codeaf-census

Overview ¶

codeaf-census is the instrument the recovery design asks for: it reads the model-call log this build always writes (`~/.codeaf/logs/calls.jsonl`, internal/calllog) and prints, as markdown, the table docs/design/recovery/DESIGN.md §1 argues from.

IT EXISTS BECAUSE EVERY WAVE OF THAT DESIGN MOVES A NUMBER, and without a committed instrument every one of those numbers is an argument about anecdotes. The first census (docs/design/recovery/census-20260910.md) was a throwaway pass over the same file; this is that pass, committed, tested on a fixture, and runnable nightly by a cron on the Spark (`make census`).

It is a developer's binary and never a verb on codeaf, for the reason cmd/codeaf-changes and cmd/codeaf-demo-home are: the shipped binary is on a checked-in byte budget (SIZE-BUDGET) and a measuring tool must not spend the product's weight.

IT READS OLD ROWS AS WELL AS NEW ONES. The log is a ten-day rolling record written by whichever build was running, so every reading below tolerates a field that is absent, and the one field this wave renamed is read under both of its spellings (row.go's hazardCeiling).

Usage:

codeaf-census [flags] [path]

  -top N        how many error signatures to print (default 25)
  -days N       the window lane health is measured over (default 3)
  -min N        the fewest finishes a (model, machine) pair needs to be
                listed in lane health (default 25)
  -chains N     how many of the longest retry chains to print (default 10)

With no path it reads the log this machine's profile writes.