coverage-report

name	coverage-report
description	Measure how comprehensive and current the list is — entries per section and per year, with thin and stale sections flagged. Use before planning content expansion or periodically as a health check.

Coverage Report Skill

Computes coverage statistics for README.md in the awesome-recursive-language-models list, so growth effort goes where the list is weakest.

When to use

• Before a content-expansion pass, to pick target sections.
• Periodically, to check the list is keeping pace with the field.
• After a large PR, to see how coverage shifted.

How to compute

Every entry is one line matching - [Title](URL) - Description. (Year). For each taxonomy section of README.md:

1. Count the entry lines.
2. Extract each trailing (Year) and record the newest year.
3. Note prose cross-references ("See the core papers above…") — they mitigate a low count but do not replace entries.

Example one-liner for the raw counts (run from the repo root):

awk '/^## /{section=$0} /^- \[/{count[section]++} END{for (s in count) print count[s], s}' README.md | sort -rn

Interpreting results

• Thin — fewer than 5 entries: a target for the paper-scout agent.
• Stale — newest entry older than ~12 months: the field has likely moved; sweep for recent work.
• Uncovered pattern — a row of the Taxonomy table with no example entries anywhere: highest-priority gap.

A section being thin is not by itself a problem — per CONTRIBUTING.md, a smaller set of well-verified resources beats broad coverage with weak relevance. Use the report to direct search effort, not to justify padding.

Output

A markdown table — section | entries | newest year | status (ok / thin / stale) — followed by a short list of recommended targets for the next expansion pass.