colt-related-work

npx skills add brycewang-stanford/Awesome-Journal-Skills --skill colt-related-work --agent claude-code

Same command for any agent — swap --agent for codex, cursor, copilot.

# COLT Related Work At COLT, related work is quantitative: the reviewer wants to know exactly which bound, in exactly which model, your result improves, matches, generalizes, or separates. Prose adjacency ("much work has studied bandits") is filler; a rates table is evidence. Use this skill to build the comparison and to keep it eligible under the current CFP's overlap rules. ## The known-vs-new ledger For each of your main results, fill one row per nearest prior result: | Prior result | Model / assumptions | Their bound | Your bound | Delta type | |---|---|---|---|---| | [Cite] Thm 4 | oblivious adversary, K arms | $O(\sqrt{TK\log K})$ | $O(\sqrt{TK})$ | log-factor removal | | [Cite] Thm 1 | i.i.d., realizable | $O(d/\epsilon)$ | same rate, weaker assumption | assumption weakening | | [Cite] | same model | lower bound $\Omega(\sqrt{TK})$ | upper matches | closes their gap | Delta types COLT reviewers recognize as contributions: closing an upper/lower gap, removing a log factor with a new technique, weakening assumptions at the same rate, a new model with a separation from an old one, a simpler proof of a known result (yes — if genuinely simpler, say so plainly), and resolving a po