From 0b6fce6f9c7d37021b3d7d342dc88f6792b23250 Mon Sep 17 00:00:00 2001
From: Drew Stone <drewstone329@gmail.com>
Date: Fri, 24 Jul 2026 19:19:41 -0600
Subject: [PATCH 1/4] feat(bench): check model availability before dispatch

---
 src/runtime/run-benchmark.test.ts | 245 ++++++++++++++++++++++++++++++
 src/runtime/run-benchmark.ts      |  42 +++++
 2 files changed, 287 insertions(+)
 create mode 100644 src/runtime/run-benchmark.test.ts

diff --git a/src/runtime/run-benchmark.test.ts b/src/runtime/run-benchmark.test.ts
new file mode 100644
index 00000000..8ae5cc54
--- /dev/null
+++ b/src/runtime/run-benchmark.test.ts
@@ -0,0 +1,245 @@
+import { afterEach, describe, expect, it, vi } from 'vitest'
+import { runBenchmark } from './run-benchmark'
+import {
+  type AgenticOptions,
+  type AgenticSurface,
+  type AgenticTask,
+  defineStrategy,
+} from './strategy'
+
+interface ChatRequest {
+  model?: string
+  messages?: Array<{ role?: string; content?: string }>
+  temperature?: number
+  max_tokens?: number
+}
+
+const task: AgenticTask = {
+  id: 'task-1',
+  systemPrompt: 'Use the test surface.',
+  userPrompt: 'Complete the task.',
+}
+
+const worker: AgenticOptions = {
+  routerBaseUrl: 'http://router.test/v1',
+  routerKey: 'test-key',
+  model: 'worker-model',
+  maxTokens: 8,
+}
+
+const oneShot = defineStrategy('one-shot', async ({ shot }) => {
+  const out = await shot()
+  return {
+    score: out?.score ?? 0,
+    resolved: out?.score === 1,
+    completions: out?.completions ?? 0,
+    progression: out ? [out.score] : [],
+    shots: 1,
+  }
+})
+
+function surface(events: string[]): AgenticSurface {
+  let sequence = 0
+  return {
+    name: 'test',
+    async open() {
+      events.push('open')
+      sequence += 1
+      return { id: `handle-${sequence}`, surface: 'test' }
+    },
+    async tools() {
+      return []
+    },
+    async call() {
+      return 'ok'
+    },
+    async score() {
+      return { passes: 1, total: 1, errored: 0 }
+    },
+    async close() {},
+  }
+}
+
+function okResponse(): Response {
+  return Response.json({
+    choices: [{ message: { content: 'DONE' } }],
+    usage: { prompt_tokens: 2, completion_tokens: 1 },
+  })
+}
+
+function stubRouter(
+  requests: ChatRequest[],
+  events: string[],
+  respond: (request: ChatRequest, index: number) => Response = () => okResponse(),
+): void {
+  vi.stubGlobal(
+    'fetch',
+    vi.fn(async (_url: string | URL | Request, init?: RequestInit) => {
+      const request = JSON.parse(String(init?.body ?? '{}')) as ChatRequest
+      requests.push(request)
+      events.push(`request:${request.model}:${request.max_tokens ?? 'default'}`)
+      return respond(request, requests.length - 1)
+    }),
+  )
+}
+
+function isModelCheck(request: ChatRequest): boolean {
+  return request.messages?.[0]?.content === 'Reply OK.'
+}
+
+afterEach(() => {
+  vi.unstubAllGlobals()
+})
+
+describe('runBenchmark model availability', () => {
+  it('checks every unique model once before opening any task', async () => {
+    const requests: ChatRequest[] = []
+    const events: string[] = []
+    stubRouter(requests, events)
+
+    await runBenchmark({
+      environment: surface(events),
+      tasks: [task, { ...task, id: 'task-2' }],
+      worker: { ...worker, analystModel: 'analyst-model' },
+      strategies: [oneShot],
+      budget: 1,
+      concurrency: 2,
+    })
+
+    const checks = requests.filter(isModelCheck)
+    expect(checks.map((request) => request.model).sort()).toEqual(['analyst-model', 'worker-model'])
+    expect(checks.every((request) => request.max_tokens === 1)).toBe(true)
+    expect(events.indexOf('open')).toBe(2)
+  })
+
+  it('deduplicates identical worker and analyst models', async () => {
+    const checked: string[] = []
+    const events: string[] = []
+
+    await runBenchmark({
+      environment: surface(events),
+      tasks: [task],
+      worker: {
+        ...worker,
+        analystModel: worker.model,
+        complete: async () => okResponse().json(),
+      },
+      strategies: [oneShot],
+      budget: 1,
+      modelPreflight: async (model) => {
+        checked.push(model)
+      },
+    })
+
+    expect(checked).toEqual(['worker-model'])
+  })
+
+  it('rejects an unavailable model before task dispatch', async () => {
+    const requests: ChatRequest[] = []
+    const events: string[] = []
+    const onTask = vi.fn()
+    stubRouter(requests, events, () => new Response('model has been deprecated', { status: 404 }))
+
+    await expect(
+      runBenchmark({
+        environment: surface(events),
+        tasks: [task],
+        worker,
+        strategies: [oneShot],
+        budget: 1,
+        onTask,
+      }),
+    ).rejects.toThrow(
+      'Benchmark model "worker-model" preflight failed: router 404: model has been deprecated',
+    )
+
+    expect(requests).toHaveLength(1)
+    expect(events).not.toContain('open')
+    expect(onTask).not.toHaveBeenCalled()
+  })
+
+  it('skips the check for an injected completion transport', async () => {
+    const requests: ChatRequest[] = []
+    const events: string[] = []
+    const complete = vi.fn(async (body: Record<string, unknown>) => {
+      requests.push(body as ChatRequest)
+      events.push('complete')
+      return okResponse().json()
+    })
+
+    await runBenchmark({
+      environment: surface(events),
+      tasks: [task],
+      worker: { ...worker, complete },
+      strategies: [oneShot],
+      budget: 1,
+    })
+
+    expect(complete).toHaveBeenCalledTimes(1)
+    expect(requests.some(isModelCheck)).toBe(false)
+    expect(events[0]).toBe('open')
+  })
+
+  it('allows the live check to be disabled explicitly', async () => {
+    const requests: ChatRequest[] = []
+    const events: string[] = []
+    stubRouter(requests, events)
+
+    await runBenchmark({
+      environment: surface(events),
+      tasks: [task],
+      worker,
+      strategies: [oneShot],
+      budget: 1,
+      modelPreflight: false,
+    })
+
+    expect(requests.some(isModelCheck)).toBe(false)
+    expect(events[0]).toBe('open')
+  })
+
+  it('runs a custom check for injected transports', async () => {
+    const events: string[] = []
+    const checked: string[] = []
+
+    await runBenchmark({
+      environment: surface(events),
+      tasks: [task],
+      worker: {
+        ...worker,
+        analystModel: 'analyst-model',
+        complete: async () => okResponse().json(),
+      },
+      strategies: [oneShot],
+      budget: 1,
+      modelPreflight: async (model) => {
+        events.push(`check:${model}`)
+        checked.push(model)
+      },
+    })
+
+    expect(checked.sort()).toEqual(['analyst-model', 'worker-model'])
+    expect(events.indexOf('open')).toBe(2)
+  })
+
+  it('retries at temperature one when the provider requires it', async () => {
+    const requests: ChatRequest[] = []
+    const events: string[] = []
+    stubRouter(requests, events, (_request, index) =>
+      index === 0
+        ? new Response('only temperature 1 is allowed for this model', { status: 400 })
+        : okResponse(),
+    )
+
+    await runBenchmark({
+      environment: surface(events),
+      tasks: [task],
+      worker,
+      strategies: [oneShot],
+      budget: 1,
+    })
+
+    expect(requests.slice(0, 2).map((request) => request.temperature)).toEqual([0.2, 1])
+    expect(events.indexOf('open')).toBe(2)
+  })
+})
diff --git a/src/runtime/run-benchmark.ts b/src/runtime/run-benchmark.ts
index ac1c3dfa..ff5dee1f 100644
--- a/src/runtime/run-benchmark.ts
+++ b/src/runtime/run-benchmark.ts
@@ -15,6 +15,7 @@
 
 import { pairedBootstrap, paretoFrontier } from '@tangle-network/agent-eval'
 import type { RuntimeHooks } from '../runtime-hooks'
+import { routerChatWithUsage } from './router-client'
 import {
   type AgenticOptions,
   type AgenticSurface,
@@ -49,6 +50,14 @@ export interface BenchmarkConfig {
   /** Lifecycle observability — every spawn/settle of every cell's shots/analysts streams
    *  here live (the watchdog/route-auditor seam, passed through to `runAgentic`). */
   hooks?: RuntimeHooks
+  /**
+   * Model availability check before tasks start.
+   *
+   * By default, live router workers send one one-token request per unique worker and analyst
+   * model. Injected `worker.complete` transports skip the check. Pass `false` to disable it or a
+   * callback to check each unique model through a custom transport.
+   */
+  modelPreflight?: false | ((model: string, worker: Readonly<AgenticOptions>) => Promise<void>)
 }
 
 export interface BenchmarkLift {
@@ -127,6 +136,37 @@ async function pool<T, R>(
   return out
 }
 
+async function preflightModels(cfg: BenchmarkConfig): Promise<void> {
+  if (cfg.modelPreflight === false) return
+  if (cfg.worker.complete && !cfg.modelPreflight) return
+
+  const models = [...new Set([cfg.worker.model, cfg.worker.analystModel ?? cfg.worker.model])]
+  const check =
+    cfg.modelPreflight ??
+    (async (model: string, worker: Readonly<AgenticOptions>) => {
+      await routerChatWithUsage(
+        {
+          routerBaseUrl: worker.routerBaseUrl,
+          routerKey: worker.routerKey,
+          model,
+        },
+        [{ role: 'user', content: 'Reply OK.' }],
+        { maxTokens: 1 },
+      )
+    })
+
+  await Promise.all(
+    models.map(async (model) => {
+      try {
+        await check(model, cfg.worker)
+      } catch (cause) {
+        const message = cause instanceof Error ? cause.message : String(cause)
+        throw new Error(`Benchmark model "${model}" preflight failed: ${message}`, { cause })
+      }
+    }),
+  )
+}
+
 /** Run the requested strategies over the tasks, scored by the Environment's own check.
  *  Resilient: a task whose rollouts fail (transient infra) is excluded from the stats but
  *  reported in `perTask` with the error — never silently dropped. */
@@ -135,6 +175,8 @@ export async function runBenchmark(cfg: BenchmarkConfig): Promise<BenchmarkRepor
   const budget = cfg.budget ?? 3
   const concurrency = cfg.concurrency ?? 3
 
+  await preflightModels(cfg)
+
   let settled = 0
   const perTask = await pool(cfg.tasks, concurrency, async (task): Promise<BenchmarkTaskRow> => {
     const cells: Record<string, BenchmarkCell> = {}

From 1ba4b8069ee667b1e9020da8bf15caab03295680 Mon Sep 17 00:00:00 2001
From: Drew Stone <drewstone329@gmail.com>
Date: Fri, 24 Jul 2026 19:34:31 -0600
Subject: [PATCH 2/4] docs(api): refresh benchmark model preflight reference

---
 docs/api/runtime.md | 94 +++++++++++++++++++++++++--------------------
 1 file changed, 53 insertions(+), 41 deletions(-)

diff --git a/docs/api/runtime.md b/docs/api/runtime.md
index ce47c7ca..6e3b10bb 100644
--- a/docs/api/runtime.md
+++ b/docs/api/runtime.md
@@ -6104,7 +6104,7 @@ The full conversation after the loop (seed + every assistant/tool turn). Lets a
 
 ### BenchmarkConfig
 
-Defined in: [src/runtime/run-benchmark.ts:32](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L32)
+Defined in: [src/runtime/run-benchmark.ts:33](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L33)
 
 #### Properties
 
@@ -6112,7 +6112,7 @@ Defined in: [src/runtime/run-benchmark.ts:32](https://github.com/tangle-network/
 
 > **environment**: [`AgenticSurface`](#agenticsurface)
 
-Defined in: [src/runtime/run-benchmark.ts:34](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L34)
+Defined in: [src/runtime/run-benchmark.ts:35](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L35)
 
 The task domain (5 hooks).
 
@@ -6120,7 +6120,7 @@ The task domain (5 hooks).
 
 > **tasks**: [`AgenticTask`](#agentictask)[]
 
-Defined in: [src/runtime/run-benchmark.ts:36](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L36)
+Defined in: [src/runtime/run-benchmark.ts:37](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L37)
 
 The tasks to score across.
 
@@ -6128,7 +6128,7 @@ The tasks to score across.
 
 > **worker**: [`AgenticOptions`](#agenticoptions)
 
-Defined in: [src/runtime/run-benchmark.ts:38](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L38)
+Defined in: [src/runtime/run-benchmark.ts:39](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L39)
 
 The worker: model + router + (optional) the critic's instruction (the steerer knob).
 
@@ -6136,7 +6136,7 @@ The worker: model + router + (optional) the critic's instruction (the steerer kn
 
 > `optional` **strategies?**: [`Strategy`](#strategy-3)\<[`StrategyResult`](#strategyresult-1)\>[]
 
-Defined in: [src/runtime/run-benchmark.ts:41](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L41)
+Defined in: [src/runtime/run-benchmark.ts:42](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L42)
 
 Which strategies to compare. Pass the built-ins (`refine`, `sample`) or your own.
  Default: [sample, refine].
@@ -6145,7 +6145,7 @@ Which strategies to compare. Pass the built-ins (`refine`, `sample`) or your own
 
 > `optional` **budget?**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:43](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L43)
+Defined in: [src/runtime/run-benchmark.ts:44](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L44)
 
 Shots (refine) / width (sample) — the equal compute budget per strategy. Default 3.
 
@@ -6153,7 +6153,7 @@ Shots (refine) / width (sample) — the equal compute budget per strategy. Defau
 
 > `optional` **concurrency?**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:45](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L45)
+Defined in: [src/runtime/run-benchmark.ts:46](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L46)
 
 Tasks scored in parallel. Default 3.
 
@@ -6161,7 +6161,7 @@ Tasks scored in parallel. Default 3.
 
 > `optional` **onTask?**: (`row`, `done`, `total`) => `void`
 
-Defined in: [src/runtime/run-benchmark.ts:48](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L48)
+Defined in: [src/runtime/run-benchmark.ts:49](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L49)
 
 Progress hook — fires as each task settles (the live-monitoring seam: append to a
  progress file, render a tree, stream to a dashboard). `done` counts settled tasks.
@@ -6188,16 +6188,28 @@ Progress hook — fires as each task settles (the live-monitoring seam: append t
 
 > `optional` **hooks?**: [`RuntimeHooks`](index.md#runtimehooks)
 
-Defined in: [src/runtime/run-benchmark.ts:51](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L51)
+Defined in: [src/runtime/run-benchmark.ts:52](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L52)
 
 Lifecycle observability — every spawn/settle of every cell's shots/analysts streams
  here live (the watchdog/route-auditor seam, passed through to `runAgentic`).
 
+##### modelPreflight?
+
+> `optional` **modelPreflight?**: `false` \| ((`model`, `worker`) => `Promise`\<`void`\>)
+
+Defined in: [src/runtime/run-benchmark.ts:60](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L60)
+
+Model availability check before tasks start.
+
+By default, live router workers send one one-token request per unique worker and analyst
+model. Injected `worker.complete` transports skip the check. Pass `false` to disable it or a
+callback to check each unique model through a custom transport.
+
 ***
 
 ### BenchmarkLift
 
-Defined in: [src/runtime/run-benchmark.ts:54](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L54)
+Defined in: [src/runtime/run-benchmark.ts:63](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L63)
 
 #### Properties
 
@@ -6205,7 +6217,7 @@ Defined in: [src/runtime/run-benchmark.ts:54](https://github.com/tangle-network/
 
 > **mean**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:56](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L56)
+Defined in: [src/runtime/run-benchmark.ts:65](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L65)
 
 Mean of paired deltas (refine − sample).
 
@@ -6213,25 +6225,25 @@ Mean of paired deltas (refine − sample).
 
 > **low**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:57](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L57)
+Defined in: [src/runtime/run-benchmark.ts:66](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L66)
 
 ##### high
 
 > **high**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:58](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L58)
+Defined in: [src/runtime/run-benchmark.ts:67](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L67)
 
 ##### n
 
 > **n**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:59](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L59)
+Defined in: [src/runtime/run-benchmark.ts:68](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L68)
 
 ***
 
 ### BenchmarkCell
 
-Defined in: [src/runtime/run-benchmark.ts:63](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L63)
+Defined in: [src/runtime/run-benchmark.ts:72](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L72)
 
 One strategy's outcome on one task — the per-task cell an optimizer consumes.
 
@@ -6241,19 +6253,19 @@ One strategy's outcome on one task — the per-task cell an optimizer consumes.
 
 > **score**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:64](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L64)
+Defined in: [src/runtime/run-benchmark.ts:73](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L73)
 
 ##### resolved
 
 > **resolved**: `boolean`
 
-Defined in: [src/runtime/run-benchmark.ts:65](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L65)
+Defined in: [src/runtime/run-benchmark.ts:74](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L74)
 
 ##### progression
 
 > **progression**: `number`[]
 
-Defined in: [src/runtime/run-benchmark.ts:67](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L67)
+Defined in: [src/runtime/run-benchmark.ts:76](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L76)
 
 The progress curve (refine: score per shot; sample: best-so-far per rollout).
 
@@ -6261,19 +6273,19 @@ The progress curve (refine: score per shot; sample: best-so-far per rollout).
 
 > **usd**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:68](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L68)
+Defined in: [src/runtime/run-benchmark.ts:77](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L77)
 
 ##### ms
 
 > **ms**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:69](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L69)
+Defined in: [src/runtime/run-benchmark.ts:78](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L78)
 
 ##### tokens
 
 > **tokens**: `object`
 
-Defined in: [src/runtime/run-benchmark.ts:70](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L70)
+Defined in: [src/runtime/run-benchmark.ts:79](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L79)
 
 ###### input
 
@@ -6287,7 +6299,7 @@ Defined in: [src/runtime/run-benchmark.ts:70](https://github.com/tangle-network/
 
 ### BenchmarkTaskRow
 
-Defined in: [src/runtime/run-benchmark.ts:73](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L73)
+Defined in: [src/runtime/run-benchmark.ts:82](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L82)
 
 #### Properties
 
@@ -6295,13 +6307,13 @@ Defined in: [src/runtime/run-benchmark.ts:73](https://github.com/tangle-network/
 
 > **taskId**: `string`
 
-Defined in: [src/runtime/run-benchmark.ts:74](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L74)
+Defined in: [src/runtime/run-benchmark.ts:83](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L83)
 
 ##### cells?
 
 > `optional` **cells?**: `Record`\<`string`, [`BenchmarkCell`](#benchmarkcell)\>
 
-Defined in: [src/runtime/run-benchmark.ts:76](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L76)
+Defined in: [src/runtime/run-benchmark.ts:85](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L85)
 
 Per-strategy cells; absent when the task errored before completing all strategies.
 
@@ -6309,7 +6321,7 @@ Per-strategy cells; absent when the task errored before completing all strategie
 
 > `optional` **errors?**: `Record`\<`string`, `string`\>
 
-Defined in: [src/runtime/run-benchmark.ts:80](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L80)
+Defined in: [src/runtime/run-benchmark.ts:89](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L89)
 
 Per-strategy failures on this task: the strategy competed, threw, and scored an
  honest zero — it loses, it does not poison the row. The message is kept so a later
@@ -6319,7 +6331,7 @@ Per-strategy failures on this task: the strategy competed, threw, and scored an
 
 > `optional` **error?**: `string`
 
-Defined in: [src/runtime/run-benchmark.ts:82](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L82)
+Defined in: [src/runtime/run-benchmark.ts:91](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L91)
 
 Why the task was excluded (infra/setup failure) — never silently dropped.
 
@@ -6327,7 +6339,7 @@ Why the task was excluded (infra/setup failure) — never silently dropped.
 
 ### BenchmarkStrategySummary
 
-Defined in: [src/runtime/run-benchmark.ts:85](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L85)
+Defined in: [src/runtime/run-benchmark.ts:94](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L94)
 
 #### Properties
 
@@ -6335,7 +6347,7 @@ Defined in: [src/runtime/run-benchmark.ts:85](https://github.com/tangle-network/
 
 > **score**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:87](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L87)
+Defined in: [src/runtime/run-benchmark.ts:96](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L96)
 
 Mean verifier score (0..1).
 
@@ -6343,7 +6355,7 @@ Mean verifier score (0..1).
 
 > **resolved**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:89](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L89)
+Defined in: [src/runtime/run-benchmark.ts:98](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L98)
 
 Fraction of tasks fully resolved.
 
@@ -6351,7 +6363,7 @@ Fraction of tasks fully resolved.
 
 > **usd**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:91](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L91)
+Defined in: [src/runtime/run-benchmark.ts:100](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L100)
 
 Mean cost vector per task.
 
@@ -6359,13 +6371,13 @@ Mean cost vector per task.
 
 > **ms**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:92](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L92)
+Defined in: [src/runtime/run-benchmark.ts:101](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L101)
 
 ***
 
 ### BenchmarkReport
 
-Defined in: [src/runtime/run-benchmark.ts:96](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L96)
+Defined in: [src/runtime/run-benchmark.ts:105](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L105)
 
 Benchmark output: per-strategy means plus the full per-task × per-strategy losses table an optimizer mines.
 
@@ -6375,19 +6387,19 @@ Benchmark output: per-strategy means plus the full per-task × per-strategy loss
 
 > **n**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:97](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L97)
+Defined in: [src/runtime/run-benchmark.ts:106](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L106)
 
 ##### excluded
 
 > **excluded**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:98](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L98)
+Defined in: [src/runtime/run-benchmark.ts:107](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L107)
 
 ##### perStrategy
 
 > **perStrategy**: `Record`\<`string`, [`BenchmarkStrategySummary`](#benchmarkstrategysummary)\>
 
-Defined in: [src/runtime/run-benchmark.ts:100](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L100)
+Defined in: [src/runtime/run-benchmark.ts:109](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L109)
 
 Per-strategy means (keyed by strategy.name).
 
@@ -6395,7 +6407,7 @@ Per-strategy means (keyed by strategy.name).
 
 > **perTask**: [`BenchmarkTaskRow`](#benchmarktaskrow)[]
 
-Defined in: [src/runtime/run-benchmark.ts:103](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L103)
+Defined in: [src/runtime/run-benchmark.ts:112](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L112)
 
 The full per-task × per-strategy table — the LOSSES an optimizer (GEPA, a
  strategy-author, an operator) consumes. Includes errored tasks with the reason.
@@ -6404,7 +6416,7 @@ The full per-task × per-strategy table — the LOSSES an optimizer (GEPA, a
 
 > **pareto**: `string`[]
 
-Defined in: [src/runtime/run-benchmark.ts:106](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L106)
+Defined in: [src/runtime/run-benchmark.ts:115](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L115)
 
 The non-dominated strategies on (score ↑, $/task ↓) — collapse-last, per the canon:
  a strategy that ties on score at half the cost WINS and a scalar would hide it.
@@ -6413,7 +6425,7 @@ The non-dominated strategies on (score ↑, $/task ↓) — collapse-last, per t
 
 > `optional` **refineVsSample?**: [`BenchmarkLift`](#benchmarklift)
 
-Defined in: [src/runtime/run-benchmark.ts:108](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L108)
+Defined in: [src/runtime/run-benchmark.ts:117](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L117)
 
 The headline when both `refine` and `sample` ran: paired-bootstrap lift of refine over sample.
 
@@ -18124,7 +18136,7 @@ Defined in: [src/runtime/personify/wave-types.ts:619](https://github.com/tangle-
 
 > **Environment** = [`AgenticSurface`](#agenticsurface)
 
-Defined in: [src/runtime/run-benchmark.ts:30](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L30)
+Defined in: [src/runtime/run-benchmark.ts:31](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L31)
 
 A checkable task domain — implement these 5 hooks and the suite does the rest. The
  same seam as `AgenticSurface`; `Environment` is the RL/gym-standard name for it.
@@ -21168,7 +21180,7 @@ The turnkey production brain — tests script a mock `ToolLoopChat`; production
 
 > **runBenchmark**(`cfg`): `Promise`\<[`BenchmarkReport`](#benchmarkreport)\>
 
-Defined in: [src/runtime/run-benchmark.ts:133](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L133)
+Defined in: [src/runtime/run-benchmark.ts:173](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L173)
 
 Run the requested strategies over the tasks, scored by the Environment's own check.
  Resilient: a task whose rollouts fail (transient infra) is excluded from the stats but
@@ -21190,7 +21202,7 @@ Run the requested strategies over the tasks, scored by the Environment's own che
 
 > **printBenchmarkReport**(`report`): `void`
 
-Defined in: [src/runtime/run-benchmark.ts:232](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L232)
+Defined in: [src/runtime/run-benchmark.ts:274](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L274)
 
 Pretty-print a report — the "free optimization" verdict, with the cost vector.
 

From c6737d49004369df4d359b6309a7d719b1b50d83 Mon Sep 17 00:00:00 2001
From: Drew Stone <drewstone329@gmail.com>
Date: Fri, 24 Jul 2026 19:37:03 -0600
Subject: [PATCH 3/4] fix(bench): reuse model checks across evolution phases

---
 docs/api/runtime.md                    | 163 +++++++++++++------------
 src/runtime/run-benchmark.test.ts      |   2 +
 src/runtime/run-benchmark.ts           |   2 +-
 src/runtime/strategy-evolution.ts      |  14 ++-
 tests/loops/strategy-evolution.test.ts |   5 +
 5 files changed, 108 insertions(+), 78 deletions(-)

diff --git a/docs/api/runtime.md b/docs/api/runtime.md
index 6e3b10bb..bb295299 100644
--- a/docs/api/runtime.md
+++ b/docs/api/runtime.md
@@ -7849,7 +7849,7 @@ Defined in: [src/runtime/strategy-author.ts:141](https://github.com/tangle-netwo
 
 ### EvolutionAuthor
 
-Defined in: [src/runtime/strategy-evolution.ts:46](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L46)
+Defined in: [src/runtime/strategy-evolution.ts:47](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L47)
 
 #### Properties
 
@@ -7857,7 +7857,7 @@ Defined in: [src/runtime/strategy-evolution.ts:46](https://github.com/tangle-net
 
 > **chat**: `ChatClient`
 
-Defined in: [src/runtime/strategy-evolution.ts:48](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L48)
+Defined in: [src/runtime/strategy-evolution.ts:49](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L49)
 
 The model-call seam (agent-eval `createChatClient`).
 
@@ -7865,31 +7865,31 @@ The model-call seam (agent-eval `createChatClient`).
 
 > `optional` **model?**: `string`
 
-Defined in: [src/runtime/strategy-evolution.ts:49](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L49)
+Defined in: [src/runtime/strategy-evolution.ts:50](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L50)
 
 ##### fallbackModel?
 
 > `optional` **fallbackModel?**: `string`
 
-Defined in: [src/runtime/strategy-evolution.ts:50](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L50)
+Defined in: [src/runtime/strategy-evolution.ts:51](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L51)
 
 ##### temperature?
 
 > `optional` **temperature?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:51](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L51)
+Defined in: [src/runtime/strategy-evolution.ts:52](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L52)
 
 ##### maxTokens?
 
 > `optional` **maxTokens?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:52](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L52)
+Defined in: [src/runtime/strategy-evolution.ts:53](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L53)
 
 ***
 
 ### StrategyEvolutionConfig
 
-Defined in: [src/runtime/strategy-evolution.ts:57](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L57)
+Defined in: [src/runtime/strategy-evolution.ts:58](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L58)
 
 #### Properties
 
@@ -7897,13 +7897,13 @@ Defined in: [src/runtime/strategy-evolution.ts:57](https://github.com/tangle-net
 
 > **environment**: [`AgenticSurface`](#agenticsurface)
 
-Defined in: [src/runtime/strategy-evolution.ts:58](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L58)
+Defined in: [src/runtime/strategy-evolution.ts:59](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L59)
 
 ##### tasks
 
 > **tasks**: (`offset`, `n`) => `Promise`\<[`AgenticTask`](#agentictask)[]\>
 
-Defined in: [src/runtime/strategy-evolution.ts:62](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L62)
+Defined in: [src/runtime/strategy-evolution.ts:63](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L63)
 
 Task supply by DISJOINT slice: `(offset, n)` must return n tasks unique to that
  offset range. Train draws [0, trainN); the holdout draws [trainN + holdoutOffset,
@@ -7927,19 +7927,19 @@ Task supply by DISJOINT slice: `(offset, n)` must return n tasks unique to that
 
 > **trainN**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:63](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L63)
+Defined in: [src/runtime/strategy-evolution.ts:64](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L64)
 
 ##### holdoutN
 
 > **holdoutN**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:64](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L64)
+Defined in: [src/runtime/strategy-evolution.ts:65](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L65)
 
 ##### holdoutOffset?
 
 > `optional` **holdoutOffset?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:66](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L66)
+Defined in: [src/runtime/strategy-evolution.ts:67](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L67)
 
 Extra offset past the train slice for the holdout draw (rotate across runs).
 
@@ -7947,19 +7947,30 @@ Extra offset past the train slice for the holdout draw (rotate across runs).
 
 > **worker**: [`AgenticOptions`](#agenticoptions)
 
-Defined in: [src/runtime/strategy-evolution.ts:67](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L67)
+Defined in: [src/runtime/strategy-evolution.ts:68](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L68)
+
+##### modelPreflight?
+
+> `optional` **modelPreflight?**: `false` \| ((`model`, `worker`) => `Promise`\<`void`\>)
+
+Defined in: [src/runtime/strategy-evolution.ts:75](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L75)
+
+Model availability check before the first benchmark phase.
+
+A successful check is reused for the remaining phases in this evolution run.
+See `BenchmarkConfig.modelPreflight`.
 
 ##### author
 
 > **author**: [`EvolutionAuthor`](#evolutionauthor)
 
-Defined in: [src/runtime/strategy-evolution.ts:68](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L68)
+Defined in: [src/runtime/strategy-evolution.ts:76](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L76)
 
 ##### budget?
 
 > `optional` **budget?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:70](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L70)
+Defined in: [src/runtime/strategy-evolution.ts:78](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L78)
 
 Rollouts (sample) / shots (refine) per strategy per task. Default 3.
 
@@ -7967,13 +7978,13 @@ Rollouts (sample) / shots (refine) per strategy per task. Default 3.
 
 > `optional` **concurrency?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:71](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L71)
+Defined in: [src/runtime/strategy-evolution.ts:79](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L79)
 
 ##### generations?
 
 > `optional` **generations?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:73](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L73)
+Defined in: [src/runtime/strategy-evolution.ts:81](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L81)
 
 Author→tournament rounds after gen0. Default 2.
 
@@ -7981,7 +7992,7 @@ Author→tournament rounds after gen0. Default 2.
 
 > `optional` **populationSize?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:75](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L75)
+Defined in: [src/runtime/strategy-evolution.ts:83](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L83)
 
 Authored candidates per generation. Default 2.
 
@@ -7989,7 +8000,7 @@ Authored candidates per generation. Default 2.
 
 > `optional` **baselines?**: [`Strategy`](#strategy-3)\<[`StrategyResult`](#strategyresult-1)\>[]
 
-Defined in: [src/runtime/strategy-evolution.ts:77](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L77)
+Defined in: [src/runtime/strategy-evolution.ts:85](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L85)
 
 The gen0 field. Default [sample, refine, sampleThenRefine].
 
@@ -7997,7 +8008,7 @@ The gen0 field. Default [sample, refine, sampleThenRefine].
 
 > `optional` **objective?**: `"score"` \| `"cost"`
 
-Defined in: [src/runtime/strategy-evolution.ts:83](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L83)
+Defined in: [src/runtime/strategy-evolution.ts:91](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L91)
 
 What "better" means for PROMOTION. 'score' (default): the candidate must beat the
  incumbent's score (superiority gate). 'cost': the candidate must prove score
@@ -8009,7 +8020,7 @@ What "better" means for PROMOTION. 'score' (default): the candidate must beat th
 
 > `optional` **scoreTolerance?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:85](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L85)
+Defined in: [src/runtime/strategy-evolution.ts:93](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L93)
 
 Cost objective: the score CI lower bound must clear −scoreTolerance. Default 0.05.
 
@@ -8017,7 +8028,7 @@ Cost objective: the score CI lower bound must clear −scoreTolerance. Default 0
 
 > `optional` **champion?**: [`ChampionPolicy`](#championpolicy)
 
-Defined in: [src/runtime/strategy-evolution.ts:87](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L87)
+Defined in: [src/runtime/strategy-evolution.ts:95](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L95)
 
 Search-side champion selection. Default 'costAware'.
 
@@ -8025,7 +8036,7 @@ Search-side champion selection. Default 'costAware'.
 
 > `optional` **championEpsilon?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:89](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L89)
+Defined in: [src/runtime/strategy-evolution.ts:97](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L97)
 
 Score band treated as a tie under 'costAware'. Default 0.01.
 
@@ -8033,7 +8044,7 @@ Score band treated as a tie under 'costAware'. Default 0.01.
 
 > **outDir**: `string`
 
-Defined in: [src/runtime/strategy-evolution.ts:91](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L91)
+Defined in: [src/runtime/strategy-evolution.ts:99](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L99)
 
 Where authored modules are written.
 
@@ -8041,7 +8052,7 @@ Where authored modules are written.
 
 > `optional` **minPairedTasks?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:93](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L93)
+Defined in: [src/runtime/strategy-evolution.ts:101](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L101)
 
 Promotion-gate evidence floor (paired holdout tasks).
 
@@ -8049,7 +8060,7 @@ Promotion-gate evidence floor (paired holdout tasks).
 
 > `optional` **band?**: `object`
 
-Defined in: [src/runtime/strategy-evolution.ts:102](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L102)
+Defined in: [src/runtime/strategy-evolution.ts:110](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L110)
 
 BAND-AWARE scoring — concentrate the measurement where lift is possible.
  Holdout: draw `holdoutPoolN` candidate tasks and run `baselines[0]` once at the run
@@ -8075,7 +8086,7 @@ Keep holdout tasks where the reference scores ≤ this. Default 0.99 — drop on
 
 > `optional` **lossesDetail?**: `"exact"` \| `"binary"`
 
-Defined in: [src/runtime/strategy-evolution.ts:111](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L111)
+Defined in: [src/runtime/strategy-evolution.ts:119](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L119)
 
 What the author learns from a tournament. 'exact' (default) = scores + progressions
  per task; 'binary' = pass/fail only — the leakage-bounded channel (one bit per cell
@@ -8085,7 +8096,7 @@ What the author learns from a tournament. 'exact' (default) = scores + progressi
 
 > `optional` **reproducerCheck?**: `object`
 
-Defined in: [src/runtime/strategy-evolution.ts:118](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L118)
+Defined in: [src/runtime/strategy-evolution.ts:126](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L126)
 
 Reproducer certification (arXiv:2606.11045): when the final champion is AUTHORED,
  compress it to a short natural-language summary, have a fresh author re-implement
@@ -8111,7 +8122,7 @@ Reproduction counts as faithful when reproducedScore ≥ championScore − toler
 
 > `optional` **checkpoint?**: `object`
 
-Defined in: [src/runtime/strategy-evolution.ts:128](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L128)
+Defined in: [src/runtime/strategy-evolution.ts:136](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L136)
 
 Endurance: write the run state after every completed phase; with `resume`, a
  restart skips completed phases (authored modules re-imported from their files).
@@ -8129,7 +8140,7 @@ Endurance: write the run state after every completed phase; with `resume`, a
 
 > `optional` **onPhase?**: (`phase`) => `Promise`\<`void`\>
 
-Defined in: [src/runtime/strategy-evolution.ts:135](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L135)
+Defined in: [src/runtime/strategy-evolution.ts:143](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L143)
 
 Called before each benchmark phase (gen0, gen1…, band-screen, holdout, reproduce).
  The seam for environment recycling — no artifacts span phases, so a runner may
@@ -8149,7 +8160,7 @@ Called before each benchmark phase (gen0, gen1…, band-screen, holdout, reprodu
 
 > `optional` **onTask?**: (`phase`, `row`, `done`, `total`) => `void`
 
-Defined in: [src/runtime/strategy-evolution.ts:136](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L136)
+Defined in: [src/runtime/strategy-evolution.ts:144](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L144)
 
 ###### Parameters
 
@@ -8177,13 +8188,13 @@ Defined in: [src/runtime/strategy-evolution.ts:136](https://github.com/tangle-ne
 
 > `optional` **hooks?**: [`RuntimeHooks`](index.md#runtimehooks)
 
-Defined in: [src/runtime/strategy-evolution.ts:137](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L137)
+Defined in: [src/runtime/strategy-evolution.ts:145](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L145)
 
 ***
 
 ### ChampionPick
 
-Defined in: [src/runtime/strategy-evolution.ts:152](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L152)
+Defined in: [src/runtime/strategy-evolution.ts:160](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L160)
 
 #### Properties
 
@@ -8191,25 +8202,25 @@ Defined in: [src/runtime/strategy-evolution.ts:152](https://github.com/tangle-ne
 
 > **name**: `string`
 
-Defined in: [src/runtime/strategy-evolution.ts:153](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L153)
+Defined in: [src/runtime/strategy-evolution.ts:161](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L161)
 
 ##### score
 
 > **score**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:154](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L154)
+Defined in: [src/runtime/strategy-evolution.ts:162](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L162)
 
 ##### usd
 
 > **usd**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:155](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L155)
+Defined in: [src/runtime/strategy-evolution.ts:163](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L163)
 
 ***
 
 ### EvolutionCandidate
 
-Defined in: [src/runtime/strategy-evolution.ts:158](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L158)
+Defined in: [src/runtime/strategy-evolution.ts:166](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L166)
 
 #### Properties
 
@@ -8217,31 +8228,31 @@ Defined in: [src/runtime/strategy-evolution.ts:158](https://github.com/tangle-ne
 
 > **name**: `string`
 
-Defined in: [src/runtime/strategy-evolution.ts:159](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L159)
+Defined in: [src/runtime/strategy-evolution.ts:167](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L167)
 
 ##### file?
 
 > `optional` **file?**: `string`
 
-Defined in: [src/runtime/strategy-evolution.ts:160](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L160)
+Defined in: [src/runtime/strategy-evolution.ts:168](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L168)
 
 ##### gzipBits?
 
 > `optional` **gzipBits?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:161](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L161)
+Defined in: [src/runtime/strategy-evolution.ts:169](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L169)
 
 ##### codeChars?
 
 > `optional` **codeChars?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:162](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L162)
+Defined in: [src/runtime/strategy-evolution.ts:170](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L170)
 
 ##### error?
 
 > `optional` **error?**: `string`
 
-Defined in: [src/runtime/strategy-evolution.ts:164](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L164)
+Defined in: [src/runtime/strategy-evolution.ts:172](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L172)
 
 Present when this author attempt failed (recorded, never silent).
 
@@ -8249,7 +8260,7 @@ Present when this author attempt failed (recorded, never silent).
 
 ### EvolutionGeneration
 
-Defined in: [src/runtime/strategy-evolution.ts:167](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L167)
+Defined in: [src/runtime/strategy-evolution.ts:175](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L175)
 
 #### Properties
 
@@ -8257,31 +8268,31 @@ Defined in: [src/runtime/strategy-evolution.ts:167](https://github.com/tangle-ne
 
 > **generation**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:168](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L168)
+Defined in: [src/runtime/strategy-evolution.ts:176](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L176)
 
 ##### candidates
 
 > **candidates**: [`EvolutionCandidate`](#evolutioncandidate)[]
 
-Defined in: [src/runtime/strategy-evolution.ts:169](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L169)
+Defined in: [src/runtime/strategy-evolution.ts:177](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L177)
 
 ##### report
 
 > **report**: [`BenchmarkReport`](#benchmarkreport)
 
-Defined in: [src/runtime/strategy-evolution.ts:170](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L170)
+Defined in: [src/runtime/strategy-evolution.ts:178](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L178)
 
 ##### champion
 
 > **champion**: [`ChampionPick`](#championpick)
 
-Defined in: [src/runtime/strategy-evolution.ts:171](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L171)
+Defined in: [src/runtime/strategy-evolution.ts:179](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L179)
 
 ***
 
 ### EvolutionArchiveNode
 
-Defined in: [src/runtime/strategy-evolution.ts:174](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L174)
+Defined in: [src/runtime/strategy-evolution.ts:182](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L182)
 
 #### Properties
 
@@ -8289,25 +8300,25 @@ Defined in: [src/runtime/strategy-evolution.ts:174](https://github.com/tangle-ne
 
 > **name**: `string`
 
-Defined in: [src/runtime/strategy-evolution.ts:175](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L175)
+Defined in: [src/runtime/strategy-evolution.ts:183](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L183)
 
 ##### source
 
 > **source**: `"baseline"` \| `"authored"`
 
-Defined in: [src/runtime/strategy-evolution.ts:176](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L176)
+Defined in: [src/runtime/strategy-evolution.ts:184](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L184)
 
 ##### generation
 
 > **generation**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:177](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L177)
+Defined in: [src/runtime/strategy-evolution.ts:185](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L185)
 
 ##### parent?
 
 > `optional` **parent?**: `string`
 
-Defined in: [src/runtime/strategy-evolution.ts:179](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L179)
+Defined in: [src/runtime/strategy-evolution.ts:187](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L187)
 
 The champion whose tournament losses this candidate was authored from.
 
@@ -8315,19 +8326,19 @@ The champion whose tournament losses this candidate was authored from.
 
 > `optional` **gzipBits?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:180](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L180)
+Defined in: [src/runtime/strategy-evolution.ts:188](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L188)
 
 ##### file?
 
 > `optional` **file?**: `string`
 
-Defined in: [src/runtime/strategy-evolution.ts:181](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L181)
+Defined in: [src/runtime/strategy-evolution.ts:189](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L189)
 
 ##### score
 
 > **score**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:184](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L184)
+Defined in: [src/runtime/strategy-evolution.ts:192](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L192)
 
 Latest measured tournament result — 0 until the node's first tournament settles
  (an authored node is created before its generation's benchmark runs).
@@ -8336,13 +8347,13 @@ Latest measured tournament result — 0 until the node's first tournament settle
 
 > **usd**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:185](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L185)
+Defined in: [src/runtime/strategy-evolution.ts:193](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L193)
 
 ***
 
 ### EvolutionBandInfo
 
-Defined in: [src/runtime/strategy-evolution.ts:204](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L204)
+Defined in: [src/runtime/strategy-evolution.ts:212](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L212)
 
 #### Properties
 
@@ -8350,7 +8361,7 @@ Defined in: [src/runtime/strategy-evolution.ts:204](https://github.com/tangle-ne
 
 > **screened**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:206](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L206)
+Defined in: [src/runtime/strategy-evolution.ts:214](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L214)
 
 Tasks screened by the reference on the holdout pool.
 
@@ -8358,7 +8369,7 @@ Tasks screened by the reference on the holdout pool.
 
 > **inBand**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:208](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L208)
+Defined in: [src/runtime/strategy-evolution.ts:216](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L216)
 
 Tasks kept (reference score ≤ maxRefScore) before truncating to holdoutN.
 
@@ -8366,7 +8377,7 @@ Tasks kept (reference score ≤ maxRefScore) before truncating to holdoutN.
 
 > **refScores**: `object`[]
 
-Defined in: [src/runtime/strategy-evolution.ts:210](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L210)
+Defined in: [src/runtime/strategy-evolution.ts:218](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L218)
 
 Reference scores per screened task (the screening record).
 
@@ -8382,7 +8393,7 @@ Reference scores per screened task (the screening record).
 
 ### EvolutionReport
 
-Defined in: [src/runtime/strategy-evolution.ts:213](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L213)
+Defined in: [src/runtime/strategy-evolution.ts:221](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L221)
 
 #### Properties
 
@@ -8390,49 +8401,49 @@ Defined in: [src/runtime/strategy-evolution.ts:213](https://github.com/tangle-ne
 
 > **gen0**: [`BenchmarkReport`](#benchmarkreport)
 
-Defined in: [src/runtime/strategy-evolution.ts:214](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L214)
+Defined in: [src/runtime/strategy-evolution.ts:222](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L222)
 
 ##### gen0Champion
 
 > **gen0Champion**: [`ChampionPick`](#championpick)
 
-Defined in: [src/runtime/strategy-evolution.ts:215](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L215)
+Defined in: [src/runtime/strategy-evolution.ts:223](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L223)
 
 ##### generations
 
 > **generations**: [`EvolutionGeneration`](#evolutiongeneration)[]
 
-Defined in: [src/runtime/strategy-evolution.ts:216](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L216)
+Defined in: [src/runtime/strategy-evolution.ts:224](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L224)
 
 ##### archive
 
 > **archive**: [`EvolutionArchiveNode`](#evolutionarchivenode)[]
 
-Defined in: [src/runtime/strategy-evolution.ts:217](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L217)
+Defined in: [src/runtime/strategy-evolution.ts:225](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L225)
 
 ##### finalChampion
 
 > **finalChampion**: [`ChampionPick`](#championpick)
 
-Defined in: [src/runtime/strategy-evolution.ts:218](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L218)
+Defined in: [src/runtime/strategy-evolution.ts:226](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L226)
 
 ##### holdout
 
 > **holdout**: [`BenchmarkReport`](#benchmarkreport)
 
-Defined in: [src/runtime/strategy-evolution.ts:219](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L219)
+Defined in: [src/runtime/strategy-evolution.ts:227](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L227)
 
 ##### verdict
 
 > **verdict**: [`PromotionVerdict`](#promotionverdict)
 
-Defined in: [src/runtime/strategy-evolution.ts:220](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L220)
+Defined in: [src/runtime/strategy-evolution.ts:228](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L228)
 
 ##### band?
 
 > `optional` **band?**: [`EvolutionBandInfo`](#evolutionbandinfo)
 
-Defined in: [src/runtime/strategy-evolution.ts:223](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L223)
+Defined in: [src/runtime/strategy-evolution.ts:231](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L231)
 
 Present when band screening ran — the verdict's estimand is then "paired lift on
  headroom tasks" (band membership fixed by the reference screen, pre-registered).
@@ -8441,7 +8452,7 @@ Present when band screening ran — the verdict's estimand is then "paired lift
 
 > `optional` **reproduction?**: `ReproductionCheck`
 
-Defined in: [src/runtime/strategy-evolution.ts:225](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L225)
+Defined in: [src/runtime/strategy-evolution.ts:233](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L233)
 
 Present when reproducerCheck ran (final champion was authored).
 
@@ -8449,7 +8460,7 @@ Present when reproducerCheck ran (final champion was authored).
 
 > **trajectory**: `object`[]
 
-Defined in: [src/runtime/strategy-evolution.ts:230](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L230)
+Defined in: [src/runtime/strategy-evolution.ts:238](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L238)
 
 SEARCH TELEMETRY, not evidence: each entry is that generation's own train-slice
  re-measurement, so cross-generation deltas mix true drift with run-to-run variance
@@ -18260,7 +18271,7 @@ fixtures, feature names) and replace only the instruction.
 
 > **ChampionPolicy** = `"score"` \| `"costAware"`
 
-Defined in: [src/runtime/strategy-evolution.ts:55](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L55)
+Defined in: [src/runtime/strategy-evolution.ts:56](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L56)
 
 ***
 
@@ -21801,7 +21812,7 @@ Author + load a strategy from losses. Throws when the author emits no loadable m
 
 > **discriminatingMeans**(`report`, `fieldOrder`): `Record`\<`string`, \{ `score`: `number`; `usd`: `number`; \}\> \| `null`
 
-Defined in: [src/runtime/strategy-evolution.ts:237](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L237)
+Defined in: [src/runtime/strategy-evolution.ts:245](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L245)
 
 Strategy means recomputed over the DISCRIMINATING tasks only — tasks where the field
  strategies did not all score identically. Zero-spread tasks (everyone 1.0, everyone
@@ -21828,7 +21839,7 @@ Strategy means recomputed over the DISCRIMINATING tasks only — tasks where the
 
 > **pickChampion**(`means`, `fieldOrder`, `policy`, `epsilon`): [`ChampionPick`](#championpick)
 
-Defined in: [src/runtime/strategy-evolution.ts:262](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L262)
+Defined in: [src/runtime/strategy-evolution.ts:270](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L270)
 
 The champion pick over a means table. 'score' takes the best mean score (ties →
  field order). 'costAware' treats scores within `epsilon` of the best as tied and
@@ -21862,7 +21873,7 @@ The champion pick over a means table. 'score' takes the best mean score (ties 
 
 > **selectChampion**(`report`, `fieldOrder`, `policy`, `epsilon`): [`ChampionPick`](#championpick)
 
-Defined in: [src/runtime/strategy-evolution.ts:285](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L285)
+Defined in: [src/runtime/strategy-evolution.ts:293](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L293)
 
 Search-side champion selection over a tournament report.
 
@@ -21894,7 +21905,7 @@ Search-side champion selection over a tournament report.
 
 > **runStrategyEvolution**(`cfg`): `Promise`\<[`EvolutionReport`](#evolutionreport)\>
 
-Defined in: [src/runtime/strategy-evolution.ts:365](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L365)
+Defined in: [src/runtime/strategy-evolution.ts:373](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L373)
 
 Multi-generation strategy search: author candidates from tournament losses, play them against the incumbent at equal budget, promote via `promotionGate` on an untouched holdout slice.
 
diff --git a/src/runtime/run-benchmark.test.ts b/src/runtime/run-benchmark.test.ts
index 8ae5cc54..34244764 100644
--- a/src/runtime/run-benchmark.test.ts
+++ b/src/runtime/run-benchmark.test.ts
@@ -12,6 +12,7 @@ interface ChatRequest {
   messages?: Array<{ role?: string; content?: string }>
   temperature?: number
   max_tokens?: number
+  reasoning_effort?: string
 }
 
 const task: AgenticTask = {
@@ -109,6 +110,7 @@ describe('runBenchmark model availability', () => {
     const checks = requests.filter(isModelCheck)
     expect(checks.map((request) => request.model).sort()).toEqual(['analyst-model', 'worker-model'])
     expect(checks.every((request) => request.max_tokens === 1)).toBe(true)
+    expect(checks.every((request) => request.reasoning_effort === 'none')).toBe(true)
     expect(events.indexOf('open')).toBe(2)
   })
 
diff --git a/src/runtime/run-benchmark.ts b/src/runtime/run-benchmark.ts
index ff5dee1f..d3fbe853 100644
--- a/src/runtime/run-benchmark.ts
+++ b/src/runtime/run-benchmark.ts
@@ -151,7 +151,7 @@ async function preflightModels(cfg: BenchmarkConfig): Promise<void> {
           model,
         },
         [{ role: 'user', content: 'Reply OK.' }],
-        { maxTokens: 1 },
+        { maxTokens: 1, reasoningEffort: 'none' },
       )
     })
 
diff --git a/src/runtime/strategy-evolution.ts b/src/runtime/strategy-evolution.ts
index ae000929..b535181e 100644
--- a/src/runtime/strategy-evolution.ts
+++ b/src/runtime/strategy-evolution.ts
@@ -28,6 +28,7 @@ import type { ChatClient } from '@tangle-network/agent-eval'
 import type { RuntimeHooks } from '../runtime-hooks'
 import { type PromotionVerdict, promotionGate } from './promotion-gate'
 import {
+  type BenchmarkConfig,
   type BenchmarkReport,
   type BenchmarkTaskRow,
   type Environment,
@@ -65,6 +66,13 @@ export interface StrategyEvolutionConfig {
   /** Extra offset past the train slice for the holdout draw (rotate across runs). */
   holdoutOffset?: number
   worker: AgenticOptions
+  /**
+   * Model availability check before the first benchmark phase.
+   *
+   * A successful check is reused for the remaining phases in this evolution run.
+   * See `BenchmarkConfig.modelPreflight`.
+   */
+  modelPreflight?: BenchmarkConfig['modelPreflight']
   author: EvolutionAuthor
   /** Rollouts (sample) / shots (refine) per strategy per task. Default 3. */
   budget?: number
@@ -404,20 +412,24 @@ export async function runStrategyEvolution(cfg: StrategyEvolutionConfig): Promis
       writeFileSync(cfg.checkpoint.path, JSON.stringify({ ...state, fingerprint }, null, 1))
   }
 
+  let modelsPreflighted = false
   const bench = async (phase: string, tasks: AgenticTask[], strategies: Strategy[]) => {
     await cfg.onPhase?.(phase)
-    return runBenchmark({
+    const report = await runBenchmark({
       environment: cfg.environment,
       tasks,
       worker: cfg.worker,
       strategies,
       budget,
       concurrency,
+      modelPreflight: modelsPreflighted ? false : cfg.modelPreflight,
       ...(cfg.onTask
         ? { onTask: (row, done, total) => cfg.onTask?.(phase, row, done, total) }
         : {}),
       ...(cfg.hooks ? { hooks: cfg.hooks } : {}),
     })
+    modelsPreflighted = true
+    return report
   }
 
   const train = await cfg.tasks(0, cfg.trainN)
diff --git a/tests/loops/strategy-evolution.test.ts b/tests/loops/strategy-evolution.test.ts
index 806c4d36..93487104 100644
--- a/tests/loops/strategy-evolution.test.ts
+++ b/tests/loops/strategy-evolution.test.ts
@@ -162,12 +162,16 @@ describe('runStrategyEvolution', () => {
     stubWorkerRouter()
     const { chat } = scriptedChat([fenced(twoShotDepthModule)])
     const sliceCalls: Array<{ offset: number; n: number }> = []
+    const checkedModels: string[] = []
     const report = await runStrategyEvolution({
       environment: shotCountingSurface(),
       tasks: sliceTasks(sliceCalls),
       trainN: 8,
       holdoutN: 8,
       worker,
+      modelPreflight: async (model) => {
+        checkedModels.push(model)
+      },
       author: { chat, model: 'author-model' },
       budget: 3,
       concurrency: 2,
@@ -178,6 +182,7 @@ describe('runStrategyEvolution', () => {
     })
 
     expect(report.gen0Champion.name).toBe('sample')
+    expect(checkedModels).toEqual(['test-model'])
     expect(report.finalChampion.name).toBe('two-shot-depth')
     expect(report.verdict.promoted).toBe(true)
     expect(report.verdict.reason).toBe('significant')

From d73908f73801ab0e79f121295288ee00f99972bb Mon Sep 17 00:00:00 2001
From: Drew Stone <drewstone329@gmail.com>
Date: Fri, 24 Jul 2026 19:41:32 -0600
Subject: [PATCH 4/4] fix(bench): bound and aggregate model checks

---
 docs/api/runtime.md               | 206 ++++++++++++++++--------------
 src/runtime/run-benchmark.test.ts |  43 +++++++
 src/runtime/run-benchmark.ts      |  49 ++++++-
 src/runtime/strategy-evolution.ts |   3 +
 4 files changed, 199 insertions(+), 102 deletions(-)

diff --git a/docs/api/runtime.md b/docs/api/runtime.md
index bb295299..f5a9415d 100644
--- a/docs/api/runtime.md
+++ b/docs/api/runtime.md
@@ -6195,7 +6195,7 @@ Lifecycle observability — every spawn/settle of every cell's shots/analysts st
 
 ##### modelPreflight?
 
-> `optional` **modelPreflight?**: `false` \| ((`model`, `worker`) => `Promise`\<`void`\>)
+> `optional` **modelPreflight?**: `false` \| ((`model`, `worker`, `signal`) => `Promise`\<`void`\>)
 
 Defined in: [src/runtime/run-benchmark.ts:60](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L60)
 
@@ -6205,11 +6205,19 @@ By default, live router workers send one one-token request per unique worker and
 model. Injected `worker.complete` transports skip the check. Pass `false` to disable it or a
 callback to check each unique model through a custom transport.
 
+##### modelPreflightTimeoutMs?
+
+> `optional` **modelPreflightTimeoutMs?**: `number`
+
+Defined in: [src/runtime/run-benchmark.ts:64](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L64)
+
+Maximum time for each model availability check. Default 30 seconds.
+
 ***
 
 ### BenchmarkLift
 
-Defined in: [src/runtime/run-benchmark.ts:63](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L63)
+Defined in: [src/runtime/run-benchmark.ts:67](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L67)
 
 #### Properties
 
@@ -6217,7 +6225,7 @@ Defined in: [src/runtime/run-benchmark.ts:63](https://github.com/tangle-network/
 
 > **mean**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:65](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L65)
+Defined in: [src/runtime/run-benchmark.ts:69](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L69)
 
 Mean of paired deltas (refine − sample).
 
@@ -6225,25 +6233,25 @@ Mean of paired deltas (refine − sample).
 
 > **low**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:66](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L66)
+Defined in: [src/runtime/run-benchmark.ts:70](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L70)
 
 ##### high
 
 > **high**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:67](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L67)
+Defined in: [src/runtime/run-benchmark.ts:71](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L71)
 
 ##### n
 
 > **n**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:68](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L68)
+Defined in: [src/runtime/run-benchmark.ts:72](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L72)
 
 ***
 
 ### BenchmarkCell
 
-Defined in: [src/runtime/run-benchmark.ts:72](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L72)
+Defined in: [src/runtime/run-benchmark.ts:76](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L76)
 
 One strategy's outcome on one task — the per-task cell an optimizer consumes.
 
@@ -6253,19 +6261,19 @@ One strategy's outcome on one task — the per-task cell an optimizer consumes.
 
 > **score**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:73](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L73)
+Defined in: [src/runtime/run-benchmark.ts:77](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L77)
 
 ##### resolved
 
 > **resolved**: `boolean`
 
-Defined in: [src/runtime/run-benchmark.ts:74](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L74)
+Defined in: [src/runtime/run-benchmark.ts:78](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L78)
 
 ##### progression
 
 > **progression**: `number`[]
 
-Defined in: [src/runtime/run-benchmark.ts:76](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L76)
+Defined in: [src/runtime/run-benchmark.ts:80](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L80)
 
 The progress curve (refine: score per shot; sample: best-so-far per rollout).
 
@@ -6273,19 +6281,19 @@ The progress curve (refine: score per shot; sample: best-so-far per rollout).
 
 > **usd**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:77](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L77)
+Defined in: [src/runtime/run-benchmark.ts:81](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L81)
 
 ##### ms
 
 > **ms**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:78](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L78)
+Defined in: [src/runtime/run-benchmark.ts:82](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L82)
 
 ##### tokens
 
 > **tokens**: `object`
 
-Defined in: [src/runtime/run-benchmark.ts:79](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L79)
+Defined in: [src/runtime/run-benchmark.ts:83](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L83)
 
 ###### input
 
@@ -6299,7 +6307,7 @@ Defined in: [src/runtime/run-benchmark.ts:79](https://github.com/tangle-network/
 
 ### BenchmarkTaskRow
 
-Defined in: [src/runtime/run-benchmark.ts:82](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L82)
+Defined in: [src/runtime/run-benchmark.ts:86](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L86)
 
 #### Properties
 
@@ -6307,13 +6315,13 @@ Defined in: [src/runtime/run-benchmark.ts:82](https://github.com/tangle-network/
 
 > **taskId**: `string`
 
-Defined in: [src/runtime/run-benchmark.ts:83](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L83)
+Defined in: [src/runtime/run-benchmark.ts:87](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L87)
 
 ##### cells?
 
 > `optional` **cells?**: `Record`\<`string`, [`BenchmarkCell`](#benchmarkcell)\>
 
-Defined in: [src/runtime/run-benchmark.ts:85](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L85)
+Defined in: [src/runtime/run-benchmark.ts:89](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L89)
 
 Per-strategy cells; absent when the task errored before completing all strategies.
 
@@ -6321,7 +6329,7 @@ Per-strategy cells; absent when the task errored before completing all strategie
 
 > `optional` **errors?**: `Record`\<`string`, `string`\>
 
-Defined in: [src/runtime/run-benchmark.ts:89](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L89)
+Defined in: [src/runtime/run-benchmark.ts:93](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L93)
 
 Per-strategy failures on this task: the strategy competed, threw, and scored an
  honest zero — it loses, it does not poison the row. The message is kept so a later
@@ -6331,7 +6339,7 @@ Per-strategy failures on this task: the strategy competed, threw, and scored an
 
 > `optional` **error?**: `string`
 
-Defined in: [src/runtime/run-benchmark.ts:91](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L91)
+Defined in: [src/runtime/run-benchmark.ts:95](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L95)
 
 Why the task was excluded (infra/setup failure) — never silently dropped.
 
@@ -6339,7 +6347,7 @@ Why the task was excluded (infra/setup failure) — never silently dropped.
 
 ### BenchmarkStrategySummary
 
-Defined in: [src/runtime/run-benchmark.ts:94](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L94)
+Defined in: [src/runtime/run-benchmark.ts:98](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L98)
 
 #### Properties
 
@@ -6347,7 +6355,7 @@ Defined in: [src/runtime/run-benchmark.ts:94](https://github.com/tangle-network/
 
 > **score**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:96](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L96)
+Defined in: [src/runtime/run-benchmark.ts:100](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L100)
 
 Mean verifier score (0..1).
 
@@ -6355,7 +6363,7 @@ Mean verifier score (0..1).
 
 > **resolved**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:98](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L98)
+Defined in: [src/runtime/run-benchmark.ts:102](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L102)
 
 Fraction of tasks fully resolved.
 
@@ -6363,7 +6371,7 @@ Fraction of tasks fully resolved.
 
 > **usd**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:100](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L100)
+Defined in: [src/runtime/run-benchmark.ts:104](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L104)
 
 Mean cost vector per task.
 
@@ -6371,13 +6379,13 @@ Mean cost vector per task.
 
 > **ms**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:101](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L101)
+Defined in: [src/runtime/run-benchmark.ts:105](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L105)
 
 ***
 
 ### BenchmarkReport
 
-Defined in: [src/runtime/run-benchmark.ts:105](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L105)
+Defined in: [src/runtime/run-benchmark.ts:109](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L109)
 
 Benchmark output: per-strategy means plus the full per-task × per-strategy losses table an optimizer mines.
 
@@ -6387,19 +6395,19 @@ Benchmark output: per-strategy means plus the full per-task × per-strategy loss
 
 > **n**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:106](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L106)
+Defined in: [src/runtime/run-benchmark.ts:110](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L110)
 
 ##### excluded
 
 > **excluded**: `number`
 
-Defined in: [src/runtime/run-benchmark.ts:107](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L107)
+Defined in: [src/runtime/run-benchmark.ts:111](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L111)
 
 ##### perStrategy
 
 > **perStrategy**: `Record`\<`string`, [`BenchmarkStrategySummary`](#benchmarkstrategysummary)\>
 
-Defined in: [src/runtime/run-benchmark.ts:109](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L109)
+Defined in: [src/runtime/run-benchmark.ts:113](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L113)
 
 Per-strategy means (keyed by strategy.name).
 
@@ -6407,7 +6415,7 @@ Per-strategy means (keyed by strategy.name).
 
 > **perTask**: [`BenchmarkTaskRow`](#benchmarktaskrow)[]
 
-Defined in: [src/runtime/run-benchmark.ts:112](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L112)
+Defined in: [src/runtime/run-benchmark.ts:116](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L116)
 
 The full per-task × per-strategy table — the LOSSES an optimizer (GEPA, a
  strategy-author, an operator) consumes. Includes errored tasks with the reason.
@@ -6416,7 +6424,7 @@ The full per-task × per-strategy table — the LOSSES an optimizer (GEPA, a
 
 > **pareto**: `string`[]
 
-Defined in: [src/runtime/run-benchmark.ts:115](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L115)
+Defined in: [src/runtime/run-benchmark.ts:119](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L119)
 
 The non-dominated strategies on (score ↑, $/task ↓) — collapse-last, per the canon:
  a strategy that ties on score at half the cost WINS and a scalar would hide it.
@@ -6425,7 +6433,7 @@ The non-dominated strategies on (score ↑, $/task ↓) — collapse-last, per t
 
 > `optional` **refineVsSample?**: [`BenchmarkLift`](#benchmarklift)
 
-Defined in: [src/runtime/run-benchmark.ts:117](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L117)
+Defined in: [src/runtime/run-benchmark.ts:121](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L121)
 
 The headline when both `refine` and `sample` ran: paired-bootstrap lift of refine over sample.
 
@@ -7951,7 +7959,7 @@ Defined in: [src/runtime/strategy-evolution.ts:68](https://github.com/tangle-net
 
 ##### modelPreflight?
 
-> `optional` **modelPreflight?**: `false` \| ((`model`, `worker`) => `Promise`\<`void`\>)
+> `optional` **modelPreflight?**: `false` \| ((`model`, `worker`, `signal`) => `Promise`\<`void`\>)
 
 Defined in: [src/runtime/strategy-evolution.ts:75](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L75)
 
@@ -7960,17 +7968,25 @@ Model availability check before the first benchmark phase.
 A successful check is reused for the remaining phases in this evolution run.
 See `BenchmarkConfig.modelPreflight`.
 
+##### modelPreflightTimeoutMs?
+
+> `optional` **modelPreflightTimeoutMs?**: `number`
+
+Defined in: [src/runtime/strategy-evolution.ts:77](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L77)
+
+Maximum time for each model availability check. Default 30 seconds.
+
 ##### author
 
 > **author**: [`EvolutionAuthor`](#evolutionauthor)
 
-Defined in: [src/runtime/strategy-evolution.ts:76](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L76)
+Defined in: [src/runtime/strategy-evolution.ts:78](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L78)
 
 ##### budget?
 
 > `optional` **budget?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:78](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L78)
+Defined in: [src/runtime/strategy-evolution.ts:80](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L80)
 
 Rollouts (sample) / shots (refine) per strategy per task. Default 3.
 
@@ -7978,13 +7994,13 @@ Rollouts (sample) / shots (refine) per strategy per task. Default 3.
 
 > `optional` **concurrency?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:79](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L79)
+Defined in: [src/runtime/strategy-evolution.ts:81](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L81)
 
 ##### generations?
 
 > `optional` **generations?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:81](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L81)
+Defined in: [src/runtime/strategy-evolution.ts:83](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L83)
 
 Author→tournament rounds after gen0. Default 2.
 
@@ -7992,7 +8008,7 @@ Author→tournament rounds after gen0. Default 2.
 
 > `optional` **populationSize?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:83](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L83)
+Defined in: [src/runtime/strategy-evolution.ts:85](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L85)
 
 Authored candidates per generation. Default 2.
 
@@ -8000,7 +8016,7 @@ Authored candidates per generation. Default 2.
 
 > `optional` **baselines?**: [`Strategy`](#strategy-3)\<[`StrategyResult`](#strategyresult-1)\>[]
 
-Defined in: [src/runtime/strategy-evolution.ts:85](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L85)
+Defined in: [src/runtime/strategy-evolution.ts:87](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L87)
 
 The gen0 field. Default [sample, refine, sampleThenRefine].
 
@@ -8008,7 +8024,7 @@ The gen0 field. Default [sample, refine, sampleThenRefine].
 
 > `optional` **objective?**: `"score"` \| `"cost"`
 
-Defined in: [src/runtime/strategy-evolution.ts:91](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L91)
+Defined in: [src/runtime/strategy-evolution.ts:93](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L93)
 
 What "better" means for PROMOTION. 'score' (default): the candidate must beat the
  incumbent's score (superiority gate). 'cost': the candidate must prove score
@@ -8020,7 +8036,7 @@ What "better" means for PROMOTION. 'score' (default): the candidate must beat th
 
 > `optional` **scoreTolerance?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:93](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L93)
+Defined in: [src/runtime/strategy-evolution.ts:95](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L95)
 
 Cost objective: the score CI lower bound must clear −scoreTolerance. Default 0.05.
 
@@ -8028,7 +8044,7 @@ Cost objective: the score CI lower bound must clear −scoreTolerance. Default 0
 
 > `optional` **champion?**: [`ChampionPolicy`](#championpolicy)
 
-Defined in: [src/runtime/strategy-evolution.ts:95](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L95)
+Defined in: [src/runtime/strategy-evolution.ts:97](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L97)
 
 Search-side champion selection. Default 'costAware'.
 
@@ -8036,7 +8052,7 @@ Search-side champion selection. Default 'costAware'.
 
 > `optional` **championEpsilon?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:97](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L97)
+Defined in: [src/runtime/strategy-evolution.ts:99](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L99)
 
 Score band treated as a tie under 'costAware'. Default 0.01.
 
@@ -8044,7 +8060,7 @@ Score band treated as a tie under 'costAware'. Default 0.01.
 
 > **outDir**: `string`
 
-Defined in: [src/runtime/strategy-evolution.ts:99](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L99)
+Defined in: [src/runtime/strategy-evolution.ts:101](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L101)
 
 Where authored modules are written.
 
@@ -8052,7 +8068,7 @@ Where authored modules are written.
 
 > `optional` **minPairedTasks?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:101](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L101)
+Defined in: [src/runtime/strategy-evolution.ts:103](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L103)
 
 Promotion-gate evidence floor (paired holdout tasks).
 
@@ -8060,7 +8076,7 @@ Promotion-gate evidence floor (paired holdout tasks).
 
 > `optional` **band?**: `object`
 
-Defined in: [src/runtime/strategy-evolution.ts:110](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L110)
+Defined in: [src/runtime/strategy-evolution.ts:112](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L112)
 
 BAND-AWARE scoring — concentrate the measurement where lift is possible.
  Holdout: draw `holdoutPoolN` candidate tasks and run `baselines[0]` once at the run
@@ -8086,7 +8102,7 @@ Keep holdout tasks where the reference scores ≤ this. Default 0.99 — drop on
 
 > `optional` **lossesDetail?**: `"exact"` \| `"binary"`
 
-Defined in: [src/runtime/strategy-evolution.ts:119](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L119)
+Defined in: [src/runtime/strategy-evolution.ts:121](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L121)
 
 What the author learns from a tournament. 'exact' (default) = scores + progressions
  per task; 'binary' = pass/fail only — the leakage-bounded channel (one bit per cell
@@ -8096,7 +8112,7 @@ What the author learns from a tournament. 'exact' (default) = scores + progressi
 
 > `optional` **reproducerCheck?**: `object`
 
-Defined in: [src/runtime/strategy-evolution.ts:126](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L126)
+Defined in: [src/runtime/strategy-evolution.ts:128](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L128)
 
 Reproducer certification (arXiv:2606.11045): when the final champion is AUTHORED,
  compress it to a short natural-language summary, have a fresh author re-implement
@@ -8122,7 +8138,7 @@ Reproduction counts as faithful when reproducedScore ≥ championScore − toler
 
 > `optional` **checkpoint?**: `object`
 
-Defined in: [src/runtime/strategy-evolution.ts:136](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L136)
+Defined in: [src/runtime/strategy-evolution.ts:138](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L138)
 
 Endurance: write the run state after every completed phase; with `resume`, a
  restart skips completed phases (authored modules re-imported from their files).
@@ -8140,7 +8156,7 @@ Endurance: write the run state after every completed phase; with `resume`, a
 
 > `optional` **onPhase?**: (`phase`) => `Promise`\<`void`\>
 
-Defined in: [src/runtime/strategy-evolution.ts:143](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L143)
+Defined in: [src/runtime/strategy-evolution.ts:145](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L145)
 
 Called before each benchmark phase (gen0, gen1…, band-screen, holdout, reproduce).
  The seam for environment recycling — no artifacts span phases, so a runner may
@@ -8160,7 +8176,7 @@ Called before each benchmark phase (gen0, gen1…, band-screen, holdout, reprodu
 
 > `optional` **onTask?**: (`phase`, `row`, `done`, `total`) => `void`
 
-Defined in: [src/runtime/strategy-evolution.ts:144](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L144)
+Defined in: [src/runtime/strategy-evolution.ts:146](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L146)
 
 ###### Parameters
 
@@ -8188,13 +8204,13 @@ Defined in: [src/runtime/strategy-evolution.ts:144](https://github.com/tangle-ne
 
 > `optional` **hooks?**: [`RuntimeHooks`](index.md#runtimehooks)
 
-Defined in: [src/runtime/strategy-evolution.ts:145](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L145)
+Defined in: [src/runtime/strategy-evolution.ts:147](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L147)
 
 ***
 
 ### ChampionPick
 
-Defined in: [src/runtime/strategy-evolution.ts:160](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L160)
+Defined in: [src/runtime/strategy-evolution.ts:162](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L162)
 
 #### Properties
 
@@ -8202,25 +8218,25 @@ Defined in: [src/runtime/strategy-evolution.ts:160](https://github.com/tangle-ne
 
 > **name**: `string`
 
-Defined in: [src/runtime/strategy-evolution.ts:161](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L161)
+Defined in: [src/runtime/strategy-evolution.ts:163](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L163)
 
 ##### score
 
 > **score**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:162](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L162)
+Defined in: [src/runtime/strategy-evolution.ts:164](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L164)
 
 ##### usd
 
 > **usd**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:163](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L163)
+Defined in: [src/runtime/strategy-evolution.ts:165](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L165)
 
 ***
 
 ### EvolutionCandidate
 
-Defined in: [src/runtime/strategy-evolution.ts:166](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L166)
+Defined in: [src/runtime/strategy-evolution.ts:168](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L168)
 
 #### Properties
 
@@ -8228,31 +8244,31 @@ Defined in: [src/runtime/strategy-evolution.ts:166](https://github.com/tangle-ne
 
 > **name**: `string`
 
-Defined in: [src/runtime/strategy-evolution.ts:167](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L167)
+Defined in: [src/runtime/strategy-evolution.ts:169](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L169)
 
 ##### file?
 
 > `optional` **file?**: `string`
 
-Defined in: [src/runtime/strategy-evolution.ts:168](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L168)
+Defined in: [src/runtime/strategy-evolution.ts:170](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L170)
 
 ##### gzipBits?
 
 > `optional` **gzipBits?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:169](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L169)
+Defined in: [src/runtime/strategy-evolution.ts:171](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L171)
 
 ##### codeChars?
 
 > `optional` **codeChars?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:170](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L170)
+Defined in: [src/runtime/strategy-evolution.ts:172](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L172)
 
 ##### error?
 
 > `optional` **error?**: `string`
 
-Defined in: [src/runtime/strategy-evolution.ts:172](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L172)
+Defined in: [src/runtime/strategy-evolution.ts:174](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L174)
 
 Present when this author attempt failed (recorded, never silent).
 
@@ -8260,7 +8276,7 @@ Present when this author attempt failed (recorded, never silent).
 
 ### EvolutionGeneration
 
-Defined in: [src/runtime/strategy-evolution.ts:175](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L175)
+Defined in: [src/runtime/strategy-evolution.ts:177](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L177)
 
 #### Properties
 
@@ -8268,31 +8284,31 @@ Defined in: [src/runtime/strategy-evolution.ts:175](https://github.com/tangle-ne
 
 > **generation**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:176](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L176)
+Defined in: [src/runtime/strategy-evolution.ts:178](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L178)
 
 ##### candidates
 
 > **candidates**: [`EvolutionCandidate`](#evolutioncandidate)[]
 
-Defined in: [src/runtime/strategy-evolution.ts:177](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L177)
+Defined in: [src/runtime/strategy-evolution.ts:179](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L179)
 
 ##### report
 
 > **report**: [`BenchmarkReport`](#benchmarkreport)
 
-Defined in: [src/runtime/strategy-evolution.ts:178](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L178)
+Defined in: [src/runtime/strategy-evolution.ts:180](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L180)
 
 ##### champion
 
 > **champion**: [`ChampionPick`](#championpick)
 
-Defined in: [src/runtime/strategy-evolution.ts:179](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L179)
+Defined in: [src/runtime/strategy-evolution.ts:181](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L181)
 
 ***
 
 ### EvolutionArchiveNode
 
-Defined in: [src/runtime/strategy-evolution.ts:182](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L182)
+Defined in: [src/runtime/strategy-evolution.ts:184](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L184)
 
 #### Properties
 
@@ -8300,25 +8316,25 @@ Defined in: [src/runtime/strategy-evolution.ts:182](https://github.com/tangle-ne
 
 > **name**: `string`
 
-Defined in: [src/runtime/strategy-evolution.ts:183](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L183)
+Defined in: [src/runtime/strategy-evolution.ts:185](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L185)
 
 ##### source
 
 > **source**: `"baseline"` \| `"authored"`
 
-Defined in: [src/runtime/strategy-evolution.ts:184](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L184)
+Defined in: [src/runtime/strategy-evolution.ts:186](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L186)
 
 ##### generation
 
 > **generation**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:185](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L185)
+Defined in: [src/runtime/strategy-evolution.ts:187](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L187)
 
 ##### parent?
 
 > `optional` **parent?**: `string`
 
-Defined in: [src/runtime/strategy-evolution.ts:187](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L187)
+Defined in: [src/runtime/strategy-evolution.ts:189](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L189)
 
 The champion whose tournament losses this candidate was authored from.
 
@@ -8326,19 +8342,19 @@ The champion whose tournament losses this candidate was authored from.
 
 > `optional` **gzipBits?**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:188](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L188)
+Defined in: [src/runtime/strategy-evolution.ts:190](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L190)
 
 ##### file?
 
 > `optional` **file?**: `string`
 
-Defined in: [src/runtime/strategy-evolution.ts:189](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L189)
+Defined in: [src/runtime/strategy-evolution.ts:191](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L191)
 
 ##### score
 
 > **score**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:192](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L192)
+Defined in: [src/runtime/strategy-evolution.ts:194](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L194)
 
 Latest measured tournament result — 0 until the node's first tournament settles
  (an authored node is created before its generation's benchmark runs).
@@ -8347,13 +8363,13 @@ Latest measured tournament result — 0 until the node's first tournament settle
 
 > **usd**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:193](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L193)
+Defined in: [src/runtime/strategy-evolution.ts:195](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L195)
 
 ***
 
 ### EvolutionBandInfo
 
-Defined in: [src/runtime/strategy-evolution.ts:212](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L212)
+Defined in: [src/runtime/strategy-evolution.ts:214](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L214)
 
 #### Properties
 
@@ -8361,7 +8377,7 @@ Defined in: [src/runtime/strategy-evolution.ts:212](https://github.com/tangle-ne
 
 > **screened**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:214](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L214)
+Defined in: [src/runtime/strategy-evolution.ts:216](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L216)
 
 Tasks screened by the reference on the holdout pool.
 
@@ -8369,7 +8385,7 @@ Tasks screened by the reference on the holdout pool.
 
 > **inBand**: `number`
 
-Defined in: [src/runtime/strategy-evolution.ts:216](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L216)
+Defined in: [src/runtime/strategy-evolution.ts:218](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L218)
 
 Tasks kept (reference score ≤ maxRefScore) before truncating to holdoutN.
 
@@ -8377,7 +8393,7 @@ Tasks kept (reference score ≤ maxRefScore) before truncating to holdoutN.
 
 > **refScores**: `object`[]
 
-Defined in: [src/runtime/strategy-evolution.ts:218](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L218)
+Defined in: [src/runtime/strategy-evolution.ts:220](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L220)
 
 Reference scores per screened task (the screening record).
 
@@ -8393,7 +8409,7 @@ Reference scores per screened task (the screening record).
 
 ### EvolutionReport
 
-Defined in: [src/runtime/strategy-evolution.ts:221](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L221)
+Defined in: [src/runtime/strategy-evolution.ts:223](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L223)
 
 #### Properties
 
@@ -8401,49 +8417,49 @@ Defined in: [src/runtime/strategy-evolution.ts:221](https://github.com/tangle-ne
 
 > **gen0**: [`BenchmarkReport`](#benchmarkreport)
 
-Defined in: [src/runtime/strategy-evolution.ts:222](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L222)
+Defined in: [src/runtime/strategy-evolution.ts:224](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L224)
 
 ##### gen0Champion
 
 > **gen0Champion**: [`ChampionPick`](#championpick)
 
-Defined in: [src/runtime/strategy-evolution.ts:223](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L223)
+Defined in: [src/runtime/strategy-evolution.ts:225](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L225)
 
 ##### generations
 
 > **generations**: [`EvolutionGeneration`](#evolutiongeneration)[]
 
-Defined in: [src/runtime/strategy-evolution.ts:224](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L224)
+Defined in: [src/runtime/strategy-evolution.ts:226](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L226)
 
 ##### archive
 
 > **archive**: [`EvolutionArchiveNode`](#evolutionarchivenode)[]
 
-Defined in: [src/runtime/strategy-evolution.ts:225](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L225)
+Defined in: [src/runtime/strategy-evolution.ts:227](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L227)
 
 ##### finalChampion
 
 > **finalChampion**: [`ChampionPick`](#championpick)
 
-Defined in: [src/runtime/strategy-evolution.ts:226](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L226)
+Defined in: [src/runtime/strategy-evolution.ts:228](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L228)
 
 ##### holdout
 
 > **holdout**: [`BenchmarkReport`](#benchmarkreport)
 
-Defined in: [src/runtime/strategy-evolution.ts:227](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L227)
+Defined in: [src/runtime/strategy-evolution.ts:229](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L229)
 
 ##### verdict
 
 > **verdict**: [`PromotionVerdict`](#promotionverdict)
 
-Defined in: [src/runtime/strategy-evolution.ts:228](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L228)
+Defined in: [src/runtime/strategy-evolution.ts:230](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L230)
 
 ##### band?
 
 > `optional` **band?**: [`EvolutionBandInfo`](#evolutionbandinfo)
 
-Defined in: [src/runtime/strategy-evolution.ts:231](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L231)
+Defined in: [src/runtime/strategy-evolution.ts:233](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L233)
 
 Present when band screening ran — the verdict's estimand is then "paired lift on
  headroom tasks" (band membership fixed by the reference screen, pre-registered).
@@ -8452,7 +8468,7 @@ Present when band screening ran — the verdict's estimand is then "paired lift
 
 > `optional` **reproduction?**: `ReproductionCheck`
 
-Defined in: [src/runtime/strategy-evolution.ts:233](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L233)
+Defined in: [src/runtime/strategy-evolution.ts:235](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L235)
 
 Present when reproducerCheck ran (final champion was authored).
 
@@ -8460,7 +8476,7 @@ Present when reproducerCheck ran (final champion was authored).
 
 > **trajectory**: `object`[]
 
-Defined in: [src/runtime/strategy-evolution.ts:238](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L238)
+Defined in: [src/runtime/strategy-evolution.ts:240](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L240)
 
 SEARCH TELEMETRY, not evidence: each entry is that generation's own train-slice
  re-measurement, so cross-generation deltas mix true drift with run-to-run variance
@@ -21191,7 +21207,7 @@ The turnkey production brain — tests script a mock `ToolLoopChat`; production
 
 > **runBenchmark**(`cfg`): `Promise`\<[`BenchmarkReport`](#benchmarkreport)\>
 
-Defined in: [src/runtime/run-benchmark.ts:173](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L173)
+Defined in: [src/runtime/run-benchmark.ts:208](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L208)
 
 Run the requested strategies over the tasks, scored by the Environment's own check.
  Resilient: a task whose rollouts fail (transient infra) is excluded from the stats but
@@ -21213,7 +21229,7 @@ Run the requested strategies over the tasks, scored by the Environment's own che
 
 > **printBenchmarkReport**(`report`): `void`
 
-Defined in: [src/runtime/run-benchmark.ts:274](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L274)
+Defined in: [src/runtime/run-benchmark.ts:309](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/run-benchmark.ts#L309)
 
 Pretty-print a report — the "free optimization" verdict, with the cost vector.
 
@@ -21812,7 +21828,7 @@ Author + load a strategy from losses. Throws when the author emits no loadable m
 
 > **discriminatingMeans**(`report`, `fieldOrder`): `Record`\<`string`, \{ `score`: `number`; `usd`: `number`; \}\> \| `null`
 
-Defined in: [src/runtime/strategy-evolution.ts:245](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L245)
+Defined in: [src/runtime/strategy-evolution.ts:247](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L247)
 
 Strategy means recomputed over the DISCRIMINATING tasks only — tasks where the field
  strategies did not all score identically. Zero-spread tasks (everyone 1.0, everyone
@@ -21839,7 +21855,7 @@ Strategy means recomputed over the DISCRIMINATING tasks only — tasks where the
 
 > **pickChampion**(`means`, `fieldOrder`, `policy`, `epsilon`): [`ChampionPick`](#championpick)
 
-Defined in: [src/runtime/strategy-evolution.ts:270](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L270)
+Defined in: [src/runtime/strategy-evolution.ts:272](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L272)
 
 The champion pick over a means table. 'score' takes the best mean score (ties →
  field order). 'costAware' treats scores within `epsilon` of the best as tied and
@@ -21873,7 +21889,7 @@ The champion pick over a means table. 'score' takes the best mean score (ties 
 
 > **selectChampion**(`report`, `fieldOrder`, `policy`, `epsilon`): [`ChampionPick`](#championpick)
 
-Defined in: [src/runtime/strategy-evolution.ts:293](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L293)
+Defined in: [src/runtime/strategy-evolution.ts:295](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L295)
 
 Search-side champion selection over a tournament report.
 
@@ -21905,7 +21921,7 @@ Search-side champion selection over a tournament report.
 
 > **runStrategyEvolution**(`cfg`): `Promise`\<[`EvolutionReport`](#evolutionreport)\>
 
-Defined in: [src/runtime/strategy-evolution.ts:373](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L373)
+Defined in: [src/runtime/strategy-evolution.ts:375](https://github.com/tangle-network/agent-runtime/blob/main/src/runtime/strategy-evolution.ts#L375)
 
 Multi-generation strategy search: author candidates from tournament losses, play them against the incumbent at equal budget, promote via `promotionGate` on an untouched holdout slice.
 
diff --git a/src/runtime/run-benchmark.test.ts b/src/runtime/run-benchmark.test.ts
index 34244764..a03cc9e2 100644
--- a/src/runtime/run-benchmark.test.ts
+++ b/src/runtime/run-benchmark.test.ts
@@ -160,6 +160,49 @@ describe('runBenchmark model availability', () => {
     expect(onTask).not.toHaveBeenCalled()
   })
 
+  it('reports every unavailable model in one failure', async () => {
+    const events: string[] = []
+
+    await expect(
+      runBenchmark({
+        environment: surface(events),
+        tasks: [task],
+        worker: { ...worker, analystModel: 'analyst-model' },
+        strategies: [oneShot],
+        budget: 1,
+        modelPreflight: async (model) => {
+          throw new Error(`${model} is unavailable`)
+        },
+      }),
+    ).rejects.toThrow(
+      'Benchmark model "worker-model" preflight failed: worker-model is unavailable; ' +
+        'Benchmark model "analyst-model" preflight failed: analyst-model is unavailable',
+    )
+
+    expect(events).not.toContain('open')
+  })
+
+  it('bounds a model check that never settles', async () => {
+    const events: string[] = []
+
+    await expect(
+      runBenchmark({
+        environment: surface(events),
+        tasks: [task],
+        worker,
+        strategies: [oneShot],
+        budget: 1,
+        modelPreflightTimeoutMs: 5,
+        modelPreflight: async (_model, _worker, signal) =>
+          new Promise<void>((_, reject) => {
+            signal.addEventListener('abort', () => reject(signal.reason), { once: true })
+          }),
+      }),
+    ).rejects.toThrow('Benchmark model "worker-model" preflight failed: timed out after 5 ms')
+
+    expect(events).not.toContain('open')
+  })
+
   it('skips the check for an injected completion transport', async () => {
     const requests: ChatRequest[] = []
     const events: string[] = []
diff --git a/src/runtime/run-benchmark.ts b/src/runtime/run-benchmark.ts
index d3fbe853..ea008d82 100644
--- a/src/runtime/run-benchmark.ts
+++ b/src/runtime/run-benchmark.ts
@@ -57,7 +57,11 @@ export interface BenchmarkConfig {
    * model. Injected `worker.complete` transports skip the check. Pass `false` to disable it or a
    * callback to check each unique model through a custom transport.
    */
-  modelPreflight?: false | ((model: string, worker: Readonly<AgenticOptions>) => Promise<void>)
+  modelPreflight?:
+    | false
+    | ((model: string, worker: Readonly<AgenticOptions>, signal: AbortSignal) => Promise<void>)
+  /** Maximum time for each model availability check. Default 30 seconds. */
+  modelPreflightTimeoutMs?: number
 }
 
 export interface BenchmarkLift {
@@ -141,9 +145,12 @@ async function preflightModels(cfg: BenchmarkConfig): Promise<void> {
   if (cfg.worker.complete && !cfg.modelPreflight) return
 
   const models = [...new Set([cfg.worker.model, cfg.worker.analystModel ?? cfg.worker.model])]
+  const timeoutMs = cfg.modelPreflightTimeoutMs ?? 30_000
+  if (!Number.isFinite(timeoutMs) || timeoutMs <= 0)
+    throw new Error('modelPreflightTimeoutMs must be a positive finite number')
   const check =
     cfg.modelPreflight ??
-    (async (model: string, worker: Readonly<AgenticOptions>) => {
+    (async (model: string, worker: Readonly<AgenticOptions>, signal: AbortSignal) => {
       await routerChatWithUsage(
         {
           routerBaseUrl: worker.routerBaseUrl,
@@ -151,20 +158,48 @@ async function preflightModels(cfg: BenchmarkConfig): Promise<void> {
           model,
         },
         [{ role: 'user', content: 'Reply OK.' }],
-        { maxTokens: 1, reasoningEffort: 'none' },
+        { maxTokens: 1, reasoningEffort: 'none', signal },
       )
     })
 
-  await Promise.all(
+  const results = await Promise.allSettled(
     models.map(async (model) => {
+      const controller = new AbortController()
+      let timeoutError: Error | undefined
+      let timer: ReturnType<typeof setTimeout> | undefined
       try {
-        await check(model, cfg.worker)
+        await Promise.race([
+          check(model, cfg.worker, controller.signal),
+          new Promise<never>((_, reject) => {
+            timer = setTimeout(() => {
+              timeoutError = new Error(`timed out after ${timeoutMs} ms`)
+              controller.abort(timeoutError)
+              reject(timeoutError)
+            }, timeoutMs)
+          }),
+        ])
       } catch (cause) {
-        const message = cause instanceof Error ? cause.message : String(cause)
-        throw new Error(`Benchmark model "${model}" preflight failed: ${message}`, { cause })
+        const reported = timeoutError ?? cause
+        const message = reported instanceof Error ? reported.message : String(reported)
+        throw new Error(`Benchmark model "${model}" preflight failed: ${message}`, {
+          cause: reported,
+        })
+      } finally {
+        if (timer) clearTimeout(timer)
       }
     }),
   )
+
+  const failures = results.flatMap((result) =>
+    result.status === 'rejected' ? [result.reason] : [],
+  )
+  if (failures.length === 1) throw failures[0]
+  if (failures.length > 1) {
+    const message = failures
+      .map((failure) => (failure instanceof Error ? failure.message : String(failure)))
+      .join('; ')
+    throw new AggregateError(failures, message)
+  }
 }
 
 /** Run the requested strategies over the tasks, scored by the Environment's own check.
diff --git a/src/runtime/strategy-evolution.ts b/src/runtime/strategy-evolution.ts
index b535181e..fcb16c88 100644
--- a/src/runtime/strategy-evolution.ts
+++ b/src/runtime/strategy-evolution.ts
@@ -73,6 +73,8 @@ export interface StrategyEvolutionConfig {
    * See `BenchmarkConfig.modelPreflight`.
    */
   modelPreflight?: BenchmarkConfig['modelPreflight']
+  /** Maximum time for each model availability check. Default 30 seconds. */
+  modelPreflightTimeoutMs?: BenchmarkConfig['modelPreflightTimeoutMs']
   author: EvolutionAuthor
   /** Rollouts (sample) / shots (refine) per strategy per task. Default 3. */
   budget?: number
@@ -423,6 +425,7 @@ export async function runStrategyEvolution(cfg: StrategyEvolutionConfig): Promis
       budget,
       concurrency,
       modelPreflight: modelsPreflighted ? false : cfg.modelPreflight,
+      modelPreflightTimeoutMs: cfg.modelPreflightTimeoutMs,
       ...(cfg.onTask
         ? { onTask: (row, done, total) => cfg.onTask?.(phase, row, done, total) }
         : {}),