@playerthree/bench/campaign
Benchmark campaigns (SPEC §20 Fixed-reference benchmark publications): each replicate authors the candidate's
policies independently, the frozen artifacts are evaluated through the series runner, and the record exports as a
consented publication. State lives in outDir, so re-running resumes.
Interfaces
BenchmarkSuiteMetadata
Defined in: packages/bench/src/campaign.ts:20
Public suite metadata for createBenchmarkSuite; candidateId names the entrant whose policies each replicate authors.
Properties
| Property | Type | Defined in |
|---|---|---|
blocks | object[] | packages/bench/src/campaign.ts:22 |
bootstrapReplicates? | number | packages/bench/src/campaign.ts:23 |
bootstrapSeed? | number | packages/bench/src/campaign.ts:23 |
build | object | packages/bench/src/campaign.ts:24 |
build.gameDigest | string | packages/playerthree/src/benchmark.ts:12 |
build.runnerDigest | string | packages/playerthree/src/benchmark.ts:12 |
candidateId | string | packages/bench/src/campaign.ts:21 |
harness | object | packages/bench/src/campaign.ts:23 |
harness.id | string | packages/playerthree/src/benchmark.ts:20 |
harness.version | string | packages/playerthree/src/benchmark.ts:20 |
id | string | packages/bench/src/campaign.ts:21 |
references | object[] | packages/bench/src/campaign.ts:22 |
replicates | number | packages/bench/src/campaign.ts:23 |
roles | string[] | packages/bench/src/campaign.ts:22 |
strata | object[] | packages/bench/src/campaign.ts:22 |
title | string | packages/bench/src/campaign.ts:21 |
track | "authored-policy" | "submitted-policy" | packages/bench/src/campaign.ts:21 |
version | number | packages/bench/src/campaign.ts:21 |
CampaignAttempt
Defined in: packages/bench/src/campaign.ts:35
Private ledger row: every authoring attempt and its spend, including infrastructure failures the public run omits.
Properties
| Property | Type | Description | Defined in |
|---|---|---|---|
costUsd | string | null | Runner-estimated USD; null while running or after a crash, when the full per-run ceiling is charged instead. | packages/bench/src/campaign.ts:39 |
finishedAt | string | null | - | packages/bench/src/campaign.ts:36 |
number | number | - | packages/bench/src/campaign.ts:36 |
outcome | "artifact" | "model_failure" | "infrastructure" | "interrupted" | null | - | packages/bench/src/campaign.ts:37 |
reason | string | null | - | packages/bench/src/campaign.ts:37 |
runId | string | - | packages/bench/src/campaign.ts:36 |
startedAt | string | - | packages/bench/src/campaign.ts:36 |
CampaignConfig
Defined in: packages/bench/src/campaign.ts:27
A campaign: the bench config and sealed suite, the candidate, the source trees whose digests must match the suite's build, and the publication consent.
Properties
CampaignOptions
Defined in: packages/bench/src/campaign.ts:46
Properties
| Property | Type | Description | Defined in |
|---|---|---|---|
log? | (message) => void | - | packages/bench/src/campaign.ts:52 |
maxCampaignUsd? | number | Cumulative USD ceiling over every authoring attempt of this campaign directory. | packages/bench/src/campaign.ts:51 |
outDir | string | - | packages/bench/src/campaign.ts:47 |
signal? | AbortSignal | Operator stop: authoring in progress becomes an infrastructure attempt; evaluation suspends without aborting its series. | packages/bench/src/campaign.ts:49 |
CampaignRecord
Defined in: packages/bench/src/campaign.ts:42
The private campaign record (campaign.json): suite, experiment, scored runs and every authoring attempt.
Properties
| Property | Type | Defined in |
|---|---|---|
attempts | CampaignAttempt[] | packages/bench/src/campaign.ts:44 |
experiment | BenchmarkExperiment | packages/bench/src/campaign.ts:43 |
format | "playerthree.benchmark-campaign" | packages/bench/src/campaign.ts:43 |
runs | BenchmarkRun[] | packages/bench/src/campaign.ts:44 |
scheduleDigest | string | null | packages/bench/src/campaign.ts:44 |
suite | BenchmarkSuite | packages/bench/src/campaign.ts:43 |
version | 1 | packages/bench/src/campaign.ts:43 |
Functions
createBenchmarkSuite()
function createBenchmarkSuite(config, metadata): BenchmarkSuite;Defined in: packages/bench/src/campaign.ts:59
Seals and validates a suite for config's series with metadata; the candidate entrant is generalized to a policy so the suite digest does not depend on it.
Parameters
| Parameter | Type |
|---|---|
config | BenchConfig |
metadata | BenchmarkSuiteMetadata |
Returns
exportCampaignPublication()
function exportCampaignPublication(
record,
consent,
publishedAt?
): BenchmarkPublication;Defined in: packages/bench/src/campaign.ts:254
Explicit consented public projection, separate from operational series retention.
Parameters
| Parameter | Type |
|---|---|
record | CampaignRecord |
consent | { artifacts: false; attribution: boolean; license: string; research: boolean; results: true; training: boolean; } |
consent.artifacts | false |
consent.attribution | boolean |
consent.license | string |
consent.research | boolean |
consent.results | true |
consent.training | boolean |
publishedAt? | string |
Returns
runCampaign()
function runCampaign(config, options): Promise<CampaignRecord>;Defined in: packages/bench/src/campaign.ts:78
Each replicate authors independently; evaluation only loads durable frozen artifacts and uses SeriesRunner.
Parameters
| Parameter | Type |
|---|---|
config | CampaignConfig |
options | CampaignOptions |
Returns
Promise<CampaignRecord>