fix(2284): give bugbot-gate an inventory row, so the audit can read the fleet again (backend#2284)

[LukasWodka]

What broke

.github#305 landed bugbot-gate.yml as a reusable — plus 1,485 lines of script,
selftest and mutation suite — and no row in repo-inventory.yml. The org audit on
develop went red 26 seconds after #293 merged:

run 32638444392 · 2026-08-23T12:06:53Z · exit 2

reusable workflow(s) ['bugbot-gate.yml'] exist in ./.github/workflows but are absent
from the inventory's `reusables` list, so they are checked against no repo and reported
by nothing.
Refusing to report on caller drift from an incomplete read.

Exit 2 is the fail-closed arm — "could not evaluate", explicitly NOT an
all-clear. The harness worked: a reusable checked against no repo is reported by
nothing, and it treats that as a finding rather than a silence.

Two things it cost, both worth naming rather than quietly fixing:

1. It refused before evaluating ANY per-repo state — including the stale-sweep flip
   that had merged 26 seconds earlier. #1979's conformance was unverified for as long
   as this was red, not confirmed-and-fine.
2. It blocked #306, which touches this guarded file and so needs a green audit on
   the same SHA via conformance-gate.

Why #293's own audit was green

#293's audit was re-run at ~11:38 and passed. #305 merged after that, then #293
merged. So #293's green verdict was stale with respect to develop — the re-run
trap this repo's README already documents under Reading the signals (required checks
are latest-per-context). Nothing in the PR was wrong; the sequencing was.

The change

19 exempt rows, mirroring every existing blocked-gate.yml row at its own indent,
plus a bugbot_gate_rollout_pending anchor in shared_reasons.

Modelled deliberately on blocked_gate_rollout_pending, because this is the same
situation: a reusable whose derivation is the whole risk surface, landing ahead of
its callers so that derivation can be reviewed alone. The two-step is forced by this
file's own header — .github's caller state is read from its audit branch over the
API, so adding a caller and flipping its entry to required in one PR fails on a
caller not yet on the audit branch. That is exactly what split #307 from #293.

The anchor self-destructs. It states that if it is still here once the rollout PR
has merged, it has become the same finding as wip_limit_check_has_no_callers — a
reusable that shipped, was never wired up, and had a written excuse — and is to be
treated as a defect, not a justification. The stale_backlog_migration_in_flight
anchor this repo just deleted is the precedent for that clause working.

bugbot-gate still gates nothing, and this PR does not change that

Measured: zero callers anywhere in the org. backend#2284's premise was "Bugbot's
verdict gates nothing", and that is still true — now with more code behind it. This PR
makes the fleet readable, not conformant-in-substance. The rollout PR that adds the
callers and flips the entries to required is the one that closes #2284.

No Closes keyword, on purpose — see backend#2364. This PR does not close
#2284, and a title reference would not have linked it anyway.

Verification

check	result
yaml.safe_load	parses; all 19 rows resolve as exempt dicts; bugbot-gate.yml in reusables (12 total)
scripts/tests/caller-drift-selftest.py	197 pass, 0 fail
real audit, this branch, against the live org	exit 0 — "No drift. Every repo read, every entry matched." 19 of 19 audited, every family OK

That last row is the one that matters: with this change the audit completes, which
is what re-establishes that the 17 stale-backlog.yml: required entries from #293
actually match the fleet. callers reads OK for all 19 repos.

---

Note

Low Risk
Inventory-only YAML: no callers, no branch protection, and no gate behavior change. Risk is limited to a wrong exemption blocking a later required flip.

Overview
Registers bugbot-gate.yml in repo-inventory.yml so the org caller-drift audit can complete again. The reusable shipped without an inventory row, which made the harness fail closed (exit 2) before evaluating any per-repo state.

Adds the reusable to the global list and an exempt row in every repo, citing a new bugbot_gate_rollout_pending reason (same two-step as blocked-gate: derivation first, callers later). This does not wire callers or make Bugbot a required check; it only makes the fleet readable until the rollout PR flips those entries to required.

^{Reviewed by Cursor Bugbot for commit 492a688. Bugbot is set up for automated code reviews on this repo. Configure here.}

[LukasWodka]

bugbot run

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 492a688. Configure here.}

[saqlainsyed007]

Verified against the inventory and the audit result, not the description.

The break and the fix. #305 landed bugbot-gate.yml as a reusable but no reusables row, so caller-drift hit its fail-closed "a reusable checked against no repo is reported by nothing" arm (exit 2, "incomplete read") and reddened develop's audit 26s after #293 merged — before it evaluated any per-repo state. This adds bugbot-gate.yml to the reusables list and an exempt: *bugbot_gate_rollout_pending entry on every repo, which is exactly the two-step blocked-gate.yml already uses (each new bugbot-gate.yml line sits right under the matching blocked-gate.yml one). The anchor is honest STAGED-not-PARKED: zero callers yet by design (the derivation-from-findings is the risk surface and lands for review first; callers follow in the rollout PR), and it carries the self-deleting clause — if it outlives the rollout it becomes the wip_limit_check_has_no_callers finding, not a justification.

Decisive check: audit and gate are green on this PR — the same checks that went red on develop — so the inventory read is complete again and every repo's bugbot-gate.yml entry resolves. Bugbot pass, no threads, mergeable (BLOCKED is review/ordering, not a conflict). This restores a red develop audit; LGTM.