refactor(evaluation): remove legacy loops and reduce gate overhead

docs/public-api.md

@@ -7,11 +7,11 @@ Generated by `pnpm api:census` on demand — this is a dated reading, not a gate
 | measure | count |
 | --- | --- |
 | export subpaths | 28 |
-| published value exports (subpath x symbol) | 1403 |
-| distinct symbols | 1231 |
+| published value exports (subpath x symbol) | 1400 |
+| distinct symbols | 1228 |
 | production | 944 |
-| planned | 218 |
-| none | 241 |
+| planned | 216 |
+| none | 240 |
 
 Type-only exports are not listed: a type binds no runtime surface, and removing one cannot break a caller at run time.
 
@@ -1289,25 +1289,22 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 
 ### `./supervisor-run`
 
-33 value exports — 23 production, 4 planned, 6 none.
+30 value exports — 23 production, 2 planned, 5 none.
 
 | symbol | consumer | evidence |
 | --- | --- | --- |
 | `analyzeSupervisorRun` | production | discovery-lab:pursuits/agent-authored-runtime-control-deepseek-20260812h/workspaces/runtime-control-deps/runtime-control.mjs |
 | `analyzeSupervisorRunIntegrity` | production | discovery-lab:pursuits/final-agent-authored-campaign-20260810/materialize-candidates.mjs |
 | `analyzeSupervisorRunSources` | production | discovery-lab:experiments/0005-run-visibility/probe.mjs |
 | `claudeCodeSupervisorRunReader` | none | only this package's tests: src/supervisor-run/claude-code-reader.test.ts:6 |
-| `findSupervisorRunDirIn` | none | only this package's tests: src/supervisor-run/loops-reader.test.ts:11 |
 | `findSupervisorRunDirs` | production | traces:src/cli.ts |
 | `isRuntimeSupervisorRunDir` | production | discovery-lab:tools/disco.mjs |
 | `isUnavailable` | production | discovery-lab:tools/disco.mjs |
-| `loopsSupervisorRunReader` | planned | named in traces (bind not in the import graph) |
 | `NO_SOURCE_LIMITS` | production | traces:src/supervisor-run-context.ts |
 | `NO_TERMINAL_RECORD` | none | only this package's tests: src/supervisor-run/terminal-record.test.ts:8 |
 | `parsePatch` | planned | named in browser-agent-driver (bind not in the import graph) |
 | `parseSupervisorTree` | production | discovery-lab:experiments/0005-run-visibility/probe.mjs |
 | `readClaudeCodeSupervisorRun` | none | only this package's tests: src/supervisor-run/claude-code-reader.test.ts:6 |
-| `readLoopsSupervisorRun` | planned | named in discovery-lab (bind not in the import graph) |
 | `readRuntimeSupervisorRun` | production | discovery-lab:pursuits/final-agent-authored-campaign-20260810/materialize-candidates.mjs |
 | `readTerminalRecord` | production | this package: src/supervisor-run/analyze.ts:18 |
 | `renderSupervisorRollupMarkdown` | production | traces:src/cli.ts |

src/analyst/benchmark-implementation.ts

@@ -139,7 +139,7 @@ export const ANALYST_BENCHMARK_IMPLEMENTATION_FILES = Object.freeze([
 ])
 
 export const ANALYST_BENCHMARK_IMPLEMENTATION_SHA256 =
-  'e499ff9c5a3b24104d1a9dbc0c3742ffd08b67ff1b852922b423d38231954254'
+  '53b4c9f453ed05e46968ed39a8e365f8593163b264c0b1a770f7781005d3aae7'
 
 export function analystBenchmarkImplementationDigest() {
   return ANALYST_BENCHMARK_IMPLEMENTATION_SHA256

src/statistics/paired-binary.ts

@@ -439,7 +439,11 @@ export function pairedRiskDifferenceScore(
   // the right endpoint is always inside and the bisection is well posed.
   let lo = -1
   let hi = riskDifference
-  for (let i = 0; i < 200; i++) {
+  // 64 halvings resolve the [-1, 1] search interval below 6e-20, past the
+  // precision available to the score calculation. More iterations only
+  // repeat identical floating-point values while multiplying every gate
+  // decision's cost.
+  for (let i = 0; i < 64; i++) {
     const mid = (lo + hi) / 2
     if (tangoScore(b, c, n, mid) > z) lo = mid
     else hi = mid
@@ -449,7 +453,7 @@ export function pairedRiskDifferenceScore(
   // Upper bound: root of score(delta) = -z on [riskDifference, 1].
   let ulo = riskDifference
   let uhi = 1
-  for (let i = 0; i < 200; i++) {
+  for (let i = 0; i < 64; i++) {
     const mid = (ulo + uhi) / 2
     if (tangoScore(b, c, n, mid) > -z) ulo = mid
     else uhi = mid

src/statistics/rank-tests.ts

@@ -1,7 +1,12 @@
 import { ValidationError } from '../errors'
 import { normalCdf } from '../math/normal'
 import { lnGamma } from '../math/special-functions'
-import { assertFiniteSample, makeRng, symmetricTwoSampleSeed } from './internal'
+import {
+  assertFiniteSample,
+  binomialSignTwoSided,
+  makeRng,
+  symmetricTwoSampleSeed,
+} from './internal'
 
 // ── Rank tests: exact by default ─────────────────────────────────────
 //
@@ -213,6 +218,26 @@ export function wilcoxonSignedRank(
   const n = diffs.length
   if (n === 0) return { w: 0, p: 1, method: 'exact', pFloor: 1, nNonZero: 0 }
 
+  // When every non-zero difference has the same magnitude, the signed-rank
+  // null reduces exactly to a binomial sign distribution. This is the common
+  // pass/fail shape ({-1, 0, +1}); recognizing it avoids a 100k-draw
+  // permutation for every repeated gate evaluation while preserving the exact
+  // p-value and attainable floor. Explicit method requests still retain their
+  // documented behavior.
+  const abs = Math.abs(diffs[0]!)
+  const allEqualAbs = diffs.every((d) => Math.abs(d) === abs)
+  if ((opts.method === undefined || opts.method === 'auto') && allEqualAbs) {
+    const positive = diffs.filter((d) => d > 0).length
+    const negative = n - positive
+    return {
+      w: (positive * (n + 1)) / 2,
+      p: binomialSignTwoSided(positive, negative),
+      method: 'exact',
+      pFloor: Math.min(1, 2 ** (1 - n)),
+      nNonZero: n,
+    }
+  }
+
   const order = diffs.map((d, i) => ({ abs: Math.abs(d), i })).sort((x, y) => x.abs - y.abs)
   const { midranks, tieTerm } = midranksWithTieTerm(order.map((entry) => entry.abs))
   const ranks: number[] = new Array(n)

src/supervisor-run/analyze.test.ts

@@ -334,7 +334,7 @@ describe('analyzeSupervisorRun — decision quality and economics', () => {
     // UNAVAILABLE != ZERO: an older supervisor wrote no tap, so truncation cannot be ruled out.
     expect(r.economics.brainTruncations).toEqual({
       unavailable:
-        'brain.jsonl absent — loops predates the brain-call tap, so truncation cannot be ruled out',
+        'brain.jsonl absent — this source has no brain-call tap, so truncation cannot be ruled out',
     })
   })
 

src/supervisor-run/analyze.ts

@@ -2,7 +2,7 @@
  * The pure analyzer. Takes already-read bytes (`SupervisorRunSources`) and
  * returns the report — every metric derivable from a synthetic journal string
  * with no filesystem, no process, and no network. All I/O lives in a reader
- * (`loops-reader.ts` is one).
+ * (`reader.ts` is one).
  */
 
 import { summarizeNumberSeries } from '../statistics'
@@ -75,16 +75,13 @@ export function analyzeSupervisorRunSources(
 
   const journalMissing =
     src.journalMissingReason ??
-    (src.supRunDir === null
-      ? 'no supervisor run dir under <ws>/.agent/supervisor (or legacy <ws>/.loops/supervisor)'
-      : 'journal.jsonl absent')
+    (src.supRunDir === null ? 'no Runtime supervisor run directory' : 'journal.jsonl absent')
   const haveJournal = src.journal !== null
   const tree = parseSupervisorTree(src)
   const state = tree.state
   const result = parseJson(src.result)
   const judge = parseJson(src.judge)
-  // Runtime's settle record outranks the legacy loops documents; the record
-  // that answered is named on the report so a status never arrives unlabeled.
+  // Runtime's terminal record is named on the report so a status never arrives unlabeled.
   const terminal = readTerminalRecord({
     state,
     result,
@@ -767,8 +764,8 @@ export function analyzeSupervisorRunSources(
             'brain.brainTruncations',
             src.brainLogMissingReason ??
               (src.supRunDir === null
-                ? 'no supervisor run dir under <ws>/.agent/supervisor (or legacy <ws>/.loops/supervisor)'
-                : 'brain.jsonl absent — loops predates the brain-call tap, so truncation cannot be ruled out'),
+                ? 'no Runtime supervisor run directory'
+                : 'brain.jsonl absent — this source has no brain-call tap, so truncation cannot be ruled out'),
           )
         : brainCalls.filter((c) => c.finish_reason === 'length').length,
     workers: {

src/supervisor-run/claude-code-reader.test.ts

@@ -397,7 +397,7 @@ describe('claudeCodeSupervisorRunReader', () => {
     // A worker with no retained transcript reports null tokens, not 0.
     const pruned = workers.find((w) => w.artifacts.transcript_ref === null)
     expect(pruned?.cost.tokens_in).toBeNull()
-    // The root row points at the real session transcript, not a loops path.
+    // The root row points at the real session transcript, not a Runtime path.
     expect(root?.artifacts.transcript_ref).toBe(s.transcriptPath)
   })
 
@@ -411,7 +411,7 @@ describe('claudeCodeSupervisorRunReader', () => {
     expect(isUnavailable(report.orchestration.steers)).toBe(true)
   })
 
-  it('exposes the SupervisorRunReader contract loops implements', async () => {
+  it('exposes the SupervisorRunReader contract Runtime implements', async () => {
     const s = await writeSession()
     const reader = claudeCodeSupervisorRunReader({
       transcriptPath: s.transcriptPath,

src/supervisor-run/claude-code-reader.ts

@@ -1,7 +1,7 @@
 /**
  * Supervision-tree reader over a THIRD-PARTY harness: Claude Code.
  *
- * `loops-reader.ts` reads a supervisor we wrote, whose journal was designed
+ * The Runtime reader reads a supervisor journal designed for this analysis.
  * for this analysis. This reader reads a harness we do not control, whose
  * transcript was designed for replaying a chat — and recovers the same tree
  * from it. If both produce a `SupervisorRunSources`, the tree model is a
@@ -31,12 +31,12 @@
  * reports `unavailable — <reason>` instead of the $0 / 0-accepted that summing
  * an empty field would produce. See `SourceLimits`.
  *
- * ## Metric coverage vs the loops journal
+ * ## Metric coverage vs the Runtime journal
  *
  * Measured on a real 52-agent session (fixture:
  * `tests/fixtures/supervisor-run/claude-code-session-*`).
  *
- * | Metric | loops | Claude Code | Why |
+ * | Metric | Runtime | Claude Code | Why |
  * |---|---|---|---|
  * | workersSpawned / Settled / Cancelled | full | full | spawn tool_use + task-notification + TaskStop |
  * | steers / steersDelivered / steersByWorker | full | full | `SendMessage`; delivery from its tool_result |
@@ -642,7 +642,7 @@ export async function readClaudeCodeSupervisorRun(
   }
 }
 
-/** A `SupervisorRunReader` over a Claude Code session — the same contract loops implements. */
+/** A `SupervisorRunReader` over a Claude Code session — the same contract Runtime implements. */
 export function claudeCodeSupervisorRunReader(opts: ClaudeCodeReaderOptions): SupervisorRunReader {
   return {
     runRef: opts.runRef ?? opts.transcriptPath,

src/supervisor-run/fixtures.ts

@@ -192,7 +192,7 @@ export function fixtureSources(over: Partial<SupervisorRunSources> = {}): Superv
     runRef: '/tmp/cell/runs/inst-1/ARM',
     instanceId: 'inst-1',
     arm: 'ARM',
-    supRunDir: '/tmp/cell/ws/.loops/supervisor/sup-1-test',
+    supRunDir: '/tmp/cell/runtime-run',
     journal: null,
     brainLog: null,
     state: null,

src/supervisor-run/index.ts

@@ -45,16 +45,12 @@ export {
 } from './integrity'
 export {
   analyzeSupervisorRun,
-  findSupervisorRunDirIn,
   findSupervisorRunDirs,
-  type LoopsReaderOptions,
-  loopsSupervisorRunReader,
-  readLoopsSupervisorRun,
   reportSupervisorRound,
   type WriteSupervisorRunOptions,
   writeSupervisorRunReport,
   writeSupervisorRunReportSafe,
-} from './loops-reader'
+} from './reader'
 export {
   renderSupervisorRollupMarkdown,
   renderSupervisorRunHeadline,

src/supervisor-run/integrity.test.ts

@@ -79,6 +79,7 @@ function completedControlSource(inbox: string | null, events: string | null): Su
   return fixtureSources({
     journal: fixtureJournal({ workers: [['worker', 1, 4]] }),
     state: fixtureState({ startSec: 0, endSec: 5 }),
+    result: JSON.stringify({ kind: 'winner', tree: { root: 'sup-1-test', nodes: [] } }),
     workers: [
       {
         workerId: 'sup-1-test:s0',