CoNEXT Reproducibility
Build the reproducible-networking story alongside the experiments, not after acceptance. CoNEXT runs
a dedicated reproducibility committee that awards optional ACM badges, and the single most
missed rule is that badge eligibility requires opting in before the paper submission deadline —
you cannot bolt it on later. Even for authors who skip badging, a reproducible artifact strengthens
a double-anonymous, one-shot-revision review.
The timing trap (read this first)
[Before the submission deadline] OPT IN for ACM badging (required for eligibility)
[At submission] anonymized, runnable artifact referenced from the paper
[Within ~1 week of acceptance] send a ONE-PAGE artifact description to the reproducibility
committee with pointers to code and other artifacts
[Camera-ready / post-accept] committee evaluates for Available / Functional / Reusable /
Reproduced badges (see conext-artifact-evaluation)
Miss the opt-in and the strongest artifact in the world cannot earn a badge this cycle.
What "reproducible" means for a networking paper
Networking reproducibility is harder than "here is the code," because the result depends on an
environment:
- Traces: the exact captures, with vantage points, dates, and anonymization documented, plus the
extraction scripts that turn raw captures into the paper's inputs.
- Configs: the exact parameters, topology, and settings used for each figure — not a
representative example.
- Environment: hardware models, firmware, kernel/OS versions, link rates, and buffer depths for
a testbed; container/VM images where feasible.
- Pipeline: a scripted path from raw data/trace to each table and figure, so the numbers
regenerate rather than being hand-copied.
Pin provenance at collection time
You cannot reconstruct provenance after the campaign ends:
- Record capture vantage points, timestamps, and anonymization method as you collect.
- Snapshot configs and firmware/OS versions at run time; a later "we think it was v2.1" is not
reproducible.
- For ML-for-networking, pin model identifiers and dates and cache raw model outputs — a package
that needs live API calls re-samples rather than reproduces.
Honest data-availability posture
- If you can release traces/code, do — with a DOI-issuing archive (Zenodo/figshare/Software
Heritage) and an open license.
- If operator agreements or privacy limits block release, say so and why, and release what you
can (aggregate data, synthetic traces, the analysis pipeline). "Available upon request" reads as a
scored weakness, not a neutral placeholder.
- Keep the availability statement honest and matched to what the artifact actually contains — a
reviewer or the committee will check.
Anonymity of the artifact (double-anonymous review)
- Re-host the artifact behind an anonymizing service before submission; scrub commit metadata,
internal hostnames, and owner-identifying paths.
- Topology diagrams and configs can leak an operator or institution — sanitize AS numbers, hostnames,
and IP ranges you own.
- The reproducibility material referenced at review time must not de-anonymize you.
The one-page artifact description
After acceptance, the committee wants a one-page description that lets an evaluator start
quickly:
- What the artifact contains (code, traces, configs, testbed scripts) and pointers to each.
- The hardware/software the evaluator needs, and any hardware you must provide access to (some
networking artifacts need specific switches/NICs — flag this early).
- A short "getting started" path and the claims the artifact supports.
Pre-submission reproducibility audit
[Opt-in] badge opt-in done BEFORE the submission deadline? (if badging) yes/no
[Traces] captures + vantage points + dates + extraction scripts present? yes/no
[Configs] exact per-figure configs and topology recorded? yes/no
[Environment] hardware/firmware/OS versions pinned; image where feasible? yes/no
[Pipeline] raw -> figure regenerates by script? yes/no
[Availability] honest statement; DOI archive or a documented reason not to release? yes/no
[Anonymity] artifact re-hosted anonymously; metadata scrubbed? yes/no
Output format
[Reproducibility status] ready / gaps
[Badge intent] opt-in before submission? yes/no/n-a
[Provenance] traces/configs/environment pinned at collection time
[Availability] what is released, where (DOI), and any documented restriction
[Anonymity] artifact anonymized for double-anonymous review
[One-pager] artifact description drafted for the committee (post-accept)
1---2name: conext-reproducibility3description: Use when building the reproducibility story for an ACM CoNEXT paper — pinned traces and configs, a runnable artifact, an honest data-availability posture, and the one-page artifact description the CoNEXT reproducibility committee needs — remembering that the ACM badge opt-in is due before the submission deadline.4metadata:5 source: https://skillmd.com/skills/brycewang-stanford/conext-reproducibility6---78# CoNEXT Reproducibility910Build the reproducible-networking story alongside the experiments, not after acceptance. CoNEXT runs11a dedicated **reproducibility committee** that awards **optional ACM badges**, and the single most12missed rule is that **badge eligibility requires opting in before the paper submission deadline** —13you cannot bolt it on later. Even for authors who skip badging, a reproducible artifact strengthens14a double-anonymous, one-shot-revision review.1516## The timing trap (read this first)1718```text19[Before the submission deadline] OPT IN for ACM badging (required for eligibility)20[At submission] anonymized, runnable artifact referenced from the paper21[Within ~1 week of acceptance] send a ONE-PAGE artifact description to the reproducibility22 committee with pointers to code and other artifacts23[Camera-ready / post-accept] committee evaluates for Available / Functional / Reusable /24 Reproduced badges (see conext-artifact-evaluation)25```2627Miss the opt-in and the strongest artifact in the world cannot earn a badge this cycle.2829## What "reproducible" means for a networking paper3031Networking reproducibility is harder than "here is the code," because the result depends on an32environment:3334- **Traces:** the exact captures, with vantage points, dates, and anonymization documented, plus the35 extraction scripts that turn raw captures into the paper's inputs.36- **Configs:** the exact parameters, topology, and settings used for each figure — not a37 representative example.38- **Environment:** hardware models, firmware, kernel/OS versions, link rates, and buffer depths for39 a testbed; container/VM images where feasible.40- **Pipeline:** a scripted path from raw data/trace to each table and figure, so the numbers41 regenerate rather than being hand-copied.4243## Pin provenance at collection time4445You cannot reconstruct provenance after the campaign ends:4647- Record capture vantage points, timestamps, and anonymization method **as you collect**.48- Snapshot configs and firmware/OS versions **at run time**; a later "we think it was v2.1" is not49 reproducible.50- For ML-for-networking, pin model identifiers and dates and **cache raw model outputs** — a package51 that needs live API calls re-samples rather than reproduces.5253## Honest data-availability posture5455- If you can release traces/code, do — with a DOI-issuing archive (Zenodo/figshare/Software56 Heritage) and an open license.57- If operator agreements or privacy limits block release, **say so and why**, and release what you58 can (aggregate data, synthetic traces, the analysis pipeline). "Available upon request" reads as a59 scored weakness, not a neutral placeholder.60- Keep the availability statement **honest and matched** to what the artifact actually contains — a61 reviewer or the committee will check.6263## Anonymity of the artifact (double-anonymous review)6465- Re-host the artifact behind an **anonymizing service** before submission; scrub commit metadata,66 internal hostnames, and owner-identifying paths.67- Topology diagrams and configs can leak an operator or institution — sanitize AS numbers, hostnames,68 and IP ranges you own.69- The reproducibility material referenced at review time must not de-anonymize you.7071## The one-page artifact description7273After acceptance, the committee wants a **one-page** description that lets an evaluator start74quickly:7576- What the artifact contains (code, traces, configs, testbed scripts) and pointers to each.77- The hardware/software the evaluator needs, and any hardware you must provide access to (some78 networking artifacts need specific switches/NICs — flag this early).79- A short "getting started" path and the claims the artifact supports.8081## Pre-submission reproducibility audit8283```text84[Opt-in] badge opt-in done BEFORE the submission deadline? (if badging) yes/no85[Traces] captures + vantage points + dates + extraction scripts present? yes/no86[Configs] exact per-figure configs and topology recorded? yes/no87[Environment] hardware/firmware/OS versions pinned; image where feasible? yes/no88[Pipeline] raw -> figure regenerates by script? yes/no89[Availability] honest statement; DOI archive or a documented reason not to release? yes/no90[Anonymity] artifact re-hosted anonymously; metadata scrubbed? yes/no91```9293## Output format9495```text96[Reproducibility status] ready / gaps97[Badge intent] opt-in before submission? yes/no/n-a98[Provenance] traces/configs/environment pinned at collection time99[Availability] what is released, where (DOI), and any documented restriction100[Anonymity] artifact anonymized for double-anonymous review101[One-pager] artifact description drafted for the committee (post-accept)102```