SebaSOFT · SebaSOFT · Jun 23, 2026 · Jun 23, 2026 · Jun 23, 2026
diff --git a/README.md b/README.md
@@ -43,12 +43,22 @@ Do not use `neuron-js` when a simple hardcoded condition is clearer and rarely c
 
 ## Public proof assets
 
-Neuron-JS proof material is published as methodology and inspectability artifacts first. Benchmark result claims remain blocked until the benchmark harness emits real measured output with `claims_allowed: true`.
+Neuron-JS proof material is published as methodology, measured benchmarks, and inspectability artifacts. Benchmark numbers come from a real `actual_benchmark` harness run only — reproduce them with `yarn benchmark`.
 
 <p align="center">
   <img src="docs/benchmarks/assets/generated/explainability-trace-diagram.svg" alt="Neuron-JS diagram showing rule JSON and business input flowing through schema validation, developer registry, deterministic Synapse evaluation, result output, explanation trace, and audit-ready decision." width="760">
 </p>
 
+### Benchmark results
+
+Measured throughput, cold-start, bundle-size, validation, and explanation overhead across the pricing, eligibility, and workflow-routing scenarios, compared against `json-rules-engine`, `json-logic-js`, a hand-coded TypeScript baseline, and `rule-engine-js`. See the full charts, results table, and provenance:
+
+- Benchmark results (charts + data + provenance): [`docs/benchmarks/results.md`](docs/benchmarks/results.md)
+- Reproduce locally: `yarn benchmark` then `yarn benchmark:charts`
+- Raw measured output: [`benchmarks/results/latest.actual.json`](benchmarks/results/latest.actual.json)
+
+### Methodology and inspectability
+
 - Benchmark methodology and result contract: [`docs/benchmarks/methodology.md`](docs/benchmarks/methodology.md)
 - Explainability proof metadata and alt text: [`docs/benchmarks/assets/generated/explainability-trace-diagram.md`](docs/benchmarks/assets/generated/explainability-trace-diagram.md)
 - Visual proof system and publication guardrails: [`docs/benchmarks/visual-proof-system.md`](docs/benchmarks/visual-proof-system.md)

diff --git a/benchmarks/charts/generate.ts b/benchmarks/charts/generate.ts
@@ -0,0 +1,293 @@
+// Renders the five benchmark charts from real harness output.
+//
+// Every visible number is read verbatim from benchmarks/results/latest.actual.json
+// (no averaging/derivation), so charts cannot drift from measured data. Comparison
+// metrics use the pricing-discount / medium row; cold-start and bundle size are
+// engine-level constants. Run: yarn benchmark:charts
+import { readFileSync, writeFileSync } from "node:fs";
+import { fileURLToPath } from "node:url";
+
+const RESULTS_PATH = fileURLToPath(
+  new URL("../results/latest.actual.json", import.meta.url),
+);
+const OUT_DIR = fileURLToPath(
+  new URL("../../docs/benchmarks/assets/generated/", import.meta.url),
+);
+const SOURCE_REF = "benchmarks/results/latest.actual.json";
+
+interface Row {
+  engine: string;
+  scenario: string;
+  input_size: string;
+  throughput_decisions_per_second: number;
+  p50_ms: number;
+  p95_ms: number;
+  cold_start_ms: number;
+  bundle_size_minified_bytes: number;
+  validation_overhead_ms: number;
+  explanation_overhead_ms: number;
+  node_version: string;
+  package_version: string;
+  commit_sha: string;
+}
+
+interface ResultsFile {
+  generated_at: string;
+  results: Row[];
+}
+
+const ENGINE_COLOR: Record<string, string> = {
+  "@sebasoft/neuron-js": "#22d3ee",
+  "json-rules-engine": "#a78bfa",
+  "json-logic-js": "#fbbf24",
+  "hand-coded-typescript": "#34d399",
+  "rule-engine-js": "#94a3b8",
+};
+const ENGINES = Object.keys(ENGINE_COLOR);
+
+const data = JSON.parse(readFileSync(RESULTS_PATH, "utf8")) as ResultsFile;
+
+function rowFor(engine: string, scenario: string, size: string): Row {
+  const row = data.results.find(
+    (item) =>
+      item.engine === engine &&
+      item.scenario === scenario &&
+      item.input_size === size,
+  );
+  if (!row) throw new Error(`Missing row ${engine}/${scenario}/${size}`);
+  return row;
+}
+
+const provenance = rowFor(ENGINES[0], "pricing-discount", "medium");
+
+function escapeXml(value: string): string {
+  return value.replace(/[<>&]/g, (c) =>
+    c === "<" ? "&lt;" : c === ">" ? "&gt;" : "&amp;",
+  );
+}
+
+const groupSep = (n: number): string => n.toLocaleString("en-US");
+
+interface ChartSpec {
+  file: string;
+  title: string;
+  metric: keyof Row & string;
+  unit: string;
+  direction: "higher is better" | "lower is better";
+  subtitle: string;
+  format: (value: number) => string;
+  /** Note rendered when most engines are zero (differentiator metrics). */
+  differentiator?: string;
+}
+
+const charts: ChartSpec[] = [
+  {
+    file: "benchmark-chart-throughput.svg",
+    title: "Throughput comparison",
+    metric: "throughput_decisions_per_second",
+    unit: "decisions / second",
+    direction: "higher is better",
+    subtitle: "pricing-discount · medium (10,000 decisions)",
+    format: (v) => `${groupSep(Math.round(v))} dec/s`,
+  },
+  {
+    file: "benchmark-chart-cold-start.svg",
+    title: "Cold-start comparison",
+    metric: "cold_start_ms",
+    unit: "milliseconds",
+    direction: "lower is better",
+    subtitle: "engine import + first decision, fresh process (median of 5)",
+    format: (v) => `${v} ms`,
+  },
+  {
+    file: "benchmark-chart-bundle-size.svg",
+    title: "Bundle-size comparison",
+    metric: "bundle_size_minified_bytes",
+    unit: "minified bytes (esbuild, node platform)",
+    direction: "lower is better",
+    subtitle: "minified bundle of the engine's public surface",
+    format: (v) =>
+      v === 0 ? "0 B (no engine dependency)" : `${groupSep(v)} B`,
+  },
+  {
+    file: "benchmark-chart-validation-overhead.svg",
+    title: "Validation overhead",
+    metric: "validation_overhead_ms",
+    unit: "milliseconds per decision",
+    direction: "lower is better",
+    subtitle: "pricing-discount · validateScript delta · medium",
+    format: (v) => `${v} ms`,
+    differentiator:
+      "Neuron-JS differentiator: competitors provide no schema-validation step (0).",
+  },
+  {
+    file: "benchmark-chart-explanation-overhead.svg",
+    title: "Explanation overhead",
+    metric: "explanation_overhead_ms",
+    unit: "milliseconds per decision",
+    direction: "lower is better",
+    subtitle: "pricing-discount · explainExecution delta · medium",
+    format: (v) => `${v} ms`,
+    differentiator:
+      "Neuron-JS differentiator: competitors provide no explanation trace (0).",
+  },
+];
+
+const W = 1200;
+const H = 675;
+const BAR_X = 250;
+const BAR_W = 760;
+const FIRST_Y = 196;
+const ROW_H = 70;
+const BAR_H = 40;
+
+// Comparison row: pricing-discount / medium. cold-start and bundle-size are
+// engine-level constants, so this row carries their values too.
+const CHART_SCENARIO = "pricing-discount";
+
+function render(spec: ChartSpec): string {
+  const rows = ENGINES.map((engine) => ({
+    engine,
+    value: rowFor(engine, CHART_SCENARIO, "medium")[spec.metric] as number,
+  }));
+  const max = Math.max(...rows.map((r) => r.value), 1);
+
+  const bars = rows
+    .map((r, i) => {
+      const y = FIRST_Y + i * ROW_H;
+      const width = max > 0 ? (r.value / max) * BAR_W : 0;
+      const color = ENGINE_COLOR[r.engine];
+      const labelInside = width > 220;
+      const valueX = labelInside ? BAR_X + width - 12 : BAR_X + width + 12;
+      const valueAnchor = labelInside ? "end" : "start";
+      const valueFill = labelInside ? "#020617" : "#e5e7eb";
+      return `  <g>
+    <text x="${BAR_X - 16}" y="${y + BAR_H / 2 + 6}" text-anchor="end" class="mono engine">${escapeXml(r.engine)}</text>
+    <rect x="${BAR_X}" y="${y}" width="${BAR_W}" height="${BAR_H}" rx="8" fill="#0f172a" stroke="#1e293b"/>
+    <rect x="${BAR_X}" y="${y}" width="${width.toFixed(1)}" height="${BAR_H}" rx="8" fill="${color}"/>
+    <text x="${valueX.toFixed(1)}" y="${y + BAR_H / 2 + 6}" text-anchor="${valueAnchor}" class="mono value" fill="${valueFill}">${escapeXml(spec.format(r.value))}</text>
+  </g>`;
+    })
+    .join("\n");
+
+  const note = spec.differentiator
+    ? `<text x="${BAR_X}" y="${FIRST_Y + ENGINES.length * ROW_H + 24}" class="mono note" fill="#22d3ee">${escapeXml(spec.differentiator)}</text>`
+    : "";
+
+  const footer = `Source: ${SOURCE_REF} · commit ${provenance.commit_sha.slice(0, 10)} · ${provenance.node_version} · ${data.generated_at.slice(0, 10)}`;
+  const altDesc = `${spec.title} (${spec.unit}, ${spec.direction}) by engine from measured Neuron-JS benchmark output: ${rows.map((r) => `${r.engine} ${spec.format(r.value)}`).join(", ")}.`;
+
+  return `<svg xmlns="http://www.w3.org/2000/svg" width="${W}" height="${H}" viewBox="0 0 ${W} ${H}" role="img" aria-labelledby="t d">
+  <title id="t">${escapeXml(spec.title)}</title>
+  <desc id="d">${escapeXml(altDesc)}</desc>
+  <metadata>Asset: NJS-GROWTH-07 ${spec.file}. Source: ${SOURCE_REF}. Generated by benchmarks/charts/generate.ts from measured actual_benchmark output. ${footer}</metadata>
+  <defs>
+    <pattern id="grid" width="40" height="40" patternUnits="userSpaceOnUse"><path d="M40 0H0v40" fill="none" stroke="#1e293b" stroke-width="0.7"/></pattern>
+    <style><![CDATA[
+      .text{font-family:Inter,system-ui,-apple-system,"Segoe UI",sans-serif;fill:#e5e7eb}
+      .mono{font-family:"JetBrains Mono",SFMono-Regular,Menlo,Consolas,monospace}
+      .title{font-size:38px;font-weight:750;letter-spacing:-.03em;fill:#e5e7eb}
+      .sub{font-size:19px;font-weight:500;fill:#94a3b8}
+      .engine{font-size:17px;fill:#cbd5e1}
+      .value{font-size:17px;font-weight:600}
+      .note{font-size:16px}
+      .foot{font-size:15px;fill:#64748b}
+      .metric{font-size:16px;fill:#94a3b8}
+    ]]></style>
+  </defs>
+  <rect width="${W}" height="${H}" fill="#020617"/>
+  <rect width="${W}" height="${H}" fill="url(#grid)" opacity=".6"/>
+  <rect x="24" y="24" width="${W - 48}" height="${H - 48}" rx="22" fill="rgba(15,23,42,.55)" stroke="#1e293b" stroke-width="2"/>
+  <text x="${BAR_X - 16}" y="92" text-anchor="end" class="text title">${escapeXml(spec.title)}</text>
+  <text x="${BAR_X - 16}" y="124" text-anchor="end" class="text sub">${escapeXml(spec.subtitle)}</text>
+  <text x="${BAR_X}" y="92" class="metric">metric: ${escapeXml(spec.metric)}</text>
+  <text x="${BAR_X}" y="124" class="metric">unit: ${escapeXml(spec.unit)} · ${spec.direction}</text>
+${bars}
+  ${note}
+  <text x="${BAR_X - 16}" y="${H - 40}" text-anchor="end" class="mono foot">Measured · no fabricated values</text>
+  <text x="${BAR_X}" y="${H - 40}" class="mono foot">${escapeXml(footer)}</text>
+</svg>
+`;
+}
+
+for (const spec of charts) {
+  const svg = render(spec);
+  writeFileSync(`${OUT_DIR}${spec.file}`, svg, "utf8");
+  process.stdout.write(`wrote ${spec.file}\n`);
+}
+
+// Generated showcase page (docs/benchmarks/results.md), built from the same data
+// so chart and table values can never drift from the measured source.
+const PAGE_PATH = fileURLToPath(
+  new URL("../../docs/benchmarks/results.md", import.meta.url),
+);
+
+function tableRows(): string {
+  const lines: string[] = [];
+  for (const engine of ENGINES) {
+    for (const scenario of [
+      "pricing-discount",
+      "eligibility-approval",
+      "workflow-routing",
+    ]) {
+      const r = rowFor(engine, scenario, "medium");
+      lines.push(
+        `| \`${r.engine}\` | ${r.scenario} | ${groupSep(Math.round(r.throughput_decisions_per_second))} | ${r.p50_ms} | ${r.p95_ms} | ${r.cold_start_ms} | ${groupSep(r.bundle_size_minified_bytes)} | ${r.validation_overhead_ms} | ${r.explanation_overhead_ms} |`,
+      );
+    }
+  }
+  return lines.join("\n");
+}
+
+const chartSection = charts
+  .map(
+    (spec) => `## ${spec.title}
+
+metric: \`${spec.metric}\` · ${spec.unit} · _${spec.direction}_
+
+![${spec.title} by engine, measured Neuron-JS benchmark output.](./assets/generated/${spec.file})
+${spec.differentiator ? `\n${spec.differentiator}\n` : ""}`,
+  )
+  .join("\n");
+
+const page = `<!-- GENERATED by benchmarks/charts/generate.ts from benchmarks/results/latest.actual.json. Do not edit by hand; run \`yarn benchmark && yarn benchmark:charts\`. -->
+# Benchmark results
+
+These are **measured** results produced by the Neuron-JS benchmark harness
+(\`yarn benchmark\`). They compare \`@sebasoft/neuron-js\` against \`json-rules-engine\`,
+\`json-logic-js\`, a hand-coded TypeScript baseline, and \`rule-engine-js\` across the
+pricing, eligibility, and workflow-routing scenarios. See the
+[methodology](./methodology) for how each metric is collected.
+
+Numbers reflect a single machine, Node version, and commit; reproduce locally before
+citing. No value on this page is hand-entered — charts and the table below are generated
+from the same source file.
+
+## Provenance
+
+| Field | Value |
+| --- | --- |
+| Generated | \`${data.generated_at}\` |
+| Node | \`${provenance.node_version}\` |
+| Commit | \`${provenance.commit_sha}\` |
+| Neuron-JS version | \`${provenance.package_version}\` |
+| Command | \`yarn benchmark\` |
+| Raw source | \`${SOURCE_REF}\` |
+
+${chartSection}
+## Full results (medium · 10,000 decisions)
+
+Throughput is decisions/second (higher is better); all latency, cold-start, and overhead
+columns are milliseconds (lower is better); bundle size is minified bytes.
+
+| Engine | Scenario | Throughput | p50 ms | p95 ms | Cold start ms | Bundle B | Validation ms | Explanation ms |
+| --- | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+${tableRows()}
+
+Validation and explanation overhead are Neuron-JS capabilities (\`validateScript\`,
+\`explainExecution\`); the other engines provide no equivalent step, so their measured
+delta is \`0\`.
+`;
+
+writeFileSync(PAGE_PATH, page, "utf8");
+process.stdout.write(`wrote ${PAGE_PATH}\n`);
diff --git a/benchmarks/harness/adapters/hand-coded-typescript.ts b/benchmarks/harness/adapters/hand-coded-typescript.ts
@@ -0,0 +1,20 @@
+import { readPath } from "../read-path.ts";
+import type { Adapter, Decision, Runner, ScenarioDef } from "../types.ts";
+
+/**
+ * Hand-coded TypeScript baseline. No engine: a direct `>=` comparison, the
+ * floor cost any rules engine is measured against.
+ */
+export const handCodedAdapter: Adapter = {
+  engine: "hand-coded-typescript",
+  prepare(scenario: ScenarioDef): Runner {
+    return (): Decision => {
+      const value = readPath(scenario.data, scenario.factPath);
+      const matched = typeof value === "number" && value >= scenario.threshold;
+      return scenario.decide(matched);
+    };
+  },
+};
+
+/** Canonical export consumed by the cold-start probe. */
+export const adapter = handCodedAdapter;
diff --git a/benchmarks/harness/adapters/json-logic-js.ts b/benchmarks/harness/adapters/json-logic-js.ts
@@ -0,0 +1,23 @@
+import jsonLogic from "json-logic-js";
+import type { Adapter, Decision, Runner, ScenarioDef } from "../types.ts";
+
+/**
+ * JsonLogic (json-logic-js) adapter. Expresses each scenario as a `>=` predicate
+ * over the nested data and derives the canonical decision from the boolean result.
+ */
+export const jsonLogicAdapter: Adapter = {
+  engine: "json-logic-js",
+  prepare(scenario: ScenarioDef): Runner {
+    const rule = {
+      ">=": [{ var: scenario.factPath }, scenario.threshold],
+    };
+
+    return (): Decision => {
+      const matched = jsonLogic.apply(rule, scenario.data) === true;
+      return scenario.decide(matched);
+    };
+  },
+};
+
+/** Canonical export consumed by the cold-start probe. */
+export const adapter = jsonLogicAdapter;
diff --git a/benchmarks/harness/adapters/json-rules-engine.ts b/benchmarks/harness/adapters/json-rules-engine.ts
@@ -0,0 +1,34 @@
+import { Engine } from "json-rules-engine";
+import type { Adapter, Decision, Runner, ScenarioDef } from "../types.ts";
+
+/**
+ * json-rules-engine adapter. Builds one Engine with a single `>=` rule per
+ * scenario, then resolves the canonical decision from the fired events. The
+ * engine API is promise-based, so the runner is async.
+ */
+export const jsonRulesEngineAdapter: Adapter = {
+  engine: "json-rules-engine",
+  prepare(scenario: ScenarioDef): Runner {
+    const engine = new Engine();
+    engine.addRule({
+      conditions: {
+        all: [
+          {
+            fact: scenario.flatFactName,
+            operator: "greaterThanInclusive",
+            value: scenario.threshold,
+          },
+        ],
+      },
+      event: { type: scenario.id },
+    });
+
+    return async (): Promise<Decision> => {
+      const { events } = await engine.run(scenario.flatFacts);
+      return scenario.decide(events.length > 0);
+    };
+  },
+};
+
+/** Canonical export consumed by the cold-start probe. */
+export const adapter = jsonRulesEngineAdapter;