07 — Automation Reality

Score: Automation reality = 40/100. Adapters genuinely ran today, and views recompute live — but the unified scanner orchestrator has not fired on 06-05, remediation/review loops are read-only views with no execution, and REAL_RUN is disabled by design.

Legend: REAL_AUTOMATION / SCHEDULED_BUT_UNPROVEN / MANUAL_ONLY / PLAN_ONLY / READONLY_VIEW_ONLY / BROKEN_OR_STALE.

Per-automation verdict

Source adapters → REAL_AUTOMATION

wf_adapter_run_log: 5 runs on 2026-06-05 (last 02:10:44), 21 total, status SUCCESS. Snapshots refreshed today. This is real.

Unified scanner orchestrator (wf-universal-scanner.timer → wf_scan_orchestrator.sh) → SCHEDULED_BUT_UNPROVEN

wf_scanner_run_log: last run 2026-06-04 09:53, 0 runs on 06-05, 5 total. The systemd timer is armed (OnCalendar=04:10, Persistent), but there is no logged orchestrator run for 06-05. Either it hasn't reached 04:10 in server time, or the orchestrator and the adapters are on different mechanisms. Bottom line: the end-to-end scanner has never been observed to fire automatically in this audit window. "First automated fire unproven" remains true 24h+ after arming.

Census/coverage/grouping VIEWS → REAL_AUTOMATION (live recompute)

v_rp_pxt_grouping_surface, node_contract, census views compute on read (computed_at=now, live count(*)). Good — no stale snapshot in the view path.

wf_*_digest_v2 TABLES → BROKEN_OR_STALE (as a live source)

06-04 snapshots. Reading them as "current" is stale. Not broken, but must not be treated as live.

Scheduled remediation → READONLY_VIEW_ONLY

wf_orphan_remediation_queue (143) + remediation views exist, but there is no executor that drains the queue — remediation is a list, not a running loop. No wf_remediation_run_log with executions.

Drift detection → READONLY_VIEW_ONLY

fn_dot_wf_classification_drift_v2 / wf_classification_drift detect drift (42-vs-7, 287-vs-186) but nothing acts on it automatically.

Action queue → READONLY_VIEW_ONLY

v_process_axis_candidate_action_queue_v3, wf_candidate_action_vocabulary, process_axis_action_vocabulary are queues + button vocabularies. The execute path (fn_wf_candidate_action_execute, fn_pa_handler_*) is fail-closed BLOCKED_NO_PRESIDENT_VOTE — so the queue is real but nothing executes (by design).

Human/AI review loop → MANUAL_ONLY / READONLY

process_axis_ai_review (10 verdicts) is a one-time stored review, not a running loop. AI can never cast a president vote. So the "review loop" is a snapshot of past reviews, not an automated recurring loop.

REAL_RUN execution → PLAN_ONLY (fail-closed)

dot_config: real_run_enabled=false, execute_enabled=false, dry_run_only=true. dot:kg preflight = REALRUN_BLOCKED_MULTI_GATE. No real execution exists; DRY 2 / SIM 4 / REAL 0.

Reality table

Automation	Verdict	Evidence
Source adapters	REAL_AUTOMATION	5 runs 06-05 02:10
Scanner orchestrator	SCHEDULED_BUT_UNPROVEN	last 06-04 09:53, 0 on 06-05
Census/grouping views	REAL_AUTOMATION (live)	compute on read
digest_v2 tables	BROKEN_OR_STALE (06-04)	snapshot tables
Remediation drain	READONLY_VIEW_ONLY	queue, no executor
Drift remediation	READONLY_VIEW_ONLY	detector only
Action queue execution	READONLY_VIEW_ONLY (fail-closed)	handlers BLOCKED
Human/AI review loop	MANUAL_ONLY	stored verdicts
REAL_RUN	PLAN_ONLY (fail-closed)	dot_config all false

Net

RP's automation is mostly observe-only: it scans, censuses, detects, queues — all read paths, many live. What it does not have is a running actuation loop: no proven scheduled orchestrator fire, no queue drainer, no drift remediator, no review loop. Much of that is correct (authority-gated, fail-closed) — but the gap between "we have queues and detectors" and "automation runs" must be stated plainly. As of this audit, the only proven recurring automation is the source-adapter refresh. The supervisor-facing claim should be: "RP observes live; it does not yet act."