fix(supervisor-run): read Runtime file run contexts

.evolve/skill-runs.jsonl

@@ -40,3 +40,5 @@
 {"skill":"/agent-eval","ts":"2026-07-30T11:07:02Z","project":"agent-eval-analyst-proof-20260730","target":"crash-safe resumable public analyst benchmark command","operatorPrompt":"","durationMin":null,"verdict":"PASS","dispatchedTo":"/verify","operatorOverride":null,"transcriptPath":null,"traceDir":null}
 {"skill":"/verify","ts":"2026-07-30T11:07:07Z","project":"agent-eval-analyst-proof-20260730","target":"benchmark command recovery, artifacts, and AgentRx metrics","operatorPrompt":"","durationMin":null,"verdict":"PASS","dispatchedTo":"/stop","operatorOverride":null,"transcriptPath":null,"traceDir":null}
 {"skill":"/semgrep","ts":"2026-07-30T15:47:18Z","project":"agent-eval-analyst-proof-20260730","target":"agent-eval public analyst benchmark, 664 files, 492 rules","operatorPrompt":"","durationMin":null,"verdict":"PASS","dispatchedTo":"/stop","operatorOverride":null,"transcriptPath":null,"traceDir":null}
+{"skill":"/agent-eval","ts":"2026-07-30T17:36:56Z","project":"agent-eval-runtime-run-reader","target":"Runtime FileRunContext supervisor-run reader","operatorPrompt":"","durationMin":null,"verdict":"PASS","dispatchedTo":"/verify","operatorOverride":null,"transcriptPath":null,"traceDir":null}
+{"skill":"/verify","ts":"2026-07-30T17:36:56Z","project":"agent-eval-runtime-run-reader","target":"Runtime FileRunContext reader, real traces CLI path, and package","operatorPrompt":"","durationMin":null,"verdict":"PASS","dispatchedTo":"/stop","operatorOverride":null,"transcriptPath":null,"traceDir":null}

CHANGELOG.md

@@ -28,6 +28,9 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
 
 ### Fixed
 
+- Supervisor-run analysis now reads agent-runtime `createFileRunContext(dir)` directories directly.
+  It composes nested journal envelopes by exact node identity, preserves profile digests, runtime tags, failures, and metered spend, and leaves unrecorded roles, terminal times, policy, transcripts, verification, and judge evidence unavailable.
+  A direct run path produces one report, while a parent directory containing multiple direct runs produces a rollup.
 - Canonical `trace://<trace>/span/<span>` evidence is classified as span evidence instead of artifact evidence.
 - Public model output selects positive integer assistant step ids.
   The runner builds canonical trace URIs and exact action excerpts from those spans, and rejects missing, non-assistant, or empty steps.

README.md

@@ -372,7 +372,7 @@ See [concepts](./docs/concepts.md), [customer paths](./docs/customer-journeys.md
 | `@tangle-network/agent-eval/reporting` | Statistical comparisons and report rendering. |
 | `@tangle-network/agent-eval/analyst` | Built-in and custom trace analysts, labeled comparison, costs, and reports. |
 | `@tangle-network/agent-eval/traces` | Store, replay, and inspect structured traces. |
-| `@tangle-network/agent-eval/supervisor-run` | Read and analyze recursive supervisor runs without collapsing missing measurements to zero. |
+| `@tangle-network/agent-eval/supervisor-run` | Read loops or agent-runtime recursive run directories without collapsing missing measurements to zero. |
 | `@tangle-network/agent-eval/benchmarks` | Benchmark adapters and retrieval metrics. |
 | `@tangle-network/agent-eval/rl` | Export rewards, preferences, and training rows. |
 | `@tangle-network/agent-eval/wire` | HTTP and RPC schemas for other languages. |

docs/trace-analysis.md

@@ -121,6 +121,14 @@ for (const c of overview.error_clusters) {
 
 `CONTROL_INTEGRITY_ANALYST` checks the existing `SupervisorRunSources` or `SupervisorRunTree` directly.
 It does not define another run format.
+`analyzeSupervisorRun(runDir)` detects both the loops layout and an agent-runtime `createFileRunContext(dir)` layout.
+Runtime directories contain `spawn-journal.jsonl` plus optional `result.json` and `trajectory.json`.
+The Runtime reader unwraps `{ kind: 'begin' | 'event', root, event }` storage envelopes and feeds the normalized events into the same source parser and analyzer used by every other reader.
+Nested tree root markers are joined to their parent spawn by exact node id.
+Profile digests, runtime tags, terminal reasons, structured verdicts, and metered spend are retained exactly when recorded.
+The reader does not derive invocation roles, outcome quality, completion time, provider policy, worker transcripts, or verification results from labels or artifacts.
+Those fields remain unavailable when their source is absent.
+
 Register it as a custom-input analyst and pass the existing value under its stable id:
 
 ```ts
@@ -129,10 +137,10 @@ import {
   CONTROL_INTEGRITY_ANALYST,
 } from '@tangle-network/agent-eval/analyst'
 import {
-  readLoopsSupervisorRun,
+  readRuntimeSupervisorRun,
 } from '@tangle-network/agent-eval/supervisor-run'
 
-const sources = await readLoopsSupervisorRun(runDir)
+const sources = await readRuntimeSupervisorRun(runDir)
 const registry = new AnalystRegistry()
 registry.register(CONTROL_INTEGRITY_ANALYST)
 
@@ -142,6 +150,7 @@ const result = await registry.run('run-123', {
 ```
 
 Pass `SupervisorRunSources` when it is available.
+Use `readLoopsSupervisorRun()` when the input is explicitly the loops layout.
 A `SupervisorRunTree` does not retain raw journal multiplicity or worker request and acknowledgement rows, so tree input explicitly reports those checks as unavailable.
 
 The deterministic pass can prove only facts represented by these two existing surfaces.

src/index.ts

@@ -557,12 +557,15 @@ export {
   analyzeSupervisorRunIntegrity,
   analyzeSupervisorRunSources,
   claudeCodeSupervisorRunReader,
+  isRuntimeSupervisorRunDir,
   isUnavailable,
   type Measured,
   readClaudeCodeSupervisorRun,
+  readRuntimeSupervisorRun,
   renderSupervisorRunHeadline,
   renderSupervisorRunMarkdown,
   rollupSupervisorRuns,
+  runtimeSupervisorRunReader,
   type SourceLimits,
   SUPERVISOR_RUN_INTEGRITY_SCHEMA,
   SUPERVISOR_RUN_SCHEMA,

src/supervisor-run/analyze.ts

@@ -77,6 +77,8 @@ export function analyzeSupervisorRunSources(
   const result = parseJson(src.result)
   const judge = parseJson(src.judge)
   const { rootId, workerSpawns, workerCloses, startedAt, completedAt } = tree
+  const rootSpawn =
+    rootId === null ? null : (tree.spawns.find((spawn) => spawn.id === rootId) ?? null)
   const spawnById = new Map(workerSpawns.map((spawn) => [spawn.id, spawn]))
   const spawnsByLabel = new Map<string, SpawnRow[]>()
   for (const spawn of workerSpawns) {
@@ -410,9 +412,13 @@ export function analyzeSupervisorRunSources(
   }
 
   // ── economics ──────────────────────────────────────────────────────────
-  const journalWorkerIn = workerCloses.reduce((a, c) => a + c.spend.tokens.input, 0)
-  const journalWorkerOut = workerCloses.reduce((a, c) => a + c.spend.tokens.output, 0)
-  const journalWorkerUsd = workerCloses.reduce((a, c) => a + c.spend.usd, 0)
+  const rootChildIds = new Set(
+    workerSpawns.filter((spawn) => spawn.parent === rootId).map((spawn) => spawn.id),
+  )
+  const rootChildCloses = workerCloses.filter((close) => rootChildIds.has(close.id))
+  const journalWorkerIn = rootChildCloses.reduce((a, c) => a + c.spend.tokens.input, 0)
+  const journalWorkerOut = rootChildCloses.reduce((a, c) => a + c.spend.tokens.output, 0)
+  const journalWorkerUsd = rootChildCloses.reduce((a, c) => a + c.spend.usd, 0)
   const workerUsdById = new Map<string, number>()
   for (const c of workerCloses) {
     workerUsdById.set(c.id, (workerUsdById.get(c.id) ?? 0) + c.spend.usd)
@@ -466,13 +472,29 @@ export function analyzeSupervisorRunSources(
     const close = spawn === null ? null : (closeById.get(spawn.id) ?? null)
     const passed = close?.valid ?? f?.passed ?? null
     const matchingRoles = new Set(matchingSpawns.map((candidate) => candidate.role))
+    const matchingRuntimes = new Set(matchingSpawns.map((candidate) => candidate.runtime))
+    const matchingProfiles = new Set(matchingSpawns.map((candidate) => candidate.profileDigest))
+    const journalWallMs =
+      spawn?.at !== null &&
+      spawn?.at !== undefined &&
+      close?.at !== null &&
+      close?.at !== undefined &&
+      close.at >= spawn.at
+        ? close.at - spawn.at
+        : null
     return {
       workerId: w.workerId ?? null,
       worker: w.label,
       role: matchingRoles.size === 1 ? (matchingSpawns[0]?.role ?? null) : null,
-      wallMs: f?.started != null && f.finishedAt != null ? f.finishedAt - f.started : null,
-      tokensIn: w.tokensIn ?? null,
-      tokensOut: w.tokensOut ?? null,
+      runtime: matchingRuntimes.size === 1 ? (matchingSpawns[0]?.runtime ?? null) : null,
+      profileDigest:
+        matchingProfiles.size === 1 ? (matchingSpawns[0]?.profileDigest ?? null) : null,
+      status: close?.status ?? null,
+      failure: close?.reason ?? null,
+      infra: close?.infra ?? null,
+      wallMs: f?.started != null && f.finishedAt != null ? f.finishedAt - f.started : journalWallMs,
+      tokensIn: w.tokensIn ?? (close?.hasSpend ? close.spend.tokens.input : null),
+      tokensOut: w.tokensOut ?? (close?.hasSpend ? close.spend.tokens.output : null),
       usd:
         usdLimit !== null
           ? null
@@ -535,9 +557,10 @@ export function analyzeSupervisorRunSources(
       src.brainLog === null
         ? gap(
             'brain.brainTruncations',
-            src.supRunDir === null
-              ? 'no supervisor run dir under <ws>/.loops/supervisor'
-              : 'brain.jsonl absent — loops predates the brain-call tap, so truncation cannot be ruled out',
+            src.brainLogMissingReason ??
+              (src.supRunDir === null
+                ? 'no supervisor run dir under <ws>/.loops/supervisor'
+                : 'brain.jsonl absent — loops predates the brain-call tap, so truncation cannot be ruled out'),
           )
         : brainCalls.filter((c) => c.finish_reason === 'length').length,
     workers: {
@@ -573,7 +596,7 @@ export function analyzeSupervisorRunSources(
       usdLimit !== null
         ? usdLimit
         : stateUsd !== null
-          ? `state.json result.spentUsd${journalWorkerUsd === 0 ? ' — brain-priced only; worker CLI inference is unpriced (see worker token counts)' : ''}`
+          ? `state.json result.spentUsd${rootChildCloses.length > 0 && journalWorkerUsd === 0 ? ' — brain-priced only; worker CLI inference is unpriced (see worker token counts)' : ''}`
           : haveJournal
             ? 'journal metered + settled usd'
             : journalMissing,
@@ -663,6 +686,10 @@ export function analyzeSupervisorRunSources(
     instanceId: src.instanceId,
     arm: src.arm,
     supervisorId: rootId !== null ? rootId : unavailable(journalMissing),
+    supervisorProfileDigest:
+      rootSpawn?.profileDigest !== null && rootSpawn?.profileDigest !== undefined
+        ? rootSpawn.profileDigest
+        : gap('supervisorProfileDigest', 'root spawned event has no profile digest'),
     generatedAt: new Date(now()).toISOString(),
     orchestration,
     decision,

src/supervisor-run/fixtures.ts

@@ -26,6 +26,7 @@ export function fixtureJournal(plan: JournalPlan): string {
       kind: 'spawned',
       id: root,
       label: 'root',
+      role: 'supervisor',
       budget: {},
       runtime: 'inline',
       seq: 0,
@@ -42,6 +43,7 @@ export function fixtureJournal(plan: JournalPlan): string {
         id,
         parent: root,
         label,
+        role: 'worker',
         budget: {},
         runtime: 'inline',
         seq: i,

src/supervisor-run/index.test.ts

@@ -1,7 +1,15 @@
 import { describe, expect, it } from 'vitest'
-import type { SupervisorRunNodeRole as RootSupervisorRunNodeRole } from '../index'
 import {
+  type SupervisorRunNodeRole as RootSupervisorRunNodeRole,
+  isRuntimeSupervisorRunDir as rootIsRuntimeSupervisorRunDir,
+  readRuntimeSupervisorRun as rootReadRuntimeSupervisorRun,
+  runtimeSupervisorRunReader as rootRuntimeSupervisorRunReader,
+} from '../index'
+import {
+  isRuntimeSupervisorRunDir,
   NO_SOURCE_LIMITS,
+  readRuntimeSupervisorRun,
+  runtimeSupervisorRunReader,
   type SourceLimits,
   type SupervisorRunNodeRole,
   type WorkerLogSource,
@@ -23,5 +31,11 @@ describe('supervisor-run public contract', () => {
     expect(rootRole).toBe('supervisor')
     expect(worker.workerId).toBe('worker-1')
     expect(limits).toMatchObject({ managerTokens: null, workerTokens: null })
+    expect(typeof readRuntimeSupervisorRun).toBe('function')
+    expect(typeof runtimeSupervisorRunReader).toBe('function')
+    expect(typeof isRuntimeSupervisorRunDir).toBe('function')
+    expect(rootReadRuntimeSupervisorRun).toBe(readRuntimeSupervisorRun)
+    expect(rootRuntimeSupervisorRunReader).toBe(runtimeSupervisorRunReader)
+    expect(rootIsRuntimeSupervisorRunDir).toBe(isRuntimeSupervisorRunDir)
   })
 })

src/supervisor-run/index.ts

@@ -62,6 +62,11 @@ export {
   renderSupervisorRunMarkdown,
 } from './render'
 export { type SupervisorRolloutOptions, supervisorRunRolloutLines } from './rollout-nodes'
+export {
+  isRuntimeSupervisorRunDir,
+  readRuntimeSupervisorRun,
+  runtimeSupervisorRunReader,
+} from './runtime-reader'
 export {
   type DecisionMetrics,
   type EconomicsMetrics,

src/supervisor-run/loops-reader.ts

@@ -25,6 +25,7 @@ import {
   renderSupervisorRunHeadline,
   renderSupervisorRunMarkdown,
 } from './render'
+import { isRuntimeSupervisorRunDir, readRuntimeSupervisorRun } from './runtime-reader'
 import {
   NO_SOURCE_LIMITS,
   type SupervisorRunReader,
@@ -270,7 +271,11 @@ export async function analyzeSupervisorRun(
   opts: LoopsReaderOptions = {},
 ): Promise<SupervisorRunReport> {
   if (typeof input === 'string') {
-    return analyzeSupervisorRunSources(await readLoopsSupervisorRun(input, opts))
+    return analyzeSupervisorRunSources(
+      (await isRuntimeSupervisorRunDir(input))
+        ? await readRuntimeSupervisorRun(input)
+        : await readLoopsSupervisorRun(input, opts),
+    )
   }
   if (isReader(input)) return analyzeSupervisorRunSources(await input.read())
   return analyzeSupervisorRunSources(input)
@@ -303,7 +308,9 @@ export async function writeSupervisorRunReport(
   runDir: string,
   opts: WriteSupervisorRunOptions = {},
 ): Promise<SupervisorRunReport> {
-  const sources = await readLoopsSupervisorRun(runDir, opts)
+  const sources = (await isRuntimeSupervisorRunDir(runDir))
+    ? await readRuntimeSupervisorRun(runDir)
+    : await readLoopsSupervisorRun(runDir, opts)
   const report = analyzeSupervisorRunSources(sources)
   const md = renderSupervisorRunMarkdown(report)
   const dest = opts.reportDir ?? runDir
@@ -389,8 +396,19 @@ export async function reportSupervisorRound(
   return rollup
 }
 
-/** Every `<...>/runs/<iid>/<arm>` directory under `root`. */
+/**
+ * Every loops or Runtime supervisor run below `root`.
+ *
+ * When `root` itself is one run, return no children so callers can distinguish
+ * a single report from a parent-directory rollup.
+ */
 export async function findSupervisorRunDirs(root: string): Promise<string[]> {
+  if (
+    (await isRuntimeSupervisorRunDir(root)) ||
+    (await findSupervisorRunDirIn(join(root, 'ws'))) !== null
+  ) {
+    return []
+  }
   const found: string[] = []
   const walk = async (dir: string, depth: number): Promise<void> => {
     if (depth > 8) return
@@ -399,13 +417,11 @@ export async function findSupervisorRunDirs(root: string): Promise<string[]> {
       if (!e.isDirectory()) continue
       if (e.name === 'node_modules' || e.name === '.git') continue
       const full = join(dir, e.name)
-      if (e.name === 'runs') {
-        const iids = await readdir(full, { withFileTypes: true }).catch(() => [])
-        for (const iid of iids) {
-          if (!iid.isDirectory()) continue
-          const arms = await readdir(join(full, iid.name), { withFileTypes: true }).catch(() => [])
-          for (const arm of arms) if (arm.isDirectory()) found.push(join(full, iid.name, arm.name))
-        }
+      if (
+        (await isRuntimeSupervisorRunDir(full)) ||
+        (await findSupervisorRunDirIn(join(full, 'ws'))) !== null
+      ) {
+        found.push(full)
         continue
       }
       await walk(full, depth + 1)

src/supervisor-run/render.ts

@@ -62,6 +62,7 @@ export function renderSupervisorRunMarkdown(r: SupervisorRunReport): string {
   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('')
 
@@ -161,12 +162,12 @@ export function renderSupervisorRunMarkdown(r: SupervisorRunReport): string {
   out.push('')
   if (!isUnavailable(e.perWorker) && e.perWorker.length > 0) {
     out.push(
-      '| Worker id | Label | Role | Wall | Tokens in | Tokens out | Patch bytes | Verify passed | Score |',
+      '| Worker id | Label | Role | Runtime | Profile digest | Status | Failure | Infra | Wall | Tokens in | Tokens out | Patch bytes | Verify passed | Score |',
     )
-    out.push('|---|---|---|---:|---:|---:|---:|---|---:|')
+    out.push('|---|---|---|---|---|---|---|---|---:|---:|---:|---:|---|---:|')
     for (const w of e.perWorker) {
       out.push(
-        `| ${w.workerId === null ? 'unavailable — legacy label join' : `\`${w.workerId}\``} | \`${w.worker}\` | ${w.role ?? 'unavailable — no journal join'} | ${w.wallMs === null ? 'unavailable — no start/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'} |`,
+        `| ${w.workerId === null ? 'unavailable — legacy label join' : `\`${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('')