aamas-artifact-evaluation

1# AAMAS Artifact Evaluation
2 
3Use this for evidence packaging around AAMAS. Because the venue is about interaction, an
4artifact must make a *multiagent* claim inspectable: the game, the other agents, and the
5protocol, not just a single trained model.
6 
7## Artifact plan
8 
9- Decide what a reviewer needs to believe the interaction claim: game or environment code,
10  opponent/population definitions, the training regime, seeds, payoff logs, proofs, or
11  qualitative episode traces.
12- Keep decision-critical evidence in the main paper or appendix; optional bulk runs can live in
13  the supplementary zip.
14- Anonymize repository history, paths, environment names, license headers, cluster paths, and
15  commit authors for the review version.
16- Include a minimal reproduction map: environment build, dependencies, hardware, commands,
17  expected outputs, per-run wall-clock, seeds, and known nondeterminism (especially in
18  self-play).
19- For a deployed or human-subject setting, give enough provenance for credible reproduction
20  without violating data-use terms.
21- After acceptance, replace anonymous archives with a public, licensed, citable artifact.
22 
23## What AAMAS evidence reviewers open first
24 
25The single fact that shapes packaging: a reviewer will re-run a small **game** far sooner than
26they will retrain a large policy, so make the strategic core turnkey before polishing anything.
27 
28| Claim type | First artifact inspected | Common failure caught |
29|---|---|---|
30…