autoresearcher 0.1.5 → 0.1.7

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "autoresearcher",
-  "version": "0.1.5",
+  "version": "0.1.7",
   "description": "Benchmark-driven autonomous research CLI for post-quantum and blockchain R&D",
   "type": "module",
   "bin": {
@@ -29,7 +29,7 @@
     "src",
     "README.md"
   ],
-  "homepage": "https://autoresearcher.multivmlabs.com",
+  "homepage": "https://autoresearcher.org",
   "license": "MIT",
   "dependencies": {
     "ralph-starter": "^0.4.5"

package/src/run-loop.js

@@ -140,6 +140,76 @@ function escapeCell(value) {
   return String(value).replaceAll('|', '\\|').replaceAll('\n', ' ');
 }
 
+function asFiniteNumber(value) {
+  return typeof value === 'number' && Number.isFinite(value) ? value : null;
+}
+
+function summarizeMetricEvolution(iterationEntries, direction) {
+  const points = iterationEntries
+    .map((entry) => ({
+      iteration: entry.iteration,
+      metric: asFiniteNumber(entry.metric),
+    }))
+    .filter((point) => point.metric != null);
+
+  if (points.length === 0) {
+    return null;
+  }
+
+  const baseline = points[0].metric;
+  const final = points[points.length - 1].metric;
+  const delta = final - baseline;
+  const relativeDeltaPct = baseline !== 0 ? (delta / baseline) * 100 : null;
+  const improvedFromBaseline = direction === 'min' ? final < baseline : final > baseline;
+  const monotonic = points.every((point, index) => {
+    if (index === 0) return true;
+    const prev = points[index - 1].metric;
+    return direction === 'min' ? point.metric <= prev : point.metric >= prev;
+  });
+
+  return {
+    baseline,
+    baselineIteration: points[0].iteration,
+    final,
+    finalIteration: points[points.length - 1].iteration,
+    delta,
+    relativeDeltaPct,
+    improvedFromBaseline,
+    monotonic,
+  };
+}
+
+function formatSigned(value) {
+  if (!Number.isFinite(value)) return 'n/a';
+  const sign = value > 0 ? '+' : '';
+  return `${sign}${value}`;
+}
+
+function sanitizeObjectiveText(value, maxChars = 4000) {
+  if (!value || typeof value !== 'string') {
+    return '';
+  }
+
+  const normalized = value.trim().replaceAll('\r', '');
+  if (!normalized) return '';
+  if (normalized.length <= maxChars) {
+    return normalized;
+  }
+
+  return `${normalized.slice(0, maxChars)}\n...[truncated]`;
+}
+
+function renderProgressBar(current, total, width = 24) {
+  if (!Number.isFinite(current) || !Number.isFinite(total) || total <= 0) {
+    return '[????????????????????????]';
+  }
+
+  const ratio = Math.max(0, Math.min(1, current / total));
+  const filled = Math.round(ratio * width);
+  const empty = Math.max(0, width - filled);
+  return `[${'#'.repeat(filled)}${'-'.repeat(empty)}]`;
+}
+
 function renderFinalReport({
   runId,
   startedAt,
@@ -153,11 +223,14 @@ function renderFinalReport({
   bestIteration,
   iterationEntries,
   runLogRelativePath,
+  agentObjective,
 }) {
   const keepCount = iterationEntries.filter((entry) => entry.status === 'keep').length;
   const rejectCount = iterationEntries.filter((entry) => entry.status === 'reject').length;
   const benchmarkFailedCount = iterationEntries.filter((entry) => entry.status === 'benchmark_failed').length;
   const agentFailedCount = iterationEntries.filter((entry) => entry.status === 'agent_failed').length;
+  const metricSummary = summarizeMetricEvolution(iterationEntries, direction);
+  const objectiveText = sanitizeObjectiveText(agentObjective);
 
   const lines = [
     '# Autoresearcher Final Report',
@@ -179,6 +252,20 @@ function renderFinalReport({
     `- Agent failures: ${agentFailedCount}`,
     `- Run log: \`${runLogRelativePath}\``,
     '',
+    '## Metric Comparison',
+    '',
+    `- Baseline metric: ${metricSummary ? formatMetric(metricSummary.baseline) : 'n/a'}${metricSummary ? ` (iteration ${metricSummary.baselineIteration})` : ''}`,
+    `- Final metric: ${metricSummary ? formatMetric(metricSummary.final) : 'n/a'}${metricSummary ? ` (iteration ${metricSummary.finalIteration})` : ''}`,
+    `- Delta (final-baseline): ${metricSummary ? formatSigned(metricSummary.delta) : 'n/a'}`,
+    `- Relative change: ${metricSummary && metricSummary.relativeDeltaPct != null ? `${formatSigned(Number(metricSummary.relativeDeltaPct.toFixed(2)))}%` : 'n/a'}`,
+    `- Improved from baseline: ${metricSummary ? (metricSummary.improvedFromBaseline ? 'yes' : 'no') : 'n/a'}`,
+    `- Monotonic trend: ${metricSummary ? (metricSummary.monotonic ? 'yes' : 'no') : 'n/a'}`,
+    '',
+    '## Research Objective Snapshot',
+    '',
+    objectiveText ? '```text' : '_No objective text captured._',
+    ...(objectiveText ? [objectiveText, '```'] : []),
+    '',
     '## Iteration Results',
     '',
     '| Iteration | Status | Metric | Best Metric | Agent Exit | Benchmark Exit | Timestamp |',
@@ -372,6 +459,7 @@ export async function runResearchLoop(config, cliOverrides = {}) {
 
   let bestMetric = null;
   let bestIteration = 0;
+  let baselineMetric = null;
 
   console.log(`Starting run ${runId}`);
   console.log(`Agent mode: ${agentMode}`);
@@ -393,6 +481,8 @@ export async function runResearchLoop(config, cliOverrides = {}) {
 
   for (let i = 1; i <= iterations; i++) {
     console.log(`\n--- Iteration ${i}/${iterations} ---`);
+    const progressPercent = Number(((i / iterations) * 100).toFixed(1));
+    console.log(`Progress: ${renderProgressBar(i, iterations)} ${progressPercent}%`);
     const beforeCommit = await getGitCommit(cwd);
 
     const iterationAgentPrompt = benchmarkFeedback
@@ -464,7 +554,7 @@ export async function runResearchLoop(config, cliOverrides = {}) {
     const nonFatalInternalMaxIterations = isNonFatalInternalMaxIterations(agentStep.agentMode, agentResult);
 
     if (nonFatalInternalMaxIterations) {
-      console.log('Agent step reached backend max iterations; continuing to benchmark...');
+      console.log('Agent step: backend iteration budget reached; continuing to benchmark...');
     }
 
     if (agentResult.code !== 0 && !nonFatalInternalMaxIterations) {
@@ -520,7 +610,15 @@ export async function runResearchLoop(config, cliOverrides = {}) {
     }
 
     const improved = isBetter(metric, bestMetric, direction);
-    console.log(`Metric: ${metric}${bestMetric == null ? ' (baseline)' : ` | best: ${bestMetric}`}`);
+    if (baselineMetric == null) {
+      baselineMetric = metric;
+    }
+
+    const baselineDelta = metric - baselineMetric;
+    const baselineDeltaDisplay = `${baselineDelta > 0 ? '+' : ''}${baselineDelta}`;
+    console.log(
+      `Metric: ${metric}${bestMetric == null ? ' (baseline)' : ` | best: ${bestMetric}`} | delta_baseline: ${baselineDeltaDisplay}`
+    );
 
     const nextBestMetric = improved ? metric : bestMetric;
     benchmarkFeedback = buildBenchmarkFeedback({
@@ -596,6 +694,7 @@ export async function runResearchLoop(config, cliOverrides = {}) {
     bestIteration,
     iterationEntries,
     runLogRelativePath,
+    agentObjective: resolvedPrompt.prompt,
   });
 
   if (merged.onCompleteCommand) {