agentsclimarketplace

Fast supplementary

Skill brycewang-stanford/Awesome-Journal-Skills/FAST-Skills/skills/fast-supplementary

Use when deciding what belongs in a USENIX FAST paper body versus its artifact and appendices, covering the USENIX two-column page budget (references excluded), the rule that decision-critical storage evidence stays inside the reviewed pages, double-blind supplementary material, and how to split a storage paper between body and package.From its SKILL.md

Install
npx -y skills add brycewang-stanford/Awesome-Journal-Skills --skill fast-supplementary

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

SKILL.md

4.6 KB, 955 tokens by cl100k_base, as published. Nobody here has run it

FAST Supplementary

Use this when assembling FAST supplementary material. The governing rule is simple and strict: the paper must be judgeable from the reviewed pages alone. The artifact, traces, and any appendix support the paper; they do not hold the argument. Reviewers open the artifact at their discretion, so anything the decision depends on lives in the body.

What goes where

ContentBody (within page budget)Artifact / appendix
The storage claim and its headline resultYes
Core design or study protocolYesFull parameter sweeps, extra configs
The testbed: device models, firmware, stateYes (summary table)The complete machine/firmware inventory
Headline storage metrics + the invariant testYes (WA/tail latency/crash test)Full distributions, per-run logs
Crash-consistency evidence for a durability claimThe result + methodThe fault-injection / replay harness
Workloads and tracesNamed + why chosenThe trace archive and replay scripts
Raw device counters, logs, scriptsYes
Reproduction instructionsA pointerThe README and run scripts

If a reviewer would need to open the artifact to know whether to accept, the paper is mis-partitioned — move that evidence into the body.

The page-budget discipline

FAST's USENIX two-column budget (FAST '27: ≤12 pages long / ≤6 short, references excluded) is firm, and it counts figures and tables. References are uncapped, so a rich bibliography costs no body space, but everything else competes. Consequences:

  • Keep the headline storage metric, the device table, and the durability/consistency check in the body — they are decision-critical.
  • Move long parameter sweeps, per-device breakdowns, and secondary microbenchmarks to the artifact with explicit forward references.
  • Do not use the artifact to smuggle in a result that would not fit — an argument that only closes with material outside the reviewed pages reads as unreviewable.

Double-blind supplementary rules

[No identity]   no authors, affiliations, acknowledgements, funders, hostnames, cluster paths, or
                repo owners anywhere in the archive or appendix
[Anonymized links] the availability link points at an anonymizing host, not a personal/institutional repo
[Scrub device dumps] SMART/firmware/config dumps can carry serials, hostnames, or datacenter tags — redact them
[Clean archive]  no .git history, .DS_Store, credentials, caches, or huge irrelevant trace dumps
[Opens clean]    verify the archive unzips and the README orients a reader in one minute on a fresh machine

Appendix and artifact-appendix architecture

  • Order supplementary sections to mirror the paper's claim/RQ order; reviewers navigate by claim, not page number.
  • Keep each supplementary section referenced from the body at least once — orphaned material is invisible under discretionary review.
  • Distinguish the review-time artifact (anonymized, for the paper's reviewers) from the artifact appendix added post-acceptance for the AEC and badges (fast-artifact-evaluation); do not conflate them.
  • Confirm on the current call whether any in-paper appendix counts against the page budget; this varies and changes the split.

Vignette: splitting a file-system design paper

A paper presenting a new file-system layout plus an evaluation: the body keeps the design, the headline results (write amplification and tail latency on named devices at steady state), the crash-consistency test summary, and the caveats; the artifact holds the full parameter sweeps, the complete device/firmware inventory, the aging-workload scripts, the crash-injection harness, the archived traces with replay scripts, and the analysis notebooks. Nothing decision-critical lives only in the artifact, because artifact inspection is discretionary and the reviewers judge from the pages.

Output format

[Supplement status] ready / needs fixes / not ready
[Partition check] anything decision-critical outside the body? <none / move: what>
[Page budget] body within the 12/6-page limit? references excluded correctly? yes/no
[Anonymity] archive + device dumps clean of identity/metadata? passed/issues
[Body dependency] <what a reviewer can decide without opening the artifact>

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.