docs: factuality sweep — fix 3 broken links + 12 count claims (audit findings 2026-05-05)

Per the cowork/docs-audit-2026-05-05/ end-to-end factuality audit
(20 confirmed findings across 76 docs, 7 parallel subagents +
audit-of-the-audit). Hot + Warm tier fixes ship here; STALE
findings (qa-test-suite.md test-count snapshot) need 'make
qa-stats' which is operator-side.

BROKEN links repaired (3):
- docs/reference/api.md L195: [Quick Start](quickstart.md) →
  ../getting-started/quickstart.md (404 pre-fix)
- docs/reference/api.md L196: [Connector Guide](connectors.md) →
  connectors/index.md (Phase 4 rename, was 404 pre-fix)
- docs/reference/protocols/scep-intune.md L377:
  [legacy-est-scep.md](legacy-est-scep.md) → scep-server.md
  (file was deleted in Phase 7 commit e9b1510)

INCORRECT count claims repaired (12):
- api.md L5 + L18-19 + L155: '78 API operations' / '# 78' /
  'all 78 documented operations' → re-derive via
  grep -cE '^\s+operationId:' (actual at HEAD: 144)
- architecture.md L66 (Mermaid label) + L502 + L1047 + L1253:
  '8 always-on + 4 optional loops' / '12-loop topology' →
  9 always-on + 5 opt-in loops (14 total). Always-on/opt-in
  breakdown derived from cmd/server/main.go startup wiring:
  always-on are agentHealthCheck, crlGeneration, jobProcessor,
  jobRetry, jobTimeout, notificationProcess, notificationRetry,
  renewalCheck, shortLivedExpiryCheck (9); opt-in are
  networkScan, digest, healthCheck, cloudDiscovery, acmeGC (5).
  Re-derive count via grep -cE '^func \(s \*Scheduler\)
  [a-zA-Z]+Loop' internal/scheduler/scheduler.go.
- configuration.md L31: '12 loops, 8 always-on + 4 opt-in' →
  '14 loops, 9 always-on + 5 opt-in'. Self-introduced regression
  from commit 3275f9f (2026-05-05).
- mcp.md L11 + L65: 'all 78 API endpoints' / '78 available tools'
  → re-derive via grep -cE 'mcp\.AddTool\(' (actual at HEAD:
  87 MCP tools, 144 API operations).
- connectors/index.md L111: '9 built-in' issuer connectors →
  '12 built-in', extending the inline enumeration to include
  Entrust, GlobalSign, EJBCA (which had been added since the
  L111 prose was written). Local-CA framing extended to mention
  tree mode + ADCS sub-CA mode-doc.
- connectors/index.md L112: '14 built-in' target connectors →
  '15 built-in', adding AWS ACM target + Azure Key Vault target
  (which had been added since the L112 prose was written).
- why-certctl.md L37 + the inline list: 'Nine issuer connectors
  ship today' → 'Twelve issuer connectors', adding
  AWS ACM PCA, Entrust, GlobalSign, EJBCA to the list and
  removing the misleading 'EST enrollment' bullet (EST is a
  protocol surface, not an issuer; clarified in trailing note).
- why-certctl.md L66: '13 deployment targets' → '15', adding
  Kubernetes Secrets, AWS ACM, and Azure KV to the inline list.
- why-certctl.md L92: 'supports 9 issuer types' → '12 issuer
  types'.
- quickstart.md L135: '35 demo certificates across 5 issuers'
  → re-derive cert count via 'grep -oE "mc-[a-z0-9_-]+"
  migrations/seed_demo.sql | sort -u | wc -l' (actual: 32,
  matches README L86; quickstart was off-by-3).
- quickstart.md L452 (Demo Data Reference table): Certificates
  '35' → '32' (matches the cert count from seed_demo.sql).

Verification:
- grep confirms no remaining stale refs across the touched
  files (8 files, 31 insertions / 28 deletions).
- All 24 ci-guards/*.sh pass locally.
- The audit's STALE findings (S-1, S-2 qa-test-suite.md
  Bundle-P snapshot) are operator-side: run 'make qa-stats'
  to refresh the Test Suite Health table.

Companion: cowork/docs-audit-2026-05-05/RESULTS.md captures
the full audit with subagent false positives and missed
findings called out.

docs/reference/architecture.md

@@ -63,7 +63,7 @@ flowchart TB
         API["REST API\n(Go net/http, :8443)"]
         SVC["Service Layer"]
         REPO["Repository Layer\n(database/sql + lib/pq)"]
-        SCHED["Background Scheduler\n8 always-on + 4 optional loops"]
+        SCHED["Background Scheduler\n9 always-on + 5 opt-in loops"]
         DASH["Web Dashboard\n(React SPA)"]
     end
 
@@ -499,7 +499,7 @@ For incident-response events requiring fleet-wide revocation (key compromise, CA
 
 ### 4. Automatic Renewal
 
-The control plane runs a scheduler with 8 always-on loops plus up to 4 optional loops (enabled by configuration). `internal/scheduler/scheduler.go:262-265` is the authoritative count.
+The control plane runs a scheduler with 9 always-on loops plus up to 5 opt-in loops (enabled by configuration). Re-derive the count via `grep -cE '^func \(s \*Scheduler\) [a-zA-Z]+Loop' internal/scheduler/scheduler.go`; the opt-in gating lives in `cmd/server/main.go` startup wiring (`cfg.NetworkScan.Enabled`, `digestService != nil`, `healthCheckService != nil`, `cloudDiscoveryService != nil`, `cfg.ACMEServer.Enabled && cfg.ACMEServer.GCInterval > 0`).
 
 ```mermaid
 flowchart LR
@@ -1044,7 +1044,7 @@ For deployments that need JWT/OIDC/mTLS, the standard pattern is to put an authe
 
 ### Concurrency Safety
 
-The background scheduler uses `sync/atomic.Bool` idempotency guards on every loop (8 always-on plus up to 4 optional) — if a tick fires while the previous iteration is still running, it skips. A `sync.WaitGroup` tracks all in-flight goroutines. `WaitForCompletion(timeout)` blocks during shutdown until all work finishes or the timeout expires, preventing state corruption from mid-flight database operations during process exit.
+The background scheduler uses `sync/atomic.Bool` idempotency guards on every loop (9 always-on plus up to 5 opt-in) — if a tick fires while the previous iteration is still running, it skips. A `sync.WaitGroup` tracks all in-flight goroutines. `WaitForCompletion(timeout)` blocks during shutdown until all work finishes or the timeout expires, preventing state corruption from mid-flight database operations during process exit.
 
 The job-processor tick fans the per-job work out across up to `CERTCTL_RENEWAL_CONCURRENCY` goroutines (default 25), gated by `golang.org/x/sync/semaphore.Weighted`. The cap is the operator's lever for "how many concurrent CA calls per scheduler tick" — operators with permissive upstream limits and large fleets (>10k certs) can bump to 100; operators with strict limits or async-CA-heavy fleets should stay at 25 or lower. Values ≤ 0 normalise to 1 (sequential). The Acquire is ctx-aware so a shutdown-driven ctx cancel interrupts the dispatch loop promptly; in-flight goroutines drain via Wait before the tick returns. Closes the #9 acquisition-readiness blocker from the 2026-05-01 issuer coverage audit (pre-fix the fan-out had no cap, so a 5,000-cert sweep tripped DigiCert / Entrust / Sectigo rate limits and the next tick re-fanned-out the same calls).
 
@@ -1250,7 +1250,7 @@ flowchart TB
 
 1. **Pluggable sources** — Each cloud provider implements the `DiscoverySource` interface (Name, Type, Discover, ValidateConfig). Three built-in sources: AWS Secrets Manager, Azure Key Vault, GCP Secret Manager
 2. **CloudDiscoveryService orchestrator** — Iterates registered sources, calls `Discover()` on each, feeds reports into `ProcessDiscoveryReport()`. Errors from one source don't prevent other sources from running
-3. **Scheduler integration** — opt-in cloud discovery scheduler loop (6h default; see `docs/architecture.md` 12-loop topology), runs immediately on startup, `atomic.Bool` idempotency guard
+3. **Scheduler integration** — opt-in cloud discovery scheduler loop (6h default; one of the 14 loops in the scheduler topology — see the Background Scheduler section above), runs immediately on startup, `atomic.Bool` idempotency guard
 4. **Sentinel agents** — Each source uses its own sentinel agent ID (`cloud-aws-sm`, `cloud-azure-kv`, `cloud-gcp-sm`) for dedup and triage filtering
 5. **Source path format** — `aws-sm://{region}/{secret}`, `azure-kv://{cert-name}/{version}`, `gcp-sm://{project}/{secret}`
 6. **No new schema** — Reuses existing `discovered_certificates` and `discovery_scans` tables. Sentinel agent IDs leverage existing `(fingerprint_sha256, agent_id, source_path)` dedup constraint