Skills Development Verifying Reproducibility of Theoretical Papers

Verifying Reproducibility of Theoretical Papers

v20260724
pods-reproducibility
This guide outlines the rigorous standards for academic paper reproducibility, particularly for theoretical venues. It mandates a complete, self-contained claim-to-proof map, ensuring every stated theorem or bound is fully provable within the submitted document, thereby preventing reliance on external or deferred proofs.
Get Skill
143 downloads
Overview

PODS Reproducibility

Use this before submission and again before camera-ready. For a theory venue, "reproducibility" means verifiability: a competent reader can follow every proof and confirm every stated theorem. PODS makes this concrete by requiring the proof appendix to be incorporated at submission — there are no online/external appendices — so the reviewers can check the mathematics now, not on trust.

Claim-to-proof map

  • Map each theorem, lemma, corollary, and reported bound to a complete proof location — the body or the appendix. No stated result may lack a full proof somewhere in the submitted PDF.
  • For each proof, state its assumptions, cite the exact prior lemma or theorem it uses, and do not defer a key step to "the full version" or "a routine argument" when the argument is not routine.
  • For constructions and algorithms, give enough that a reader could re-derive them: the invariant, the data structure, the parameter settings, and the complexity accounting.
  • Keep the assumptions ledger honest: every conditional result (ETH, OMv, #P-hardness) is labeled where it is used, and the model (data vs. combined complexity, cost model) is fixed.
  • Keep the paper and its proofs consistent: an abstract that claims more than the theorems prove, or a corollary that does not follow from the stated theorem, is the contradiction reviewers read as carelessness.

Verifiability audit

Claim in the paper Weak (reject-prone) form PODS-ready form
"Theorem: the problem is hard" Hardness "sketched"; no reduction given A complete reduction in the appendix, with the source problem cited
"Our algorithm is optimal" Upper bound only Upper bound + matching lower bound, both proved
"This holds for all such queries" Proof for the examples shown A general proof covering every case, or a scoped, honest claim
"It is easy to see that..." The hard step waved away The argument written out, or a precise appendix pointer
"Full proofs in the full version" Nothing in the appendix Complete proofs in the at-submission appendix; arXiv full version optional-but-consistent

"Proof omitted" with no appendix proof is treated at PODS the way "available on request" is treated at a systems venue — as not provided. The appendix exists so nothing decision-critical is off-paper.

Assumptions and scope pinning

[Model]        fix data model, query class, and complexity/cost measure; use them consistently
[Assumptions]  label every conditional bound with its conjecture at the point of use
[Scope]        state exactly which cases the result covers and which remain open — do not overclaim
[Dependencies] each proof cites the exact prior result it invokes; no circular lemma chains
[Constants]    expose hidden dependence on query size / schema / fixed parameters

Degrees of verifiability (state the one you achieved)

  • Complete: every theorem has a full, self-contained proof in the submitted PDF (body + appendix). This is the PODS default and expectation.
  • Scoped: the main theorems are fully proved; a clearly labeled extension or a corollary is stated with a proof idea and deferred to the journal full version — acceptable only if nothing decision-critical is deferred.
  • Conjectural: a statement is offered as a conjecture with partial evidence — allowed as such, never dressed as a theorem.

The full-version-on-arXiv norm

PODS is an extended-abstract venue: the community norm is to post a full version on arXiv (with all proofs and any extended development) and, at camera-ready, DOI-link it to the PACMMOD paper. At submission the arXiv version must not break double-anonymity; keep it anonymized or withhold the link until acceptance. The at-submission appendix, not the arXiv link, is what the reviewers must be able to verify.

Vignette: a dichotomy paper

Consider a paper proving a PTIME/hard dichotomy for a query class. Its verifiability spine: the exact model and query class in preliminaries; the tractable direction as a proved algorithm with complexity; the hard direction as a complete reduction from a cited problem, under a named assumption if conditional; a lemma establishing that the two directions partition the class (completeness); and a scope sentence naming the one extension (e.g. self-joins) left open — every proof in the body or the at-submission appendix, with a consistent arXiv full version prepared for later.

Consistency and camera-ready pass

  • Before submission: every stated result traces to a complete proof in the PDF; assumptions are labeled; the appendix is incorporated and anonymized.
  • Before camera-ready: post the (de-anonymized) full version on arXiv and add the DOI link; ensure the arXiv version's theorems and the PACMMOD version's theorems agree exactly.

Output format

[Claim inventory] <theorem/lemma -> complete proof location (body/appendix)>
[Verifiability] complete / scoped / conjectural, stated honestly
[Assumption gaps] <any conditional bound unlabeled? any circular dependency?>
[Scope honesty] <open cases stated? overclaim in the abstract? yes/no>
[Paper fixes] <proofs to complete or pointers to add before submission>
[Full-version plan] <arXiv version prepared, anonymized now, DOI-linked at camera-ready>
Info
Category Development
Name pods-reproducibility
Version v20260724
Size 5.65KB
Updated At 2026-07-29
Language