Skills Data Science Guidelines for Artifact Reproducibility and Release

Guidelines for Artifact Reproducibility and Release

v20260724
imc-artifact-evaluation
A comprehensive guide for researchers on preparing, archiving, and declaring the availability of research artifacts (datasets, code, and platforms) for academic publication. It details best practices for provenance tracking, ensuring reproducibility, and meeting the standards required for community contribution awards.
Get Skill
386 downloads
Overview

IMC Artifact Evaluation

Use this for IMC's artifact and dataset story. Unlike venues with a separate badge-granting committee, IMC's mechanism is centered on an availability declaration at submission plus shepherding after acceptance to ensure the promised data, code, or platform actually becomes available. The venue-defining incentive is the Community Contribution Award, which exists to honor a released dataset, tool, or open platform. Whether IMC also awards formal ACM reproducibility badges in a given edition is 待核实 — confirm on the current call.

Two artifacts, two audiences

  • Review artifact (at submission): anonymized for the reviewers — no owner strings, testbed or AS identifiers, probe-account IDs, cluster paths, or lab-domain links. It backs the paper's claims during double-blind review.
  • Public release (after acceptance): de-anonymized, licensed, permanently archived. This is what the shepherd checks, what the camera-ready cites, and what the Community Contribution Award evaluates.

The availability declaration, delivered

At submission you declared full / partial / none. After acceptance, the shepherd holds you to it:

Declared What the shepherd expects Common failure caught
Full The dataset/tool/platform, publicly retrievable and documented Link promised, never published; broken archive
Partial The shareable subset + a stated reason for the rest "Partial" used to avoid work; unclear what is shared
None A specific, legitimate justification (proprietary/privacy/legal) Vague "on request"; no reason given

"Available on request" is not availability. If law or privacy blocks release, say precisely why and, where possible, release a derived/aggregated safe version.

What a strong measurement dataset release contains

[Data]        the measured dataset itself (or a documented, privacy-safe derivative)
[Schema]      a documented schema/dictionary: every field, unit, and its meaning
[Provenance]  vantage points (locations, ASes, probe types), measurement dates and durations,
              target lists with capture dates, tool versions, sampling and rate limits
[Method]      the collection scripts/tooling and how to re-run the *method* (data will differ)
[Ethics]      the privacy/anonymization applied to the release; disclosure status
[License]     an explicit open license (e.g., CC-BY for data, OSI license for code)
[Archive]     a DOI-issuing repository (Zenodo, figshare, Software Heritage) or a durable
              community archive — not a personal homepage

Reproducibility for a moving Internet

Measurement data cannot be re-collected identically — the Internet changes. So the release must make the captured data and its provenance the reproducible core, and the method re-runnable to produce new comparable data. Ship the analysis scripts that turn the released data into the paper's figures; a number in the PDF that no script regenerates from the released data is the contradiction a shepherd (and a reader) will flag.

Community Contribution Award eligibility

The award recognizes an outstanding dataset, source-code distribution, open platform, or service to the community. To be eligible, the data/code/tool must be publicly available and usable by the time of the camera-ready deadline — not "coming soon." Aim for it by making the release:

  • Usable by a stranger: documented schema, a README, and a runnable example.
  • Durable: a DOI and a maintenance/versioning note.
  • Reusable: an open license and, for platforms, an access path others can actually use.

Worked vignette: releasing a longitudinal scan dataset

A paper contributes a two-year scan of a protocol's deployment. For a strong, award-eligible release: publish the per-scan records with a documented schema; include vantage-point and timing metadata for every scan; apply and document IP/host anonymization consistent with the Ethics section; ship the analysis notebooks that regenerate each figure from the released data; deposit in a DOI-issuing archive with a CC-BY license; and state honestly which raw captures cannot be shared for privacy reasons and what safe derivative replaces them.

Calibration

  • Availability is judged at camera-ready time for the award and during shepherding for the accepted paper — plan the release before, not after, acceptance.
  • Whether formal ACM badges are offered, and the exact shepherding process, vary by edition — confirm on the current call (待核实).

Output format

[Artifact role] anonymized review artifact / public release
[Declaration] full / partial / none (+ justification)
[Contents] <data/schema/provenance/method/ethics/license/archive>
[Reproducibility] scripts regenerate figures from released data? yes/no
[Award eligibility] public + usable by camera-ready? yes/no
[Fixes before release] <ordered list>
Info
Category Data Science
Name imc-artifact-evaluation
Version v20260724
Size 5.27KB
Updated At 2026-07-28
Language