@tangle-network/agent-eval 0.216.1 → 0.217.1

package/dist/{report-command-SC9nqb4M.js → render-C-w2EZ6p.js}

@@ -1,10 +1,9 @@
-import { l as errorMessage } from "./errors-DVZiqS7K.js";
-import { a as hashCanonical } from "./canonical-B2RqLofv.js";
-import { o as summarizeNumberSeries } from "./descriptive-DboxPZrG.js";
+import { a as hashCanonical } from "./canonical-Dq2hOQ8L.js";
+import { o as summarizeNumberSeries } from "./descriptive-CVNy48rR.js";
 import { t as roundDecimal } from "./round-decimal-COrKglMh.js";
-import { a as nonEmptyString, c as parseSupervisorTree, d as NO_SOURCE_LIMITS, g as unavailable, h as showMeasured, i as asRecord, l as recordOrNull, m as isUnavailable, o as parseJson, p as SUPERVISOR_RUN_SCHEMA, r as readTerminalRecord, s as parseJsonl, u as workerSourceKey } from "./terminal-record-DZyQZ3pz.js";
-import { join, resolve } from "node:path";
-import { appendFile, mkdir, readFile, readdir, stat, writeFile } from "node:fs/promises";
+import { a as nonEmptyString, c as parseSupervisorTree, d as NO_SOURCE_LIMITS, g as unavailable, h as showMeasured, i as asRecord, l as recordOrNull, m as isUnavailable, o as parseJsonRecord, p as SUPERVISOR_RUN_SCHEMA, r as readTerminalRecord, s as parseJsonl, u as workerSourceKey } from "./terminal-record-78g6E4vh.js";
+import { join } from "node:path";
+import { readFile, readdir, stat } from "node:fs/promises";
 //#region src/supervisor-run/patch-stats.ts
 /** Unified-diff stats. Counts `+++ b/<path>` targets, body +/- lines, and test-file touches. */
 function parsePatch(text) {
@@ -60,12 +59,12 @@ function analyzeSupervisorRunSources(src, now = Date.now) {
 	const haveJournal = src.journal !== null;
 	const tree = parseSupervisorTree(src);
 	const state = tree.state;
-	const result = parseJson(src.result);
-	const judge = parseJson(src.judge);
+	const result = parseJsonRecord(src.result);
+	const judge = parseJsonRecord(src.judge);
 	const terminal = readTerminalRecord({
 		state,
 		result,
-		failure: src.failure === void 0 ? void 0 : parseJson(src.failure)
+		failure: src.failure === void 0 ? void 0 : parseJsonRecord(src.failure)
 	});
 	const { rootId, workerSpawns, workerCloses, startedAt, completedAt } = tree;
 	const rootSpawn = rootId === null ? null : tree.spawns.find((spawn) => spawn.id === rootId) ?? null;
@@ -482,212 +481,6 @@ function pickString(rec, key) {
 	return typeof v === "string" ? v : null;
 }
 //#endregion
-//#region src/supervisor-run/render.ts
-/**
-* Human-readable renderings of a supervisor-run report. Zero and unavailable
-* render differently on purpose (`0` vs `unavailable — <reason>`), because the
-* two have driven opposite conclusions about the same architecture.
-*/
-/**
-* The block appended to a run log after every run — the answers an operator asks
-* for, in the log tail, with no extra command.
-*/
-function renderSupervisorRunHeadline(r) {
-	const o = r.orchestration;
-	const steerNote = isUnavailable(o.steers) ? `unavailable — ${o.steers.unavailable}` : o.steers === 0 ? "0 (spawn→wait→respawn only; no mid-task steering)" : `${o.steers} queued / ${showMeasured(o.steersDelivered)} delivered`;
-	return [
-		`RUN-REPORT ${r.instanceId ?? "?"} [${r.arm ?? "?"}]`,
-		`  status=${showMeasured(r.outcome.supStatus)} source=${showMeasured(r.outcome.supStatusSource)} reason=${showMeasured(r.outcome.supReason)} failure=${fmtFailure(r.outcome.failure)}`,
-		`  steers=${steerNote}`,
-		`  waves=${showMeasured(o.waves)} sizes=${isUnavailable(o.waveSizes) ? `unavailable — ${o.waveSizes.unavailable}` : `[${o.waveSizes.join(",")}]`} workers=${showMeasured(o.workersSpawned)} settled=${showMeasured(o.workersSettled)} cancelled=${showMeasured(o.workersCancelled)}`,
-		`  concurrency max=${showMeasured(o.maxConcurrency)} utilization=${showMeasured(o.workerUtilization)} idle=${fmtMs(o.idleMs)} (${showMeasured(o.idlePct)}%) wall=${fmtMs(o.supervisorWallMs)}${o.supervisorWallSource === "journal-span" ? " (journal-span lower bound)" : ""}`,
-		`  respawns=${showMeasured(o.respawns)} evidence→respawn=${showMeasured(r.decision.observeThenRespawn)} blind-respawn=${showMeasured(r.decision.respawnWithoutEvidence)} depth=${showMeasured(o.delegationDepth)}`,
-		`  accepted=${showMeasured(r.decision.accepted)} rejected=${showMeasured(r.decision.rejected)} empty-pass=${showMeasured(r.decision.emptyPass)}`,
-		`  brain=$${showMeasured(r.economics.brain.usd)} total=$${showMeasured(r.economics.totalUsd)} judge.resolved=${showMeasured(r.outcome.judgeResolved)} score=${showMeasured(r.outcome.judgeScore)} verify=${showMeasured(r.outcome.verifyPass)}`,
-		r.gaps.length > 0 ? `  gaps(${r.gaps.length}): ${r.gaps.join("; ")}` : "  gaps: none"
-	].join("\n");
-}
-/**
-* The recorded error on one line; `none recorded` and `unavailable` stay
-* distinct, and a throw the run outlived is labeled so it is not read as the
-* outcome of the settled result beside it.
-*/
-function fmtFailure(v) {
-	if (isUnavailable(v)) return `unavailable — ${v.unavailable}`;
-	if (v === null) return "none recorded";
-	const name = v.name ?? "unavailable — record has no error.name";
-	const message = v.message ?? "unavailable — record has no error.message";
-	const attempt = v.earlierAttempt ? ", earlier attempt" : "";
-	return `${name}: ${message} [${v.source}${v.at === null ? "" : ` at ${v.at}`}${attempt}]`;
-}
-function resourceRows(records) {
-	if (isUnavailable(records)) return [`unavailable — ${records.unavailable}`, ""];
-	const out = [
-		"Receipts are source records, not additive totals.",
-		"A false completeness flag means the amount is a recorded subtotal, not complete usage.",
-		"",
-		"| Node | Source | Resource | Unit | Amount | Known |",
-		"|---|---|---|---|---:|---|"
-	];
-	const cell = (value) => value.replace(/\|/g, "\\|").replace(/\r?\n/g, " ");
-	for (const record of records) {
-		const prefix = `| ${cell(record.nodeId ?? "unavailable")} | ${cell(record.source)} |`;
-		if (isUnavailable(record.resources)) out.push(`${prefix} ${cell(showMeasured(record.resources))} | — | — | — |`);
-		else if (record.resources.length === 0) out.push(`${prefix} recorded empty map | — | — | — |`);
-		else for (const receipt of record.resources) out.push(`${prefix} ${cell(showMeasured(receipt.name))} | ${cell(showMeasured(receipt.unit))} | ${cell(showMeasured(receipt.amount))} | ${cell(showMeasured(receipt.known))} |`);
-	}
-	if (!records.length) out.push("| — | — | no spend records | — | — | — |");
-	out.push("");
-	return out;
-}
-function fmtMs(v) {
-	if (isUnavailable(v)) return `unavailable — ${v.unavailable}`;
-	if (v < 1e3) return `${v}ms`;
-	const s = v / 1e3;
-	if (s < 120) return `${roundDecimal(s, 1)}s`;
-	return `${roundDecimal(s / 60, 1)}min`;
-}
-function renderSupervisorRunMarkdown(r) {
-	const o = r.orchestration;
-	const d = r.decision;
-	const e = r.economics;
-	const out = [];
-	out.push(`# Run report — ${r.instanceId ?? "unknown instance"} [${r.arm ?? "unknown arm"}]`);
-	out.push("");
-	out.push("```");
-	out.push(renderSupervisorRunHeadline(r));
-	out.push("```");
-	out.push("");
-	out.push(`- Run: \`${r.runRef}\``);
-	out.push(`- Supervisor: \`${showMeasured(r.supervisorId)}\``);
-	out.push(`- Supervisor profile: \`${showMeasured(r.supervisorProfileDigest)}\``);
-	out.push(`- Generated: ${r.generatedAt}`);
-	out.push("");
-	out.push("## Orchestration");
-	out.push("");
-	out.push("| Metric | Value |");
-	out.push("|---|---|");
-	out.push(`| Workers spawned | ${showMeasured(o.workersSpawned)} |`);
-	out.push(`| Workers settled | ${showMeasured(o.workersSettled)} |`);
-	out.push(`| Workers cancelled | ${showMeasured(o.workersCancelled)} |`);
-	out.push(`| **Steers (mid-task messages to live workers)** | **${showMeasured(o.steers)}** |`);
-	out.push(`| Steers delivered | ${showMeasured(o.steersDelivered)} |`);
-	out.push(`| Outer-driver \`supervisor_steer\` calls | ${showMeasured(o.driverSteerCalls)} |`);
-	out.push(`| Spawn waves | ${showMeasured(o.waves)} |`);
-	out.push(`| Wave sizes | ${isUnavailable(o.waveSizes) ? showMeasured(o.waveSizes) : `[${o.waveSizes.join(", ")}]`} |`);
-	out.push(`| Max concurrency | ${showMeasured(o.maxConcurrency)} |`);
-	out.push(`| Respawns (after same parent's first settle) | ${showMeasured(o.respawns)} |`);
-	out.push(`| Repeated labels | ${isUnavailable(o.repeatedLabels) ? showMeasured(o.repeatedLabels) : o.repeatedLabels.length === 0 ? "none" : o.repeatedLabels.join(", ")} |`);
-	out.push(`| Delegation depth | ${showMeasured(o.delegationDepth)} |`);
-	out.push(`| Time to first spawn | ${fmtMs(o.timeToFirstSpawnMs)} |`);
-	out.push(`| Supervisor wall | ${fmtMs(o.supervisorWallMs)}${isUnavailable(o.supervisorWallSource) ? "" : ` (source: ${o.supervisorWallSource}${o.supervisorWallSource === "journal-span" ? ", lower bound" : ""})`} |`);
-	out.push(`| Idle (zero live workers) | ${fmtMs(o.idleMs)} (${showMeasured(o.idlePct)}%) |`);
-	out.push(`| Worker utilization (Σ worker wall ÷ supervisor wall) | ${showMeasured(o.workerUtilization)} |`);
-	out.push("");
-	if (!isUnavailable(o.steersByWorker) && o.steersByWorker.length > 0) {
-		out.push("### Steers per worker");
-		out.push("");
-		out.push("| Worker id | Label | Queued | Delivered |");
-		out.push("|---|---|---:|---:|");
-		for (const s of o.steersByWorker) out.push(`| ${s.workerId === null ? "unavailable — worker id absent; label join used" : `\`${s.workerId}\``} | \`${s.worker}\` | ${s.queued} | ${s.delivered} |`);
-		out.push("");
-	} else if (isUnavailable(o.steersByWorker)) out.push(`### Steers per worker\n\nunavailable — ${o.steersByWorker.unavailable}\n`);
-	out.push("## Decision quality");
-	out.push("");
-	out.push("| Metric | Value |");
-	out.push("|---|---|");
-	out.push(`| Settled by status | ${fmtCounts(d.settledByStatus)} |`);
-	out.push(`| Settled verdicts | ${fmtCounts(d.settledVerdicts)} |`);
-	out.push(`| Accepted (verify green + patch bytes) | ${showMeasured(d.accepted)} |`);
-	out.push(`| Rejected (verify red) | ${showMeasured(d.rejected)} |`);
-	out.push(`| Empty pass (green, no patch) | ${showMeasured(d.emptyPass)} |`);
-	out.push(`| Evidence → respawn sequences | ${showMeasured(d.observeThenRespawn)} |`);
-	out.push(`| Respawn with no same-parent settled evidence in front | ${showMeasured(d.respawnWithoutEvidence)} |`);
-	out.push(`| Review actions (steers + worker questions) | ${showMeasured(d.reviewActions)} |`);
-	out.push(`| Worker evidence returned | ${isUnavailable(d.workerEvidenceBytes) ? showMeasured(d.workerEvidenceBytes) : `${d.workerEvidenceBytes} bytes`} |`);
-	out.push("");
-	out.push("## Economics");
-	out.push("");
-	out.push("| Role | Tokens in | Tokens out | Cache read | Cache write | USD | Source |");
-	out.push("|---|---:|---:|---:|---:|---:|---|");
-	out.push(`| brain | ${showMeasured(e.brain.tokensIn)} | ${showMeasured(e.brain.tokensOut)} | ${showMeasured(e.brain.cacheRead)} | ${showMeasured(e.brain.cacheWrite)} | ${showMeasured(e.brain.usd)} | ${e.brain.source} |`);
-	out.push(`| workers | ${showMeasured(e.workers.tokensIn)} | ${showMeasured(e.workers.tokensOut)} | ${showMeasured(e.workers.cacheRead)} | ${showMeasured(e.workers.cacheWrite)} | ${showMeasured(e.workers.usd)} | ${e.workers.source} |`);
-	out.push("");
-	if (!isUnavailable(e.brainTruncations) && e.brainTruncations > 0) out.push(`- **BRAIN OUTPUT TRUNCATED: ${e.brainTruncations} completion(s) hit \`finish_reason: "length"\`** — the supervisor acted on a half-written plan. Its output ceiling is too low; see \`brain.jsonl\` for the per-call \`req_max_tokens\`.`);
-	else out.push(`- Brain completions truncated (finish_reason=length): ${showMeasured(e.brainTruncations)}`);
-	out.push(`- Total USD: ${showMeasured(e.totalUsd)} (source: ${e.totalUsdSource})`);
-	out.push(`- Spend measured two ways: journal-derived $${showMeasured(e.spend.journalDerived.usd)} over ${e.spend.journalDerived.records} journal record(s) (execution accounting) · close-record $${showMeasured(e.spend.closeRecord.usd)} over ${e.spend.closeRecord.records} close record(s) (billing-shaped). Divergence is a signal, not an error.`);
-	out.push(`- Cost per accepted patch: ${showMeasured(e.costPerAcceptedPatchUsd)}`);
-	if (isUnavailable(e.workerWallMsDistribution)) out.push(`- Worker wall distribution: unavailable — ${e.workerWallMsDistribution.unavailable}`);
-	else {
-		const w = e.workerWallMsDistribution;
-		out.push(`- Worker wall (n=${w.n}): min ${fmtMs(w.min)} / p50 ${fmtMs(w.p50)} / p90 ${fmtMs(w.p90)} / max ${fmtMs(w.max)} / Σ ${fmtMs(w.sum)}`);
-	}
-	out.push("");
-	if (!isUnavailable(e.perWorker) && e.perWorker.length > 0) {
-		out.push("| Worker id | Label | Role | Runtime | Profile digest | Status | Failure | Infra | Wall | Tokens in | Tokens out | Patch bytes | Verify passed | Score |");
-		out.push("|---|---|---|---|---|---|---|---|---:|---:|---:|---:|---|---:|");
-		for (const w of e.perWorker) out.push(`| ${w.workerId === null ? "unavailable — worker id absent; label join used" : `\`${w.workerId}\``} | \`${w.worker}\` | ${w.role ?? "unavailable — source recorded no role"} | ${w.runtime ?? "unavailable — source recorded no runtime"} | ${w.profileDigest === null ? "unavailable — source recorded no profile digest" : `\`${w.profileDigest}\``} | ${w.status ?? "unavailable — no terminal event"} | ${w.failure ?? "none recorded"} | ${w.infra ?? "unavailable"} | ${w.wallMs === null ? "unavailable — no spawn/finish pair" : fmtMs(w.wallMs)} | ${w.tokensIn ?? "unavailable — store does not attribute tokens per worker"} | ${w.tokensOut ?? "unavailable — store does not attribute tokens per worker"} | ${w.patchBytes ?? "unavailable — no worker patch file"} | ${w.passed === null ? "unavailable — no verdict" : String(w.passed)} | ${w.score ?? "unavailable — no numeric score"} |`);
-		out.push("");
-	}
-	if (e.resourceRecords !== void 0) out.push("## Named resource receipts", "", ...resourceRows(e.resourceRecords));
-	out.push("## Outcome");
-	out.push("");
-	out.push("| Metric | Value |");
-	out.push("|---|---|");
-	out.push(`| Supervisor status | ${showMeasured(r.outcome.supStatus)} |`);
-	out.push(`| Status source | ${showMeasured(r.outcome.supStatusSource)} |`);
-	out.push(`| Terminal reason | ${showMeasured(r.outcome.supReason)} |`);
-	out.push(`| Failure | ${fmtFailure(r.outcome.failure)} |`);
-	out.push(`| Supervisor verdict | ${showMeasured(r.outcome.supVerdict)} |`);
-	out.push(`| Delivered | ${showMeasured(r.outcome.delivered)} |`);
-	out.push(`| Judge resolved | ${showMeasured(r.outcome.judgeResolved)} |`);
-	out.push(`| Judge score | ${showMeasured(r.outcome.judgeScore)} |`);
-	out.push(`| Judge passed / total | ${showMeasured(r.outcome.judgePassed)} / ${showMeasured(r.outcome.judgeTotal)} |`);
-	out.push(`| Judge source | ${r.outcome.judgeSource ?? "unavailable — no judge.json and no ledger row"} |`);
-	out.push(`| Verify gate | pass=${showMeasured(r.outcome.verifyPass)} rc=${showMeasured(r.outcome.verifyRc)} |`);
-	if (isUnavailable(r.outcome.patch)) out.push(`| Patch | unavailable — ${r.outcome.patch.unavailable} |`);
-	else {
-		const p = r.outcome.patch;
-		out.push(`| Patch | ${p.files} file(s), +${p.linesAdded}/-${p.linesRemoved}, test files touched: ${p.testFilesTouched.length === 0 ? "none" : p.testFilesTouched.join(", ")} |`);
-	}
-	out.push("");
-	out.push("## Gaps");
-	out.push("");
-	if (r.gaps.length === 0) out.push("None — every metric above is backed by a present artifact.");
-	else for (const g of r.gaps) out.push(`- ${g}`);
-	out.push("");
-	out.push(`> Harness-session view of the same run (model calls, stuck loops, tool errors): \`${r.traceCommand}\``);
-	out.push("");
-	return out.join("\n");
-}
-function fmtCounts(v) {
-	if (isUnavailable(v)) return showMeasured(v);
-	const entries = Object.entries(v);
-	return entries.length === 0 ? "none" : entries.map(([k, n]) => `${k}=${n}`).join(", ");
-}
-function renderSupervisorRollupMarkdown(rollup, title = "Round rollup") {
-	const out = [];
-	out.push(`# ${title}`);
-	out.push("");
-	out.push(`- Cells: ${rollup.cells}`);
-	out.push(`- **Steers across all cells: ${showMeasured(rollup.steersTotal)}** (cells with ≥1 steer: ${showMeasured(rollup.cellsWithSteers)}; cells where the steer count is unavailable: ${rollup.cellsWithUnavailableSteers})`);
-	out.push(`- Waves per cell (mean): ${showMeasured(rollup.wavesMean)}`);
-	out.push(`- Max concurrency observed: ${showMeasured(rollup.maxConcurrencyMax)}`);
-	out.push(`- Worker utilization (mean): ${showMeasured(rollup.utilizationMean)}`);
-	out.push(`- Idle share (mean): ${showMeasured(rollup.idlePctMean)}%`);
-	out.push(`- Workers spawned: ${showMeasured(rollup.workersSpawnedTotal)} · accepted: ${showMeasured(rollup.acceptedTotal)}`);
-	out.push(`- Spend: $${showMeasured(rollup.usdTotal)} · judged resolved: ${showMeasured(rollup.resolvedCount)}/${rollup.cells}`);
-	out.push(`- Spend measured two ways: journal-derived $${showMeasured(rollup.spendUsd.journalDerived.value)} over ${rollup.spendUsd.journalDerived.runs}/${rollup.cells} runs (execution accounting) · close-record $${showMeasured(rollup.spendUsd.closeRecord.value)} over ${rollup.spendUsd.closeRecord.runs}/${rollup.cells} runs (billing-shaped)`);
-	out.push("");
-	out.push("| Instance | Arm | Steers | Waves | Utilization | Idle % | Resolved | USD |");
-	out.push("|---|---|---:|---:|---:|---:|---|---:|");
-	for (const c of rollup.perCell) out.push(`| ${c.instanceId ?? "?"} | ${c.arm ?? "?"} | ${showMeasured(c.steers)} | ${showMeasured(c.waves)} | ${showMeasured(c.utilization)} | ${showMeasured(c.idlePct)} | ${showMeasured(c.resolved)} | ${showMeasured(c.usd)} |`);
-	out.push("");
-	for (const cell of rollup.perCell) if (cell.resourceRecords !== void 0) out.push(`## Resource receipts — ${cell.instanceId ?? "?"} [${cell.arm ?? "?"}]`, "", ...resourceRows(cell.resourceRecords));
-	return out.join("\n");
-}
-//#endregion
 //#region src/supervisor-run/coordination-reader.ts
 /**
 * Optional Runtime run-dir files, read as bytes.
@@ -1368,13 +1161,6 @@ async function readRuntimeSupervisorRun(runDir, opts = {}) {
 		traceCommand: nativeTraceSummary(workers, transcripts.get(normalized.root))
 	};
 }
-/** The agent-runtime file-backed layout as a `SupervisorRunReader`. */
-function runtimeSupervisorRunReader(runDir, opts = {}) {
-	return {
-		runRef: runDir,
-		read: () => readRuntimeSupervisorRun(runDir, opts)
-	};
-}
 /**
 * True when a directory holds any file only Runtime's durable layer writes:
 * the spawn journal, the observer journal, or the failure record. A run that
@@ -1413,56 +1199,6 @@ function isReader(input) {
 	return typeof input.read === "function";
 }
 /**
-* Read a completed run, write `run-report.json` + `run-report.md` beside its
-* artifacts, and append the headline block to the run log. Never throws on a
-* missing artifact — a run that produced nothing still yields a report whose
-* every metric says why.
-*/
-async function writeSupervisorRunReport(runDir, opts = {}) {
-	const report = analyzeSupervisorRunSources(await readRuntimeSupervisorRun(runDir));
-	const md = renderSupervisorRunMarkdown(report);
-	const dest = opts.reportDir ?? runDir;
-	const stem = opts.reportDir === void 0 ? "run-report" : supervisorReportStem(runDir);
-	if (opts.reportDir !== void 0) await mkdir(opts.reportDir, { recursive: true }).catch(() => {});
-	await writeFile(join(dest, `${stem}.json`), JSON.stringify(report, null, 1)).catch(() => {});
-	await writeFile(join(dest, `${stem}.md`), md).catch(() => {});
-	const headline = renderSupervisorRunHeadline(report);
-	if (opts.appendHeadlineTo !== void 0) await appendFile(opts.appendHeadlineTo, `${headline}\n`).catch(() => {});
-	if (opts.echo !== false) console.log(headline);
-	return report;
-}
-/**
-* File stem for out-of-tree reports. Built from the run path's identifying
-* segments — candidate tag (the segment under `arm-runs/`), rep, instance, arm
-* — so two runs of the same instance from different candidates/reps never
-* overwrite each other.
-*/
-function supervisorReportStem(runDir) {
-	const parts = runDir.split("/").filter(Boolean);
-	const arm = parts[parts.length - 1] ?? "cell";
-	const iid = parts[parts.length - 2] ?? "instance";
-	const rep = parts.find((p) => /^rep-\d+$/.test(p));
-	const armRunsIdx = parts.indexOf("arm-runs");
-	return [
-		armRunsIdx >= 0 ? parts[armRunsIdx + 1] : void 0,
-		rep,
-		iid,
-		arm
-	].filter((s) => s !== void 0 && s !== "runs").join(".").replace(/[^A-Za-z0-9._-]/g, "_");
-}
-/**
-* Best-effort wrapper for a hot path: a reporting failure must never kill a run
-* that already produced real work. Returns null and logs the reason instead.
-*/
-async function writeSupervisorRunReportSafe(runDir, opts = {}) {
-	try {
-		return await writeSupervisorRunReport(runDir, opts);
-	} catch (err) {
-		console.log(`RUN-REPORT failed for ${runDir}: ${err instanceof Error ? err.message : String(err)}`);
-		return null;
-	}
-}
-/**
 * Every Runtime supervisor run below `root`.
 *
 * When `root` itself is one run, return no children so callers can distinguish
@@ -1489,106 +1225,212 @@ async function findSupervisorRunDirs(root) {
 	return found.sort();
 }
 //#endregion
-//#region src/supervisor-run/report-command.ts
+//#region src/supervisor-run/render.ts
 /**
-* `agent-eval supervisor-run report <runDir>` takes one run directory and
-* prints the report the module already renders. The command reads through
-* `analyzeSupervisorRun`, so a Runtime run directory takes the
-* same path and the status comes from the record `terminal-record.ts` names.
-*
-* Exit codes follow the other self-parsing subcommands: 2 for a usage error,
-* 1 when the directory cannot be read (missing, not a directory, or a corrupt
-* journal or terminal record), 0 after the report was written to stdout.
+* Human-readable renderings of a supervisor-run report. Zero and unavailable
+* render differently on purpose (`0` vs `unavailable — <reason>`), because the
+* two have driven opposite conclusions about the same architecture.
 */
-const FORMATS = [
-	"headline",
-	"markdown",
-	"json"
-];
-const SUPERVISOR_RUN_USAGE = "usage: agent-eval supervisor-run report <runDir> [--format headline|markdown|json]";
-const PROCESS_IO = {
-	stdout: (text) => {
-		process.stdout.write(text);
-	},
-	stderr: (text) => {
-		process.stderr.write(text);
-	}
-};
-function isFormat(value) {
-	return FORMATS.includes(value);
-}
-function parseReportArgs(argv) {
-	let runDir = null;
-	let format = "headline";
-	for (let i = 0; i < argv.length; i++) {
-		const arg = argv[i];
-		if (arg === "--format") {
-			const raw = argv[++i];
-			if (raw === void 0 || !isFormat(raw)) throw new Error(`--format expects one of ${FORMATS.join("|")}, got ${JSON.stringify(raw ?? "")}`);
-			format = raw;
-		} else if (arg.startsWith("--format=")) {
-			const raw = arg.slice(9);
-			if (!isFormat(raw)) throw new Error(`--format expects one of ${FORMATS.join("|")}, got ${JSON.stringify(raw)}`);
-			format = raw;
-		} else if (arg.startsWith("--")) throw new Error(`unknown flag "${arg}"`);
-		else if (runDir === null) runDir = arg;
-		else throw new Error(`unexpected argument "${arg}"`);
-	}
-	if (runDir === null) throw new Error(SUPERVISOR_RUN_USAGE);
-	return {
-		runDir,
-		format
-	};
+/**
+* The block appended to a run log after every run — the answers an operator asks
+* for, in the log tail, with no extra command.
+*/
+function renderSupervisorRunHeadline(r) {
+	const o = r.orchestration;
+	const steerNote = isUnavailable(o.steers) ? `unavailable — ${o.steers.unavailable}` : o.steers === 0 ? "0 (spawn→wait→respawn only; no mid-task steering)" : `${o.steers} queued / ${showMeasured(o.steersDelivered)} delivered`;
+	return [
+		`RUN-REPORT ${r.instanceId ?? "?"} [${r.arm ?? "?"}]`,
+		`  status=${showMeasured(r.outcome.supStatus)} source=${showMeasured(r.outcome.supStatusSource)} reason=${showMeasured(r.outcome.supReason)} failure=${fmtFailure(r.outcome.failure)}`,
+		`  steers=${steerNote}`,
+		`  waves=${showMeasured(o.waves)} sizes=${isUnavailable(o.waveSizes) ? `unavailable — ${o.waveSizes.unavailable}` : `[${o.waveSizes.join(",")}]`} workers=${showMeasured(o.workersSpawned)} settled=${showMeasured(o.workersSettled)} cancelled=${showMeasured(o.workersCancelled)}`,
+		`  concurrency max=${showMeasured(o.maxConcurrency)} utilization=${showMeasured(o.workerUtilization)} idle=${fmtMs(o.idleMs)} (${showMeasured(o.idlePct)}%) wall=${fmtMs(o.supervisorWallMs)}${o.supervisorWallSource === "journal-span" ? " (journal-span lower bound)" : ""}`,
+		`  respawns=${showMeasured(o.respawns)} evidence→respawn=${showMeasured(r.decision.observeThenRespawn)} blind-respawn=${showMeasured(r.decision.respawnWithoutEvidence)} depth=${showMeasured(o.delegationDepth)}`,
+		`  accepted=${showMeasured(r.decision.accepted)} rejected=${showMeasured(r.decision.rejected)} empty-pass=${showMeasured(r.decision.emptyPass)}`,
+		`  brain=$${showMeasured(r.economics.brain.usd)} total=$${showMeasured(r.economics.totalUsd)} judge.resolved=${showMeasured(r.outcome.judgeResolved)} score=${showMeasured(r.outcome.judgeScore)} verify=${showMeasured(r.outcome.verifyPass)}`,
+		r.gaps.length > 0 ? `  gaps(${r.gaps.length}): ${r.gaps.join("; ")}` : "  gaps: none"
+	].join("\n");
 }
 /**
-* CLI driver for `agent-eval supervisor-run`. `argv` is everything after the
-* command name, so `['report', runDir, ...flags]`. Returns the process exit code.
+* The recorded error on one line; `none recorded` and `unavailable` stay
+* distinct, and a throw the run outlived is labeled so it is not read as the
+* outcome of the settled result beside it.
 */
-async function runSupervisorRunCommand(argv, io = PROCESS_IO) {
-	const [verb, ...rest] = argv;
-	if (verb === void 0 || verb === "--help" || verb === "-h" || verb === "help") {
-		io.stdout(`${SUPERVISOR_RUN_USAGE}\n`);
-		return verb === void 0 ? 2 : 0;
-	}
-	if (verb !== "report") {
-		io.stderr(`unknown supervisor-run subcommand "${verb}"\n${SUPERVISOR_RUN_USAGE}\n`);
-		return 2;
-	}
-	if (rest.includes("--help") || rest.includes("-h")) {
-		io.stdout(`${SUPERVISOR_RUN_USAGE}\n`);
-		return 0;
+function fmtFailure(v) {
+	if (isUnavailable(v)) return `unavailable — ${v.unavailable}`;
+	if (v === null) return "none recorded";
+	const name = v.name ?? "unavailable — record has no error.name";
+	const message = v.message ?? "unavailable — record has no error.message";
+	const attempt = v.earlierAttempt ? ", earlier attempt" : "";
+	return `${name}: ${message} [${v.source}${v.at === null ? "" : ` at ${v.at}`}${attempt}]`;
+}
+function resourceRows(records) {
+	if (isUnavailable(records)) return [`unavailable — ${records.unavailable}`, ""];
+	const out = [
+		"Receipts are source records, not additive totals.",
+		"A false completeness flag means the amount is a recorded subtotal, not complete usage.",
+		"",
+		"| Node | Source | Resource | Unit | Amount | Known |",
+		"|---|---|---|---|---:|---|"
+	];
+	const cell = (value) => value.replace(/\|/g, "\\|").replace(/\r?\n/g, " ");
+	for (const record of records) {
+		const prefix = `| ${cell(record.nodeId ?? "unavailable")} | ${cell(record.source)} |`;
+		if (isUnavailable(record.resources)) out.push(`${prefix} ${cell(showMeasured(record.resources))} | — | — | — |`);
+		else if (record.resources.length === 0) out.push(`${prefix} recorded empty map | — | — | — |`);
+		else for (const receipt of record.resources) out.push(`${prefix} ${cell(showMeasured(receipt.name))} | ${cell(showMeasured(receipt.unit))} | ${cell(showMeasured(receipt.amount))} | ${cell(showMeasured(receipt.known))} |`);
 	}
-	let args;
-	try {
-		args = parseReportArgs(rest);
-	} catch (error) {
-		io.stderr(`${errorMessage(error)}\n`);
-		return 2;
+	if (!records.length) out.push("| — | — | no spend records | — | — | — |");
+	out.push("");
+	return out;
+}
+function fmtMs(v) {
+	if (isUnavailable(v)) return `unavailable — ${v.unavailable}`;
+	if (v < 1e3) return `${v}ms`;
+	const s = v / 1e3;
+	if (s < 120) return `${roundDecimal(s, 1)}s`;
+	return `${roundDecimal(s / 60, 1)}min`;
+}
+function renderSupervisorRunMarkdown(r) {
+	const o = r.orchestration;
+	const d = r.decision;
+	const e = r.economics;
+	const out = [];
+	out.push(`# Run report — ${r.instanceId ?? "unknown instance"} [${r.arm ?? "unknown arm"}]`);
+	out.push("");
+	out.push("```");
+	out.push(renderSupervisorRunHeadline(r));
+	out.push("```");
+	out.push("");
+	out.push(`- Run: \`${r.runRef}\``);
+	out.push(`- Supervisor: \`${showMeasured(r.supervisorId)}\``);
+	out.push(`- Supervisor profile: \`${showMeasured(r.supervisorProfileDigest)}\``);
+	out.push(`- Generated: ${r.generatedAt}`);
+	out.push("");
+	out.push("## Orchestration");
+	out.push("");
+	out.push("| Metric | Value |");
+	out.push("|---|---|");
+	out.push(`| Workers spawned | ${showMeasured(o.workersSpawned)} |`);
+	out.push(`| Workers settled | ${showMeasured(o.workersSettled)} |`);
+	out.push(`| Workers cancelled | ${showMeasured(o.workersCancelled)} |`);
+	out.push(`| **Steers (mid-task messages to live workers)** | **${showMeasured(o.steers)}** |`);
+	out.push(`| Steers delivered | ${showMeasured(o.steersDelivered)} |`);
+	out.push(`| Outer-driver \`supervisor_steer\` calls | ${showMeasured(o.driverSteerCalls)} |`);
+	out.push(`| Spawn waves | ${showMeasured(o.waves)} |`);
+	out.push(`| Wave sizes | ${isUnavailable(o.waveSizes) ? showMeasured(o.waveSizes) : `[${o.waveSizes.join(", ")}]`} |`);
+	out.push(`| Max concurrency | ${showMeasured(o.maxConcurrency)} |`);
+	out.push(`| Respawns (after same parent's first settle) | ${showMeasured(o.respawns)} |`);
+	out.push(`| Repeated labels | ${isUnavailable(o.repeatedLabels) ? showMeasured(o.repeatedLabels) : o.repeatedLabels.length === 0 ? "none" : o.repeatedLabels.join(", ")} |`);
+	out.push(`| Delegation depth | ${showMeasured(o.delegationDepth)} |`);
+	out.push(`| Time to first spawn | ${fmtMs(o.timeToFirstSpawnMs)} |`);
+	out.push(`| Supervisor wall | ${fmtMs(o.supervisorWallMs)}${isUnavailable(o.supervisorWallSource) ? "" : ` (source: ${o.supervisorWallSource}${o.supervisorWallSource === "journal-span" ? ", lower bound" : ""})`} |`);
+	out.push(`| Idle (zero live workers) | ${fmtMs(o.idleMs)} (${showMeasured(o.idlePct)}%) |`);
+	out.push(`| Worker utilization (Σ worker wall ÷ supervisor wall) | ${showMeasured(o.workerUtilization)} |`);
+	out.push("");
+	if (!isUnavailable(o.steersByWorker) && o.steersByWorker.length > 0) {
+		out.push("### Steers per worker");
+		out.push("");
+		out.push("| Worker id | Label | Queued | Delivered |");
+		out.push("|---|---|---:|---:|");
+		for (const s of o.steersByWorker) out.push(`| ${s.workerId === null ? "unavailable — worker id absent; label join used" : `\`${s.workerId}\``} | \`${s.worker}\` | ${s.queued} | ${s.delivered} |`);
+		out.push("");
+	} else if (isUnavailable(o.steersByWorker)) out.push(`### Steers per worker\n\nunavailable — ${o.steersByWorker.unavailable}\n`);
+	out.push("## Decision quality");
+	out.push("");
+	out.push("| Metric | Value |");
+	out.push("|---|---|");
+	out.push(`| Settled by status | ${fmtCounts(d.settledByStatus)} |`);
+	out.push(`| Settled verdicts | ${fmtCounts(d.settledVerdicts)} |`);
+	out.push(`| Accepted (verify green + patch bytes) | ${showMeasured(d.accepted)} |`);
+	out.push(`| Rejected (verify red) | ${showMeasured(d.rejected)} |`);
+	out.push(`| Empty pass (green, no patch) | ${showMeasured(d.emptyPass)} |`);
+	out.push(`| Evidence → respawn sequences | ${showMeasured(d.observeThenRespawn)} |`);
+	out.push(`| Respawn with no same-parent settled evidence in front | ${showMeasured(d.respawnWithoutEvidence)} |`);
+	out.push(`| Review actions (steers + worker questions) | ${showMeasured(d.reviewActions)} |`);
+	out.push(`| Worker evidence returned | ${isUnavailable(d.workerEvidenceBytes) ? showMeasured(d.workerEvidenceBytes) : `${d.workerEvidenceBytes} bytes`} |`);
+	out.push("");
+	out.push("## Economics");
+	out.push("");
+	out.push("| Role | Tokens in | Tokens out | Cache read | Cache write | USD | Source |");
+	out.push("|---|---:|---:|---:|---:|---:|---|");
+	out.push(`| brain | ${showMeasured(e.brain.tokensIn)} | ${showMeasured(e.brain.tokensOut)} | ${showMeasured(e.brain.cacheRead)} | ${showMeasured(e.brain.cacheWrite)} | ${showMeasured(e.brain.usd)} | ${e.brain.source} |`);
+	out.push(`| workers | ${showMeasured(e.workers.tokensIn)} | ${showMeasured(e.workers.tokensOut)} | ${showMeasured(e.workers.cacheRead)} | ${showMeasured(e.workers.cacheWrite)} | ${showMeasured(e.workers.usd)} | ${e.workers.source} |`);
+	out.push("");
+	if (!isUnavailable(e.brainTruncations) && e.brainTruncations > 0) out.push(`- **BRAIN OUTPUT TRUNCATED: ${e.brainTruncations} completion(s) hit \`finish_reason: "length"\`** — the supervisor acted on a half-written plan. Its output ceiling is too low; see \`brain.jsonl\` for the per-call \`req_max_tokens\`.`);
+	else out.push(`- Brain completions truncated (finish_reason=length): ${showMeasured(e.brainTruncations)}`);
+	out.push(`- Total USD: ${showMeasured(e.totalUsd)} (source: ${e.totalUsdSource})`);
+	out.push(`- Spend measured two ways: journal-derived $${showMeasured(e.spend.journalDerived.usd)} over ${e.spend.journalDerived.records} journal record(s) (execution accounting) · close-record $${showMeasured(e.spend.closeRecord.usd)} over ${e.spend.closeRecord.records} close record(s) (billing-shaped). Divergence is a signal, not an error.`);
+	out.push(`- Cost per accepted patch: ${showMeasured(e.costPerAcceptedPatchUsd)}`);
+	if (isUnavailable(e.workerWallMsDistribution)) out.push(`- Worker wall distribution: unavailable — ${e.workerWallMsDistribution.unavailable}`);
+	else {
+		const w = e.workerWallMsDistribution;
+		out.push(`- Worker wall (n=${w.n}): min ${fmtMs(w.min)} / p50 ${fmtMs(w.p50)} / p90 ${fmtMs(w.p90)} / max ${fmtMs(w.max)} / Σ ${fmtMs(w.sum)}`);
 	}
-	const runDir = resolve(args.runDir);
-	const entry = await stat(runDir).catch(() => null);
-	if (entry === null || !entry.isDirectory()) {
-		io.stderr(`[agent-eval] supervisor-run report: ${runDir} is not a directory\n`);
-		return 1;
+	out.push("");
+	if (!isUnavailable(e.perWorker) && e.perWorker.length > 0) {
+		out.push("| Worker id | Label | Role | Runtime | Profile digest | Status | Failure | Infra | Wall | Tokens in | Tokens out | Patch bytes | Verify passed | Score |");
+		out.push("|---|---|---|---|---|---|---|---|---:|---:|---:|---:|---|---:|");
+		for (const w of e.perWorker) out.push(`| ${w.workerId === null ? "unavailable — worker id absent; label join used" : `\`${w.workerId}\``} | \`${w.worker}\` | ${w.role ?? "unavailable — source recorded no role"} | ${w.runtime ?? "unavailable — source recorded no runtime"} | ${w.profileDigest === null ? "unavailable — source recorded no profile digest" : `\`${w.profileDigest}\``} | ${w.status ?? "unavailable — no terminal event"} | ${w.failure ?? "none recorded"} | ${w.infra ?? "unavailable"} | ${w.wallMs === null ? "unavailable — no spawn/finish pair" : fmtMs(w.wallMs)} | ${w.tokensIn ?? "unavailable — store does not attribute tokens per worker"} | ${w.tokensOut ?? "unavailable — store does not attribute tokens per worker"} | ${w.patchBytes ?? "unavailable — no worker patch file"} | ${w.passed === null ? "unavailable — no verdict" : String(w.passed)} | ${w.score ?? "unavailable — no numeric score"} |`);
+		out.push("");
 	}
-	try {
-		const report = await analyzeSupervisorRun(runDir);
-		switch (args.format) {
-			case "headline":
-				io.stdout(`${renderSupervisorRunHeadline(report)}\n`);
-				break;
-			case "markdown":
-				io.stdout(renderSupervisorRunMarkdown(report));
-				break;
-			case "json": io.stdout(`${JSON.stringify(report, null, 2)}\n`);
-		}
-		return 0;
-	} catch (error) {
-		io.stderr(`[agent-eval] supervisor-run report failed for ${runDir}: ${errorMessage(error)}\n`);
-		return 1;
+	if (e.resourceRecords !== void 0) out.push("## Named resource receipts", "", ...resourceRows(e.resourceRecords));
+	out.push("## Outcome");
+	out.push("");
+	out.push("| Metric | Value |");
+	out.push("|---|---|");
+	out.push(`| Supervisor status | ${showMeasured(r.outcome.supStatus)} |`);
+	out.push(`| Status source | ${showMeasured(r.outcome.supStatusSource)} |`);
+	out.push(`| Terminal reason | ${showMeasured(r.outcome.supReason)} |`);
+	out.push(`| Failure | ${fmtFailure(r.outcome.failure)} |`);
+	out.push(`| Supervisor verdict | ${showMeasured(r.outcome.supVerdict)} |`);
+	out.push(`| Delivered | ${showMeasured(r.outcome.delivered)} |`);
+	out.push(`| Judge resolved | ${showMeasured(r.outcome.judgeResolved)} |`);
+	out.push(`| Judge score | ${showMeasured(r.outcome.judgeScore)} |`);
+	out.push(`| Judge passed / total | ${showMeasured(r.outcome.judgePassed)} / ${showMeasured(r.outcome.judgeTotal)} |`);
+	out.push(`| Judge source | ${r.outcome.judgeSource ?? "unavailable — no judge.json and no ledger row"} |`);
+	out.push(`| Verify gate | pass=${showMeasured(r.outcome.verifyPass)} rc=${showMeasured(r.outcome.verifyRc)} |`);
+	if (isUnavailable(r.outcome.patch)) out.push(`| Patch | unavailable — ${r.outcome.patch.unavailable} |`);
+	else {
+		const p = r.outcome.patch;
+		out.push(`| Patch | ${p.files} file(s), +${p.linesAdded}/-${p.linesRemoved}, test files touched: ${p.testFilesTouched.length === 0 ? "none" : p.testFilesTouched.join(", ")} |`);
 	}
+	out.push("");
+	out.push("## Gaps");
+	out.push("");
+	if (r.gaps.length === 0) out.push("None — every metric above is backed by a present artifact.");
+	else for (const g of r.gaps) out.push(`- ${g}`);
+	out.push("");
+	out.push(`> Harness-session view of the same run (model calls, stuck loops, tool errors): \`${r.traceCommand}\``);
+	out.push("");
+	return out.join("\n");
+}
+function fmtCounts(v) {
+	if (isUnavailable(v)) return showMeasured(v);
+	const entries = Object.entries(v);
+	return entries.length === 0 ? "none" : entries.map(([k, n]) => `${k}=${n}`).join(", ");
+}
+function renderSupervisorRollupMarkdown(rollup, title = "Round rollup") {
+	const out = [];
+	out.push(`# ${title}`);
+	out.push("");
+	out.push(`- Cells: ${rollup.cells}`);
+	out.push(`- **Steers across all cells: ${showMeasured(rollup.steersTotal)}** (cells with ≥1 steer: ${showMeasured(rollup.cellsWithSteers)}; cells where the steer count is unavailable: ${rollup.cellsWithUnavailableSteers})`);
+	out.push(`- Waves per cell (mean): ${showMeasured(rollup.wavesMean)}`);
+	out.push(`- Max concurrency observed: ${showMeasured(rollup.maxConcurrencyMax)}`);
+	out.push(`- Worker utilization (mean): ${showMeasured(rollup.utilizationMean)}`);
+	out.push(`- Idle share (mean): ${showMeasured(rollup.idlePctMean)}%`);
+	out.push(`- Workers spawned: ${showMeasured(rollup.workersSpawnedTotal)} · accepted: ${showMeasured(rollup.acceptedTotal)}`);
+	out.push(`- Spend: $${showMeasured(rollup.usdTotal)} · judged resolved: ${showMeasured(rollup.resolvedCount)}/${rollup.cells}`);
+	out.push(`- Spend measured two ways: journal-derived $${showMeasured(rollup.spendUsd.journalDerived.value)} over ${rollup.spendUsd.journalDerived.runs}/${rollup.cells} runs (execution accounting) · close-record $${showMeasured(rollup.spendUsd.closeRecord.value)} over ${rollup.spendUsd.closeRecord.runs}/${rollup.cells} runs (billing-shaped)`);
+	out.push("");
+	out.push("| Instance | Arm | Steers | Waves | Utilization | Idle % | Resolved | USD |");
+	out.push("|---|---|---:|---:|---:|---:|---|---:|");
+	for (const c of rollup.perCell) out.push(`| ${c.instanceId ?? "?"} | ${c.arm ?? "?"} | ${showMeasured(c.steers)} | ${showMeasured(c.waves)} | ${showMeasured(c.utilization)} | ${showMeasured(c.idlePct)} | ${showMeasured(c.resolved)} | ${showMeasured(c.usd)} |`);
+	out.push("");
+	for (const cell of rollup.perCell) if (cell.resourceRecords !== void 0) out.push(`## Resource receipts — ${cell.instanceId ?? "?"} [${cell.arm ?? "?"}]`, "", ...resourceRows(cell.resourceRecords));
+	return out.join("\n");
 }
 //#endregion
-export { writeSupervisorRunReportSafe as a, runtimeSupervisorRunReader as c, renderSupervisorRunMarkdown as d, analyzeSupervisorRunSources as f, writeSupervisorRunReport as i, renderSupervisorRollupMarkdown as l, analyzeSupervisorRun as n, isRuntimeSupervisorRunDir as o, parsePatch as p, findSupervisorRunDirs as r, readRuntimeSupervisorRun as s, runSupervisorRunCommand as t, renderSupervisorRunHeadline as u };
+export { findSupervisorRunDirs as a, analyzeSupervisorRunSources as c, analyzeSupervisorRun as i, parsePatch as l, renderSupervisorRunHeadline as n, isRuntimeSupervisorRunDir as o, renderSupervisorRunMarkdown as r, readRuntimeSupervisorRun as s, renderSupervisorRollupMarkdown as t };
 
-//# sourceMappingURL=report-command-SC9nqb4M.js.map
+//# sourceMappingURL=render-C-w2EZ6p.js.map