ci: merge_group slice_count follows vars.CI_MERGE_GROUP_SLICES (4..16, unset=8) (t_f447539b)

[Kyzcreig]

Dynamic CI slots arming (card t_f447539b, parent t_cb901847). merge_group slice_count now reads vars.CI_MERGE_GROUP_SLICES, clamped 4..16. When it is unset it is 8, the current pin, so nothing changes until ci-placement is armed. The only writer is ci-placement through scripts/ci_routing_ledger.py (ledgered). pull_request stays 16.

Companion: Kyzcreig/fleet-ops-scripts#65 (placement writes T + the var). The ci-overflow-controller slice_counts arm + TEST-REPIN lands after this merges.

Verified: actionlint shows only the two pre-existing detect-changes input findings at ci.yaml:134-135. Nothing was reported on the changed line.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[Kyzcreig]

🤖 merged-by: apollo · lane: alerts-sweep · gate: BYPASS: FleetReview paused by Ace 2026-09-22 (state/fleetreview-pause marker present) · why: t_f447539b Apollo review: ci.yaml merge_group slice_count reads vars.CI_MERGE_GROUP_SLICES clamped 4..16, unset=8 (behavior-identical until set); ci-reviewed; CI green

[Kyzcreig]

Apollo (09-25 19:1x PT): dequeued. This PR's merge_group slice_count expression (fromJSON(vars.CI_MERGE_GROUP_SLICES) < 4 && ...) cannot be evaluated by the ci-overflow-controller (fos#69 evaluator has no comparison/fromJSON/vars support), so every merge_group batch containing it is refused and Placement waits 180 s then goes all-local. Once on main, that would be every run. Re-enqueue after the controller fix lands (kanban card linked from t_dfa3bf56).

[ang-fleet-lander]

🤖 merged-by: apollo · lane: alerts-sweep · gate: BYPASS: FleetReview paused by Ace 2026-09-22 (state/fleetreview-pause marker present) · why: re-enqueue: #1117 (ci.yaml reads CI_MERGE_GROUP_SLICES, t_f447539b) was removed from the MQ 02:13Z 'manual' by an unidentified lane on the shared Kyzcreig login; CLEAN, CI green on 804bad3

[ang-fleet-lander]

🤖 merged-by: apollo · lane: discord · gate: BYPASS: FR paused by Ace 2026-09-22; Argus off card review (Ace 09-24 13:08); gate = CI green + Apollo read; PR was previously queued by the sibling lane · why: ci: merge_group slice_count follows vars.CI_MERGE_GROUP_SLICES (t_f447539b). Re-enqueued by Apollo after fos#78 deployed: the ci-overflow-controller now evaluates this expression (was dequeued 19:1x because every batch carrying it was refused id-domain)

[Kyzcreig]

Apollo (09-25 20:3x PT): re-enqueued. fleet-ops-scripts#78 deployed on the Studio controller — it now evaluates this PR's slice_count expression (vars.*, fromJSON, comparisons) and refuses fast if it ever can't.