Taking a repo's reviewer attention baseline¶

Reviewer attention per merged agent PR means little without a number from before. So record a baseline for a repo before hangar's first job there. Repo 1 needs one before hangar is used on it.

The baseline is the same measure on the repo's own merged PRs: review events per merged PR. It covers the four ISO weeks (UTC) before hangar starts in the repo. It leaves out PRs a bot opened, and reviews and comments by bots. The PR's author counts, as on agent PRs.

Take it¶

Run this with the gh CLI, signed in with read access to the repo. Set the repo and the four weeks first.

REPO=owner/repo-1 SINCE=2026-10-05 UNTIL=2026-11-01   # four ISO weeks, UTC, before hangar's first job

count() {  # $1 PR number, $2 reviews|comments
  gh api --paginate "repos/$REPO/pulls/$1/$2" \
    --jq '.[] | select(.user.type != "Bot" and .state != "PENDING") | .id' | wc -l
}

gh pr list -R "$REPO" --state merged --limit 1000 --search "merged:$SINCE..$UNTIL" \
  --json number,author --jq '.[] | select(.author.is_bot | not) | .number' |
while read -r n; do
  echo "$n $(count "$n" reviews) $(count "$n" comments)"
done | tee baseline.txt |
awk '{ e += $2 + $3 } END { printf "%d PRs, %d review events, %.2f per merged PR\n", NR, e, NR ? e / NR : 0 }'

baseline.txt keeps one line per PR: its number, its reviews and its review comments.

Record it¶

Put the summary line, the dates and baseline.txt on the repo's onboarding ticket in Jira. Then compare it with reviewer_attention.events_per_merged_pr from hangar metrics weekly --repo owner/repo-1.

The two differ in one way. The baseline counts humans' PRs, and hangar counts agent PRs. Change both if the definition changes.