cscw-artifact-evaluation
GitHub指导CSCW论文 artifacts 的打包与发布，强调社区数据伦理。提供不同 artifact 类型的审查与公开策略，执行社区重识别测试以保护隐私，并定义最小可发布包结构及发布备忘录要求。
Trigger Scenarios
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill cscw-artifact-evaluation -g -y
SKILL.md
Frontmatter
{
"name": "cscw-artifact-evaluation",
"description": "Use when packaging what stands behind a CSCW paper — systems, analysis pipelines, codebooks, instruments, datasets from real communities — for review-time scrutiny and post-acceptance release, where community-data ethics constrain release more than any badge checklist."
}
CSCW Artifact Packaging
CSCW has no artifact-evaluation committee stamping badges; the "evaluators" of your artifacts are the reviewers deciding whether to trust your methods, and later the researchers and the studied communities themselves who encounter the released materials. That second audience is the venue's distinctive constraint: a CSCW artifact release is an act with consequences for real people who never signed your consent form.
Artifact classes and their release logic
| Artifact | Review-time form | Post-acceptance form | Governing constraint |
|---|---|---|---|
| Study instruments (guides, surveys) | Anonymized in supplement | Public archive | Institution redaction only |
| Codebooks | Anonymized in supplement | Public archive | Paraphrased exemplars only |
| Analysis code | Anonymized archive | Public repository + DOI archive | Strip identity from history |
| Built systems / prototypes | Screenshots/video + code where feasible | Open source if maintainable | Third-party assets, API keys |
| Trace datasets | Aggregates or synthetic sample | Aggregates; raw only if terms + ethics allow | Platform ToS + re-identification risk |
| Interview/field data | Never | Almost never | Consent scope is absolute |
The community re-identification test
Before releasing any dataset or exemplar text derived from an online community, run the adversary exercise — assume a motivated actor (a journalist, a harasser, a platform admin, a community member with a grudge):
- Quote search: can any released text snippet be pasted into a search engine and land on the original post? If yes, paraphrase or drop it.
- Join attack: can released fields (timestamps + community + activity counts) be joined against public APIs or archives to recover usernames? Coarsen until the join fails.
- Small-population exposure: would identifying the community (not any member) expose it to raids, deplatforming, or press attention it has not chosen? If the community is small or marginalized, community-level anonymity is part of the ethical contract — name it only with its agreement.
- Future-context check: platforms change hands and moderation regimes change; would this release be safe under a hostile future owner of the platform's data?
Document the outcome in a short release memo; it becomes the honest core of the
paper's data-availability statement (cscw-reproducibility).
Minimum viable released package
release/
├── README.md # what claims each artifact supports; setup in ≤ 5 steps
├── LICENSE # code license + data terms, which may differ
├── instruments/ # guides, surveys, recruitment text (redacted)
├── codebook/ # definitions + paraphrased exemplars
├── pipeline/ # collection + analysis code, config, environment spec
├── data/
│ ├── aggregates/ # tables behind each figure
│ └── synthetic/ # optional: structure-preserving fake sample
└── ETHICS.md # consent scope, what is withheld and WHY, contact path
ETHICS.md is the CSCW-specific file: it tells reusers what they may not do
(re-identify, recontact, join against other datasets) and why the withheld parts
are withheld. A release that explains its own limits earns more trust than a
maximal dump.
Review-time discipline
- Everything reviewers can open must be anonymous — including git metadata, archive paths, and hosting URLs (use anonymized repository services, not a lab server whose domain resolves the institution).
- Never put decision-critical evidence only in the artifact; the paper must stand if a reviewer opens nothing.
- Keep the review package and the eventual public release as separate builds from
one source tree; the deanonymization diff at acceptance should be mechanical
(
cscw-camera-ready).
Release audit
[Claims map] each artifact → the claim it supports (no orphan uploads)
[Adversary test] quote-search / join / community-exposure / future-context: pass?
[Consent gate] every released item inside consent + ToS scope? y/n
[Two builds] review build anonymous; release build attributed; same tree? y/n
[ETHICS.md] withholdings explained, reuse limits stated? y/n
No CSCW badge program existed at the 2026-07-08 check; if one appears in a future cycle, its checklist supplements — never replaces — the community-protection tests above.
Version History
- 9f86f09 Current 2026-07-19 14:41
