Isca Workflow

ISCA Workflow

An ISCA year has one gate. Miss mid-November and the next identical opportunity is twelve months away, which is why experienced groups treat the ISCA calendar as an engineering schedule with the deadline as tape-out. This skill lays out that schedule using the verified ISCA 2026 dates as the template; as of 2026-07-08 the ISCA 2027 CFP was not yet posted, so every 2027 date below is an expectation to confirm against iscaconf.org/isca2027/ the moment it exists (待核实).

The verified 2026-cycle skeleton

Industry-track submissions ran about a month behind (abstracts December 5, papers December 12, 2025). Final notification date for 2026 was not captured in this pack's verification — 待核实.

Working backward from November

Count in weeks from the expected paper deadline (call it P, likely mid-November 2026 for ISCA 2027):

• P minus 16 weeks (late July): venue decision final (isca-topic-selection); simulation or prototype infrastructure chosen and validated (isca-experiments); the one-sentence claim written on the whiteboard.
• P minus 10 weeks: baseline configurations frozen and reproduced from scratch by a second person. Baseline drift discovered later invalidates every downstream number.
• P minus 6 weeks: main-result experiments complete. From here, compute goes to sensitivity studies and reviewer-anticipation runs, not to the headline.
• P minus 4 weeks: full draft exists — bad prose, complete structure. Internal "mock PC" read by two colleagues who did not build the system.
• P minus 1 week (abstract deadline): title/abstract/topics/conflicts final in HotCRP. The abstract steers reviewer assignment; write it for bidding, not as a placeholder.
• P: upload per isca-submission, then freeze the artifact snapshot that matches the submitted numbers (isca-reproducibility).

The winter: silence, then three loud weeks

Between November and mid-February the authors' only jobs are (a) keep the evaluation environment resurrectable and (b) pre-build rebuttal ammunition: the five most likely objections, each with a ready experiment or a pointer into the submitted PDF. The 2026 cycle then opened a combined rebuttal/revision period (February 16 - March 6) — unusually long by conference standards precisely because ISCA lets authors change the paper, not just argue about it. Staff those three weeks like a sprint: compute reserved, coauthors on call, no conference travel booked. Details of what to do inside the window live in isca-author-response.

Branch plans after notification

• Accept: camera-ready under the edition's publisher workflow (isca-camera-ready), voluntary artifact submission (isca-artifact-evaluation), talk and travel for late June.
• Reject with salvageable reviews: map every misunderstanding to the sentence that permitted it; decide MICRO (spring gate) vs HPCA (summer gate) within two weeks while the reviews are fresh.
• Reject with a fit objection: rerun isca-topic-selection honestly — the committee may have told you the paper's true venue.

Team ledger

Keep a single tracking file in the paper repo; every risk has one owner.

# isca-campaign.yaml
target: ISCA 2027            # 54th by arithmetic - ordinal 待核实
gate:
  abstract: TBD-2026-11      # confirm on iscaconf.org/isca2027 when live
  paper:    TBD-2026-11
  source:   "not yet posted as of 2026-07-08"
owners:
  claim_and_story:   <name>   # one sentence, defended weekly
  simulation_infra:  <name>   # versions pinned, configs archived
  baselines:         <name>   # frozen P-10w, independently reproduced
  writing:           <name>   # draft complete P-4w
  submission_admin:  <name>   # HotCRP, conflicts, template, metadata
  rebuttal_reserve:  <name>   # compute + calendar hold, mid-Feb to early Mar
checks:
  - abstract written for reviewer bidding, not filler
  - artifact snapshot tagged at submission commit
  - mock-PC read completed and objections filed
  - fallback venue named with trigger condition

Failure modes this calendar prevents

• The November collision: abstract and paper a week apart means groups that planned "we'll write after the abstract" write the whole paper in seven days. The P-4w draft rule exists for this.
• The dead environment: rebuttal-window experiments that fail because the simulator tree no longer builds in February. Tag and containerize at submission.
• The unstaffed revision: ISCA's revision option is leverage only for teams that can produce new results in days; treat the window as scheduled work.
• The stale retarget: rejected papers resubmitted unchanged to MICRO carry their ISCA reviews' fingerprints; committees overlap. Budget a real revision.

Milestone-to-skill routing

All dates above are the 2026-cycle record (../../resources/official-source-map.md, checked 2026-07-08). Confirm the 2027 calendar, page rules, and review mechanics on the official site before acting on any of them.

Source: brycewang-stanford/Awesome-Journal-Skills → ISCA-Skills/skills/isca-workflow/SKILL.md