ASE Supplementary
Split content by decision-criticality. ASE gives you 10 pages of body (plus 2 for
references), and reviewers judge the paper on those pages. The artifact and any supplementary
material support reproduction and reuse — they do not carry the argument. The failure mode is
exiling a decision-critical result to the artifact and assuming reviewers will dig for it; they
judge what is in the body.
The decision-criticality rule
- In the body (must be within 10 pages): anything a reviewer needs to accept the paper — the
automated task, the technique, the headline results on real subjects, the closest baseline
comparison, the key ablation, and the central threats argument.
- In the artifact / supplement: everything needed to reproduce or reuse — full configurations,
complete result tables, extra subjects, proofs, additional plots, raw logs, and cached model
outputs.
If a result decides acceptance, it lives in the body even if space is tight; cut elsewhere.
What each container is for
| Content |
Where |
Why |
| Task, technique, headline evaluation, closest-baseline comparison, key ablation |
Body (≤10 pp) |
Decision-critical; reviewers score these |
| Central threats-to-validity argument |
Body |
Construct/oracle validity is judged, not reproduced |
| Data Availability Statement |
Body, after Conclusions, inside 10 pp |
Mandatory and reviewed |
| Full config files, seeds, environment |
Artifact |
Reproduction detail, not an argument |
| Complete/extra result tables, extra subjects |
Artifact / supplement |
Support, not headline |
| Proofs, derivations too long for the body |
Artifact / supplement (if the venue permits) |
Verification detail |
| Raw logs, cached LLM outputs, dataset snapshots |
Artifact |
Provenance and reuse |
The reviewer-behavior reality
- Reviewers may open the artifact to check a claim, but they will not reconstruct a headline
result from raw logs. State every acceptance-deciding number in the body with a pointer to where
the artifact backs it.
- A pointer to the artifact is a promise the body's claim is verifiable — not a substitute for
stating the claim.
- Keep the body self-contained: a reader who never opens the artifact should still be able to
judge the contribution.
Anonymity across containers
- The supplement and artifact are part of the double-anonymous submission: scrub tool names that
encode your identity, repository owners, and file paths.
- A result table in the artifact that reveals your institution's cluster paths is as much a leak as
a name on the PDF.
Space-recovery moves (before cutting content)
[Figures] merge redundant plots; one figure that carries the argument beats three that decorate.
[Tables] move exhaustive per-subject tables to the artifact; keep the summary with effect sizes.
[Prose] delete literature-tour sentences; related work is a delta argument, not a survey.
[Config] move full configurations to the artifact; keep the one setting that matters in the body.
[Roadmap] cut over-signposting; the arc should be visible without a paragraph announcing it.
Output format
[Decision-critical in body?] task / technique / headline results / baseline / key ablation / threats — all present within 10pp?
[Data Availability] in body, after Conclusions, inside 10pp?
[Artifact-only] configs / extra tables / proofs / logs / cached outputs -> not decision-critical?
[Self-contained] judgeable without opening the artifact? yes/no
[Anonymity] supplement + artifact scrubbed of identity? yes/no
1---2name: ase-supplementary3description: Use when deciding what belongs in the 10-page ASE (IEEE/ACM Automated Software Engineering) paper body versus the artifact and supplementary material, splitting content by decision-criticality so nothing that decides acceptance lives outside the reviewed pages under the 10+2 budget.4---56# ASE Supplementary78Split content by **decision-criticality**. ASE gives you **10 pages** of body (plus 2 for9references), and reviewers judge the paper on those pages. The artifact and any supplementary10material *support* reproduction and reuse — they do not carry the argument. The failure mode is11exiling a decision-critical result to the artifact and assuming reviewers will dig for it; they12judge what is in the body.1314## The decision-criticality rule1516- **In the body (must be within 10 pages):** anything a reviewer needs to accept the paper — the17 automated task, the technique, the headline results on real subjects, the closest baseline18 comparison, the key ablation, and the central threats argument.19- **In the artifact / supplement:** everything needed to *reproduce or reuse* — full configurations,20 complete result tables, extra subjects, proofs, additional plots, raw logs, and cached model21 outputs.2223If a result decides acceptance, it lives in the body even if space is tight; cut elsewhere.2425## What each container is for2627| Content | Where | Why |28|---|---|---|29| Task, technique, headline evaluation, closest-baseline comparison, key ablation | Body (≤10 pp) | Decision-critical; reviewers score these |30| Central threats-to-validity argument | Body | Construct/oracle validity is judged, not reproduced |31| Data Availability Statement | Body, after Conclusions, inside 10 pp | Mandatory and reviewed |32| Full config files, seeds, environment | Artifact | Reproduction detail, not an argument |33| Complete/extra result tables, extra subjects | Artifact / supplement | Support, not headline |34| Proofs, derivations too long for the body | Artifact / supplement (if the venue permits) | Verification detail |35| Raw logs, cached LLM outputs, dataset snapshots | Artifact | Provenance and reuse |3637## The reviewer-behavior reality3839- Reviewers may open the artifact to check a claim, but they will **not reconstruct a headline40 result** from raw logs. State every acceptance-deciding number in the body with a pointer to where41 the artifact backs it.42- A pointer to the artifact is a promise the body's claim is verifiable — not a substitute for43 stating the claim.44- Keep the body **self-contained**: a reader who never opens the artifact should still be able to45 judge the contribution.4647## Anonymity across containers4849- The supplement and artifact are part of the double-anonymous submission: scrub tool names that50 encode your identity, repository owners, and file paths.51- A result table in the artifact that reveals your institution's cluster paths is as much a leak as52 a name on the PDF.5354## Space-recovery moves (before cutting content)5556```text57[Figures] merge redundant plots; one figure that carries the argument beats three that decorate.58[Tables] move exhaustive per-subject tables to the artifact; keep the summary with effect sizes.59[Prose] delete literature-tour sentences; related work is a delta argument, not a survey.60[Config] move full configurations to the artifact; keep the one setting that matters in the body.61[Roadmap] cut over-signposting; the arc should be visible without a paragraph announcing it.62```6364## Output format6566```text67[Decision-critical in body?] task / technique / headline results / baseline / key ablation / threats — all present within 10pp?68[Data Availability] in body, after Conclusions, inside 10pp?69[Artifact-only] configs / extra tables / proofs / logs / cached outputs -> not decision-critical?70[Self-contained] judgeable without opening the artifact? yes/no71[Anonymity] supplement + artifact scrubbed of identity? yes/no72```
brycewang-stanford