Archive · Data · swarm-routing/benchmarks
benchmarks
Immutable, reproducible benchmark datasets: every number traceable to its image, commits and seeds.
Locked: Phase 4: dataset format first
What it does
benchmarks will hold the org's evidence: 20-seed ##RUN## / ##BENCH## data, the paired statistical verdicts recomputed from it, and a manifest for every run pinning the immutable ns3-dev:<ver>-<profile>-<sha7> image, each protocol's commit, the scenario, the seeds and the exact command line. No manifest, no merge.
Merged data is immutable: a correction is a new run that names the one it supersedes. Campaigns run on the optimized image; CI legs use debug and sanitized. The Colony Arena's leaderboard (site#12) will be regenerated from these verdicts by a script, never typed by hand.
Diagrams
Package structure
The repository's layout, its licence per directory (LICENSING.md) and what exists today.
| Path | What | Licence | State |
|---|---|---|---|
runs/<yyyy-mm-dd>-<campaign>/ | raw marker lines / CSVs + manifest.json, immutable once merged | GPL-2.0-only | planned |
verdicts/ | paired comparison tables produced by ns3-tool | GPL-2.0-only | planned |
schemas/ | the marker-line schema versions the data uses | GPL-2.0-only | planned |
Where it sits in the colony
Quests
- Open: Dataset layout, manifest schema, validator (benchmarks#3)
- Open: CI validates every manifest and refuses edits to merged data
- Open: Leaderboard regenerated from the verdicts in this site's CI (site#12)
Sources
Facts as of 2026-11-08, from the repository's AGENTS.md and the org ROADMAP.