ceremony/actions/labels-reconcile/labels-reconcile.sh

#!/usr/bin/env bash
if [ "${BASH_SOURCE[0]}" = "$0" ]; then
  set -euo pipefail
else
  # Fixture tests source the pure functions and deliberately inspect failures.
  set -u
fi

# labels-reconcile.sh — the automation LABELS.md promises: state labels are
# written by machinery, never by hand. Every run derives each open PR's
# state:* from GitHub's own facts (draft flag, requested reviewers, submitted
# reviews) and converges the labels to it, so a killed run or a hand-moved
# label heals on the next pass. Stale is judged from real activity — commits,
# comments, reviews — never from label churn, or the sweep would un-stale its
# own mark every tick.
#
# The verdict contract (CONTRIBUTING.md): reviews end in approve or
# request-changes. Some live bots are comment-only and post agreement as a
# COMMENTED review — a non-verdict this machine refuses to guess about (body
# parsing is a heuristic, and a wrong guess promotes an unapproved PR). The
# judgment call belongs to the PR AUTHOR, who reads the round and escalates
# by requesting the human's review — an explicit request is a fact, and it is
# the one this machine trusts (see decide_state's top precedence). The
# machine auto-requests the human only in the no-judgment-needed case: every
# required verdict is a formal head-current approval. Any approval that counts
# must be bound to
# the CURRENT head SHA: GitHub keeps approvals alive across pushes, and a
# stale approval must never promote unreviewed code to the human.
#
# DRY_RUN=1 narrates every mutation instead of performing it (how this script
# is rehearsed against the live repo). A workflow_dispatch run also bootstraps
# the taxonomy (label create --force) — that heal is dispatch-only; the cron
# sweep tolerates a missing label rather than recreating it.
#
# The state machine below is pure (globals in, state out) and covered by
# fixture tests in test/labels-reconcile.test.sh.

HUMAN="${HUMAN_REVIEWER:-danmt}"
BOTS=()
# Per-author panels (#224): parallel arrays because the conf is tiny and an
# associative array buys nothing but a bash-4 dependency statement. One entry
# per panel[<login>]= row — PANEL_AUTHORS holds the login, PANEL_ROWS the
# space-joined reviewer set at the same index.
PANEL_AUTHORS=()
PANEL_ROWS=()
REQUIRED_BOTS=()
STATES=(state:building state:bots-reviewing state:addressing state:needs-human)
BLOCKERS=(blocker:conflict blocker:ci-red blocker:unrequested)
# The PR's current labels, set per PR by the sweep. Initialized here because
# the script runs under `set -u` even when sourced, and the pure-function
# fixtures call decide_state — which now reads has_label — without ever
# setting it (#51). An empty default keeps has_label honest for every caller.
LABELS=""
# Labels this machine used to own and no longer does. Cleared on sight so a
# retirement heals the board instead of stranding a label nothing recomputes.
RETIRED=(state:needs-rebase)
STALE_AFTER=$((48 * 3600))
# How long the facts behind blocker:unrequested must have stood still before it
# is written (#236 D2). The operator's "more than 5 minutes", measured off the
# inputs' own timestamps rather than off sweep memory — this script is
# stateless per pass and stays that way. Overridable the way this file's other
# constants are, for a caller whose round cadence is slower or faster.
RECONCILE_UNREQUESTED_GRACE="${RECONCILE_UNREQUESTED_GRACE:-300}"
# The workflow whose runs checks_state must never grade — its own (#208).
# GITHUB_WORKFLOW is ambient in every Actions step and names the CALLER (the
# consumer's PR-facing workflow, since consumers name the caller), so this
# self-serves with no workflow-file change. The explicit override exists for
# two readers: the fixtures, and #209's detached sweep caller, which will
# need to point this at the PR-facing caller's name once reconcile no longer
# runs inside it. Empty means "filter nothing" — a caller outside Actions
# (a local rehearsal, an older pin) must not silently start dropping entries.
SELF_WORKFLOW="${SELF_WORKFLOW:-${GITHUB_WORKFLOW:-}}"

# The needs-ruling invariants (#52) — one implementation for both surfaces.
# shellcheck source=lib/ruling.sh
. "$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)/../../lib/ruling.sh"
# shellcheck source=lib/forge.sh
. "$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)/../../lib/forge.sh"
# The attention target invariants (#232) — diagnosis only, both surfaces.
# shellcheck source=lib/attention.sh
. "$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)/../../lib/attention.sh"
# The guarded read and its reason line (#101) — one implementation for both
# surfaces. read_failure_reason lived here until the issue surface needed the
# identical rule (#247); a second copy of it is the failure lib/ruling.sh's
# own header was written to record.
# shellcheck source=lib/read.sh
. "$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)/../../lib/read.sh"

log() { printf 'labels: %s\n' "$*"; }

run() { # every mutation goes through here — DRY_RUN=1 logs instead of doing
  if [ -n "${DRY_RUN:-}" ]; then log "DRY_RUN: $*"; else "$@"; fi
}

blind_sweep_warning() { # $1 = unreadable PRs, $2 = all open PRs, $3 = sampled read-failure reason
  # Report, do not diagnose (#101 D5). The old text asserted the caller's
  # checks:/statuses: grants as THE cause — an inference #95 made from a
  # control case, and the merged consumer-side fix (incubator#48/PR #49)
  # left the symptom standing while the run emitting this warning held the
  # evidence that would have said so. Lead with what gh actually said this
  # sweep; the permissions hint stays, demoted to one named candidate.
  if [ "$2" -gt 0 ] && [ "$1" -eq "$2" ]; then
    local reason="${3:-}"
    if [ -n "$reason" ]; then
      echo "::warning::labels: every open PR was unreadable; sampled reason: $reason — one candidate is missing checks: read, statuses: read and actions: read in the caller (private repos do not imply them)"
    else
      echo "::warning::labels: every open PR was unreadable; no reason was captured — one candidate is missing checks: read, statuses: read and actions: read in the caller (private repos do not imply them)"
    fi
  fi
}

missing_core_labels_warning() { # $1 = declared rows, $2 = repo label names
  local rows="$1" repo_labels="$2" row name missing=""
  [ -n "$repo_labels" ] || return 0
  while IFS= read -r row; do
    [ -n "$row" ] || continue
    name="${row%%|*}"
    if ! grep -qxF "$name" <<<"$repo_labels"; then
      if [ -n "$missing" ]; then missing="$missing, $name"; else missing="$name"; fi
    fi
  done <<<"$rows"
  if [ -n "$missing" ]; then
    echo "::warning::labels: missing core label(s): $missing; bump the ceremony pin, then re-dispatch workflow_dispatch to bootstrap the taxonomy"
  fi
}

load_config() { # $1 = consumer labels.conf; panel is mandatory, scopes optional
  local conf="$1" line panel_seen=false
  [ -f "$conf" ] || {
    echo "labels: missing config: $conf (a panel= line is required)" >&2
    return 1
  }
  BOTS=()
  PANEL_AUTHORS=()
  PANEL_ROWS=()
  # shellcheck disable=SC2094 # parse_panel_author_row takes $conf for its
  # error messages only — nothing in this loop writes the file it reads
  while IFS= read -r line || [ -n "$line" ]; do
    [ -n "$line" ] || continue
    # The panel[ prefix is matched QUOTED (#224 D7): in a case pattern an
    # unquoted panel[abc]=* is a bracket expression that matches panela=…,
    # panelb=…, panelc=… — silently rerouting ordinary settings. The
    # panela= tripwire in test/labels.test.sh goes red if this regresses.
    case "$line" in
      panel=*)
        [ "$panel_seen" = false ] || {
          echo "labels: duplicate panel line in $conf" >&2
          return 1
        }
        panel_seen=true
        read -r -a BOTS <<<"${line#panel=}"
        [ "${#BOTS[@]}" -gt 0 ] || {
          echo "labels: panel must name at least one reviewer in $conf" >&2
          return 1
        }
        ;;
      "panel["*) parse_panel_author_row "$line" "$conf" || return ;;
      triage-actors=*) ;;
      *) parse_label_row "$line" >/dev/null || return ;;
    esac
  done <"$conf"
  [ "$panel_seen" = true ] || {
    echo "labels: missing panel= line in $conf" >&2
    return 1
  }
}

parse_panel_author_row() { # panel[<login>]=<space-separated logins> (#224)
  # Every failure here is a hard one that names the offending line (D3): a
  # conf error takes the whole board down, and the run log is the only place
  # the operator can read why. A malformed bracket is refused AS a bracket
  # (D4) — falling through to parse_label_row would report it as a
  # "malformed label row", the misleading diagnostic #224 was filed over.
  local line="$1" conf="$2" login rest existing
  case "$line" in
    "panel["*"]="*) ;;
    *)
      echo "labels: malformed panel[<login>]= row (expected panel[<login>]=<reviewers>): $line in $conf" >&2
      return 1
      ;;
  esac
  login="${line#panel[}"
  login="${login%%]=*}"
  [ -n "$login" ] || {
    echo "labels: empty login in panel row: $line in $conf" >&2
    return 1
  }
  # The login must be exactly one well-formed bracket pair of login
  # characters. Without this, panel[z]]=b parses: the case above only
  # establishes that SOME ]= occurs, ${login%%]=*} keeps the stray ] inside
  # the login (z]), and set_required_bots for the real z then silently falls
  # back to the base panel — the misroute D4 exists to refuse. GitHub logins
  # are [A-Za-z0-9-], per the #285 spec.
  case "$login" in
    *[!A-Za-z0-9-]*)
      echo "labels: malformed panel[<login>]= row (a login is [A-Za-z0-9-] only): $line in $conf" >&2
      return 1
      ;;
  esac
  for existing in ${PANEL_AUTHORS[@]+"${PANEL_AUTHORS[@]}"}; do
    [ "$existing" != "$login" ] || {
      echo "labels: duplicate panel[$login]= row in $conf: $line" >&2
      return 1
    }
  done
  local -a row=()
  rest="${line#*]=}"
  read -r -a row <<<"$rest"
  [ "${#row[@]}" -gt 0 ] || {
    echo "labels: panel[$login]= must name at least one reviewer in $conf: $line" >&2
    return 1
  }
  PANEL_AUTHORS+=("$login")
  PANEL_ROWS+=("${row[*]}")
}

parse_label_row() { # exact name|color|description; pipes in descriptions are refused
  local line="$1" name color desc extra
  IFS='|' read -r name color desc extra <<<"$line"
  if [ -z "$name" ] || [ -z "$color" ] || [ -z "$desc" ] || [ -n "${extra:-}" ]; then
    echo "labels: malformed label row: $line" >&2
    return 1
  fi
  printf '%s|%s|%s\n' "$name" "$color" "$desc"
}

configured_label_rows() { # validated scope rows, excluding the panel setting
  local conf="$1" line
  [ -f "$conf" ] || return 0
  while IFS= read -r line || [ -n "$line" ]; do
    [ -n "$line" ] || continue
    # "panel["* quoted for the same D7 reason as load_config's case; skipping
    # the bracketed rows (D5) keeps a dispatch bootstrap from trying to
    # create a label named panel[<login>].
    case "$line" in panel=* | "panel["* | triage-actors=*) continue ;; esac
    parse_label_row "$line" || return
  done <"$conf"
}

panel_for_author() { # $1 = author → the effective panel, space-joined (#224 D2)
  # THE resolution point: the author's panel[<login>]= row when the conf
  # defines one, the base panel= otherwise. Everything that computes a
  # required set goes through here, because two places computing the panel
  # is how the engine and the reconciler came to disagree in the first place.
  local author="$1" i
  for i in ${PANEL_AUTHORS[@]+"${!PANEL_AUTHORS[@]}"}; do
    if [ "${PANEL_AUTHORS[i]}" = "$author" ]; then
      printf '%s\n' "${PANEL_ROWS[i]}"
      return
    fi
  done
  printf '%s\n' "${BOTS[*]}"
}

set_required_bots() { # the PR author is recused by construction
  # Minus-the-author applies to WHICHEVER set panel_for_author returns (#224
  # D2's safety net): an author who mistakenly appears inside its own
  # bracketed row is still recused.
  local author="$1" bot
  local -a effective=()
  read -r -a effective <<<"$(panel_for_author "$author")"
  REQUIRED_BOTS=()
  for bot in ${effective[@]+"${effective[@]}"}; do
    [ "$bot" = "$author" ] || REQUIRED_BOTS+=("$bot")
  done
}

# ---------------------------------------------------------------------------
# The state machine. Pure functions over these globals, set per PR:
#   DRAFT        true|false
#   HEAD_SHA     the PR's current head commit
#   BASE_SHA     the PR's base branch head (the release-shape guard's ref)
#   REQUESTED    newline-separated logins with a review currently requested
#   REVIEWS_JSON JSON array of submitted (non-PENDING) reviews
#   MERGEABLE    MERGEABLE | CONFLICTING | UNKNOWN  (GitHub's own verdict)
#   CHECKS       SUCCESS | FAILURE | PENDING | NONE (the check rollup)
#   LABELS       newline-separated labels currently on the PR
#   HEAD_COMMIT_AT  the head commit's own date, ISO-8601; empty when unread
#   NOW          this sweep's epoch seconds (main sets it once per run)
# ---------------------------------------------------------------------------

requested() { grep -qxF "$1" <<<"$REQUESTED"; }

# outstanding_requests <requested-logins> — the portable "who still owes a
# verdict on THIS head" (issue #188, term 4).
#
# GitHub clears requested_reviewers when a verdict lands, so on that forge the
# field already answers this question and the filter below removes nothing.
# **Forgejo does not clear it.** Measured 2026-08-02: rig!140 listed all three
# panelists with all three verdicts in, and rig!146 still lists three while
# MERGED — the field is stale even on a closed PR, so it over-counts forever.
#
# Reading it raw on Forgejo pins a PR at state:bots-reviewing for life and
# stops blocker:unrequested from ever being true: the sweep believes a round
# is permanently live. So the requested set is intersected with "has not
# submitted a verdict for the current head", which is derived from
# /pulls/{n}/reviews — the read that is true on both forges.
#
# Pure over REVIEWS_JSON/HEAD_SHA so the fixtures can drive it; a reviewer
# whose only verdict is STALE still owes one, which is why this asks
# bot_verdict rather than merely "has any review".
outstanding_requests() {
  local login
  while IFS= read -r login; do
    [ -n "$login" ] || continue
    case "$(bot_verdict "$login")" in
      APPROVE | BLOCK | FEEDBACK) continue ;;
    esac
    printf '%s\n' "$login"
  done <<<"${1-}"
}

checks_state() { # rollup JSON on stdin → SUCCESS | FAILURE | PENDING | NONE | UNREADABLE
  # UNREADABLE is the absence of the key itself, which is what a failed fetch
  # leaves behind — distinct from a present-but-empty rollup, which honestly
  # means this PR has no checks. Collapsing the two let an API hiccup present
  # as "nothing is failing", i.e. as mergeable-by-a-human: the same
  # unknown-certified-as-green shape as the bug this machine exists to stop.
  # The caller skips the PR entirely rather than labelling on facts it did not
  # read; blocking on it instead would flap the whole board on one bad call.
  # The rollup mixes two node types with two different closed enums: CheckRun
  # carries `conclusion` (CheckConclusionState), StatusContext carries `state`
  # (StatusState). Rather than list the outcomes that block — the version that
  # shipped in this PR's first round listed four, and ERROR, CANCELLED and
  # STALE fell through its `else` into SUCCESS — this lists the outcomes that
  # DON'T, and treats everything else as blocking.
  #
  # That direction is the point. An outcome we do not recognise is one we
  # cannot certify as mergeable, and certifying the unrecognised as green is
  # the exact shape of #136. The cost of being wrong is symmetric in form and
  # not in consequence: a false FAILURE parks the PR on the agent, who looks;
  # a false SUCCESS invites a human to merge a tree that will not merge.
  #
  # The list-what-passes rule has exactly one carve-out, and it is narrower
  # than an outcome: a CANCELLED entry is discarded when its context holds at
  # least one non-cancelled sibling (#139). The reconcile job queues in one
  # repo-global concurrency group, so any repo event — a sibling PR's push,
  # triage labelling an issue — evicts the queued duplicate AFTER it has
  # attached a check run to this PR's head, and that cancelled entry became
  # the context's newest word: blocker:ci-red on a PR whose real checks were
  # all green (#133/#136, evictable only by an empty commit). A cancelled run
  # said nothing about this head; a non-cancelled sibling is a real verdict
  # about exactly these bytes, whatever order the two arrived in — and for
  # this workflow the evictor performs the duplicate's work anyway, since
  # every sweep covers every open PR. This does not widen unknown-into-green:
  # a context whose entries are ALL cancelled never reported at all (a killed
  # or timed-out required job), so it keeps CANCELLED and still blocks —
  # discard needs a surviving verdict, never an empty context.
  #
  # And one exclusion that comes before every rule above: the label machine
  # never grades its own runs (#208). Every reconcile sweep serializes
  # through one shared concurrency group, and GitHub records a displaced
  # queued run as CANCELLED — there is no "superseded" conclusion for queue
  # displacement. When the displaced run was born from a pull_request_target
  # event, that cancelled entry attaches to the victim PR while its
  # SUCCESSOR — triggered by a different PR or an issues event — attaches
  # elsewhere, so the #139 carve-out's premise (a surviving sibling on the
  # same PR) fails structurally: on the victim the newest self entry stays
  # CANCELLED, the deny-list scores it FAILURE, and the sweep sets
  # blocker:ci-red off its own corpse — then re-affirms it every cadence.
  # Proven on crew#227: every real check green, the only red rollup entry
  # the sweep's own displaced run. So drop every entry belonging to
  # $SELF_WORKFLOW before the newest-per-context collapse. Accepted
  # consequences: a rollup of ONLY self entries scores NONE (honestly: no
  # checks — never SUCCESS), and a genuine reconcile failure surfaces on the
  # Actions tab instead of as blocker:ci-red, which is right because no PR
  # edit can fix the label machinery. An empty $self filters nothing — the
  # exclusion must never widen into dropping entries on a guess.
  jq -r --arg self "$SELF_WORKFLOW" '
    if (has("statusCheckRollup") | not) then "UNREADABLE" else

    # NEUTRAL and SKIPPED satisfy branch protection — a skipped required check
    # is not a failed one, and path-filtered jobs skip constantly here.
    ["SUCCESS", "NEUTRAL", "SKIPPED"] as $passing
    # "" covers a StatusContext still reported with no state at all.
    | ["", "PENDING", "IN_PROGRESS", "QUEUED", "WAITING", "REQUESTED", "EXPECTED"] as $waiting

    # A re-run does not evict the run it superseded — the rollup keeps both.
    # This PR proved it: its own tip carried a CANCELLED `scope` (15:19:39)
    # beside the SUCCESS `scope` (15:19:45) that replaced it, same workflow.
    # Once CANCELLED blocks, judging every entry would strand this very PR in
    # needs-rebase forever, so collapse each context to its newest entry first.
    # Key on workflow + name because a bare job name is only unique within its
    # workflow.
    #
    # Dating a run is the subtle part, and getting it wrong restores the bug.
    # A run still in flight has no completion, but `gh` does not omit the
    # field: its Go struct marshals the zero time as "0001-01-01T00:00:00Z",
    # which is a string, so `//` will not fall through it. Ordering on
    # completion therefore sorted the LIVE re-run to the bottom and let `last`
    # pick the very run it superseded — reporting the old SUCCESS while a
    # replacement was still running, which is #136 again.
    #
    # So: date a run by when it BEGAN, discarding both spellings of absent
    # (null, and the zero sentinel) and falling back only if it never recorded
    # a beginning. NOT by the newest stamp of any kind: `max` compares the
    # completion of a finished run against the start of a live one, which are
    # different quantities and not an ordering on runs. A run cancelled by the
    # concurrency group does not stop the instant its replacement starts — the
    # runner has to wind down — so predecessor.completedAt > successor.startedAt
    # is the ordinary case, and `max` dated the dead predecessor newer than the
    # live run that replaced it, narrowing both failures above without closing
    # them. The list is already in preference order, so `first` IS that rule.
    #
    # An entry that carries no usable timestamp at all sorts LAST rather than
    # first — something we cannot date is most likely the thing just created,
    # and treating it as newest keeps an undateable in-flight run from being
    # discarded in favour of a stale success. Every ambiguity resolves toward
    # "not settled".
    # The #208 exclusion (header above): self entries leave the rollup here,
    # BEFORE the group_by — a self-only context must vanish entirely, never
    # survive as an all-cancelled context that still classifies FAILURE.
    | [ (.statusCheckRollup // [])[]
        | select($self == "" or (.workflowName // "") != $self)
        | { ctx: [.workflowName // "", .name // .context // ""],
            at:  ([.startedAt, .createdAt, .completedAt]
                  | map(select(type == "string" and . != ""
                               and (startswith("0001-01-01") | not)))
                  | first // ""),
            outcome: ((.conclusion // .state // "") | ascii_upcase) } ]
    # The #139 carve-out (header above): drop CANCELLED entries only when the
    # context keeps a non-cancelled survivor — BEFORE the sort, so a cancelled
    # entry that arrived newest cannot outvote the real verdict it displaced.
    # An all-cancelled context is left intact and still classifies FAILURE.
    | group_by(.ctx)
    | map( map(select(.outcome != "CANCELLED")) as $live
           | (if ($live | length) > 0 then $live else . end)
           | sort_by([(.at == ""), .at]) | last | .outcome ) as $latest

    | if   ($latest | length) == 0                            then "NONE"
      elif (($latest - $passing - $waiting) | length) > 0     then "FAILURE"
      elif (($latest - $passing) | length) > 0                then "PENDING"
      else "SUCCESS" end

    end'
}

bot_verdict() { # $1 = login → MISSING | BLOCK | APPROVE | STALE | FEEDBACK
  local review state commit
  review="$(jq -c --arg u "$1" \
    '[.[] | select(.user.login == $u)] | sort_by(.submitted_at) | last // empty' \
    <<<"$REVIEWS_JSON")"
  if [ -z "$review" ]; then echo MISSING; return; fi
  state="$(jq -r '.state' <<<"$review")"
  commit="$(jq -r '.commit_id' <<<"$review")"
  case "$state" in
    CHANGES_REQUESTED)
      # blocks at ANY head — GitHub's own semantic: only a newer review
      # from the same reviewer clears it
      echo BLOCK ;;
    APPROVED)
      if [ "$commit" = "$HEAD_SHA" ]; then echo APPROVE; else echo STALE; fi ;;
    *)
      # COMMENTED and anything else: a non-verdict. The machine does not
      # read bodies — if the comment is really an agreement, the AUTHOR
      # says so by requesting the human's review.
      echo FEEDBACK ;;
  esac
}

iso_epoch() { # $1 = ISO-8601 timestamp → epoch seconds; nothing, rc 1, when unreadable
  # An absent field reaches this as the empty string or as jq's literal "null";
  # both are "we did not read a time", and neither may be graded as one.
  local at="${1-}" epoch
  case "$at" in "" | null) return 1 ;; esac
  epoch="$(date -d "$at" +%s 2>/dev/null)" || return 1
  [ -n "$epoch" ] || return 1
  printf '%s\n' "$epoch"
}

unrequested_quiescent() { # 0 when the unrequested facts have stood for the grace (#236 D2)
  # The stall blocker's supporting facts are the head and the round's newest
  # submitted review: the ask it demands is owed only once both have stopped
  # moving. Measured off those timestamps, not off sweep memory — ceremony#235
  # was flagged inside the ~90 seconds between a round-answer push and the
  # author's re-request, because a sweep read the facts before the request
  # landed and wrote after it. That is a round in motion, not a dropped ball.
  #
  # "Newest submitted review" is any submitted review, COMMENTED included: a
  # non-verdict is still evidence the round is live, and counting it can only
  # delay a flag, never invent one.
  #
  # A timestamp we could not read refuses the blocker (the standing rule: an
  # unreadable fact never invents a verdict). This direction is deliberate and
  # asymmetric — a missed flag costs one sweep of the 15-minute cadence, a
  # false one flags a builder for doing exactly what BUILDER.md requires.
  local newest verdict_at verdict_epoch
  newest="$(iso_epoch "${HEAD_COMMIT_AT:-}")" || return 1
  verdict_at="$(jq -r '[.[].submitted_at] | max // empty' <<<"${REVIEWS_JSON:-[]}")"
  if [ -n "$verdict_at" ]; then
    # A round WITH verdicts whose newest one cannot be dated is unreadable, not
    # quiescent; a round with no verdicts at all is simply the head's clock.
    verdict_epoch="$(iso_epoch "$verdict_at")" || return 1
    [ "$verdict_epoch" -gt "$newest" ] && newest="$verdict_epoch"
  fi
  [ $((${NOW:-0} - newest)) -ge "$RECONCILE_UNREQUESTED_GRACE" ]
}

human_request_needed() { # 0 when needs-human requires a FRESH human request
  # already requested → the handoff is live; head-current human approval →
  # nothing left to ask. Anything else (never reviewed, an old comment, an
  # approval of an older head) stalls the handoff unless we request —
  # guarding on "has the human ever reviewed" wedged exactly that way.
  if requested "$HUMAN"; then return 1; fi
  if [ "$(bot_verdict "$HUMAN")" = APPROVE ]; then return 1; fi
  return 0
}

blockers() { # → the blocker:* labels this PR should carry, one per line
  # The second axis. These are FACTS ABOUT THE BRANCH, and they are mutually
  # independent — a PR can be conflicted and red and unasked at once — so they
  # are a set, not an ordering. That is the whole point of splitting them out
  # of state:*: every precedence bug this machine has had (needs-human
  # surviving a conflict, MISSING swallowing STALE) came from projecting
  # independent facts onto one totally-ordered label. A set has no precedence
  # to get wrong.
  #
  # UNKNOWN mergeability is deliberately NOT a conflict: GitHub reports it for
  # about a minute after every merge while it recomputes, and flapping every
  # open PR on each merge would be worse than the bug. Same for a failed read
  # of either fact — both default to the "do not know" value, which blocks
  # nothing. An unset global (an older fixture, a failed fetch) must never
  # invent a verdict it did not read.
  case "${MERGEABLE:-UNKNOWN}" in CONFLICTING) echo blocker:conflict ;; esac
  case "${CHECKS:-NONE}" in FAILURE) echo blocker:ci-red ;; esac

  # Nobody is on the hook for a verdict somebody still owes. Distinct from
  # bots-reviewing, which says a request is live and an answer is coming:
  # here the round is stalled because no one was ever asked, and the board
  # said "waiting on the bots" for the 48h it took `stale` to notice.
  # A draft is exempt (the bots ignore drafts by design), and so is an
  # explicit human request — a maintainer claiming a PR early is deliberate,
  # not a dropped ball.
  #
  # And so is a head whose checks have not answered yet (#236 D1). This is the
  # one blocker that names an act the author must PERFORM, so it is the one
  # that has to know when performing it is permitted: BUILDER.md's review round
  # requires a green check at the head before requesting, so a builder waiting
  # out a pending run is complying, and flagging compliance teaches its readers
  # to ignore the label. Both 2026-08-03 instances were exactly that —
  # crew#318 at ~12:44Z carried state:addressing + blocker:unrequested while
  # the head's run was IN_PROGRESS, and ceremony#235 at 12:30Z caught the
  # ~90-second gap between a round-answer push and the re-request.
  #
  # PENDING and FAILURE each already have an owner, which is why gating loses
  # no coverage: on PENDING the next move is CI's and state:addressing /
  # state:bots-reviewing already say what the PR is doing; on FAILURE
  # blocker:ci-red owns that head, and stacking a second blocker on it
  # double-flags one stall. NONE joins SUCCESS because no checks configured is
  # nothing to wait for — the same reading the request rule gives the builder.
  # UNREADABLE never arrives here: the caller skips the PR before deciding.
  local checks_permit_the_ask=false
  case "${CHECKS:-NONE}" in SUCCESS | NONE) checks_permit_the_ask=true ;; esac
  if [ "$DRAFT" != true ] && [ "$checks_permit_the_ask" = true ] && ! requested "$HUMAN"; then
    local b v owed=false any_requested=false
    for b in "${REQUIRED_BOTS[@]}"; do
      requested "$b" && any_requested=true
      # MISSING and STALE are both verdicts this head does not have: nobody
      # reviewed it, or everybody reviewed something else. The agent owes an
      # ask either way — the stale round is if anything the worse of the two,
      # since it has approvals on the page that no longer describe the tree.
      v="$(bot_verdict "$b")"
      case "$v" in MISSING | STALE) owed=true ;; esac
    done
    # The quiescence grace (#236 D2) is the last question, after the debt is
    # established: it asks whether the debt has stood long enough to be a
    # dropped ball rather than a round still in motion.
    if [ "$owed" = true ] && [ "$any_requested" = false ] && unrequested_quiescent; then
      echo blocker:unrequested
    fi
  fi
}

round_outranks_draft() { # 0 when the round's standing word survives a re-draft (#205)
  # A standing non-approving verdict outranks draft: a PR that took a round,
  # carries CHANGES_REQUESTED (or a comment owed a reply, or approvals a push
  # staled), and is then converted back to draft is a fix round in progress,
  # not a build — and hiding it behind state:building is a dropped ball the
  # staleness sweep reads as work in progress. Approvals do NOT outrank
  # draft: a re-draft after a passed round is deliberately building again,
  # and a draft must never read state:needs-human.
  #
  # A LIVE panel request on a draft also falls through — deliberately
  # surfaced, not absorbed (#205's must-not-paper-over): the bots ignore
  # drafts by design, so a draft wearing state:bots-reviewing on the board
  # is the visible symptom of a real defect (a request nobody cleared at
  # round close, or a hand-requested draft), and reading it as building
  # would hide exactly that.
  local b
  for b in "${REQUIRED_BOTS[@]}"; do
    requested "$b" && return 0
    case "$(bot_verdict "$b")" in BLOCK | FEEDBACK | STALE) return 0 ;; esac
  done
  [ "$(bot_verdict "$HUMAN")" = BLOCK ]
}

decide_state() { # → the one state:* label this PR should carry
  # Draft decides the state only when the round implies nothing else (#205):
  # a draft with no round history reads state:building exactly as it always
  # has, and round_outranks_draft is what "nothing else" means.
  if [ "$DRAFT" = true ] && ! round_outranks_draft; then
    echo state:building
    return
  fi

  local s
  s="$(round_state)"

  # A draft disqualifies needs-human unconditionally (#205, round 1): with
  # the short-circuit above now conditional, a draft carrying a live human
  # request plus a standing bot block or comment fell through to
  # round_state, whose explicit-human-request precedence sits above the
  # BLOCK/FEEDBACK cases — and GitHub cannot merge a draft at all, so
  # "a human could merge this right now" would lie no matter what the
  # round says. state:addressing is the same honest landing the blocker/
  # needs-ruling/blocked clauses below use: the round's word stands, only
  # the mergeable-now claim is off the table while the PR is a draft.
  if [ "$s" = state:needs-human ] && [ "$DRAFT" = true ]; then
    echo state:addressing
    return
  fi

  # The one rule joining the two axes: state:needs-human means a human could
  # merge this RIGHT NOW, so it requires a clear branch. Any blocker at all
  # means the work is the agent's — whatever the review round says — and the
  # blocker label says which work it is. Nothing else in this function reads
  # the branch, which is what keeps the ordering below purely about reviews.
  if [ "$s" = state:needs-human ] && [ -n "$(blockers)" ]; then
    echo state:addressing; return
  fi

  # A pending ruling disqualifies needs-human the same way (#51): while
  # `needs-ruling` is up, the human's turn lives in the THREAD — the flag
  # marks it — and "mergeable right now" must not read true beside an open
  # decision. state:addressing is the honest landing because the ball ON THE
  # PR is the builder's: the flag-setter judges when agreement is reached and
  # carries the ruling in (#50 D6). The label is deliberately NOT in BLOCKERS:
  # that array is machine-owned, and the converge loop strips every entry the
  # current facts do not re-derive — `needs-ruling` is hand-set intent the
  # machine reads and never writes (#50 D9), so parking it there would strip
  # a live escalation on the next 15-minute tick.
  if [ "$s" = state:needs-human ] && has_label needs-ruling; then
    echo state:addressing; return
  fi

  # A directed hold disqualifies it the same way (#180): `blocked` is hand-set
  # intent — triage sets it, anyone may correct it — and during the #111
  # freeze rig#126/#128 carried it beside state:needs-human, so the board said
  # "mergeable right now" about PRs a hold said must not merge (rig#126 was
  # merged seven minutes later). Not a BLOCKERS entry, deliberately: that
  # array is machine-owned and the converge loop strips whatever the facts do
  # not re-derive, so emitting the label there would strip a live hold on the
  # next 15-minute tick — the same trap #51 names for `needs-ruling`.
  # state:addressing is the accepted imprecision: under a hold the builder
  # owes nothing, but "a human could merge this now" must not lie.
  if [ "$s" = state:needs-human ] && has_label blocked; then
    echo state:addressing; return
  fi
  echo "$s"
}

round_state() { # → the state the REVIEW ROUND alone implies; knows no branch facts
  local b verdicts=""
  for b in "${REQUIRED_BOTS[@]}"; do
    if requested "$b"; then echo state:bots-reviewing; return; fi
  done
  # Collect the WHOLE round before applying any precedence. Deciding inside
  # the loop let BOTS order pick the winner: a MISSING returned immediately,
  # so a STALE belonging to a later bot was never even read, and the mixed
  # round (one approval staled by a push, another bot yet to review) came out
  # needs-human — the #136 headline shape, with zero reviews bound to the head.
  for b in "${REQUIRED_BOTS[@]}"; do
    verdicts="$verdicts $(bot_verdict "$b")"
  done
  case "$verdicts" in
    # STALE = a verdict for an older head. Unlike MISSING, this outranks the
    # human request: every approval it covers was invalidated by a push, so
    # NOBODY has reviewed this tree. Handing that to the human is the #136 case
    # where everything reads green — mergeable, CI passing, "waiting on the
    # human" — over code no reviewer has seen. The agent owes a re-request.
    # Checked before MISSING because "unfinished" must not swallow "and also
    # stale": a round that is both is a push that outran the re-requests, not
    # a maintainer deliberately claiming the PR early.
    *STALE*) echo state:addressing; return ;;
  esac
  case "$verdicts" in
    # No verdict at all from some bot, and nothing staled. An explicit human
    # request still outranks an unfinished round — a maintainer pulling a PR
    # to themselves early is a deliberate act, and the original precedence.
    #
    # Otherwise it is the AGENT's ball, not the bots'. The loop above already
    # returned for every live bot request, so reaching here with a MISSING
    # means somebody owes a verdict and nobody was asked for one — the round
    # is not running. Calling that bots-reviewing was the lie that let a
    # forgotten PR read "waiting on the reviewers" for the 48h it took the
    # stale sweep to notice. blocker:unrequested says why.
    *MISSING*)
      if requested "$HUMAN"; then echo state:needs-human; return; fi
      echo state:addressing; return ;;
  esac
  # an explicit human request outranks the remaining bot outcomes — it is the
  # final gate, and a maintainer pulling a PR to themselves early counts too
  if requested "$HUMAN"; then echo state:needs-human; return; fi
  case "$verdicts" in
    # FEEDBACK = a comment with no verdict → the agent owes the round-reply.
    *BLOCK* | *FEEDBACK*) echo state:addressing; return ;;
  esac
  # the bots all approve — but if the human's standing word is
  # changes-requested (and nobody re-requested them yet), the agent owes
  # fixes, not the human a nag
  if [ "$(bot_verdict "$HUMAN")" = BLOCK ]; then
    echo state:addressing
  else
    echo state:needs-human
  fi
}

# ---------------------------------------------------------------------------
# The sweep: fetch facts, decide, converge. One PR's failure never aborts the
# others — each PR reconciles in a subshell and a failure just logs.
# ---------------------------------------------------------------------------

core_label_rows() {
  cat <<'EOF'
state:building|FBCA04|Pre-round: the builder is still building — draft is evidence for it, not the definition
state:bots-reviewing|1D76DB|Waiting on the bot reviewers to finish the round
state:addressing|D93F0B|All bots reviewed — coding agent owes the single reply + fixes
state:needs-human|8250DF|No blockers, all bots approve — waiting on the human reviewer
blocker:conflict|B60205|Does not merge — the branch conflicts and the agent owes a rebase
blocker:ci-red|B60205|A check is failing — the agent owes a fix (not a rebase)
blocker:unrequested|E99695|Somebody still owes a verdict and nobody was asked for one
merge-next|0E8A16|Head of the merge queue — merge this one next (set by hand/agent, cleared here)
stale|B60205|No activity for 48h — needs a poke (sweep-managed)
blocked|6A737D|Waiting on another PR or issue to land first
offsite|CFD3D7|Issue deliverable is a PR in another repository — claim clock paused
needs-ruling|D4C5F9|A human decision is pending — question, options and a recommendation are in the comment
attention|D93F0B|A demand is parked here for the assignee: pick up the thread, ack by removing this label
release|0E8A16|Release flow and version/packaging work
needs-triage|FBCA04|Did not come through triage — owes normalization or conversion to a discussion
ready|0E8A16|Triaged, spec complete, unblocked — a builder can start now and succeed
claimed|1D76DB|A builder owns it: assignee set, draft PR expected shortly
post-merge|006B75|Refs-linked PR merged; post-merge criteria remain and triage owns completion
epic|5319E7|Organizes other issues via a dependency-ordered task list — builders never pick it
EOF
}

retired_label_names() { # the GitHub defaults LABELS.md retires — a `question` is a discussion
  # One registry, kept beside core_label_rows() for the same reason those rows
  # are not in labels.conf: a rule that must hold in every governed repo
  # cannot live in a per-repo file. The six names match LABELS.md exactly.
  cat <<'EOF'
duplicate
invalid
question
wontfix
help wanted
good first issue
EOF
}

bootstrap_labels() { # dispatch-only: ~20 upserts is too chatty for every cron tick
  local rows name
  rows="$(core_label_rows)"
  if [ -f "$LABELS_CONF" ]; then
    rows="$rows
$(configured_label_rows "$LABELS_CONF")"
  fi
  while IFS='|' read -r name color desc; do
    [ -n "$name" ] || continue
    run forge_label_create "$name" "$color" "$desc"
  done <<<"$rows"

  # LABELS.md publishes the defaults as deleted at bootstrap; until #93
  # nothing deleted them — incubator's first dispatch ran green and left
  # `good first issue` standing. Deletion is dispatch-only like the upserts,
  # and never fatal: `gh label delete` exits non-zero on a label that is
  # already gone, the NORMAL case from the second dispatch on, and under
  # set -e an unguarded call aborts the whole run (#91's shape). A 403
  # refusal gets the same tolerance — the bot bootstrap already 403s on
  # blocker:drill-pending, and a token that cannot delete must still get
  # the taxonomy it can create. Either way: log the name, keep going.
  while IFS= read -r name; do
    [ -n "$name" ] || continue
    run forge_label_delete "$name" \
      || log "retire: '$name' not deleted (already absent, or refused) — continuing"
  done <<<"$(retired_label_names)"
}

has_label() { grep -qxF "$1" <<<"$LABELS"; }

release_shape_warning() { # $1 = PR, $2 = head version, $3 = base version
  # The #128 incident's guard (#130): a release-shaped PR — bare X.Y.Z at
  # its head where the base says something else — reaching the board with
  # no `release` label is exactly the state whose merge would publish
  # nothing, so the sweep says so instead of letting the merge door
  # discover it. A WARNING, never a write: `release` is declared intent,
  # and the reconciler does not guess intent (LABELS.md's rule for
  # `blocked`/`release`). An unreadable version blocks nothing — the
  # sweep must not nag on facts it did not read.
  local n="$1" head_ver="$2" base_ver="$3"
  [ -n "$head_ver" ] || return 0
  grep -qE '^[0-9]+\.[0-9]+\.[0-9]+$' <<<"$head_ver" || return 0
  [ "$head_ver" != "$base_ver" ] || return 0
  echo "::warning::labels: #$n is release-shaped (version ${base_ver:-unreadable} -> $head_ver at its head) but carries no release label — the merge door reads that label as declared intent and will refuse without it; if this is the ceremony PR, apply release (#130; the #128 incident)"
}

tree_version() { # $1 = ref → that tree's version via the API, or nothing
  # Both backends, no checkout: a VERSION file first, package.json's
  # version field second (jq, not node — a read needs no npm machinery).
  # Every failure path prints nothing: the caller treats "could not read"
  # as "not release-shaped" rather than warning on a guess.
  local ref="$1" ver
  ver="$(forge_api "repos/$REPO/contents/VERSION?ref=$ref" --jq '.content' 2>/dev/null \
    | base64 -d 2>/dev/null | tr -d '[:space:]')"
  if [ -z "$ver" ]; then
    ver="$(forge_api "repos/$REPO/contents/package.json?ref=$ref" --jq '.content' 2>/dev/null \
      | base64 -d 2>/dev/null | jq -r '.version // empty' 2>/dev/null)"
  fi
  [ -z "$ver" ] || printf '%s\n' "$ver"
  return 0
}

# label_write <n> <args…> — every label mutation on this surface goes through
# here (#192). A write that did not happen must reach main's exit code, and the
# first version of this fix marked only the primary state edit: clearing
# `merge-next` and the two `stale` edits could still fail into the generic
# per-PR branch and finish with `reconciled.` and exit 0
# (@codex-reviewer-andresmgsl). One helper means a future call site cannot
# reopen that by forgetting to mark itself.
#
# The marker is a log line rather than a return code because reconcile_pr runs
# in a subshell whose STDOUT main reads — the same channel the degraded-read
# warning already travels on.
label_write() {
  local n="$1"
  shift
  if run forge_issue_edit "$n" "$@" >/dev/null; then
    return 0
  fi
  log "#$n: label edit FAILED — attempted: forge_issue_edit $n $*; the write did not happen (reason on stderr above)"
  return 1
}

reconcile_pr() { # $1 = PR number; relies on the globals set from its fetch
  local n="$1" desired remove s args last_activity last_activity_epoch age

  desired="$(decide_state)"

  # encode the runbook's last step for the no-judgment case: every required
  # verdict is a head-current approval → the human is asked, once. The guard
  # asks whether
  # a FRESH human review is needed for THIS head — never "has the human ever
  # reviewed", which wedged the handoff after any earlier human comment.
  # Idempotent (a live request suppresses it); race-free via the shared
  # concurrency group in labels.yml. With a comment-only bot on the panel
  # this path stays cold and the AUTHOR requests the human.
  if [ "$desired" = state:needs-human ] && human_request_needed; then
    run forge_request_reviewer "$n" "$HUMAN"
    log "#$n: requested $HUMAN (round passed)"
  fi

  # ---- converge both axes ----
  # state:* is exclusive (everything but $desired comes off); blocker:* is a
  # set (each one on or off on its own); RETIRED always comes off. One edit
  # call for all of it, so a PR never flickers through a half-applied board.
  local want_blockers add=""
  want_blockers="$(blockers)"

  remove=""
  for s in "${STATES[@]}"; do
    if [ "$s" != "$desired" ] && has_label "$s"; then remove="$remove,$s"; fi
  done
  for s in "${RETIRED[@]}"; do
    if has_label "$s"; then remove="$remove,$s"; fi
  done
  for s in "${BLOCKERS[@]}"; do
    if grep -qxF "$s" <<<"$want_blockers"; then
      has_label "$s" || add="$add,$s"
    else
      has_label "$s" && remove="$remove,$s"
    fi
  done
  add="${add#,}"
  remove="${remove#,}"

  # Never NAME a label the repo does not have. `gh issue edit --add-label`
  # rejects the WHOLE call on one unknown name — nothing is applied — so a
  # single missing blocker would take the state convergence down with it, on
  # exactly the PRs this change exists to fix, surfacing only as a log line.
  # Batching state and blockers into one edit for anti-flicker is what widened
  # that blast radius; filtering the add side is what closes it again.
  # Removals need no filter: they are built from has_label, so the label
  # provably exists. REPO_LABELS unreadable means no filtering rather than
  # filtering everything out — a failed read must not silently strip the board.
  local skip_edit=false
  if [ -n "${REPO_LABELS:-}" ]; then
    local kept="" missing="" want
    for want in ${add//,/ }; do
      if grep -qxF "$want" <<<"$REPO_LABELS"; then kept="$kept,$want"
      else missing="$missing $want"; fi
    done
    add="${kept#,}"
    # A missing STATE label skips only the EDIT — never the rest of this
    # function. Everything below is independent of the state:* taxonomy, and
    # returning here stranded it: `merge-next` kept claiming "merge this one
    # next" on a PR the board had moved to the agent, and the stale sweep
    # stopped running. That is the original false-invitation bug, reintroduced
    # in the very fix meant to survive a cold-start repo — and a regression
    # against the old behaviour, which failed the edit and fell through.
    if ! grep -qxF "$desired" <<<"$REPO_LABELS"; then
      log "#$n: WARNING: state label '$desired' does not exist — skipping the label edit; dispatch the workflow to bootstrap"
      skip_edit=true
    elif [ -n "$missing" ]; then
      log "#$n: WARNING: missing label(s)$missing — state still converged; dispatch the workflow to bootstrap"
    fi
  fi
  if [ "$skip_edit" = false ] && { ! has_label "$desired" || [ -n "$remove" ] || [ -n "$add" ]; }; then
    args=(--add-label "$desired${add:+,$add}")
    [ -n "$remove" ] && args+=(--remove-label "$remove")
    if label_write "$n" "${args[@]}"; then
      log "#$n: state -> $desired${add:+ +$add}${remove:+ (cleared $remove)}"
    else
      # A WRITE THAT DID NOT HAPPEN IS FATAL, not a warning (#192). This was
      # `log WARNING` and fell through, so the sweep printed `reconciled.` and
      # exited green over an edit the forge had refused — the
      # degraded-write-reports-success class #188 exists to eliminate,
      # surviving inside the reconciler that reports it.
      #
      # The old text also diagnosed a cause it had not established: it named a
      # missing label and told the operator to bootstrap, when the label was
      # present and the call had returned 500. #101's rule is report, do not
      # diagnose — so this says what was attempted and that it did not happen,
      # and leaves the backend's own stderr to say why.
      return 1
    fi
  fi

  # ---- the release-shape guard (#130): a warning, never a write --------
  # Drafts are exempt (the build phase is the builder's); the version
  # reads cost two API calls and only on PRs missing the label.
  if [ "$DRAFT" != true ] && ! has_label release; then
    release_shape_warning "$n" "$(tree_version "$HEAD_SHA")" "$(tree_version "$BASE_SHA")"
  fi

  # ---- merge-next: cleared, never set ----------------------------------
  # Queue order is INTENT — which PR should land first is a judgement about
  # conflicts and dependencies that GitHub knows nothing about, so the
  # reconciler must not guess it (LABELS.md's rule for `blocked`/`release`).
  # What it CAN do is stop the label going stale the way needs-human did:
  # the moment the PR is no longer the thing a human should merge next, the
  # claim is removed. Setting it stays with whoever owns the queue.
  if has_label merge-next && [ "$desired" != state:needs-human ]; then
    label_write "$n" --remove-label merge-next || return 1
    log "#$n: cleared merge-next (state is $desired, not mergeable-by-a-human)"
  fi

  # ---- stale: real activity only, and blocked is legitimately quiet ----
  # forge_pr_activity owns the portable half: issue comments + commits +
  # inline review comments. The flat /pulls/{n}/comments endpoint 404s on
  # Forgejo; the forgejo backend re-derives it from reviews with
  # comments_count > 0 (#188 / #4844). PR created_at and review submitted_at
  # stay here — they are already in hand and need no second fetch.
  last_activity="$(
    {
      jq -r '.created_at' <<<"$PR_JSON"
      jq -r '.[].submitted_at // empty' <<<"$REVIEWS_JSON"
      # Non-fatal degrade (pre-#188 same edge), but do NOT swallow stderr —
      # forge_api names failures loudly, and hiding them re-opens this issue's
      # silent-green class (#4879 / #101 D5).
      forge_pr_activity "$n" || true
    } | sort | tail -n1
  )"
  last_activity_epoch="$(date -d "$last_activity" +%s)"
  age=$((NOW - last_activity_epoch))
  # needs-ruling joins blocked here: waiting on a human is legitimately quiet
  # (#50 D10). The 7-day nudge is #52's, once for both surfaces.
  if has_label blocked || has_label needs-ruling || [ "$age" -le "$STALE_AFTER" ]; then
    if has_label stale; then
      label_write "$n" --remove-label stale || return 1
      log "#$n: unstale"
    fi
  elif ! has_label stale; then
    label_write "$n" --add-label stale || return 1
    log "#$n: stale ($((age / 3600))h quiet)"
  fi

  # ---- the ruling invariants (#52): the bare-flag check + the 7-day nudge --
  # The stale EXEMPTION above is #51's; these are the sweep halves that ride
  # the same real-activity computation (lib/ruling.sh, shared with the issue
  # side). Behind the flag check so flag-free PRs — all of them, almost
  # always — cost no extra API reads.
  if has_label needs-ruling; then
    reconcile_ruling "$n" "$last_activity_epoch" "$NOW"
  fi

  # `attention` belongs on the assigned issue that owns the claim, never on
  # a pull request (#232). Behind the label gate so ordinary PRs pay no read.
  if has_label attention; then
    reconcile_attention "$n" pr "$(jq '.assignees | length' <<<"$PR_JSON")" ""
  fi
}

main() {
  # BEFORE anything reads the board (#188). Every call site below is still
  # `gh`, so that is what this declares — honestly, which is the point: on
  # a Forgejo consumer the preflight refuses here instead of letting the
  # sweep run blind and print "reconciled." over zero PRs (rig run 979).
  # The forge is decided once, here, before anything reads the board, and
  # the backend that can speak it is loaded (#188). The CEREMONY_FORGE_CLIENT
  # wrapper that stood here died with the call-site port: it declared "this
  # code uses gh", which stopped being true the moment every site went
  # through the shim, and leaving it would have defaulted the forgejo path
  # into the very client its own preflight refuses.
  forge_preflight || return 1
  # "" means decide from the environment; forge_select takes an explicit
  # forge only in tests.
  forge_select "" || return 1

  REPO="${REPO:?set REPO to owner/name}"
  LABELS_CONF="${LABELS_CONF:-.github/labels.conf}"
  load_config "$LABELS_CONF"
  NOW="$(date +%s)"

  if [ "${GITHUB_EVENT_NAME:-}" = workflow_dispatch ]; then
    log "workflow_dispatch: bootstrapping the taxonomy"
    bootstrap_labels
  fi

  # The repo's label set, read ONCE per sweep — reconcile_pr filters every
  # add against it, because one unknown name fails the whole edit call.
  REPO_LABELS="$(forge_label_list 2>/dev/null || echo "")"
  [ -z "$REPO_LABELS" ] && log "WARNING: could not read the label set — applying labels unfiltered"
  missing_core_labels_warning "$(core_label_rows)" "$REPO_LABELS"

  local n output status total=0 unreadable=0 write_failures=0 sampled_reason=""
  while IFS= read -r n; do
    [ -n "$n" ] || continue
    total=$((total + 1))
    status=0
    output="$(
      (
      PR_JSON="$(forge_api "repos/$REPO/pulls/$n")"
      DRAFT="$(jq -r '.draft' <<<"$PR_JSON")"
      AUTHOR="$(jq -r '.user.login' <<<"$PR_JSON")"
      set_required_bots "$AUTHOR"
      HEAD_SHA="$(jq -r '.head.sha' <<<"$PR_JSON")"
      BASE_SHA="$(jq -r '.base.sha' <<<"$PR_JSON")"
      LABELS="$(jq -r '.labels[].name' <<<"$PR_JSON")"
      # PENDING reviews are unsubmitted drafts in someone's browser — not a verdict
      REVIEWS_JSON="$(forge_api --paginate "repos/$REPO/pulls/$n/reviews" --jq '.[]' \
        | jq -s '[.[] | select(.state != "PENDING")]')"
      # Read AFTER the reviews, because the raw field is not portable: Forgejo
      # never clears it, so it is intersected with who still owes a verdict on
      # this head (#188 term 4). A no-op on GitHub, which clears it itself.
      REQUESTED="$(outstanding_requests "$(jq -r '.requested_reviewers[].login' <<<"$PR_JSON")")"
      # mergeability + the check rollup, the two facts the state machine was
      # blind to (#136). `gh pr view` rather than the REST PR object: the API's
      # `mergeable` is a tri-state boolean that GitHub computes lazily, while
      # this returns the same MERGEABLE/CONFLICTING/UNKNOWN string the UI shows.
      # Failure to read them is NOT fatal and NOT treated as broken — an API
      # hiccup must never flap every PR into needs-rebase, so both degrade to
      # the "do not know" value that triggers nothing.
      # The WHY goes to gh's stderr, and 2>/dev/null threw it away — a
      # permanent denial and a network hiccup left byte-identical evidence,
      # and #95 had to infer a cause from a control case instead of reading
      # it off a run (wrongly, it turned out). Captured into a file (#101
      # D2), never left to interleave raw into the per-PR output block,
      # where an unlucky line could collide with a matched string.
      GH_VIEW_ERR_FILE="$(mktemp)"
      GH_VIEW="$(forge_pr_view "$n" 2>"$GH_VIEW_ERR_FILE" || echo '{}')"
      GH_VIEW_ERR="$(cat "$GH_VIEW_ERR_FILE")"
      rm -f "$GH_VIEW_ERR_FILE"
      MERGEABLE="$(jq -r '.mergeable // "UNKNOWN"' <<<"$GH_VIEW")"
      CHECKS="$(checks_state <<<"$GH_VIEW")"
      # Read failed: leave this PR exactly as it is. Recomputing on facts we
      # did not read is how an API hiccup turns into a false "merge me" —
      # and the next tick is 15 minutes away, not 15 hours.
      if [ "$CHECKS" = UNREADABLE ]; then
        # Two lines on purpose (#101 D1): the sweep detects a wholly blind
        # pass by whole-line-matching the counted line below, so the reason
        # rides its OWN line — folding it in would silently break the
        # `unreadable` counter and the wholly-blind warning #96 landed.
        log "#$n: could not read mergeability/checks — left alone this pass"
        log "#$n: read failed: $(read_failure_reason "$GH_VIEW_ERR")"
        exit 0
      fi
      # The head's own clock, for the blocker:unrequested grace (#236 D2). One
      # read, pinned to the head SHA — not `gh pr view --json commits`, which
      # asks for the FIRST hundred commits and would date a longer PR by a
      # commit that is not its head. Last of the fetches on purpose: a PR the
      # skip above walked away from must not pay for it, and neither do drafts,
      # which never reach that blocker. Empty (a failed read, or a body without
      # the field) leaves the blocker unjudged, by unrequested_quiescent.
      HEAD_COMMIT_AT=""
      if [ "$DRAFT" != true ]; then
        HEAD_COMMIT_ERR_FILE="$(mktemp)"
        HEAD_COMMIT_AT="$(forge_api "repos/$REPO/commits/$HEAD_SHA" \
          --jq '.commit.committer.date' 2>"$HEAD_COMMIT_ERR_FILE" || echo "")"
        HEAD_COMMIT_ERR="$(cat "$HEAD_COMMIT_ERR_FILE")"
        rm -f "$HEAD_COMMIT_ERR_FILE"
        case "$HEAD_COMMIT_AT" in
          "" | null)
            # Say why it degraded (#101 D2/D4), on its own line: this one
            # narrows a blocker rather than skipping the PR, so it must not
            # read as the wholly-blind shape the counted line above matches.
            HEAD_COMMIT_AT=""
            log "#$n: could not read the head commit's date: $(read_failure_reason "$HEAD_COMMIT_ERR") — blocker:unrequested not judged this pass" ;;
        esac
      fi
      reconcile_pr "$n"
      ) 2>&1
    )" || status=$?
    [ -n "$output" ] && printf '%s\n' "$output"
    if grep -qxF "labels: #$n: could not read mergeability/checks — left alone this pass" <<<"$output"; then
      unreadable=$((unreadable + 1))
      # the first observed reason stands in for the sweep in the blind warning
      if [ -z "$sampled_reason" ]; then
        sampled_reason="$(sed -n "s/^labels: #$n: read failed: //p" <<<"$output" | head -n1)"
      fi
    elif [ "$status" -ne 0 ]; then
      # The per-PR tolerance is right and stays: one bad PR must not blind the
      # sweep over the rest of the board. What was missing is the sweep-level
      # accounting — a failed WRITE has to reach main's exit code, or a builder
      # satisfies every task and the sweep still prints `reconciled.` over an
      # edit that never happened (#192, @kimi-reviewer-andresmgsl #5189).
      #
      # Reads stay tolerated: an unreadable fact is already reported by the
      # blind-sweep warning and leaves the board untouched. A write is
      # different — the board and the tree now disagree.
      if grep -q "^labels: #$n: label edit FAILED" <<<"$output"; then
        write_failures=$((write_failures + 1))
        log "#$n: reconcile failed on a WRITE — continuing the sweep, but it will not report success"
      else
        log "#$n: reconcile failed — continuing with the remaining PRs"
      fi
    fi
  done < <(forge_pr_list)
  blind_sweep_warning "$unreadable" "$total" "$sampled_reason"
  if [ "$write_failures" -gt 0 ]; then
    # The line must not contain the literal "reconciled." ANYWHERE — "NOT
    # reconciled." still does, and a consumer grepping a job-log tail for that
    # token would find it after a write that did not happen
    # (@codex-reviewer-andresmgsl). The test asserts the whole output is free
    # of it, not merely that the success prefix is absent.
    log "$write_failures label write(s) attempted did not happen — sweep incomplete"
    return 1
  fi
  log "reconciled."
}

# sourced by test/labels-reconcile.sh for the fixture tests; executed in CI
if [ "${BASH_SOURCE[0]}" = "$0" ]; then
  main "$@"
fi