micro-submission

MICRO Submission

Run this audit against the exact PDF headed to HotCRP. Anchors below are the MICRO

2026 (59th) cycle — CFP and submission guidelines at microarch.org/micro59/submit/,

checked 2026-07-08. Every number is a snapshot; reopen the current cycle's pages.

Three traps that are specifically MICRO's

1. Deadlines run on US Eastern time. MICRO 2026 closed abstracts on March 31 and

papers on April 7, each at 11:59 PM EDT. Authors trained on AoE venues lose up

to a working day by assuming AoE. Convert to every co-author's timezone in writing

during deadline week.

1. The abstract deadline is a registration gate, one week early. No abstract in

HotCRP by the first date means no paper submission at all. Register placeholders

for every plausible paper; withdrawing is free, adding late is impossible.

1. No appendix. At all. The 2026 guidelines allow at most 11 pages of content

plus unlimited references — and explicitly permit no appendix. Anything you

planned to park after the references must be cut, absorbed into the 11 pages, or

moved to an anonymized artifact repository (see micro-supplementary).

Format gate table (2026 guidelines)

| Rule | 2026 setting | Enforcement |

|---|---|---|

| Content length | ≤ 11 pages, everything except references | Inspected; overlength rejected |

| References | Unlimited pages | Must name all authors — no "et al." per entry |

| Appendix | Forbidden | No parking space exists |

| Fonts | Minimum 9pt everywhere, figures included | Checked visually, not just by tool |

| Page numbers | Required on the submission | Unusual — most venues strip them |

| Template | MICRO 2026 LaTeX template strongly encouraged | Geometry edits and \vspace squeezing barred |

| Inspection | Visual and automated | Passing the HotCRP format checker is not clearance |

The last row deserves emphasis: the guidelines state submissions can be rejected for

format violations even if the HotCRP check passed. Treat the checker as a smoke

test, then eyeball a printed copy — tiny axis labels on plots are the classic kill.

Anonymity, MICRO flavor

• Double-blind review, maintained through the rebuttal/revision window.
• Artifact links (e.g., a GitHub repository) must be **fully anonymized, or

removed** if they cannot be — the 2026 guidelines say both halves explicitly.

• No acknowledgments section in the submission: no people, no funding agencies.
• Self-citations in third person; scrub PDF metadata, embedded paths, and simulator

config files inside any linked artifact that carry usernames or lab hostnames.

Mechanical pre-flight

# page count and font floor
pdfinfo paper.pdf | grep Pages
pdffonts paper.pdf | awk 'NR>2 {print $1}' | sort -u        # spot rasterized/tiny fonts manually
# page numbers present? (render page 2 footer and look)
pdftotext -f 2 -l 2 paper.pdf - | tail -3
# reference hygiene: no "et al." allowed inside entries
grep -n "et al" references.bib && echo "FIX: expand all author lists"
# identity leakage
pdftotext paper.pdf - | grep -inE "(our (prior|previous) work|anonymous|acknowledg)"
exiftool paper.pdf | grep -iE "author|creator"

HotCRP sequencing for deadline week

1. T-8 days (abstract deadline): register title, abstract, authors, topics,

conflicts. Topic checkboxes drive reviewer assignment — pick for the mechanism,

not the application domain.

1. T-3: freeze the evaluation; regenerate all figures from the run archive so

the artifact and the PDF cannot disagree later at AE time.

1. T-1: full format pass (table above) on the final PDF; print pages with the

densest figures and check 9pt compliance by eye.

1. T-0, hours early in EDT terms: upload, confirm HotCRP shows the final PDF

(re-download it and diff checksums), and have a second author log in

independently to verify the submission state.

What the abstract registration should actually contain

The T-8 abstract is not a formality — reviewers bid on it and topic-matching runs

from it. Three failure modes to avoid at the registration gate:

• Placeholder prose ("we study caches") attracts mismatched bidders whose

eventual reviews read the paper against the wrong literature.

• Overclaiming the final numbers before the sweep finishes — the PDF must

deliver the abstract's promise, and the HotCRP abstract should be updated to

match the final PDF before the paper deadline.

• Wrong topic checkboxes. Pick the mechanism area (memory hierarchy, branch

prediction, accelerators, security) over the application domain (ML, graphs) —

the mechanism community is who you want reviewing.

Conflicts deserve the same care at registration: advisors, recent co-authors, and

institutional conflicts declared per the form's definitions. A conflict discovered

after assignment is an integrity incident, not a paperwork fix.

Also decide: main track or Industry Track

MICRO 2026 opened an inaugural Industry Track with its own CFP and a review

process tailored to production constraints (micro59/submit/industrial.php). If the

paper's value is deployment evidence rather than a new general mechanism, route it

there deliberately — do not let a production paper be judged by main-track novelty

rules. Track-specific deadlines and rules: 待核实 against the live page.

If something goes wrong at the wire

• HotCRP shows "draft" at T-minus-minutes: an unfinished required field is

the usual cause — the form, not the PDF. Fill, save, re-check the state label.

• The format checker rejects a compliant PDF: rebuild with the untouched

template class file; embedded Type-3 fonts from a figure pipeline are the

common trigger. Do not "fix" it by altering template geometry.

• A co-author spots an identity leak post-deadline: do not silently update

anything; email the program chairs immediately — self-reporting a leak is

survivable, discovery is not.

• Missed the abstract gate entirely: there is no paper this cycle. Move

straight to the lattice in micro-workflow and book the next venue's window.

Output format

[Upload verdict] go / no-go (blocking items listed)
[Clock] deadline in EDT: <local equivalents per author>
[Abstract gate] registered by T-7: yes / no
[Pages] content: N/11 · references: N (unlimited) · appendix present: must be "no"
[Format] 9pt floor / page numbers / template intact / no vspace hacks: pass-fail each
[References] entries with "et al.": N (must be 0)
[Anonymity] artifact links anonymized-or-removed / metadata clean / no acks: pass-fail
[Track] main / industry — one-line justification