Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› soda-related-work
soda-related-work
GitHub指导在SODA论文中准确定位相关工作，追溯算法界限演变。涵盖跨会议/期刊版本验证、arXiv优先权处理及双盲自引用规范，确保引用准确性与学术诚信。
Trigger Scenarios
撰写SODA等算法顶会论文的Related Work部分
核实算法结果的文献渊源与界限对比
处理arXiv预印本引用与优先级声明
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill soda-related-work -g -y
SKILL.md
Frontmatter
{
"name": "soda-related-work",
"description": "Use when positioning a SODA (ACM-SIAM Symposium on Discrete Algorithms) paper in the algorithms literature — tracing bound lineages across SODA\/STOC\/FOCS\/ESA\/ICALP and journals, handling arXiv-first priority culture, citing conference versus journal versions correctly, and self-citation under lightweight double-blind rules."
}
SODA Related Work
Positioning at SODA is quantitative: the related-work section's job is to establish, bound by bound, exactly where your result sits in a lineage the referees already know. The most common positioning failure is not a missing citation but a mis-stated bound — quoting a superseded conference version, missing a stronger journal version, or comparing against the randomized bound when a better deterministic one exists.
The five lanes to sweep
| Lane | Where it lives | What referees check |
|---|---|---|
| Direct predecessors | SODA/STOC/FOCS/ESA/ICALP proceedings, arXiv | Is the displaced bound the true state of the art? |
| Journal consolidations | SICOMP, JACM, ACM ToA, Algorithmica | Did the journal version strengthen the bound you quote? |
| Lower bounds / hardness | Same venues + ITCS, CCC | Does a conditional lower bound cap your claimed improvement's meaning? |
| Concurrent arXiv postings | arXiv (last ~18 months) | Anything simultaneous that changes the delta |
| Adjacent-model results | Distributed (PODC/DISC), parallel (SPAA), geometry (SoCG), dynamic/streaming lines | Incomparable-but-related results a subreviewer will ask about |
The concurrent lane is the one that changes late: re-sweep arXiv in the final week before the July deadline and again before the September rebuttal — a concurrent improvement discovered by the referee but unacknowledged by you turns a coexistence story into a credibility problem.
Bound-lineage verification recipe
For every row of your comparison table (soda-writing-style):
1. Find the paper's dblp entry (dblp.org) — canonical venue + year.
2. Check for a journal version listed alongside the conference version;
if one exists, read ITS abstract for the strongest stated bound.
3. Check the arXiv page for post-publication revisions or errata.
4. Quote: strongest bound, both versions cited ("[SODA 2019; SICOMP 2021]"),
with the model/caveat column filled from the actual theorem statement.
5. If two papers claim the same bound, cite both and let dblp dates
order them — do not adjudicate priority disputes in your paper.
arXiv-first culture and priority
The algorithms community treats arXiv posting as the priority event and the conference as certification. Consequences for positioning:
- Cite arXiv preprints of results you use or displace — refusing to engage with
a well-known preprint because it is "unpublished" reads as gamesmanship.
Distinguish status in the bibliography (
arXiv:2401.xxxxx, preprint). - If your result was scooped between writing and deadline, the honest structure is a "concurrent and independent work" paragraph stating both results' deltas — PCs routinely accept parallel discoveries when the relationship is cleanly disclosed.
- Your own posting strategy: SODA permits arXiv dissemination before and during review (2027 CFP, checked 2026-07-08), so establishing priority early is free.
Self-citation under lightweight double-blind
SODA's double-blind rules explicitly require that important references not be
omitted or anonymized (soda-submission). So:
- Cite your prior work in third person, fully: "extending the decomposition of [14]" — never "our decomposition [14]," and never a blanked "[anonymized]."
- A direct predecessor paper of yours belongs in the comparison table like anyone else's; suppressing it to "look more novel" is the actual violation.
- Do not cite an in-submission companion manuscript that referees cannot read; if the companion is load-bearing, post it to arXiv first.
Bibliography mechanics for dual-version results
The conference/journal duality has a concrete BibTeX answer — cite both, quote the strongest:
% Conference announcement:
@inproceedings{HLRW24soda, ... booktitle = {SODA}, year = {2024}, ... }
% If a journal version exists, add and PREFER its bound:
@article{HLRW26sicomp, ... journal = {SIAM J. Comput.}, ... }
% In text: "achieves O(m log^2 n) [HLRW24soda; journal version HLRW26sicomp]"
Hygiene rules that referees notice when violated:
- Never cite an arXiv identifier for a result that has a published version — it suggests the sweep stopped at the preprint.
- Never quote a bound from a paper's abstract; abstracts round. Quote the theorem, with its number if your comparison is load-bearing.
- Keep dblp keys or DOIs in comments next to each entry so the September rebuttal and the October camera-ready don't re-litigate attribution.
- Alphabetic-initial citation labels ([HLRW24]) are house convention in theory bibliographies; numeric labels work, but author-initial labels make your comparison table readable without flipping to the references.
Positioning prose patterns
- Lineage sentence: "For dynamic connectivity, worst-case bounds improved
from
O(sqrt n)[F] throughn^{o(1)}[NSW] to the currentpolylogamortized bounds [HdLT; journal version]." Three citations, one sentence, and the referee knows you know the field. - Incomparability paragraph: for results that trade parameters ("faster but randomized," "deterministic but amortized"), state the trade explicitly instead of claiming dominance.
- Technique genealogy: name where your tools come from ("our charging argument adapts [X]'s token scheme") — theory culture rewards disclosed intellectual debts and punishes discovered ones.
Output format
[Lineage verdict] Sound / Gaps / Mis-stated bounds found
[Lane sweep] <per-lane findings, incl. concurrent arXiv window>
[Version errors] <rows quoting superseded or weaker versions>
[Self-citation audit] <third-person compliance; nothing omitted or anonymized>
[Table repairs] <rows to add, bounds to correct, caveats to co-locate>
Version History
- 9f86f09 Current 2026-07-19 17:38
