Summary checker
Paste the body of your English summary. The thresholds are corpus medians, not taste — a low score does not mean bad writing, it usually means the judge has to work to find your conclusions.
—
—
Why these seven thresholds
Each one is taken from the distribution in the corpus, not from a list of "best practices".
Number density
Outstanding summaries carry 20–35 concrete numbers; Meritorious ones 10–15. This is the most stable gap between the two tiers.
Task coverage
Every sub-problem gets at least one sentence of method plus a quantitative result. A missing task is a direct elimination in the second round; roughly 25% of teams do it.
Length
Median 437 words, interquartile range 350–520. Anything past 600 usually means too much background.
Keywords
Four or five, and one of them must be your model name — a memorable label such as GIID or BDS.
Sensitivity
The line that recurs in judges' commentaries: teams that do sensitivity properly almost always stand out.
Limitations
End on solid ground: strengths, limitations, conclusion. Two of the three beats a conclusion alone.
Why papers get cut · incidence
Cross-read from judges' commentaries and corpus statistics; use it to decide what to fix first.
Install it into your agent
The guide ships as a skill pack. Once installed it is picked up when you draft, revise or audit a paper. You can also just read it as a handbook.
npx skills add Timevelyer/mcm-o-guide
git clone, then drop the folder into ~/.claude/skills/.