03 — AX-PROCESS Count Reliability

Registration

axis_registry AX-PROCESS: status CANDIDATE (owner GOV-MOW; reflection lens over live workflows registry; Dieu-34/MOW Option-3). substrate_resolver=fn_process_node_substrate. NOT official.

Counts (live)

metric	value
official AX-PROCESS axis_assignment	0
wf_process_candidate / candidate_universe	19
wf_process_candidate_member	143
wf_remediation_triage	143
verified_candidates_v3	1 (job:cut)
workflows table rows	2 (WF-001, WF-002)
process_run_observation	6 (all DRY/SIM)
process_component_observation	9
UI nodes (universal contract)	22

Reconciliation — PASS (exact)

22 UI nodes = 1 verified (job:cut, count 8) + 2 official_workflow (WF-001 10, WF-002 60) + 19 WPC candidates. The 19 WPC candidate member counts sum to exactly 143 == wf_process_candidate_member == wf_remediation_triage. Clean partition. job:cut: count_status verified, governance READY_FOR_PRESIDENT, flags [VERIFIED, SOLE_OFFICIAL_READY]. Sole verified process. Correct.

Reliability concerns

1. Two official workflows are axis-invisible. WF-001 ("Quy trình duyệt công việc", 10) and WF-002 ("đơn hàng xuất khẩu 60 bước", 60) are official_workflow (born in workflows) but flagged BORN_NO_AXIS_ASSIGNMENT / NO_AX_PROCESS_ASSIGNMENT. So "official AX-PROCESS = 0" is true for axis_assignment, yet 2 real workflows exist unassigned. Honestly flagged. (Likely seed/demo workflows — verify provenance before treating as real business processes.)
2. The "19 candidates" headline mixes real processes with non-processes. Lane breakdown via warning_flags:
   • DOT_IMPLEMENTATION_ONLY (Type-1): WPC-DOT-BIN-RECONCILE 18, WPC-DOT-NRM 14, WPC-DOT-HC 2.
   • PROCESS_CANDIDATE (genuine): APPROVAL-LIFECYCLE, BACKUP-DR 13, HEALTH-MON 14, PERM-GUARD 10, RECONCILE 6, CONTENT-PUBLISH 5.
   • COMPONENT_OF_EXISTING_PROCESS (not separate processes): CRON-ENV-HELPER 8, KB-SOP-LINK 2, MONITOR-BOOTSTRAP 0, PROCESS-DISCOVERY 1.
   • BACKUP_OR_NOISE: NOISE-QUARANTINE 3.
   • OS_LEVEL_INFRA: OS-INFRA 29.
   • NEEDS_MORE_EVIDENCE / NEEDS_HUMAN_REVIEW / NEEDS_OWNER: DB-SCHED-JOB 4, UNCLASSIFIED 0, INFRA-CONFIG 1, DOCKER-RUNTIME 11. → Each is individually labeled, but the raw "19" overstates genuine process candidates (~6). COUNT_UNRELIABLE as a headline unless the lane is shown. The ledger (doc 05) already collapses this to 6 birth-ready candidates.
3. Three provenance lanes feed one axis (verified job:cut from job_queue; born WF-001/002 from workflows; 19 scanned from wf_process_candidate) and are not unified. No overlap/double-count detected (job:cut not in workflows; WPC not in workflows), but the UI must reconcile three sources to answer "how many processes?".
4. 2 missing-substrate nodes: WPC-MONITOR-BOOTSTRAP (0) and WPC-UNCLASSIFIED (0) have substrate_available=false — empty placeholders, labeled.

Type 1/2/3

• Type-1 DOT-contained: the 3 DOT_IMPLEMENTATION_ONLY candidates project reliably from dot_tools/dot_bin (reconcilable).
• Type-2 multi-DOT automated: HEALTH-MON/RECONCILE/BACKUP-DR etc. — candidate-quality, evidence-gated, owner-blocked.
• Type-3 human-in-loop: not yet represented (WF-001/002 are workflow-table rows, not Type-3 instrumented). Gap.

Census denominator

v_universal_workflow_census_v2 = 16 SOURCE rows (not object universe). Object universe spans DB process-defs (dot_tools 309, dot_iu_command 54, job_queue 8, workflows 2 = 373) + host. Candidate members (143) cover the orphan/uncovered set (wf_orphan_remediation_queue=143). v_workflow_rp_missing_processes_v2=121. Denominator is explicit and live, but the headline "candidates" is a small adjudicated slice of a much larger uncovered universe — coverage is partial (see doc 08).

Score: 74/100

Internal partition exact, sole-verified honest, fully labeled. Docked for noisy candidate headline, axis-invisible official workflows, three unmerged provenance lanes, and absent Type-3 instrumentation.