release-manager — every placement into a release is a reasoned ruling (HIMMEL-6796)

For the release-manager session only (HIMMEL_RELEASE_MANAGER=1, armed from
docs/handover/rm-brief-template.md). Every verb is a thin procedure over
jira release; the engine holds the formulas, the signatures and the checks,
so the board, the tick and this skill count the same way. Run the CLI by its
absolute path from the primary checkout:
node <repo>/scripts/jira/dist/index.js release <sub> <R> ….

There are no count gates: a ruling never cites a quota, a cap or a depth
(decision 104). It weighs effort and delivery, and the sector and dependency
chain (decision 100), and names its ground.

Grounds and classes

• Ground (--ground): critical, regression, broken-operator-flow,
  chain:<name>, receipt, defer:<why>, resequence:<why> (a move within
  the release), resolved:<resolution> (a DISCHARGE). A ground explains the
  ruling; it never authorizes a placement by itself.
• Class is derived by the engine from the target version: current (the
  active version), next (a later version of the release), backlog.
• Sector (--sector) is required. A ticket with no sector: ask the chief
  to set it, then rule.
• Emergencies are always current (decision 109): an evidenced emergency is
  placed under the standing record O-<R>-emergency. Record its class and
  sector; only an operator record moves it out.

INTAKE / REVERSION / DISCHARGE — one procedure for every request

The chief writes the request and sends REQ <id>. Then:

1. release claim <R> — claims every open request, and any claimed by an
   older lease epoch. One line per request.

2. Per request, release rule <R> --req <Q-id> --outcome <O> --op <kind> --from <v|backlog> --to <v|backlog> --sector <s> --ground <g> --reason-file <f> [--supersedes <ids>] [--authority operator --ref <O-id>].

   Verb	Outcomes (op)
   INTAKE	ADMIT (add, move), BACKLOG (remove, move, none), AD-HOC (none)
   REVERSION	MOVE (move, remove), KEEP (keep)
   DISCHARGE	ACCEPT (discharge), REJECT (none)

   When a ruling (or the baseline, baseline-<v>-g<n>) is already in effect
   for the key and version, --supersedes must name it; the error names it.
   An operator ruling cites the chief's O- record, which must cover the
   exact key, op and versions; a bare decision number authorizes nothing.

3. release apply <R> <R-id> — writes Jira, reads it back, records applied,
   posts the receipt comment, records receipted, and prints the RULING line.
   Re-run it after any failure: it is idempotent and never posts twice.

4. Send the chief the printed line, verbatim:
   RULING <id> <req> <verb> <key> <op> <from>→<to> <outcome> <class> sector=<s> — <ground>: <reason>.
   Only a receipted ruling is sent.

A refusal is a fact to act on, not to route around: UNKNOWN state, a pending
intent, a missing supersedes, an uncovered authority. Report a refusal you
cannot resolve to the chief as a FINDING.

STATUS

release state <R> --json. Write release/<R>/status-<ISO date>-<HHMM>.md
under the state directory: each input and its freshness, the lease, the open
and claimed requests, the leaks (unauthorized members and ruled-in
non-members), the integrity lines, the burn-down and the trend. Send
STATUS <path>. A state that is UNKNOWN is reported as UNKNOWN, never as a
clean result.

BURN-DOWN

From the same state: per version, baseline open of baseline, admitted open,
emergency open and deferred out; then the open keys grouped by sector, with
the next keys per lane you would dispatch from. Write
release/<R>/burndown.md and send BURN-DOWN <path>. You advise; the
chief dispatches.

READY [v|R]

Every row is PASS, FAIL or UNKNOWN, and UNKNOWN fails:

1. every baseline key of v is terminal or carries an effective move out;
2. every effective in key of v is terminal;
3. no open Critical, Blocker or Highest bug is in (or emergency) for v;
4. no leaks, no open or claimed requests for v, no integrity lines;
5. every required check succeeded on the candidate sha;
6. the fold covers every Done key merged in the tag range;
7. no open PR is linked to an in key of v without a deferral ruling.

Sync the mirror first (mirror), then read the state. Write
release/<R>/ready-<v>-<sha7>.md and send
READY <v|R> <sha> TAG-READY|NOT-READY <path>. The operator tags and judges
stability; a FAIL row is overridden only by an operator record naming it.

Lease and hand-off

• Start: release lease acquire <R> --expect-epoch <n> --doc <your doc> --session <name> (the brief names n), then release claim <R>, then
  release apply every older-epoch intent it lists in the state (a successor
  resolves them by readback; one Jira never saw is voided and its request
  reopened).
• Hand-off (45 % context or HALT): apply or void every intent, release lease release <R>, write <doc stem>-RM-HANDOFF.md (release, epoch, chain heads,
  last STATUS path), append WRAPPED — handoff <path> and send the chief
  WRAPPED <path>.
• You never run release operator, release request, create or
  transition; the engine refuses them for you anyway.