Wiki: Benchmark campaign orchestration¶
Campaigns coordinate many benchmark harness runs from one campaign definition (YAML/JSON). Each member run is a normal benchmark output tree; the campaign root adds a campaign manifest, rollups, and optional alwm benchmark campaign run --dry-run planning without executing suites.
Concept + workflow page: campaign-orchestration.md.
Canonical references¶
| Topic | Location |
|---|---|
| Concept + failure modes | docs/wiki/campaign-orchestration.md |
| Walkthrough (committed examples) | docs/workflows/campaign-walkthrough.md |
| End-to-end publication (run → validate → pack → compare → publish) | docs/workflows/campaign-result-pack-publication.md |
| Workflow (CLI, fields, outputs, validation) | docs/workflows/benchmark-campaigns.md |
| Architecture decision | docs/adr/0001-campaign-orchestration.md or docs/architecture/adr/0001-benchmark-campaign-orchestration.md |
| Contract + tracking checklist | docs/tracking/campaign-orchestration.md |
| Example definitions | examples/campaigns/v1/ |
| Sample output tree | examples/campaign_runs/minimal_offline/ |
Relationship to longitudinal reporting¶
Member runs are benchmark_manifest artifacts under runs/runNNNN/. Point alwm benchmark longitudinal at runs/*/manifest.json for cross-run analysis (see longitudinal-reporting.md).