Facts as data

The skill has three moves. A census leads into design; a review of one diff stands alone.

Census

1. Run python3 <skill-folder>/scripts/buried_facts.py <paths>. It lists leads: names joined in a regex alternation or an || chain, literal lists of names, and runs of branches keyed by a literal; its --help gives the thresholds. It skips test files unless given --include-tests. A lead is a place to read; whether the fact is buried is your judgment.
2. Read the code for the buried facts no pattern shows: branches whose bodies differ only in a constant, two functions that differ only in a name, and a shared append-list (registry, module list, schema list, chained command) that every change edits.
3. Group the leads by what they mean, in domain words: "calls that pause the thread", "retry libraries", not the rule or file that holds them. A meaning found at two sites already has two owners; name both sites in its group: one of them must become the owner and the other read from it.
4. Record one row per lead site in a census file, in the format in references/census-format.md: table (it moves into a table), is-table (it already is the shared table) or not-data (its variation has no real form; see the counter-test below). Rerun with --record <census.tsv> until it exits 0. Keep the census in the issue body or the task's decision log.

The census is done when the script exits 0 against the record and every group names its meaning.

Design

For each group recorded as table: