ppopp-supplementary

PPoPP Supplementary

Decide what lives in the 10 reviewed pages (text and figures, two-column acmart) and what

moves to the artifact or an appendix. The governing rule at PPoPP is decision-criticality:

anything a reviewer must read to be convinced the result is correct under concurrency and

scales stays in the body. References are unlimited and free; body space is the scarce resource.

The decision-criticality test

For each candidate piece of content, ask: *would a reviewer's accept/reject change if they could

not see this?*

[Decides acceptance]  the correctness argument's core, the headline scaling curve, the baseline
                      comparison, the key design insight  -> BODY (inside 10 pages)
[Supports/confirms]   full linearizability proof details, every core-count data point, additional
                      workloads, sensitivity sweeps, extra GPUs  -> ARTIFACT / APPENDIX
[Bulk / mechanical]   raw logs, per-run tables, generator internals, build scripts  -> ARTIFACT

The failure mode PPoPP punishes is burying a load-bearing argument — the linearization points, the

one plot that shows saturation — outside the reviewed pages, so the reviewer cannot judge the claim

and defaults to skepticism.

What must stay in the body

• The correctness argument's spine. The property (linearizability/progress), the linearization

points or the key invariant, and why it holds under the named memory model. The full formal proof

can move; the argument a reviewer needs to believe it cannot.

• The headline scaling evidence. At least one clear speedup/throughput curve over core or thread

count, with the baseline and machine named in the caption.

• The baseline comparison that substantiates the delta over prior work.
• The design insight that makes the parallelism work — the idea, not every implementation

detail.

What can move out

• Full proofs and case analyses — keep the argument in the body, push the exhaustive case split

to an appendix or the artifact.

• Complete sweeps — the body shows the representative curve; the artifact holds every thread

count, socket configuration, input size, and GPU.

• Additional workloads and sensitivity studies that confirm rather than establish the claim.
• Reproduction machinery — Dockerfiles, scripts, raw data, and per-run numbers belong in the

artifact (see ppopp-artifact-evaluation).

Where "out" is at PPoPP

• The artifact is the primary home for bulk evidence and is evaluated post-acceptance —

reviewers of the paper are not obligated to run it, so it cannot carry a decision-critical claim

at review time.

• An appendix, if the call permits one within submission rules, is read at the reviewer's

discretion — do not put anything acceptance-deciding there and assume it was read. Confirm the

current appendix policy each cycle (待核实 if unstated).

Because the artifact is post-acceptance, the split at PPoPP is stricter than at venues that review

the artifact alongside the paper: if the reviewer must see it to say yes, it is in the 10 pages.

Keeping the body legible under the page cap

• Use small multiples to compress a core sweep into one figure rather than five.
• Move exhaustive tables to the artifact and keep a summarizing plot in the body.
• Cut prose before cutting evidence; the correctness argument and the scaling curve are the last

things to shrink.

Output format

[Body (<=10pp)] correctness spine + headline curve + baseline + design insight all present?
[Moved out] proofs-detail / full sweeps / extra workloads / repro machinery -> artifact/appendix
[Decision-critical check] anything acceptance-deciding placed outside the body? (must be none)
[Appendix policy] permitted this cycle? confirmed / 待核实
[Space recovered] figures merged / tables moved to hit 10 pages