@skillstate/bench 2.0.0

package/dist/harness.d.ts

@@ -0,0 +1,81 @@
+/**
+ * Deterministic local benchmark harness: conversation baseline vs SKILL.state.
+ *
+ * @non-paper — this harness is NOT the paper's evaluation. The paper's
+ * Table 1 / §5.2 numbers come from Gemini/Gemma runs on Warehousing tasks;
+ * this file measures a minimal, fully deterministic A/B on fixed synthetic
+ * data so anyone can reproduce OUR numbers with `npm run bench`.
+ *
+ * Method (honest A/B on IDENTICAL data):
+ * - Fixed spec P (BENCH_SPEC), fixed observation Ot (BENCH_OBSERVATION,
+ *   constant length), fixed mock-LLM response (BENCH_REASONING + valid
+ *   state_patch + action, constant length). No RNG anywhere, so no seed is
+ *   needed (BENCH_SEED documents that fact).
+ * - Mode (b) SKILL.state: each step prompts with At = (P, Σt, Ot) via the
+ *   real SkillStateRuntime (paper-exact `formatPaper`). Prompt strings are
+ *   captured from the mock LLM to measure real char lengths.
+ * - Mode (a) conversation baseline: prompt[t] is the concatenation of the
+ *   state prompts for steps 0..t — i.e. every prior turn is re-sent at full
+ *   current-turn size. This is exactly the TokenTracker.compareWithBaseline
+ *   model (cumulative conv = Σt Σi≤t promptChars[i], paper §3.3 eq.5), so
+ *   with constant-size turns the closed form is reductionFactor = (T+1)/2.
+ *   Caveat (read before quoting): P is included in every re-sent turn, so
+ *   this is an UPPER bound on real-baseline savings — a production baseline
+ *   sends P once and re-sends cheaper turn payloads.
+ *
+ * Metrics are raw string chars (paper §4.3), never tokenizer output.
+ */
+import type { Observation, ProceduralSpec, StatePatch } from '@skillstate/core';
+/** No RNG is used anywhere — determinism needs no seed; this says so. */
+export declare const BENCH_SEED = "none-no-rng-deterministic";
+/** Horizon values measured by `npm run bench`. */
+export declare const BENCH_T_VALUES: readonly number[];
+/** Minimal fixed skill: one string field, constant serialized size. */
+export declare const BENCH_SPEC: ProceduralSpec;
+/** Fixed observation payload: 64 chars, identical every step. */
+export declare const BENCH_OBSERVATION: string;
+/** Fixed mock-LLM reasoning prefix (discarded, never stored). */
+export declare const BENCH_REASONING = "bench reasoning note";
+/** Fixed mock-LLM action. */
+export declare const BENCH_ACTION = "bench-action";
+/** Fixed patch: re-assigns the schema default, so Σt never changes size. */
+export declare const BENCH_PATCH: StatePatch;
+/** One fixed mock-LLM response (valid state_patch + action). */
+export declare function benchResponse(): string;
+/** Fixed observation object (timestamp pinned for determinism). */
+export declare function benchObservation(): Observation;
+/** Per-horizon benchmark outcome (all sizes in raw chars). */
+export interface BenchResult {
+    T: number;
+    seed: string;
+    /** Per-step SKILL.state prompt sizes |At| (measured). */
+    statePerStep: number[];
+    /** Per-step conversation-baseline prompt sizes (prefix sums, measured). */
+    convPerStep: number[];
+    /** Per-step raw response sizes (measured). */
+    responsePerStep: number[];
+    stateCumulative: number;
+    convCumulative: number;
+    responseCumulative: number;
+    stateAverage: number;
+    convAverage: number;
+    /** convCumulative / stateCumulative. */
+    reductionFactor: number;
+    /** statePerStep[T-1] - statePerStep[0] (flat = 0 when constant). */
+    stateSlope: number;
+    /** convPerStep[T-1] - convPerStep[0] (linear growth). */
+    convSlope: number;
+}
+/** Closed-form expectation for constant-size turns (paper §3.3 eq.5-7). */
+export declare function expectedReduction(T: number): number;
+/**
+ * Run one deterministic horizon: drive the real SkillStateRuntime for T
+ * steps on fixed data, capture every state prompt string, then build the
+ * conversation-baseline prompts as concatenations of the state prompts.
+ */
+export declare function runScenario(T: number): Promise<BenchResult>;
+/** Run every horizon in order. */
+export declare function runAll(horizons: readonly number[]): Promise<BenchResult[]>;
+/** Render results as a human-readable text table (used by `npm run bench`). */
+export declare function formatTable(results: readonly BenchResult[]): string;
+//# sourceMappingURL=harness.d.ts.map

package/dist/harness.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"harness.d.ts","sourceRoot":"","sources":["../src/harness.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AAGH,OAAO,KAAK,EACV,WAAW,EACX,cAAc,EACd,UAAU,EACX,MAAM,kBAAkB,CAAC;AAE1B,yEAAyE;AACzE,eAAO,MAAM,UAAU,8BAA8B,CAAC;AAEtD,kDAAkD;AAClD,eAAO,MAAM,cAAc,EAAE,SAAS,MAAM,EAAuB,CAAC;AAEpE,uEAAuE;AACvE,eAAO,MAAM,UAAU,EAAE,cAaxB,CAAC;AAEF,iEAAiE;AACjE,eAAO,MAAM,iBAAiB,QAAiB,CAAC;AAEhD,iEAAiE;AACjE,eAAO,MAAM,eAAe,yBAAyB,CAAC;AAEtD,6BAA6B;AAC7B,eAAO,MAAM,YAAY,iBAAiB,CAAC;AAE3C,4EAA4E;AAC5E,eAAO,MAAM,WAAW,EAAE,UAAiC,CAAC;AAE5D,gEAAgE;AAChE,wBAAgB,aAAa,IAAI,MAAM,CAKtC;AAED,mEAAmE;AACnE,wBAAgB,gBAAgB,IAAI,WAAW,CAE9C;AAED,8DAA8D;AAC9D,MAAM,WAAW,WAAW;IAC1B,CAAC,EAAE,MAAM,CAAC;IACV,IAAI,EAAE,MAAM,CAAC;IACb,yDAAyD;IACzD,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,2EAA2E;IAC3E,WAAW,EAAE,MAAM,EAAE,CAAC;IACtB,8CAA8C;IAC9C,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,eAAe,EAAE,MAAM,CAAC;IACxB,cAAc,EAAE,MAAM,CAAC;IACvB,kBAAkB,EAAE,MAAM,CAAC;IAC3B,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,wCAAwC;IACxC,eAAe,EAAE,MAAM,CAAC;IACxB,oEAAoE;IACpE,UAAU,EAAE,MAAM,CAAC;IACnB,yDAAyD;IACzD,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,2EAA2E;AAC3E,wBAAgB,iBAAiB,CAAC,CAAC,EAAE,MAAM,GAAG,MAAM,CAEnD;AAUD;;;;GAIG;AACH,wBAAsB,WAAW,CAAC,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,WAAW,CAAC,CAwCjE;AAED,kCAAkC;AAClC,wBAAsB,MAAM,CAC1B,QAAQ,EAAE,SAAS,MAAM,EAAE,GAC1B,OAAO,CAAC,WAAW,EAAE,CAAC,CAMxB;AAED,+EAA+E;AAC/E,wBAAgB,WAAW,CAAC,OAAO,EAAE,SAAS,WAAW,EAAE,GAAG,MAAM,CAQnE"}

package/dist/harness.js

@@ -0,0 +1,132 @@
+/**
+ * Deterministic local benchmark harness: conversation baseline vs SKILL.state.
+ *
+ * @non-paper — this harness is NOT the paper's evaluation. The paper's
+ * Table 1 / §5.2 numbers come from Gemini/Gemma runs on Warehousing tasks;
+ * this file measures a minimal, fully deterministic A/B on fixed synthetic
+ * data so anyone can reproduce OUR numbers with `npm run bench`.
+ *
+ * Method (honest A/B on IDENTICAL data):
+ * - Fixed spec P (BENCH_SPEC), fixed observation Ot (BENCH_OBSERVATION,
+ *   constant length), fixed mock-LLM response (BENCH_REASONING + valid
+ *   state_patch + action, constant length). No RNG anywhere, so no seed is
+ *   needed (BENCH_SEED documents that fact).
+ * - Mode (b) SKILL.state: each step prompts with At = (P, Σt, Ot) via the
+ *   real SkillStateRuntime (paper-exact `formatPaper`). Prompt strings are
+ *   captured from the mock LLM to measure real char lengths.
+ * - Mode (a) conversation baseline: prompt[t] is the concatenation of the
+ *   state prompts for steps 0..t — i.e. every prior turn is re-sent at full
+ *   current-turn size. This is exactly the TokenTracker.compareWithBaseline
+ *   model (cumulative conv = Σt Σi≤t promptChars[i], paper §3.3 eq.5), so
+ *   with constant-size turns the closed form is reductionFactor = (T+1)/2.
+ *   Caveat (read before quoting): P is included in every re-sent turn, so
+ *   this is an UPPER bound on real-baseline savings — a production baseline
+ *   sends P once and re-sends cheaper turn payloads.
+ *
+ * Metrics are raw string chars (paper §4.3), never tokenizer output.
+ */
+import { SkillStateRuntime } from '@skillstate/core';
+/** No RNG is used anywhere — determinism needs no seed; this says so. */
+export const BENCH_SEED = 'none-no-rng-deterministic';
+/** Horizon values measured by `npm run bench`. */
+export const BENCH_T_VALUES = [10, 50, 100, 200];
+/** Minimal fixed skill: one string field, constant serialized size. */
+export const BENCH_SPEC = {
+    id: 'bench-harness',
+    name: 'BenchHarness',
+    instructions: 'Hold a steady state and emit the fixed benchmark action every step.',
+    schema: {
+        status: {
+            type: 'string',
+            default: 'steady',
+            description: 'Fixed benchmark status marker',
+        },
+    },
+    version: '1.0.0',
+};
+/** Fixed observation payload: 64 chars, identical every step. */
+export const BENCH_OBSERVATION = 'o'.repeat(64);
+/** Fixed mock-LLM reasoning prefix (discarded, never stored). */
+export const BENCH_REASONING = 'bench reasoning note';
+/** Fixed mock-LLM action. */
+export const BENCH_ACTION = 'bench-action';
+/** Fixed patch: re-assigns the schema default, so Σt never changes size. */
+export const BENCH_PATCH = { status: 'steady' };
+/** One fixed mock-LLM response (valid state_patch + action). */
+export function benchResponse() {
+    return `${BENCH_REASONING}\n\n\`\`\`json\n${JSON.stringify({
+        state_patch: BENCH_PATCH,
+        action: BENCH_ACTION,
+    })}\n\`\`\``;
+}
+/** Fixed observation object (timestamp pinned for determinism). */
+export function benchObservation() {
+    return { content: BENCH_OBSERVATION, timestamp: 0, source: 'bench' };
+}
+/** Closed-form expectation for constant-size turns (paper §3.3 eq.5-7). */
+export function expectedReduction(T) {
+    return (T + 1) / 2;
+}
+function sum(values) {
+    let total = 0;
+    for (const value of values) {
+        total += value;
+    }
+    return total;
+}
+/**
+ * Run one deterministic horizon: drive the real SkillStateRuntime for T
+ * steps on fixed data, capture every state prompt string, then build the
+ * conversation-baseline prompts as concatenations of the state prompts.
+ */
+export async function runScenario(T) {
+    const promptTexts = [];
+    const responseText = benchResponse();
+    const runtime = new SkillStateRuntime({
+        spec: BENCH_SPEC,
+        llm: async (prompt) => {
+            promptTexts.push(prompt);
+            return responseText;
+        },
+        execute: async () => benchObservation(),
+    });
+    for (let step = 0; step < T; step += 1) {
+        await runtime.step(benchObservation());
+    }
+    const statePerStep = promptTexts.map((text) => text.length);
+    const convPerStep = promptTexts.map((_text, index) => promptTexts.slice(0, index + 1).join('').length);
+    const responsePerStep = promptTexts.map(() => responseText.length);
+    const stateCumulative = sum(statePerStep);
+    const convCumulative = sum(convPerStep);
+    const responseCumulative = sum(responsePerStep);
+    return {
+        T,
+        seed: BENCH_SEED,
+        statePerStep,
+        convPerStep,
+        responsePerStep,
+        stateCumulative,
+        convCumulative,
+        responseCumulative,
+        stateAverage: stateCumulative / T,
+        convAverage: convCumulative / T,
+        reductionFactor: convCumulative / stateCumulative,
+        stateSlope: statePerStep[T - 1] - statePerStep[0],
+        convSlope: convPerStep[T - 1] - convPerStep[0],
+    };
+}
+/** Run every horizon in order. */
+export async function runAll(horizons) {
+    const results = [];
+    for (const T of horizons) {
+        results.push(await runScenario(T));
+    }
+    return results;
+}
+/** Render results as a human-readable text table (used by `npm run bench`). */
+export function formatTable(results) {
+    const header = 'T | state/step | conv/step(avg) | state cum | conv cum | reduction | state slope | conv slope';
+    const rows = results.map((result) => `${result.T} | ${result.statePerStep[0]} | ${result.convAverage.toFixed(1)} | ${result.stateCumulative} | ${result.convCumulative} | ${result.reductionFactor.toFixed(2)}x (formula (T+1)/2=${expectedReduction(result.T)}) | ${result.stateSlope} | ${result.convSlope}`);
+    return [header, ...rows].join('\n');
+}
+//# sourceMappingURL=harness.js.map

package/dist/harness.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"harness.js","sourceRoot":"","sources":["../src/harness.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AAEH,OAAO,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAOrD,yEAAyE;AACzE,MAAM,CAAC,MAAM,UAAU,GAAG,2BAA2B,CAAC;AAEtD,kDAAkD;AAClD,MAAM,CAAC,MAAM,cAAc,GAAsB,CAAC,EAAE,EAAE,EAAE,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC;AAEpE,uEAAuE;AACvE,MAAM,CAAC,MAAM,UAAU,GAAmB;IACxC,EAAE,EAAE,eAAe;IACnB,IAAI,EAAE,cAAc;IACpB,YAAY,EACV,qEAAqE;IACvE,MAAM,EAAE;QACN,MAAM,EAAE;YACN,IAAI,EAAE,QAAQ;YACd,OAAO,EAAE,QAAQ;YACjB,WAAW,EAAE,+BAA+B;SAC7C;KACF;IACD,OAAO,EAAE,OAAO;CACjB,CAAC;AAEF,iEAAiE;AACjE,MAAM,CAAC,MAAM,iBAAiB,GAAG,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC;AAEhD,iEAAiE;AACjE,MAAM,CAAC,MAAM,eAAe,GAAG,sBAAsB,CAAC;AAEtD,6BAA6B;AAC7B,MAAM,CAAC,MAAM,YAAY,GAAG,cAAc,CAAC;AAE3C,4EAA4E;AAC5E,MAAM,CAAC,MAAM,WAAW,GAAe,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC;AAE5D,gEAAgE;AAChE,MAAM,UAAU,aAAa;IAC3B,OAAO,GAAG,eAAe,mBAAmB,IAAI,CAAC,SAAS,CAAC;QACzD,WAAW,EAAE,WAAW;QACxB,MAAM,EAAE,YAAY;KACrB,CAAC,UAAU,CAAC;AACf,CAAC;AAED,mEAAmE;AACnE,MAAM,UAAU,gBAAgB;IAC9B,OAAO,EAAE,OAAO,EAAE,iBAAiB,EAAE,SAAS,EAAE,CAAC,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;AACvE,CAAC;AAyBD,2EAA2E;AAC3E,MAAM,UAAU,iBAAiB,CAAC,CAAS;IACzC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;AACrB,CAAC;AAED,SAAS,GAAG,CAAC,MAAgB;IAC3B,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;QAC3B,KAAK,IAAI,KAAK,CAAC;IACjB,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,WAAW,CAAC,CAAS;IACzC,MAAM,WAAW,GAAa,EAAE,CAAC;IACjC,MAAM,YAAY,GAAG,aAAa,EAAE,CAAC;IACrC,MAAM,OAAO,GAAG,IAAI,iBAAiB,CAAC;QACpC,IAAI,EAAE,UAAU;QAChB,GAAG,EAAE,KAAK,EAAE,MAAc,EAAmB,EAAE;YAC7C,WAAW,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;YACzB,OAAO,YAAY,CAAC;QACtB,CAAC;QACD,OAAO,EAAE,KAAK,IAA0B,EAAE,CAAC,gBAAgB,EAAE;KAC9D,CAAC,CAAC;IACH,KAAK,IAAI,IAAI,GAAG,CAAC,EAAE,IAAI,GAAG,CAAC,EAAE,IAAI,IAAI,CAAC,EAAE,CAAC;QACvC,MAAM,OAAO,CAAC,IAAI,CAAC,gBAAgB,EAAE,CAAC,CAAC;IACzC,CAAC;IAED,MAAM,YAAY,GAAa,WAAW,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IACtE,MAAM,WAAW,GAAa,WAAW,CAAC,GAAG,CAC3C,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,MAAM,CAClE,CAAC;IACF,MAAM,eAAe,GAAa,WAAW,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,YAAY,CAAC,MAAM,CAAC,CAAC;IAE7E,MAAM,eAAe,GAAG,GAAG,CAAC,YAAY,CAAC,CAAC;IAC1C,MAAM,cAAc,GAAG,GAAG,CAAC,WAAW,CAAC,CAAC;IACxC,MAAM,kBAAkB,GAAG,GAAG,CAAC,eAAe,CAAC,CAAC;IAEhD,OAAO;QACL,CAAC;QACD,IAAI,EAAE,UAAU;QAChB,YAAY;QACZ,WAAW;QACX,eAAe;QACf,eAAe;QACf,cAAc;QACd,kBAAkB;QAClB,YAAY,EAAE,eAAe,GAAG,CAAC;QACjC,WAAW,EAAE,cAAc,GAAG,CAAC;QAC/B,eAAe,EAAE,cAAc,GAAG,eAAe;QACjD,UAAU,EAAE,YAAY,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,YAAY,CAAC,CAAC,CAAC;QACjD,SAAS,EAAE,WAAW,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,WAAW,CAAC,CAAC,CAAC;KAC/C,CAAC;AACJ,CAAC;AAED,kCAAkC;AAClC,MAAM,CAAC,KAAK,UAAU,MAAM,CAC1B,QAA2B;IAE3B,MAAM,OAAO,GAAkB,EAAE,CAAC;IAClC,KAAK,MAAM,CAAC,IAAI,QAAQ,EAAE,CAAC;QACzB,OAAO,CAAC,IAAI,CAAC,MAAM,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC;IACrC,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,+EAA+E;AAC/E,MAAM,UAAU,WAAW,CAAC,OAA+B;IACzD,MAAM,MAAM,GACV,+FAA+F,CAAC;IAClG,MAAM,IAAI,GAAG,OAAO,CAAC,GAAG,CACtB,CAAC,MAAM,EAAE,EAAE,CACT,GAAG,MAAM,CAAC,CAAC,MAAM,MAAM,CAAC,YAAY,CAAC,CAAC,CAAC,MAAM,MAAM,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,MAAM,CAAC,eAAe,MAAM,MAAM,CAAC,cAAc,MAAM,MAAM,CAAC,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC,sBAAsB,iBAAiB,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,MAAM,CAAC,UAAU,MAAM,MAAM,CAAC,SAAS,EAAE,CAC5Q,CAAC;IACF,OAAO,CAAC,MAAM,EAAE,GAAG,IAAI,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACtC,CAAC"}

package/dist/index.d.ts

@@ -0,0 +1,3 @@
+export * from './harness.js';
+export * from './run.js';
+//# sourceMappingURL=index.d.ts.map

package/dist/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAKA,cAAc,cAAc,CAAC;AAC7B,cAAc,UAAU,CAAC"}

package/dist/index.js

@@ -0,0 +1,8 @@
+// @skillstate/bench — deterministic local benchmark harness.
+//
+// NOTE: re-exporting `./run.js` also evaluates its top-level `main()`
+// (the `npm run bench` entry), exactly like importing the historical
+// `dist/bench/run.js` directly.
+export * from './harness.js';
+export * from './run.js';
+//# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,6DAA6D;AAC7D,EAAE;AACF,sEAAsE;AACtE,qEAAqE;AACrE,gCAAgC;AAChC,cAAc,cAAc,CAAC;AAC7B,cAAc,UAAU,CAAC"}

package/dist/run.d.ts

@@ -0,0 +1,9 @@
+/**
+ * `npm run bench` entry point (zero-deps, runs from compiled dist).
+ *
+ * @non-paper — local deterministic harness CLI, not part of the paper.
+ */
+import type { BenchResult } from './harness.js';
+/** Run all horizons, print the table plus machine-readable JSON. */
+export declare function main(): Promise<BenchResult[]>;
+//# sourceMappingURL=run.d.ts.map

package/dist/run.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../src/run.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAGH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAEhD,oEAAoE;AACpE,wBAAsB,IAAI,IAAI,OAAO,CAAC,WAAW,EAAE,CAAC,CAKnD"}

package/dist/run.js

@@ -0,0 +1,15 @@
+/**
+ * `npm run bench` entry point (zero-deps, runs from compiled dist).
+ *
+ * @non-paper — local deterministic harness CLI, not part of the paper.
+ */
+import { BENCH_T_VALUES, runAll, formatTable } from './harness.js';
+/** Run all horizons, print the table plus machine-readable JSON. */
+export async function main() {
+    const results = await runAll(BENCH_T_VALUES);
+    console.log(formatTable(results));
+    console.log(JSON.stringify(results, null, 2));
+    return results;
+}
+void main();
+//# sourceMappingURL=run.js.map

package/dist/run.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"run.js","sourceRoot":"","sources":["../src/run.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EAAE,cAAc,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAGnE,oEAAoE;AACpE,MAAM,CAAC,KAAK,UAAU,IAAI;IACxB,MAAM,OAAO,GAAG,MAAM,MAAM,CAAC,cAAc,CAAC,CAAC;IAC7C,OAAO,CAAC,GAAG,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC;IAClC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IAC9C,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,KAAK,IAAI,EAAE,CAAC"}

package/package.json

@@ -0,0 +1,24 @@
+{
+  "name": "@skillstate/bench",
+  "version": "2.0.0",
+  "description": "Deterministic local benchmark harness for the skillstate runtime (entry only).",
+  "type": "module",
+  "main": "dist/index.js",
+  "types": "dist/index.d.ts",
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "default": "./dist/index.js"
+    },
+    "./package.json": "./package.json"
+  },
+  "files": ["dist"],
+  "sideEffects": false,
+  "engines": {
+    "node": ">=20"
+  },
+  "license": "MIT",
+  "dependencies": {
+    "@skillstate/core": "^2.0.0"
+  }
+}