brycewang-stanford PLDI Supplementary Material
PLDI's cap is on text pages excluding bibliography in the main submission (20 for the 2026 cycle); everything else — complete proofs, full benchmark tables, anonymized implementations — is supporting material whose exact submission mechanics (appendix inside the PDF vs separate upload, size limits, reviewer obligations) are reset per cycle and were not fully visible for 2026 (待核实). Decide placement by role first, then check the live CFP for the mechanics.
Placement by role
| Material | Where it belongs | Why |
|---|---|---|
| Proof sketches of main theorems | Main text | Reviewers judge soundness from the body |
| Full proofs, case analyses | Appendix/supplement | Depth on demand; body states theorem + sketch |
| Headline benchmark results | Main text | The claim lives or dies here |
| Per-benchmark raw tables, extra configurations | Supplement | Evidence of thoroughness, not the argument |
| Anonymized source code | Anonymous link or upload | Early credibility; becomes the AE artifact later |
| Semantics figures too large for the text block | Rethink first | If reviewers need it to follow §4, it is not supplementary |
The governing rule: the 20 pages must carry the whole argument. PLDI reviewers are typically not obliged to read beyond the main paper, so a proof step or baseline definition that exists only in the supplement effectively does not exist for a skeptical reviewer.
Anonymity extends to every byte
The double-blind requirement covers supplements with sharper teeth than the PDF, because archives carry metadata the paper does not:
# Sweep an anonymized code drop before upload
rm -rf pkg/.git pkg/.github
grep -rniE 'copyright|@[a-z0-9.-]+\.(edu|org|com)|university|author' pkg/ | head
find pkg/ \( -name '*.DS_Store' -o -name '*.orig' -o -path '*__pycache__*' \) -delete
tar tzf pkg.tgz | grep -Ei '/home/|/Users/' && echo "LEAK: user paths in archive"
PL-specific leaks: license headers naming the lab, a build script pulling from a personal GitHub, benchmark result CSVs with hostname columns, and proof files whose module comments name the authors. Host anonymized code on an anonymizing service, never on a "private-but-guessable" lab URL.
Consistency contract with the main PDF
- Numbers quoted in the body must be derivable from supplement tables — same runs, same aggregation. A geomean in the paper that cannot be recomputed from the supplementary CSVs reads as fabrication risk, not sloppiness.
- Theorem numbering must match between body and full-proof appendix; renumber once, late, mechanically.
- If the supplement includes a README, it obeys the same anonymity and no-forbidden-links rules as the paper.
Timing
Build the supplement with the paper, not the night after the deadline. The
supplement you assemble in a rush is the one that ships a .git directory; and
since the same package matures into the post-acceptance Zenodo artifact
(pldi-artifact-evaluation), early hygiene is compounding interest.
Output format
[Self-containment] can reviewers judge the claim from 20 pages alone? yes/no
[Placement map] item -> body / appendix / code drop
[Anonymity sweep] git dirs / metadata / paths / license headers / hosting
[Consistency] body numbers recomputable from supplement? theorem numbering aligned?
[Mechanics 待核实] appendix-vs-upload rules and size caps from the live CFP