---

title: 06 — Count Gap & Drift Decision (Branch E) date: 2026-05-31 net_gap_live: +5,347 (fluctuating, CAT-023-driven)

06 — Count Gap & Drift Decision (Branch E)

Live re-evaluation at run time. Do not "fix" drift by updating counts — no approval covers it. This branch classifies each drift row by source_model and assigns its action to a macro.

Live invariant (leaf set, 160 rows)

Σ record = 2,007,256   Σ actual = 2,001,909   net_gap = +5,347
drift rows = 3   unmeasured = 5   orphan(leaf) = 0   pivot_backed = 21/160

The 3 drift rows sum exactly to net_gap (5,216 + 146 − 15 = 5,347) → invariant closes.

The 3 drift rows (changed since last run)

code	registry	model	record	actual	gap	Δ since last run	classification
CAT-023	birth_registry	A	990,687	985,471	+5,216	gap grew from +1 → +5,216	model_a_surplus_recheck
CAT-006	dot_tools	B	309	163	+146	stable	model_b_phantom_candidate
CAT-007	ui_pages	B	37	52	−15	stable	model_b_unregistered_candidate

What changed and why (the headline)

The entire net_gap move (+132/+148 → +5,347) is CAT-023. birth_registry.record_count (live, trigger-refreshed) grew to 990,687 while actual_count (last full scan) is frozen at 985,471. The gap grows monotonically until the next scan refresh. This was re-observed inside a transaction: when the Branch-C bundle INSERT fired the refresh, PIV-500 jumped 2,007,256 → 2,007,261 (+5 births in ~2 s). This is a model-A live-write race / stale-scan — NOT a phantom.

Per-row action class & macro assignment

row	action class	what to do	macro
CAT-023	write-race / refresh	refresh actual_count scan; do NOT treat as phantom; consider live actual_count from PIV-019 instead of a frozen scan	Macro 2 (scan cadence)
CAT-006	phantom candidate	council/RG4 must define phantom (LAW_DEFINITION_GAP) before any "phantom_count"; 309 registered vs 163 file = 146 candidates	Macro 2 → RG4
CAT-007	unregistered	register the 15 pages present on disk but absent from ui_pages	Macro 2
CAT-1006..1010	scan + register	5 pg_table rows, counts NULL → run a count scan, set registry_collection, then pivot	Macro 2

Why blind record > actual is unsafe as a phantom signal (proof)

CAT-023 is record > actual by +5,216 yet is a write-race, not a phantom. CAT-006 is record > actual by +146 and is a phantom candidate. The direction is identical; only source_model distinguishes them (A=live-write/stale-scan vs B=registry-vs-source). The six-object drift_classification is therefore source_model-aware and labelled "candidate" — it never asserts phantom. RG4 owns the authoritative definition.

Decision

• No count writes this run (no approval). The drift is characterized, not fixed.
• count_integrity_status = FAILED is the honest live state (3 drift + 5 unmeasured), surfaced — not hidden behind a frontend Σ|gap|.
• Belongs in Macro 2 (scan cadence + unregistered fixes + phantom-law), not Macro 1.