ATC Review Process
Model the pipeline before interpreting any single review. ATC's process is unusual among systems
venues in 2026 because it runs in two rounds gated by a two-page extended abstract, and because
acceptance is conditional on a shepherd. The most consequential mental shift for authors
arriving from a single-round conference is that round one can end your submission before anyone
reads the full paper — the extended abstract is not a formality.
Process model
- Submission and review run on HotCRP (
atc26.hotcrp.com) with double-blind anonymity:
identities are mutually hidden through reviewing.
- Round one: each two-page extended abstract is read by two experienced reviewers with a
broad systems perspective; abstracts below the bar are rejected early. The extended abstract is
review-only and not published.
- Round two: papers that advance are read in full by 3-4 reviewers, who weigh the
significance and usefulness of the systems contribution, the soundness of the design, the honesty
and completeness of the measurements, reproducibility, and clarity.
- Rebuttal / description of changes: authors get a short (about one page) window to correct
factual misreadings and, where relevant, summarize revisions relative to a prior submission.
- Decision + shepherding: selected papers are conditionally accepted; final acceptance
requires satisfying a shepherd (a PC member) on required revisions. Accepted papers publish in
the ACM DL under Open Access.
Reading a decision against the model
| Outcome |
What it means |
Author move |
| Early reject (round one) |
The abstract did not convey a problem, contribution, and result that clear the bar |
Reframe the abstract; the paper may be fine but the pitch failed — see atc-writing-style |
| Full review, then reject |
A design or evidence gap survived the full read |
Fix the substance, not the framing; consider a sibling venue |
| Conditional accept |
Contribution holds; specific revisions required |
Treat the shepherd's list as binding; make or explicitly justify each item |
| Reject with encouragement |
Structural but repairable over a cycle |
Rebuild the evidence and resubmit next cycle or reroute |
The strategic reading: because round one judges the abstract, the highest-leverage editing on the
whole paper is often the two pages a round-one reviewer sees. Write the submission so its problem
and result are legible before the details.
How ATC differs from its siblings
- vs. OSDI / SOSP: those flagships are more selective and lean toward larger, novel systems; ATC
takes a broader, more practical slice and welcomes solid, useful, well-measured work a
flagship might call incremental. A paper bounced from OSDI on "not novel enough" can be a strong
ATC paper if the engineering and measurement are real. Neither shares ATC's two-round
extended-abstract gate.
- vs. NSDI: NSDI centers networked systems and runs its own multi-deadline model; ATC's
scope is all of systems and its 2026 calendar is a single June deadline. Route a networked-systems
design there if the network is the contribution.
- vs. EuroSys: EuroSys is the other broad, SIGOPS-affiliated systems conference; the choice is
often calendar and community pull. Under SIGOPS, ATC and EuroSys now sit in the same family, so
check both deadlines.
- vs. pre-2026 USENIX ATC: the old ATC ran a single-round review on a winter deadline with
USENIX-funded open access and no APC. The 2026 SIGOPS edition changed the round structure, the
calendar, and the cost model at once — do not carry the old process forward.
Who reads you
Round one: two broad-perspective systems reviewers judging whether the idea and result are worth a
full read. Round two: 3-4 reviewers closer to your subarea who open the artifact, check whether the
measurements are end-to-end and honest about cost, and probe whether the baselines are fair. Vague
design descriptions and average-only results get caught, not skimmed.
Where author leverage actually exists
[Before submission] topic tags -> reviewer pool; a self-standing abstract -> surviving round one
[Round-one gate] none after upload — the abstract is your only round-one argument
[Rebuttal] factual corrections, a missing number a reviewer asked for, misread-claim fixes
[Conditional accept] the shepherd round: make every required change or justify declining it, evidenced
[After reject] no appeal; reroute to a sibling systems venue or strengthen and resubmit
A rebuttal moves borderline papers when it corrects a factual misreading or supplies a measurement a
reviewer said was missing; it does not move papers by arguing taste. In the shepherd round, a
silently skipped required change is what turns a conditional accept into trouble.
Misreadings to avoid
- Treating the extended abstract as boilerplate — it is the round-one decision surface.
- Treating conditional acceptance as unconditional — the shepherd can hold the paper until the
required revisions land.
- Assuming OSDI-style novelty is required — ATC rewards solid, useful, measured systems work.
- Projecting the old USENIX ATC process — the round structure, calendar, and cost model changed
in the SIGOPS transition.
Output format
[Process stage] pre-submission / round one / round two / rebuttal / conditional accept / final
[Decision driver] the criterion carrying the outcome (significance | soundness | evidence | clarity | reproducibility)
[Round-one readiness] does the extended abstract stand alone and clear the bar? yes/no
[Leverage plan] the next-stage action that can actually change the outcome
[Forbidden moves] identity leak / unsupported new claims / skipping a shepherd-required change
Source: brycewang-stanford/Awesome-Journal-Skills → ATC-Skills/skills/atc-review-process/SKILL.md
1---2name: atc-review-process3description: Use when reasoning about how an ATC (ACM SIGOPS Annual Technical Conference, formerly USENIX ATC) submission is evaluated — the two-round extended-abstract model, the round-one gate by two reviewers, the round-two read by 3-4 reviewers, double-blind reviewing, rebuttal, conditional acceptance with shepherding, and how ATC differs from OSDI/SOSP/NSDI.4---5
6
7# ATC Review Process
8
9Model the pipeline before interpreting any single review. ATC's process is unusual among systems
10venues in 2026 because it runs in **two rounds gated by a two-page extended abstract**, and because
11acceptance is **conditional on a shepherd**. The most consequential mental shift for authors
12arriving from a single-round conference is that **round one can end your submission before anyone
13reads the full paper** — the extended abstract is not a formality.
14
15## Process model
16
17- Submission and review run on **HotCRP** (`atc26.hotcrp.com`) with **double-blind** anonymity:
18 identities are mutually hidden through reviewing.
19- **Round one:** each **two-page extended abstract** is read by **two experienced reviewers** with a
20 broad systems perspective; abstracts below the bar are **rejected early**. The extended abstract is
21 review-only and not published.
22- **Round two:** papers that advance are read in full by **3-4 reviewers**, who weigh the
23 significance and usefulness of the systems contribution, the soundness of the design, the honesty
24 and completeness of the measurements, reproducibility, and clarity.
25- **Rebuttal / description of changes:** authors get a short (about one page) window to correct
26 factual misreadings and, where relevant, summarize revisions relative to a prior submission.
27- **Decision + shepherding:** selected papers are **conditionally accepted**; final acceptance
28 requires satisfying a **shepherd** (a PC member) on required revisions. Accepted papers publish in
29 the **ACM DL under Open Access**.
30
31## Reading a decision against the model
32
33| Outcome | What it means | Author move |
34|---|---|---|
35| Early reject (round one) | The abstract did not convey a problem, contribution, and result that clear the bar | Reframe the abstract; the paper may be fine but the *pitch* failed — see `atc-writing-style` |
36| Full review, then reject | A design or evidence gap survived the full read | Fix the substance, not the framing; consider a sibling venue |
37| Conditional accept | Contribution holds; specific revisions required | Treat the shepherd's list as binding; make or explicitly justify each item |
38| Reject with encouragement | Structural but repairable over a cycle | Rebuild the evidence and resubmit next cycle or reroute |
39
40The strategic reading: because round one judges the abstract, **the highest-leverage editing on the
41whole paper is often the two pages a round-one reviewer sees**. Write the submission so its problem
42and result are legible before the details.
43
44## How ATC differs from its siblings
45
46- **vs. OSDI / SOSP:** those flagships are more selective and lean toward larger, novel systems; ATC
47 takes a **broader, more practical** slice and welcomes solid, useful, well-measured work a
48 flagship might call incremental. A paper bounced from OSDI on "not novel enough" can be a strong
49 ATC paper if the engineering and measurement are real. Neither shares ATC's two-round
50 extended-abstract gate.
51- **vs. NSDI:** NSDI centers **networked systems** and runs its own multi-deadline model; ATC's
52 scope is all of systems and its 2026 calendar is a single June deadline. Route a networked-systems
53 design there if the network is the contribution.
54- **vs. EuroSys:** EuroSys is the other broad, SIGOPS-affiliated systems conference; the choice is
55 often calendar and community pull. Under SIGOPS, ATC and EuroSys now sit in the same family, so
56 check both deadlines.
57- **vs. pre-2026 USENIX ATC:** the old ATC ran a single-round review on a winter deadline with
58 USENIX-funded open access and no APC. The 2026 SIGOPS edition changed the round structure, the
59 calendar, and the cost model at once — do not carry the old process forward.
60
61## Who reads you
62
63Round one: two broad-perspective systems reviewers judging whether the idea and result are worth a
64full read. Round two: 3-4 reviewers closer to your subarea who open the artifact, check whether the
65measurements are end-to-end and honest about cost, and probe whether the baselines are fair. Vague
66design descriptions and average-only results get caught, not skimmed.
67
68## Where author leverage actually exists
69
70```text
71[Before submission] topic tags -> reviewer pool; a self-standing abstract -> surviving round one
72[Round-one gate] none after upload — the abstract is your only round-one argument
73[Rebuttal] factual corrections, a missing number a reviewer asked for, misread-claim fixes
74[Conditional accept] the shepherd round: make every required change or justify declining it, evidenced
75[After reject] no appeal; reroute to a sibling systems venue or strengthen and resubmit
76```
77
78A rebuttal moves borderline papers when it corrects a factual misreading or supplies a measurement a
79reviewer said was missing; it does not move papers by arguing taste. In the shepherd round, a
80**silently skipped required change** is what turns a conditional accept into trouble.
81
82## Misreadings to avoid
83
84- **Treating the extended abstract as boilerplate** — it is the round-one decision surface.
85- **Treating conditional acceptance as unconditional** — the shepherd can hold the paper until the
86 required revisions land.
87- **Assuming OSDI-style novelty is required** — ATC rewards solid, useful, measured systems work.
88- **Projecting the old USENIX ATC process** — the round structure, calendar, and cost model changed
89 in the SIGOPS transition.
90
91## Output format
92
93```text
94[Process stage] pre-submission / round one / round two / rebuttal / conditional accept / final
95[Decision driver] the criterion carrying the outcome (significance | soundness | evidence | clarity | reproducibility)
96[Round-one readiness] does the extended abstract stand alone and clear the bar? yes/no
97[Leverage plan] the next-stage action that can actually change the outcome
98[Forbidden moves] identity leak / unsupported new claims / skipping a shepherd-required change
99```
100
101---
102
103**Source:** [`brycewang-stanford/Awesome-Journal-Skills`](https://github.com/brycewang-stanford/Awesome-Journal-Skills) → `ATC-Skills/skills/atc-review-process/SKILL.md`
thedixitjain