brycewang-stanford PODC Artifact Evaluation (there isn't one — read this)
If you arrived here expecting an Artifacts-Available / Functional / Reusable / Reproduced badge process, stop: PODC has no artifact-evaluation track and no ACM badge program. It is a proofs venue. The object that is "evaluated" is your theorem and its proof, and the honesty of your model and assumptions. This skill redirects artifact energy into the three things that actually decide a PODC paper's technical standing.
Why PODC has no artifact track
A PODC contribution is a model, a theorem, and a proof, not a running system or a dataset. There
is nothing for evaluators to install and run to confirm the claim — the claim is confirmed by
reading the proof. Importing a badge workflow here is a category error: it would evaluate an
artifact that is, at most, an optional illustration. (Contrast this with software-engineering or
systems venues, where the artifact is part of the evidence.) Any simulation you ship is a
figure-generator, not proof of correctness (podc-experiments, podc-reproducibility).
The three real "evaluations" at PODC
1. Proof-appendix completeness (the analogue of "Functional")
Just as an SE evaluator checks that an artifact runs, a PODC reviewer checks that the proof closes. Make it pass:
[ ] Every lemma the main theorem uses is stated and fully proved (or cited with the statement quoted)
[ ] No "it is easy to see" or "the remaining case is symmetric" hides a nontrivial step
[ ] Base cases of inductions/recursions are present
[ ] Named invariants are shown to hold initially and be preserved
[ ] The full version is self-contained: no proof depends on an external private note
2. Model/assumption rigor (the analogue of "Reusable")
An artifact is "Reusable" when others can build on it; a model is reusable when it is stated precisely enough that others can prove new results in it and trust yours:
[ ] The model box fixes network, timing, faults, adversary, randomness, and cost measure before any theorem
[ ] Every assumption the proofs consume is declared in the box (no assumption creep)
[ ] No proof silently strengthens synchrony or weakens the adversary mid-argument
[ ] Parameter regimes (ranges of n, t, diameter) are stated where results are conditional
[ ] Definitions match standard usage, or deviations are flagged explicitly
Run the assumption audit from podc-experiments; this is the single most common source of a PODC
soundness rejection.
3. Optional-simulation transparency (the analogue of "Available")
If — and only if — your paper includes a simulation, make it transparent, while remembering it establishes nothing:
[ ] Labeled as illustrative (a constant, a convergence rate, average-case behavior), not as verification
[ ] Parameters, ranges, number of trials, and random seeds reported
[ ] Environment pinned (language/version) in the full version or a linked bundle
[ ] Any repo/author-page link kept OUT of the anonymized submission (lightweight double-blind)
[ ] Added openly only in the camera-ready / arXiv full version
Mapping artifact instincts to PODC actions
| Artifact-venue instinct | PODC equivalent |
|---|---|
| "Package the code so it runs on a clean machine" | Make the proof close for a reader with only the submission |
| "Document dependencies" | Declare every assumption in the model box |
| "Provide a claim -> script -> result mapping" | Provide a theorem -> lemma -> proof map with exact cross-refs |
| "Deposit in a DOI archive for Available" | Post the full version with all proofs to arXiv |
| "Reproduce the headline numbers" | Let a reader re-derive the bound from the stated model |
What NOT to do
- Do not build a Dockerfile, a badge application, or an evaluator README expecting a PODC artifact track — there is none.
- Do not present a simulation as if it earned a "Reproduced" badge; PODC has no such badge and the proof is the evidence.
- Do not de-anonymize the submission by linking a simulation repository during review.
- Do not confuse PODC with software-engineering or systems venues (or with DISC's practice); the proofs-only evaluation model is the point.
Output format
[Reality check] confirmed: PODC has no artifact/badge track; the proof is what is evaluated
[Proof completeness] every lemma proved; no hidden steps; self-contained full version?
[Model rigor] model box complete; no assumption creep; regimes stated?
[Simulation] absent / illustrative-and-transparent (seeds/ranges), no de-anonymizing links?
[Fix queue] <proof gaps to close; assumptions to declare; simulation framing to correct>
Source: brycewang-stanford/Awesome-Journal-Skills → PODC-Skills/skills/podc-artifact-evaluation/SKILL.md