push({
+                module: 'agni',
+                severity: 'critical',
+                message: 'Cognitive cost pressure is critical.',
+                action: 'Switch to conservative mode and cap reasoning depth.',
+            });
+        }
+        if (recs.length === 0) {
+            recs.push({ module: 'agni', severity: 'info', message: 'Cognitive metabolism is efficient.' });
+        }
+        return recs;
+    }
+    getPolicy() {
+        return { ...this.policy };
+    }
+    /** Defensive copy of retained traces. Useful for introspecting retention. */
+    getTraces() {
+        return this.traces.map((t) => ({
+            ...t,
+            failures: t.failures?.map((f) => ({ ...f, metadata: f.metadata ? { ...f.metadata } : undefined })),
+        }));
+    }
+    exportState() {
+        return {
+            traces: [...this.getTraces()],
+            outputTokensSaved: this.outputTokensSaved,
+            routerReport: this.getRouterReport().map((r) => ({
+                taskClass: r.taskClass,
+                n: r.n,
+                successes: r.successes,
+            })),
+        };
+    }
+    importState(snapshot) {
+        if (!snapshot)
+            return;
+        this.traces = (snapshot.traces ?? []).map((t) => ({
+            ...t,
+            failures: t.failures?.map((f) => ({ ...f, metadata: f.metadata ? { ...f.metadata } : undefined })),
+        }));
+        this.outputTokensSaved = snapshot.outputTokensSaved ?? 0;
+        for (const row of snapshot.routerReport ?? []) {
+            for (let i = 0; i < row.n; i += 1) {
+                this.modelRouter.recordOutcome(row.taskClass, i < row.successes);
+            }
+        }
+    }
+    updatePolicy(updates) {
+        this.policy = this.validatePolicy({ ...this.policy, ...updates });
+        this.enforceTraceLimit();
+    }
+}
+exports.Agni = Agni;
+//# sourceMappingURL=index.js.map

package/dist/agni/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/agni/index.ts"],"names":[],"mappings":";AAAA;;;;GAIG;;;AAEH,oCAYkB;AAClB,iDAAwD;AACxD,6DAAgE;AAGhE;;;;;;;;;;GAUG;AACU,QAAA,uBAAuB,GAAmB;IACrD,IAAI,EAAE,QAAQ;IACd,QAAQ,CAAC,IAAY;QACnB,IAAI,OAAO,IAAI,KAAK,QAAQ,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,CAAC,CAAC;QAC5D,OAAO,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IACpC,CAAC;CACF,CAAC;AAEF,SAAS,GAAG;IACV,OAAO,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;AAClC,CAAC;AAED,SAAS,KAAK,CAAC,CAAS,EAAE,GAAG,GAAG,CAAC,EAAE,GAAG,GAAG,CAAC;IACxC,OAAO,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC;AACzC,CAAC;AAED,SAAS,WAAW,CAAC,KAAa,EAAE,MAAc;IAChD,OAAO,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,CAAC,EAAE,SAAS,EAAE,GAAG,EAAE,EAAE,MAAM,EAAE,CAAC;AAC3D,CAAC;AAED;;;GAGG;AACH,SAAS,aAAa,CAAC,KAAc;IACnC,IAAI,KAAK,KAAK,IAAI,IAAI,KAAK,KAAK,SAAS;QAAE,OAAO,EAAE,CAAC;IACrD,IAAI,CAAC;QACH,OAAO,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC;IACrC,CAAC;IAAC,MAAM,CAAC;QACP,IAAI,CAAC;YACH,OAAO,MAAM,CAAC,KAAK,CAAC,CAAC;QACvB,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,kBAAkB,CAAC;QAC5B,CAAC;IACH,CAAC;AACH,CAAC;AAyBD,MAAa,IAAI;IACP,MAAM,CAAmB;IACzB,MAAM,GAAqB,EAAE,CAAC;IACrB,cAAc,CAAiB;IAC/B,WAAW,CAAc;IACzB,iBAAiB,CAAoB;IAEtD;;;;OAIG;IACK,iBAAiB,GAAG,CAAC,CAAC;IAE9B,YAAY,SAAoC,EAAE,EAAE,UAAuB,EAAE;QAC3E,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,cAAc,CAAC,EAAE,GAAG,iCAAyB,EAAE,GAAG,MAAM,EAAE,CAAC,CAAC;QAC/E,IAAI,CAAC,cAAc,GAAG,OAAO,CAAC,cAAc,IAAI,+BAAuB,CAAC;QACxE,IAAI,CAAC,WAAW,GAAG,OAAO,CAAC,WAAW,IAAI,IAAI,qCAAsB,EAAE,CAAC;QACvE,IAAI,CAAC,iBAAiB,GAAG,OAAO,CAAC,iBAAiB,IAAI,6CAAwB,CAAC;IACjF,CAAC;IAED;;;;OAIG;IACH,qBAAqB;QACnB,OAAO,IAAI,CAAC,cAAc,CAAC,IAAI,CAAC;IAClC,CAAC;IAED,yDAAyD;IACzD,kBAAkB;QAChB,OAAO,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC;IAC/B,CAAC;IAED,+DAA+D;IAC/D,wBAAwB;QACtB,OAAO,IAAI,CAAC,iBAAiB,CAAC,IAAI,CAAC;IACrC,CAAC;IAED;;;;OAIG;IACH,iBAAiB,CAAC,SAAiB,EAAE,SAAkB;QACrD,IAAI,CAAC,WAAW,CAAC,aAAa,CAAC,SAAS,EAAE,SAAS,CAAC,CAAC;IACvD,CAAC;IAED;;;;;;OAMG;IACH,cAAc,CAAC,SAAiB;QAC9B,OAAO,IAAI,CAAC,WAAW,CAAC,SAAS,CAAC,SAAS,CAAC,CAAC;IAC/C,CAAC;IAED;;;;OAIG;IACH,eAAe;QACb,OAAO,IAAI,CAAC,WAAW,CAAC,MAAM,EAAE,CAAC;IACnC,CAAC;IAED;;;;;OAKG;IACH,eAAe,CAAC,IAAY;QAC1B,MAAM,MAAM,GAAG,IAAI,CAAC,iBAAiB,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;QACpD,IAAI,CAAC,iBAAiB,IAAI,MAAM,CAAC,aAAa,CAAC;QAC/C,OAAO,MAAM,CAAC;IAChB,CAAC;IAED;;;;;;OAMG;IACH,yBAAyB;QACvB,OAAO,IAAI,CAAC,iBAAiB,CAAC;IAChC,CAAC;IAEO,cAAc,CAAC,MAAwB;QAC7C,MAAM,MAAM,GAAG,CAAC,CAAS,EAAE,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC;QAC1E,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,mBAAmB,CAAC,IAAI,MAAM,CAAC,mBAAmB,IAAI,CAAC,EAAE,CAAC;YAC3E,MAAM,IAAI,KAAK,CAAC,4DAA4D,CAAC,CAAC;QAChF,CAAC;QACD,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,YAAY,CAAC,IAAI,MAAM,CAAC,YAAY,IAAI,CAAC,EAAE,CAAC;YAC7D,MAAM,IAAI,KAAK,CAAC,qDAAqD,CAAC,CAAC;QACzE,CAAC;QACD,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,mBAAmB,CAAC,IAAI,MAAM,CAAC,mBAAmB,IAAI,CAAC,EAAE,CAAC;YACrF,MAAM,IAAI,KAAK,CAAC,sDAAsD,CAAC,CAAC;QAC1E,CAAC;QACD,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,sBAAsB,CAAC,IAAI,MAAM,CAAC,sBAAsB,GAAG,CAAC,IAAI,MAAM,CAAC,sBAAsB,GAAG,CAAC,EAAE,CAAC;YACrH,MAAM,IAAI,KAAK,CAAC,+DAA+D,CAAC,CAAC;QACnF,CAAC;QACD,IAAI,MAAM,CAAC,SAAS,KAAK,SAAS,IAAI,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,MAAM,CAAC,SAAS,GAAG,CAAC,CAAC,EAAE,CAAC;YACpG,MAAM,IAAI,KAAK,CAAC,uDAAuD,CAAC,CAAC;QAC3E,CAAC;QACD,IAAI,MAAM,CAAC,oBAAoB,KAAK,SAAS,IAAI,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,oBAAoB,CAAC,IAAI,MAAM,CAAC,oBAAoB,IAAI,CAAC,CAAC,EAAE,CAAC;YAC5H,MAAM,IAAI,KAAK,CAAC,oEAAoE,CAAC,CAAC;QACxF,CAAC;QACD,OAAO,MAAM,CAAC;IAChB,CAAC;IAEO,iBAAiB;QACvB,MAAM,GAAG,GAAG,IAAI,CAAC,MAAM,CAAC,SAAS,IAAI,CAAC,CAAC;QACvC,IAAI,GAAG,GAAG,CAAC,IAAI,IAAI,CAAC,MAAM,CAAC,MAAM,GAAG,GAAG,EAAE,CAAC;YACxC,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,EAAE,IAAI,CAAC,MAAM,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC;QAClD,CAAC;IACH,CAAC;IAED,WAAW,CAAC,KAAqB;QAC/B,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACxB,IAAI,CAAC,iBAAiB,EAAE,CAAC;IAC3B,CAAC;IAED,uBAAuB,CAAC,SAAiB,EAAE,cAAsB;QAC/D,MAAM,QAAQ,GAAG,KAAK,CAAC,CAAC,SAAS,GAAG,GAAG,CAAC,GAAG,CAAC,cAAc,GAAG,GAAG,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC,sBAAsB,CAAC,CAAC;QACxG,IAAI,QAAQ,IAAI,IAAI;YAAE,OAAO,MAAM,CAAC;QACpC,IAAI,QAAQ,IAAI,GAAG;YAAE,OAAO,UAAU,CAAC;QACvC,OAAO,OAAO,CAAC;IACjB,CAAC;IAED,MAAM,CAAC,SAA2B,IAAI,CAAC,MAAM;QAC3C,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACxB,OAAO;gBACL,eAAe,EAAE,WAAW,CAAC,GAAG,EAAE,sBAAsB,CAAC;gBACzD,WAAW,EAAE,WAAW,CAAC,GAAG,EAAE,kBAAkB,CAAC;gBACjD,mBAAmB,EAAE,WAAW,CAAC,GAAG,EAAE,0BAA0B,CAAC;gBACjE,iBAAiB,EAAE,WAAW,CAAC,GAAG,EAAE,wBAAwB,CAAC;gBAC7D,YAAY,EAAE,CAAC;aAChB,CAAC;QACJ,CAAC;QAED,MAAM,UAAU,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,CAAC,UAAU,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;QAC5F,MAAM,WAAW,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC;QACpF,MAAM,cAAc,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,SAAS,EAAE,MAAM,IAAI,CAAC,CAAC,GAAG,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC;QAEhI,yEAAyE;QACzE,sEAAsE;QACtE,uEAAuE;QACvE,yEAAyE;QACzE,uEAAuE;QACvE,eAAe;QACf,MAAM,cAAc,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE;YAC1C,IAAI,OAAO,KAAK,CAAC,UAAU,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,UAAU,CAAC,IAAI,KAAK,CAAC,UAAU,IAAI,CAAC,EAAE,CAAC;gBACvG,OAAO,KAAK,CAAC,UAAU,CAAC;YAC1B,CAAC;YACD,MAAM,eAAe,GAAG,CAAC,KAAK,CAAC,WAAW,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,YAAY,IAAI,CAAC,CAAC,CAAC;YAC7E,IAAI,eAAe,GAAG,CAAC,IAAI,MAAM,CAAC,QAAQ,CAAC,eAAe,CAAC;gBAAE,OAAO,eAAe,CAAC;YACpF,MAAM,QAAQ,GAAG,OAAO,KAAK,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,aAAa,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;YAC5F,MAAM,SAAS,GAAG,OAAO,KAAK,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,aAAa,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;YAChG,MAAM,YAAY,GAAG,KAAK,CAAC,SAAS,IAAI,EAAE,CAAC;YAC3C,sEAAsE;YACtE,oEAAoE;YACpE,qEAAqE;YACrE,IAAI,GAAG,GAAG,CAAC,CAAC;YACZ,IAAI,CAAC;gBACH,GAAG,GAAG,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,QAAQ,CAAC;sBACxC,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,SAAS,CAAC;sBACvC,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,YAAY,CAAC,CAAC;YACjD,CAAC;YAAC,MAAM,CAAC;gBACP,qEAAqE;gBACrE,kEAAkE;gBAClE,GAAG,GAAG,+BAAuB,CAAC,QAAQ,CAAC,QAAQ,GAAG,SAAS,GAAG,YAAY,CAAC,CAAC;YAC9E,CAAC;YACD,OAAO,GAAG,CAAC;QACb,CAAC,CAAC,CAAC;QACH,MAAM,SAAS,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,cAAc,CAAC,MAAM,CAAC;QACpF,MAAM,YAAY,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE;YAChD,MAAM,CAAC,GAAG,KAAK,CAAC,SAAS,CAAC;YAC1B,OAAO,GAAG,GAAG,CAAC,OAAO,CAAC,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QAC/E,CAAC,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;QACtB,MAAM,UAAU,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE;YAC9C,MAAM,CAAC,GAAG,KAAK,CAAC,OAAO,CAAC;YACxB,OAAO,GAAG,GAAG,CAAC,OAAO,CAAC,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QAC/E,CAAC,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;QAEtB,MAAM,eAAe,GAAG,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,SAAS,GAAG,IAAI,CAAC,MAAM,CAAC,mBAAmB,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC,mBAAmB,CAAC;QACvH,MAAM,iBAAiB,GAAG,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,UAAU,GAAG,IAAI,CAAC,MAAM,CAAC,YAAY,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC,YAAY,CAAC;QAC5G,MAAM,kBAAkB,GAAG,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,YAAY,GAAG,IAAI,CAAC,MAAM,CAAC,mBAAmB,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC,mBAAmB,CAAC;QAC7H,MAAM,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,WAAW,EAAE,kBAAkB,CAAC,CAAC;QAClE,MAAM,mBAAmB,GAAG,CAAC,GAAG,cAAc,CAAC;QAC/C,MAAM,aAAa,GAAG,IAAI,CAAC,MAAM,CAAC,oBAAoB;YACpD,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,UAAU,GAAG,IAAI,CAAC,MAAM,CAAC,oBAAoB,CAAC;YAC5D,CAAC,CAAC,CAAC,CAAC;QACN,MAAM,YAAY,GAAG,KAAK,CACxB,CAAC,SAAS,GAAG,IAAI,CAAC,MAAM,CAAC,mBAAmB,CAAC,GAAG,GAAG;YACnD,CAAC,UAAU,GAAG,IAAI,CAAC,MAAM,CAAC,YAAY,CAAC,GAAG,IAAI;YAC9C,CAAC,YAAY,GAAG,IAAI,CAAC,MAAM,CAAC,mBAAmB,CAAC,GAAG,IAAI;YACvD,WAAW,GAAG,IAAI;YAClB,aAAa,GAAG,IAAI,CACrB,CAAC;QAEF,OAAO;YACL,eAAe,EAAE,WAAW,CAAC,eAAe,EAAE,sBAAsB,CAAC;YACrE,WAAW,EAAE,WAAW,CAAC,WAAW,EAAE,kBAAkB,CAAC;YACzD,mBAAmB,EAAE,WAAW,CAAC,mBAAmB,EAAE,0BAA0B,CAAC;YACjF,iBAAiB,EAAE,WAAW,CAAC,iBAAiB,EAAE,wBAAwB,CAAC;YAC3E,YAAY;SACb,CAAC;IACJ,CAAC;IAED,SAAS,CAAC,SAA2B,IAAI,CAAC,MAAM;QAC9C,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QACnC,MAAM,IAAI,GAA2B,EAAE,CAAC;QAExC,IAAI,MAAM,CAAC,eAAe,CAAC,KAAK,GAAG,GAAG,EAAE,CAAC;YACvC,IAAI,CAAC,IAAI,CAAC;gBACR,MAAM,EAAE,MAAM;gBACd,QAAQ,EAAE,SAAS;gBACnB,OAAO,EAAE,kCAAkC;gBAC3C,MAAM,EAAE,gFAAgF;aACzF,CAAC,CAAC;QACL,CAAC;QAED,IAAI,MAAM,CAAC,iBAAiB,CAAC,KAAK,GAAG,GAAG,EAAE,CAAC;YACzC,IAAI,CAAC,IAAI,CAAC;gBACR,MAAM,EAAE,MAAM;gBACd,QAAQ,EAAE,SAAS;gBACnB,OAAO,EAAE,iCAAiC;gBAC1C,MAAM,EAAE,iEAAiE;aAC1E,CAAC,CAAC;QACL,CAAC;QAED,IAAI,MAAM,CAAC,YAAY,GAAG,IAAI,EAAE,CAAC;YAC/B,IAAI,CAAC,IAAI,CAAC;gBACR,MAAM,EAAE,MAAM;gBACd,QAAQ,EAAE,UAAU;gBACpB,OAAO,EAAE,sCAAsC;gBAC/C,MAAM,EAAE,sDAAsD;aAC/D,CAAC,CAAC;QACL,CAAC;QAED,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACtB,IAAI,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,oCAAoC,EAAE,CAAC,CAAC;QACjG,CAAC;QAED,OAAO,IAAI,CAAC;IACd,CAAC;IAED,SAAS;QACP,OAAO,EAAE,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC;IAC5B,CAAC;IAED,6EAA6E;IAC7E,SAAS;QACP,OAAO,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YAC7B,GAAG,CAAC;YACJ,QAAQ,EAAE,CAAC,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,QAAQ,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC;SACnG,CAAC,CAAC,CAAC;IACN,CAAC;IAED,WAAW;QACT,OAAO;YACL,MAAM,EAAE,CAAC,GAAG,IAAI,CAAC,SAAS,EAAE,CAAC;YAC7B,iBAAiB,EAAE,IAAI,CAAC,iBAAiB;YACzC,YAAY,EAAE,IAAI,CAAC,eAAe,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;gBAC/C,SAAS,EAAE,CAAC,CAAC,SAAS;gBACtB,CAAC,EAAE,CAAC,CAAC,CAAC;gBACN,SAAS,EAAE,CAAC,CAAC,SAAS;aACvB,CAAC,CAAC;SACJ,CAAC;IACJ,CAAC;IAED,WAAW,CAAC,QAAgD;QAC1D,IAAI,CAAC,QAAQ;YAAE,OAAO;QACtB,IAAI,CAAC,MAAM,GAAG,CAAC,QAAQ,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YAChD,GAAG,CAAC;YACJ,QAAQ,EAAE,CAAC,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,QAAQ,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC;SACnG,CAAC,CAAC,CAAC;QACJ,IAAI,CAAC,iBAAiB,GAAG,QAAQ,CAAC,iBAAiB,IAAI,CAAC,CAAC;QACzD,KAAK,MAAM,GAAG,IAAI,QAAQ,CAAC,YAAY,IAAI,EAAE,EAAE,CAAC;YAC9C,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC;gBAClC,IAAI,CAAC,WAAW,CAAC,aAAa,CAAC,GAAG,CAAC,SAAS,EAAE,CAAC,GAAG,GAAG,CAAC,SAAS,CAAC,CAAC;YACnE,CAAC;QACH,CAAC;IACH,CAAC;IAED,YAAY,CAAC,OAAkC;QAC7C,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,cAAc,CAAC,EAAE,GAAG,IAAI,CAAC,MAAM,EAAE,GAAG,OAAO,EAAE,CAAC,CAAC;QAClE,IAAI,CAAC,iBAAiB,EAAE,CAAC;IAC3B,CAAC;CACF;AAtSD,oBAsSC"}

package/dist/agni/model-router.d.ts

@@ -0,0 +1,77 @@
+/**
+ * `ConfidenceRoutingTable` — default `ModelRouter` implementation.
+ *
+ * The idea: many tasks are routinely simple (running tests, formatting
+ * code, looking up a config) and don't need a flagship model. The
+ * router watches honest pass/fail outcomes per task class, computes a
+ * Wilson 95% CI on the success rate, and only recommends routing to
+ * the cheap model when the *lower bound* of the CI exceeds a
+ * configured threshold. That's "fail-closed": if data is sparse, the
+ * task class stays on the flagship model.
+ *
+ * Why Wilson instead of bootstrap (which we use elsewhere in
+ * benchmarks): the routing decision is a real-time per-task call, so
+ * an analytic CI is preferred over a 1000-resample bootstrap. Wilson
+ * has good coverage even at small n (≥ 10), which matters when a task
+ * class has just been observed for the first time.
+ *
+ * **Hard-coded safety classes**: certain task classes are never routed
+ * to cheap regardless of observed success rate. By default these are
+ * `'security'` and `'auth'`. The caller can override the list via the
+ * policy.
+ *
+ * The router itself never calls an LLM. It returns a *recommendation*;
+ * wiring it to your provider gateway is the caller's job.
+ */
+import type { ModelRouter, ModelRouterDecision, ModelRouterTaskReport } from '../types';
+export interface ConfidenceRoutingTablePolicy {
+    /**
+     * Lower-bound success-rate threshold below which cheap routing is
+     * not recommended. Default 0.80. Tighten this if a quality regression
+     * is more expensive than a few extra dollars of token spend.
+     */
+    successRateThreshold?: number;
+    /**
+     * Minimum observation count before *any* cheap routing is allowed.
+     * Default 10. Below this, every recommendation falls back to
+     * `flagship` with reason `insufficient_data`.
+     */
+    minObservations?: number;
+    /**
+     * Task classes that must always route to flagship regardless of
+     * observed success. Default `['security', 'auth']`. Matched
+     * case-insensitively as a prefix of the supplied taskClass.
+     */
+    safetyClasses?: string[];
+    /**
+     * CI coverage level. Default 0.95. Reduce for a looser bound (more
+     * cheap routing); raise for a tighter bound (more conservative).
+     */
+    ciLevel?: number;
+}
+/**
+ * Wilson score interval for a binomial proportion.
+ *
+ * Given `k` successes out of `n` trials, returns the [lo, hi] bounds at
+ * coverage `level` (default 0.95). For `n === 0` returns `null`.
+ *
+ * We use the standard form (Wilson 1927) — Agresti–Coull is the other
+ * popular choice but Wilson is more conservative at low n, which is
+ * what we want for fail-closed routing.
+ */
+export declare function wilsonInterval(successes: number, n: number, level?: number): {
+    lo: number;
+    hi: number;
+} | null;
+export declare class ConfidenceRoutingTable implements ModelRouter {
+    readonly name = "confidence-routing-table/wilson";
+    private readonly policy;
+    private readonly buckets;
+    constructor(policy?: ConfidenceRoutingTablePolicy);
+    recordOutcome(taskClass: string, succeeded: boolean): void;
+    recommend(taskClass: string): ModelRouterDecision;
+    report(): ReadonlyArray<ModelRouterTaskReport>;
+    private successCI;
+    private isSafetyClass;
+}
+//# sourceMappingURL=model-router.d.ts.map

package/dist/agni/model-router.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"model-router.d.ts","sourceRoot":"","sources":["../../src/agni/model-router.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AAEH,OAAO,KAAK,EACV,WAAW,EACX,mBAAmB,EACnB,qBAAqB,EACtB,MAAM,UAAU,CAAC;AAElB,MAAM,WAAW,4BAA4B;IAC3C;;;;OAIG;IACH,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B;;;;OAIG;IACH,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB;;;;OAIG;IACH,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;IACzB;;;OAGG;IACH,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB;AAOD;;;;;;;;;GASG;AACH,wBAAgB,cAAc,CAC5B,SAAS,EAAE,MAAM,EACjB,CAAC,EAAE,MAAM,EACT,KAAK,SAAO,GACX;IAAE,EAAE,EAAE,MAAM,CAAC;IAAC,EAAE,EAAE,MAAM,CAAA;CAAE,GAAG,IAAI,CAenC;AAED,qBAAa,sBAAuB,YAAW,WAAW;IACxD,QAAQ,CAAC,IAAI,qCAAqC;IAClD,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAyC;IAChE,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAiC;gBAE7C,MAAM,GAAE,4BAAiC;IASrD,aAAa,CAAC,SAAS,EAAE,MAAM,EAAE,SAAS,EAAE,OAAO,GAAG,IAAI;IAS1D,SAAS,CAAC,SAAS,EAAE,MAAM,GAAG,mBAAmB;IAkDjD,MAAM,IAAI,aAAa,CAAC,qBAAqB,CAAC;IAiB9C,OAAO,CAAC,SAAS;IAMjB,OAAO,CAAC,aAAa;CAMtB"}

package/dist/agni/model-router.js

@@ -0,0 +1,163 @@
+"use strict";
+/**
+ * `ConfidenceRoutingTable` — default `ModelRouter` implementation.
+ *
+ * The idea: many tasks are routinely simple (running tests, formatting
+ * code, looking up a config) and don't need a flagship model. The
+ * router watches honest pass/fail outcomes per task class, computes a
+ * Wilson 95% CI on the success rate, and only recommends routing to
+ * the cheap model when the *lower bound* of the CI exceeds a
+ * configured threshold. That's "fail-closed": if data is sparse, the
+ * task class stays on the flagship model.
+ *
+ * Why Wilson instead of bootstrap (which we use elsewhere in
+ * benchmarks): the routing decision is a real-time per-task call, so
+ * an analytic CI is preferred over a 1000-resample bootstrap. Wilson
+ * has good coverage even at small n (≥ 10), which matters when a task
+ * class has just been observed for the first time.
+ *
+ * **Hard-coded safety classes**: certain task classes are never routed
+ * to cheap regardless of observed success rate. By default these are
+ * `'security'` and `'auth'`. The caller can override the list via the
+ * policy.
+ *
+ * The router itself never calls an LLM. It returns a *recommendation*;
+ * wiring it to your provider gateway is the caller's job.
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.ConfidenceRoutingTable = void 0;
+exports.wilsonInterval = wilsonInterval;
+/**
+ * Wilson score interval for a binomial proportion.
+ *
+ * Given `k` successes out of `n` trials, returns the [lo, hi] bounds at
+ * coverage `level` (default 0.95). For `n === 0` returns `null`.
+ *
+ * We use the standard form (Wilson 1927) — Agresti–Coull is the other
+ * popular choice but Wilson is more conservative at low n, which is
+ * what we want for fail-closed routing.
+ */
+function wilsonInterval(successes, n, level = 0.95) {
+    if (n <= 0)
+        return null;
+    // z = inverse normal CDF at (1 + level) / 2. We hard-code the
+    // common values to avoid pulling in a stats library.
+    const z = level === 0.95 ? 1.959963984540054 : level === 0.99 ? 2.5758293035489004 : 1.959963984540054;
+    const p = successes / n;
+    const denom = 1 + (z * z) / n;
+    const centre = p + (z * z) / (2 * n);
+    const margin = z * Math.sqrt((p * (1 - p)) / n + (z * z) / (4 * n * n));
+    const lo = (centre - margin) / denom;
+    const hi = (centre + margin) / denom;
+    return {
+        lo: Math.max(0, Math.min(1, lo)),
+        hi: Math.max(0, Math.min(1, hi)),
+    };
+}
+class ConfidenceRoutingTable {
+    name = 'confidence-routing-table/wilson';
+    policy;
+    buckets = new Map();
+    constructor(policy = {}) {
+        this.policy = {
+            successRateThreshold: clamp01(policy.successRateThreshold ?? 0.8),
+            minObservations: Math.max(1, Math.floor(policy.minObservations ?? 10)),
+            safetyClasses: policy.safetyClasses ?? ['security', 'auth'],
+            ciLevel: clamp01(policy.ciLevel ?? 0.95),
+        };
+    }
+    recordOutcome(taskClass, succeeded) {
+        const key = normaliseTaskClass(taskClass);
+        if (key === '')
+            return;
+        const bucket = this.buckets.get(key) ?? { n: 0, successes: 0 };
+        bucket.n += 1;
+        if (succeeded)
+            bucket.successes += 1;
+        this.buckets.set(key, bucket);
+    }
+    recommend(taskClass) {
+        const key = normaliseTaskClass(taskClass);
+        // Safety classes never route cheap, no matter how good the stats are.
+        if (this.isSafetyClass(key)) {
+            return {
+                route: 'flagship',
+                reason: `safety_class:${key}`,
+                n: this.buckets.get(key)?.n ?? 0,
+                successCI: this.successCI(key),
+            };
+        }
+        const bucket = this.buckets.get(key);
+        if (!bucket || bucket.n < this.policy.minObservations) {
+            return {
+                route: 'flagship',
+                reason: `insufficient_data:${bucket?.n ?? 0}_of_${this.policy.minObservations}`,
+                n: bucket?.n ?? 0,
+                successCI: bucket ? wilsonInterval(bucket.successes, bucket.n, this.policy.ciLevel) : null,
+            };
+        }
+        const ci = wilsonInterval(bucket.successes, bucket.n, this.policy.ciLevel);
+        if (!ci) {
+            // Defensive: should be impossible given `n >= minObservations >= 1`.
+            return {
+                route: 'flagship',
+                reason: 'ci_unavailable',
+                n: bucket.n,
+                successCI: null,
+            };
+        }
+        if (ci.lo >= this.policy.successRateThreshold) {
+            return {
+                route: 'cheap',
+                reason: `wilson_lo=${ci.lo.toFixed(3)}>=${this.policy.successRateThreshold.toFixed(2)}`,
+                n: bucket.n,
+                successCI: ci,
+            };
+        }
+        return {
+            route: 'flagship',
+            reason: `wilson_lo=${ci.lo.toFixed(3)}<${this.policy.successRateThreshold.toFixed(2)}`,
+            n: bucket.n,
+            successCI: ci,
+        };
+    }
+    report() {
+        const out = [];
+        for (const [taskClass, b] of this.buckets) {
+            const ci = wilsonInterval(b.successes, b.n, this.policy.ciLevel);
+            const decision = this.recommend(taskClass);
+            out.push({
+                taskClass,
+                n: b.n,
+                successes: b.successes,
+                successRate: b.n === 0 ? 0 : b.successes / b.n,
+                successCI: ci,
+                currentRoute: decision.route,
+            });
+        }
+        return out.sort((a, b) => b.n - a.n);
+    }
+    successCI(key) {
+        const b = this.buckets.get(key);
+        if (!b)
+            return null;
+        return wilsonInterval(b.successes, b.n, this.policy.ciLevel);
+    }
+    isSafetyClass(key) {
+        for (const safe of this.policy.safetyClasses) {
+            if (key.startsWith(safe.toLowerCase()))
+                return true;
+        }
+        return false;
+    }
+}
+exports.ConfidenceRoutingTable = ConfidenceRoutingTable;
+function clamp01(v) {
+    if (!Number.isFinite(v))
+        return 0;
+    return Math.max(0, Math.min(1, v));
+}
+function normaliseTaskClass(taskClass) {
+    return (taskClass ?? '').toLowerCase().trim();
+}
+//# sourceMappingURL=model-router.js.map

package/dist/agni/model-router.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"model-router.js","sourceRoot":"","sources":["../../src/agni/model-router.ts"],"names":[],"mappings":";AAAA;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;;;AAiDH,wCAmBC;AA7BD;;;;;;;;;GASG;AACH,SAAgB,cAAc,CAC5B,SAAiB,EACjB,CAAS,EACT,KAAK,GAAG,IAAI;IAEZ,IAAI,CAAC,IAAI,CAAC;QAAE,OAAO,IAAI,CAAC;IACxB,8DAA8D;IAC9D,qDAAqD;IACrD,MAAM,CAAC,GAAG,KAAK,KAAK,IAAI,CAAC,CAAC,CAAC,iBAAiB,CAAC,CAAC,CAAC,KAAK,KAAK,IAAI,CAAC,CAAC,CAAC,kBAAkB,CAAC,CAAC,CAAC,iBAAiB,CAAC;IACvG,MAAM,CAAC,GAAG,SAAS,GAAG,CAAC,CAAC;IACxB,MAAM,KAAK,GAAG,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;IAC9B,MAAM,MAAM,GAAG,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACrC,MAAM,MAAM,GAAG,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;IACxE,MAAM,EAAE,GAAG,CAAC,MAAM,GAAG,MAAM,CAAC,GAAG,KAAK,CAAC;IACrC,MAAM,EAAE,GAAG,CAAC,MAAM,GAAG,MAAM,CAAC,GAAG,KAAK,CAAC;IACrC,OAAO;QACL,EAAE,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QAChC,EAAE,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;KACjC,CAAC;AACJ,CAAC;AAED,MAAa,sBAAsB;IACxB,IAAI,GAAG,iCAAiC,CAAC;IACjC,MAAM,CAAyC;IAC/C,OAAO,GAAG,IAAI,GAAG,EAAsB,CAAC;IAEzD,YAAY,SAAuC,EAAE;QACnD,IAAI,CAAC,MAAM,GAAG;YACZ,oBAAoB,EAAE,OAAO,CAAC,MAAM,CAAC,oBAAoB,IAAI,GAAG,CAAC;YACjE,eAAe,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,eAAe,IAAI,EAAE,CAAC,CAAC;YACtE,aAAa,EAAE,MAAM,CAAC,aAAa,IAAI,CAAC,UAAU,EAAE,MAAM,CAAC;YAC3D,OAAO,EAAE,OAAO,CAAC,MAAM,CAAC,OAAO,IAAI,IAAI,CAAC;SACzC,CAAC;IACJ,CAAC;IAED,aAAa,CAAC,SAAiB,EAAE,SAAkB;QACjD,MAAM,GAAG,GAAG,kBAAkB,CAAC,SAAS,CAAC,CAAC;QAC1C,IAAI,GAAG,KAAK,EAAE;YAAE,OAAO;QACvB,MAAM,MAAM,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC;QAC/D,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC;QACd,IAAI,SAAS;YAAE,MAAM,CAAC,SAAS,IAAI,CAAC,CAAC;QACrC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;IAChC,CAAC;IAED,SAAS,CAAC,SAAiB;QACzB,MAAM,GAAG,GAAG,kBAAkB,CAAC,SAAS,CAAC,CAAC;QAE1C,sEAAsE;QACtE,IAAI,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,EAAE,CAAC;YAC5B,OAAO;gBACL,KAAK,EAAE,UAAU;gBACjB,MAAM,EAAE,gBAAgB,GAAG,EAAE;gBAC7B,CAAC,EAAE,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC;gBAChC,SAAS,EAAE,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC;aAC/B,CAAC;QACJ,CAAC;QAED,MAAM,MAAM,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACrC,IAAI,CAAC,MAAM,IAAI,MAAM,CAAC,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC,eAAe,EAAE,CAAC;YACtD,OAAO;gBACL,KAAK,EAAE,UAAU;gBACjB,MAAM,EAAE,qBAAqB,MAAM,EAAE,CAAC,IAAI,CAAC,OAAO,IAAI,CAAC,MAAM,CAAC,eAAe,EAAE;gBAC/E,CAAC,EAAE,MAAM,EAAE,CAAC,IAAI,CAAC;gBACjB,SAAS,EAAE,MAAM,CAAC,CAAC,CAAC,cAAc,CAAC,MAAM,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC,EAAE,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI;aAC3F,CAAC;QACJ,CAAC;QAED,MAAM,EAAE,GAAG,cAAc,CAAC,MAAM,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC,EAAE,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QAC3E,IAAI,CAAC,EAAE,EAAE,CAAC;YACR,qEAAqE;YACrE,OAAO;gBACL,KAAK,EAAE,UAAU;gBACjB,MAAM,EAAE,gBAAgB;gBACxB,CAAC,EAAE,MAAM,CAAC,CAAC;gBACX,SAAS,EAAE,IAAI;aAChB,CAAC;QACJ,CAAC;QAED,IAAI,EAAE,CAAC,EAAE,IAAI,IAAI,CAAC,MAAM,CAAC,oBAAoB,EAAE,CAAC;YAC9C,OAAO;gBACL,KAAK,EAAE,OAAO;gBACd,MAAM,EAAE,aAAa,EAAE,CAAC,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,IAAI,CAAC,MAAM,CAAC,oBAAoB,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE;gBACvF,CAAC,EAAE,MAAM,CAAC,CAAC;gBACX,SAAS,EAAE,EAAE;aACd,CAAC;QACJ,CAAC;QACD,OAAO;YACL,KAAK,EAAE,UAAU;YACjB,MAAM,EAAE,aAAa,EAAE,CAAC,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC,MAAM,CAAC,oBAAoB,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE;YACtF,CAAC,EAAE,MAAM,CAAC,CAAC;YACX,SAAS,EAAE,EAAE;SACd,CAAC;IACJ,CAAC;IAED,MAAM;QACJ,MAAM,GAAG,GAA4B,EAAE,CAAC;QACxC,KAAK,MAAM,CAAC,SAAS,EAAE,CAAC,CAAC,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;YAC1C,MAAM,EAAE,GAAG,cAAc,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;YACjE,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,SAAS,CAAC,CAAC;YAC3C,GAAG,CAAC,IAAI,CAAC;gBACP,SAAS;gBACT,CAAC,EAAE,CAAC,CAAC,CAAC;gBACN,SAAS,EAAE,CAAC,CAAC,SAAS;gBACtB,WAAW,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,GAAG,CAAC,CAAC,CAAC;gBAC9C,SAAS,EAAE,EAAE;gBACb,YAAY,EAAE,QAAQ,CAAC,KAAK;aAC7B,CAAC,CAAC;QACL,CAAC;QACD,OAAO,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;IACvC,CAAC;IAEO,SAAS,CAAC,GAAW;QAC3B,MAAM,CAAC,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QAChC,IAAI,CAAC,CAAC;YAAE,OAAO,IAAI,CAAC;QACpB,OAAO,cAAc,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IAC/D,CAAC;IAEO,aAAa,CAAC,GAAW;QAC/B,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,MAAM,CAAC,aAAa,EAAE,CAAC;YAC7C,IAAI,GAAG,CAAC,UAAU,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC;gBAAE,OAAO,IAAI,CAAC;QACtD,CAAC;QACD,OAAO,KAAK,CAAC;IACf,CAAC;CACF;AAtGD,wDAsGC;AAED,SAAS,OAAO,CAAC,CAAS;IACxB,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC;QAAE,OAAO,CAAC,CAAC;IAClC,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;AACrC,CAAC;AAED,SAAS,kBAAkB,CAAC,SAAiB;IAC3C,OAAO,CAAC,SAAS,IAAI,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC,IAAI,EAAE,CAAC;AAChD,CAAC"}

package/dist/agni/response-distiller.d.ts

@@ -0,0 +1,37 @@
+/**
+ * `defaultResponseDistiller` — rule-based filler stripper for agent
+ * output.
+ *
+ * Pattern: LLM responses routinely contain ~30-40% filler — "Sure,
+ * I'd be happy to help!", "Hope this helps!", chain-of-thought
+ * meta-commentary, repeated preambles. None of it is load-bearing
+ * for the caller. Stripping it
+ * saves tokens on the output side without touching what actually
+ * matters.
+ *
+ * **Code blocks are sacrosanct.** Anything between fenced code fences
+ * (``` … ```) is preserved verbatim, byte-for-byte. We also preserve
+ * indented code blocks (4-space indent) and inline single-backtick
+ * spans.
+ *
+ * Three intensity levels:
+ *   - **`lite`**: strip only the lowest-risk preamble/closer phrases
+ *     (e.g. "Sure!", "Hope this helps!"). Won't touch substance.
+ *   - **`full`** (default): adds hedging (`I think`, `It seems`),
+ *     filler intros ("Let me take a look", "Here's what I found"), and
+ *     repeated "in summary" / "to conclude" closers.
+ *   - **`ultra`**: also collapses repeated empty lines, strips
+ *     standalone Markdown emphasis on filler phrases, and removes
+ *     "Note:" / "FYI:" prefixes from individual sentences.
+ *
+ * False-positive guard: every pattern is anchored to either start-of-
+ * string, start-of-line, end-of-string, or a sentence boundary. We
+ * never strip mid-sentence content unless it's a standalone hedge
+ * (`, I think,`) — and even then only at intensity `full`+.
+ */
+import type { ResponseDistiller } from '../types';
+export type DistillerIntensity = 'lite' | 'full' | 'ultra';
+export declare function createResponseDistiller(intensity?: DistillerIntensity): ResponseDistiller;
+/** Default `ResponseDistiller` at `full` intensity. Re-export for convenience. */
+export declare const defaultResponseDistiller: ResponseDistiller;
+//# sourceMappingURL=response-distiller.d.ts.map

package/dist/agni/response-distiller.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"response-distiller.d.ts","sourceRoot":"","sources":["../../src/agni/response-distiller.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8BG;AAEH,OAAO,KAAK,EAAE,iBAAiB,EAAyB,MAAM,UAAU,CAAC;AAEzE,MAAM,MAAM,kBAAkB,GAAG,MAAM,GAAG,MAAM,GAAG,OAAO,CAAC;AA0I3D,wBAAgB,uBAAuB,CAAC,SAAS,GAAE,kBAA2B,GAAG,iBAAiB,CA+CjG;AAED,kFAAkF;AAClF,eAAO,MAAM,wBAAwB,EAAE,iBAAmD,CAAC"}