Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› stoc-related-work
stoc-related-work
GitHub指导如何为STOC论文定位相关工作，包括追踪界限改进谱系、正确引用会议/期刊/预印本版本、处理并发工作并确保双盲评审安全。
Trigger Scenarios
撰写理论计算机科学论文的Related Work部分
需要引用STOC相关文献并处理多版本差异
处理arXiv或ECCC上的并发预印本
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill stoc-related-work -g -y
SKILL.md
Frontmatter
{
"name": "stoc-related-work",
"description": "Use when positioning a STOC (ACM Symposium on Theory of Computing) paper against the theory literature — tracing bound-improvement lineages, citing conference\/journal\/preprint version pairs correctly, handling concurrent arXiv and ECCC work, and keeping self-citation double-blind-safe."
}
STOC Related Work
Theory positioning is quantitative: reviewers expect the paper to place its bounds in an explicit lineage ("[A] gave $O(n^2)$; [B] improved to $O(n \log n)$ under assumption X; we remove X and reach $O(n)$"), and they check the lineage against their own memory of it. Errors here — a missed improvement, a misquoted exponent, a result attributed to the journal version's year instead of the conference announcement — are among the fastest credibility losses available at this venue.
The lanes a STOC reviewer checks
| Lane | What must be covered | Typical miss |
|---|---|---|
| The direct lineage | Every prior bound on your exact problem, with its model | Quoting the weakest prior bound to inflate the delta |
| The barrier literature | Known obstructions your result confronts (relativization, natural proofs, hardness reductions, integrality gaps) | Claiming a breakthrough while silent on why the barrier does not apply |
| Neighboring models | The same question in weaker/stronger models (randomized vs. deterministic, static vs. dynamic, classical vs. quantum) | "First" claims that are only first in one cell of the model grid |
| Technique ancestry | Where your machinery comes from, even from other subfields | Reviewers who invented the machinery are plausibly your reviewers |
| Concurrent work | Recent arXiv/ECCC preprints on the problem | The subreviewer knows about last month's preprint; your draft predates it |
Version-aware citation, a theory-specific craft
Most important theory results exist as a conference extended abstract, a later journal version, and a preprint — and they can differ mathematically (bounds improve; bugs get fixed in the journal version). House rules:
- Cite the version whose statement you actually use. If you invoke the strengthened journal bound, cite the journal paper, possibly alongside the conference announcement.
- Date claims by first announcement (conference or preprint), attribute proofs by where the complete proof appears.
- Keep both versions in the entry when the community habitually does:
@inproceedings{ST01-smoothed,
author = {Spielman, Daniel A. and Teng, Shang-Hua},
title = {Smoothed Analysis of Algorithms: Why the Simplex Algorithm
Usually Takes Polynomial Time},
booktitle = {Proc. 33rd ACM Symposium on Theory of Computing (STOC)},
pages = {296--305},
year = {2001},
note = {Journal version: J. ACM 51(3), 2004}
}
- dblp is the community's canonical metadata source; export entries from it rather than hand-typing venues, and prefer its conference/journal splitting over publisher aggregator records.
Concurrent and independent work
The arXiv/ECCC-first culture means collisions are visible within days, not years. Norms as practiced:
- A concurrent preprint (posted around your submission, obtained independently) gets a clearly flagged paragraph — "Independently and concurrently, [Z] obtained..." — with an honest comparison of strength, generality, and technique. Committees routinely accept both papers when the note is fair.
- A prior preprint you missed is not concurrent work; the delta must be recomputed against it, sometimes fatally. Run a fresh arXiv/ECCC sweep the week before the deadline, and again before camera-ready.
- If a collision surfaces mid-review, current practice varies on whether and how to notify the chair; check the cycle's instructions (待核实) rather than improvising contact with the PC.
Double-blind self-positioning
STOC 2026 was double-blind (checked 2026-07-08). Your prior work is often the nearest related work, so:
- Cite yourself in third person, with no ownership markers ("building on the framework of [7]" — not "our framework [7]").
- Do not cite an unpublished manuscript of yours that only the authors could possess; if the submission genuinely depends on it, it must be publicly posted (anonymity-compatibly) or its content inlined.
- Acknowledgement-style thanks to collaborators on adjacent work wait for camera-ready.
Building the comparison honestly
- State comparisons in the prior work's own parameterization before yours; translating everything into the units most flattering to your bound is spotted immediately by the lane-one reviewer.
- When your result is incomparable to a prior one (better dependence on $n$,
worse on $\varepsilon$), say "incomparable" and give the crossover point —
that sentence pre-empts the review question (
stoc-author-response's objection ledger). - A small comparison table in the introduction (result, model, bound, source) is house-standard for crowded problems and often becomes the most-read object in the paper.
The deadline-week sweep, as a procedure
- Search arXiv (
cs.DS,cs.CC, and your problem's other categories) and ECCC for the problem's standard names and their variants; theory titles are conventional enough that keyword search works. - Check the most recent STOC/FOCS/SODA accepted-paper lists — results announced there may not have preprints yet.
- Walk the dblp pages of the five authors most active on the problem; new preprints appear there within days.
- For each hit, classify: prior (delta must be recomputed), concurrent (comparison paragraph), or unrelated-despite-title (note why, for your own future reference).
- Re-run the same sweep before camera-ready; the February-to-March window regularly produces new lineage entries that the full version should cite.
Cycle-volatility warnings
- Anonymity rules for citing your own preprints follow the live CFP's double-blind wording (待核实 per cycle; the policy is new at STOC and may be refined).
- Deadline-week literature sweeps must include ECCC and the current FOCS/SODA accepted-paper lists, which shift the frontier between CFPs.
Output format
[Lineage table] complete / gaps: <missing bounds or models>
[Barrier account] addressed / silent <- fix before review
[Version hygiene] statements matched to cited versions? dblp-verified?
[Concurrency] <preprints found in final sweep; comparison note drafted?>
[Self-citation] double-blind-safe / leaks: <list>
Version History
- 9f86f09 Current 2026-07-19 17:39
