[docs] FOLLOWUPS: record CI wall-time investigation post-PR #72

docs/FOLLOWUPS.md

@@ -561,6 +561,33 @@ sites (internal/telemetry/{build_info,slo}.go) that match the
       but premature pre-v0. *Trigger:* v0 cut — integrate or write
       a one-line rationale for skipping.
 
+### CI wall-time (post-verify-split)
+
+- [ ] **Shard `coverage-check` across multiple jobs to close the
+      155s → 242s projection gap.** Investigated 2026-05-18 after
+      PR #72's verify split (wall 427s → 242s, projection was
+      155s). The gap is NOT setup-go cache: all three verify-*
+      jobs already hit the same shared cache key (verified in
+      actions logs); setup-go costs ~10s/job. The gap is
+      `verify-test` bounded at 235s by `make coverage-check`,
+      which is `go test -race -coverpkg=... ./...` across 50+
+      packages. Closing the gap requires sharding test execution
+      by directory (e.g., `./cmd/...`, `./components/...`,
+      `./internal/...`, `./tools/... ./pkg/...`) into 3–4
+      parallel jobs with a coverage-aggregator step. *Trigger:*
+      when PR throughput becomes the binding constraint, or as
+      part of a broader CI cost review. Do NOT re-chase setup-go
+      cache sharing — already optimal.
+- [ ] **Deduplicate compile work across verify-lint and
+      verify-static.** `make build-tags` (95s, runs `go vet
+      ./...` twice) duplicates verify-lint's `make vet` (83s).
+      `make build` (53s) re-compiles packages verify-lint
+      already type-checked. Moving both into verify-lint would
+      save ~17% runner-minutes but **zero wall-time** (still
+      bounded by verify-test). *Trigger:* runner-minute budget
+      pressure, or when touching the verify split for another
+      reason.
+
 ### Architecture (trigger-based)
 
 - [ ] **Re-evaluate `signalops.go` duplication.** Three near-identical