CHI's Papers track posted no separate artifact-evaluation committee or badge track for the cycles checked (2026, 2027; confirm each year — 待核实 as a standing item). That absence cuts both ways: nobody certifies your artifacts, and nobody but you ensures a skeptical reviewer can inspect them. At CHI the "artifact" is rarely just code — it is the prototype, the study instruments, the codebook, the dataset, and the video evidence of the system working. Package for two audiences with opposite needs: the anonymous reviewer with five minutes, and the future researcher with a real use for your materials.
| Artifact | Reviewer's question | Archival value after acceptance |
|---|---|---|
| Prototype / system code | "Does the claimed interaction actually exist?" | Others extend or compare against it |
| Video figure of real use | "Does it work outside the authors' hands?" | Permanent DL evidence (chi-supplementary) |
| Study instruments (guides, questionnaires, tasks) | "Was the study what the paper says?" | Direct reuse in replications |
| Codebook / analysis audit trail | "Do the themes live in the data?" | Methods teaching material |
| Dataset (de-identified) | "Do the numbers re-derive?" | Secondary analysis |
| Design files (3D prints, schematics, figma) | "Could this be rebuilt?" | Fabrication replication |
| Prompts / model configs for AI conditions | "What system did participants face?" | The only record once the API moves |
Inventory first, then decide per artifact: reviewed now, released later, or honestly
withheld with a reason (chi-reproducibility has the data-sharing ladder).
The review-phase archive rides the single September deadline with everything else. Design it so the first five minutes land:
analysis/h1_test.R →
Table 2." Three such lines are worth thirty pages of appendix.requirements.txt, renv.lock) because a reviewer will not debug pip..git, no metadata, no named accounts, anonymized
platform views only, usernames scrubbed from notebook outputs and file paths.# Cold-simulate the reviewer on the exact ZIP you will upload
rm -rf /tmp/ae && unzip -q supplement.zip -d /tmp/ae && cd /tmp/ae
cat README* | head -30 # does the claims map appear immediately?
grep -rEil 'author|university|(^|[^a-z])lab' --include='*.md' . | head
time bash run_minimal_demo.sh # the five-minute budget is literal
At the publication-ready stage (February for CHI 2027), rebuild the artifact set under real names for permanence, not for review:
A packaging note that says "the deployment used partner infrastructure we cannot ship; this archive contains the full client, the API contract, and a mock server reproducing the study conditions" earns more trust than a repo padded with dead code that hides the same gap. Reviewers at CHI read many partial artifacts; what they punish is discovering the gap themselves after the README implied completeness.
chi-experiments).chi-submission's checks.[Inventory] <artifact: reviewed / released-later / withheld+reason, per row>
[Claims map] present in README: yes/no · claims covered: <n>/<n>
[Five-minute test] cold demo ran in <time> / failed at <step>
[Anonymity] archive clean: yes/no — <channels checked>
[Release plan] home: <DL/OSF/Zenodo> · license: <code/data> · consent re-check owner: <name>