SOSP Topic Selection

Use this months before any deadline, when the routing decision is still cheap. SOSP solicits work on "the design, implementation, analysis, evaluation, and deployment of computer systems software" (2026 CFP at sigops.org/s/conferences/sosp/2026/cfp.html, checked 2026-07-08) — a broad charter that in practice is filtered through one question: is there an operating-systems principle here, embodied in something built and measured?

The three-part SOSP shape test

1. Principle — an extractable insight that outlives the artifact. Someone should be able to apply it while building a different system. "We made X fast" is not a principle; "X can be fast because responsibility R can move below interface I" is.
2. Embodiment — a real system that exists and runs. SOSP is hostile to simulation-only and design-only papers outside genuinely unusual cases; the venue's culture is build-and-measure.
3. Evidence at the claim's scale — an evaluation that could survive sosp-experiments. A principle about datacenter tails needs datacenter-shaped evidence.

A project passing all three is SOSP-shaped regardless of its subdomain: kernels, virtualization, distributed systems, storage, verification, security architecture, cloud infrastructure, and systems-for-ML have all produced recent SOSP programs.

Routing table

Routing is about the center of gravity, not exclusivity — plenty of storage and network papers appear at SOSP when the principle is systemic rather than domain-internal. The test: which PC can evaluate the paper's hardest claim?

Timing the call

• Project start: choose the target family (SOSP/OSDI-tier vs domain venue), because it dictates the evaluation scale you must budget for.
• First results: run the shape test honestly. A missing principle at this stage is fixable — often the principle exists but the team hasn't extracted it. Write the one-sentence version; if it keeps coming out as a speedup report, reroute.
• Three months out: lock the venue. Late switches leak time into reformatting and re-framing that the evaluation needed.

Shape-test worksheet

Principle (one sentence, no system name allowed): <...>
  - could a different team use it in a different system?      yes/no
Embodiment: what runs today end-to-end?                        <...>
Evidence scale: hardest claim <...> needs <workload/hardware>  budgeted? yes/no
Nearest published neighbor + venue:                            <paper, venue'yy>
Center of gravity (one): OS principle / networking / storage /
  architecture / data mgmt / ML-systems / engineering lessons
-> Route: SOSP | OSDI (timing) | <domain venue> | HotOS first

Anti-patterns this venue punishes

• The 10% paper: a measured improvement on a known design with no new understanding — competent, publishable, not SOSP.
• The unbuilt design: elegant architecture, evaluation by argument. HotOS or a workshop first; SOSP later, with the system.
• The scale flex: impressive deployment numbers, but the lesson is "we had more machines." ATC will value the experience report more than a SOSP PC will.
• The camouflaged domain paper: a database or ML paper wearing systems vocabulary; the PC detects the mismatch at the meeting when the hardest claim needs expertise the room correctly declines to fake.

Output format

[Shape test] principle / embodiment / evidence-scale — pass/fail each
[Principle] the extracted one-liner
[Route] SOSP | sibling (named) + the deciding factor
[Timing] next posted deadline for the routed venue (never an implied one)
[Risk] the anti-pattern this project is closest to