ci: allow eight hosted template legs concurrently

.github/workflows/ci.yml

@@ -786,10 +786,10 @@ jobs:
     needs: [scope, build-artifacts]
     if: ${{ (needs.scope.outputs.selection == 'full' || needs.scope.outputs.selection == 'template') }}
     # One scaffold/install/typecheck/cook per template. Every nonvisual scenario still executes;
-    # splitting identical setup across 29 jobs wastes scarce slots at max-parallel 4.
+    # Keep one leg per selected kit; eight hosted legs may run together without changing coverage.
     strategy:
       fail-fast: false
-      max-parallel: 4
+      max-parallel: 8
       matrix: ${{ fromJSON(needs.scope.outputs.plan).templateMatrix }}
     # Hosted on purpose: template legs use hosted while the balancer routes builds and tests
     # to the local pool, avoiding one routing variable sending the whole burst to a single pool.

docs/PRDs/CI/PRD-483-impact-driven-ci-without-repeated-template-setup.md

@@ -14,6 +14,7 @@ prd_contract: v1
   then whether retained tests apply now. Docs-only skips product tests. CI configuration proves its
   routing/contracts. Native runs for native and shared runtime consumers. Unknown meaningful inputs
   keep full coverage. No settings, credentials, runners or concurrency changes.
+- João, 2026-10-04 follow-up: allow up to eight existing hosted template legs at once; preserve selected coverage, fail-fast false, runner routing and timeouts. This separately authorized cap change does not alter the original cleanup decision or historical measurements.
 - Exhaustive template scenarios remain on broad changes, main qualification and scheduled audits.
   Exact kit changes run the affected kits; shared generator/render sources keep exhaustive fanout in
   this tranche. Narrowing shared sources further needs a proved dependency boundary.
@@ -46,7 +47,7 @@ There are 13 kit manifests, not a hardcoded CI count.
 | One kit | 4 full unit shards + 31 template jobs / full board | 1 scaffolder-contract unit job + 2 template jobs / same |
 | Shared runtime, generator, unknown meaningful code; main/nightly | exhaustive 31 template jobs | exhaustive 13 scenario jobs + 2 journey jobs |
 
-The template cap stays four. Baseline's nonvisual slot-work lower bound is 94.6 / 4 = 23.6 minutes;
+The hosted template cap is eight; selected kits and scenarios are unchanged. The historical baseline at cap four had a nonvisual slot-work lower bound of 94.6 / 4 = 23.6 minutes;
 removing 16 repeated setups at the workflow's documented ~65s estimates 17.3 runner-min saved,
 retaining both existing golden journeys. That gives ~83.6 template runner-min, an estimate,
 not measured after-change wall time. Individual exhaustive template jobs will be longer (shooter

scripts/__tests__/ci-structure.spec.ts

@@ -1762,7 +1762,9 @@ describe("CI pipeline structure", () => {
     expect(lane).toContain('for scenario in "${scenarios[@]}"');
     expect(lane).toContain('test "${#scenarios[@]}" -gt 0');
     expect(lane).not.toContain("SHARD");
-    expect(lane).toContain("max-parallel: 4");
+    expect(lane).toContain("max-parallel: 8");
+    expect(lane).toContain("fail-fast: false");
+    expect(lane).toContain("runs-on: ubuntu-24.04");
   });
 
   it("preserves platformer's unique production artifact in its installed scenario job", async () => {