thedixitjain ICDT Artifact Evaluation
Read this to set expectations correctly: ICDT is a pure database-theory venue with no code-artifact track and no ACM-style artifact badges. The role an artifact plays at a systems or empirical venue — letting an evaluator re-run your result — is played at ICDT by the complete, checkable proof: the marked appendix that a referee reads, and the archived full version that carries every proof for readers after publication. This skill is about making that proof-artifact airtight, not about Docker images.
If you arrived expecting the ACM "Available / Functional / Reusable / Reproduced" badges, note that those belong to venues like the co-located EDBT (systems) or SIGMOD/PODS's ACM track. ICDT publishes in LIPIcs and evaluates proofs. Whether a given EDBT/ICDT edition offers any optional artifact/reproducibility recognition for ICDT papers is 待核实 — check the current call; the default is none.
The ICDT "artifact" is the proof
| Systems-venue artifact | ICDT analogue |
|---|---|
| Runnable code + README | Complete proofs in the marked appendix, self-contained in the PDF |
| Reproducibility of numbers | Verifiability of theorems by a competent referee |
| DOI-issued code archive | The LIPIcs paper (DOI on DROPS) + the arXiv full version |
| "Works on a clean machine" | "Checks out under a careful reading" |
Design the proof-artifact so a referee can certify correctness in a bounded reading budget, exactly as a systems evaluator wants a package that runs in the first ten minutes.
The marked appendix (the reviewed artifact)
- Contains the full proofs for every theorem stated in the body, read at the PC's discretion.
- Is inside the single submission PDF — ICDT does not allow online/external appendices, so there
is no separate artifact upload for review (see
icdt-supplementary). - Is signposted per theorem so a referee checking one result finds its proof immediately.
- Is navigable, not a dump — a 40-page unstructured appendix is as unpersuasive as no proof.
The full version (the archived artifact)
After acceptance, the enduring "artifact" is the full version, typically on arXiv:
[Contents] every theorem with its complete proof; all lemmas, all cases; the constructions in full
[Consistency] identical theorem statements and bounds to the LIPIcs paper — no silent divergence
[Link] referenced from the camera-ready via \relatedversion (icdt-camera-ready)
[Openness] arXiv is open access, matching LIPIcs's CC-BY spirit; both are permanent, citable
[Timing] post/refresh it around camera-ready, after the revision's fixes are folded in
The full version is what the community actually reads and builds on; a conference paper whose full proofs never appear anywhere weakens the result's standing even after acceptance.
The rare algorithmic paper with code
If your ICDT paper contributes an algorithm and you ran an experiment (icdt-experiments), you may
choose to share code — but this is optional and not badged:
- Deposit it in a DOI-issuing archive (Zenodo / Software Heritage) for permanence, link it from the full version, and license it openly.
- Keep it anonymized for review if you reference it at all; the reviewed object is still the PDF.
- Do not present the code as the contribution — if it were, the paper likely belongs at EDBT/SIGMOD.
Output format
[Artifact type] proof-artifact (default) / optional code (algorithmic paper)
[Marked appendix] full proofs present, in-PDF, signposted per theorem? yes/no
[Full version] complete-proofs arXiv version, consistent with the paper, linked? yes/no
[Badges] none at ICDT (confirm current call); EDBT/systems path if code is the point
[Fix queue] <complete missing proofs / signpost appendix / post full version>
Source: brycewang-stanford/Awesome-Journal-Skills → ICDT-Skills/skills/icdt-artifact-evaluation/SKILL.md