03 — DOT Rehearsal Results

All rehearsal was read-only. No clone was needed: the cleanup operates on the live mapping view + governed truth, and the decisive blocker (no update-DOT) is a tooling fact, not a data-state question. A BEGIN..ROLLBACK rehearsal of UPDATEs was deliberately not run — it would model the Đ26-forbidden manual path and fire the pivot triggers inside the transaction. SQL: sql/03_rehearsal_readonly.sql.

R1 — Target-state preview (9 actionable rows)

pivot_code	species	current_comp	target_comp	comp_status	current_group	lane	is_active
MTX-TEST	—	matrix	(null)	no_ref	test	inactive_retire	false
PIV-020	—	atom	(null)	no_ref	default	inactive_retire	false
PIV-001	catalog	molecule	atom	mismatch	cấu_trúc	gov_home	true
PIV-016	species	atom	meta	mismatch	default	rp_local	true
PIV-021	checkpoint_support	atom	molecule	mismatch	default	rp_local	true
PIV-015	governance_infra	atom	atom	match	default	rp_local	true
PIV-017	governance_infra	atom	atom	match	default	rp_local	true
PIV-018	governance_infra	atom	atom	match	default	rp_local	true
PIV-019	governance_infra	atom	atom	match	default	rp_local	true

• Action A = the 3 mismatch rows; target_comp is the governed value.
• Action D = the 2 inactive_retire rows — both already is_active=false.
• Action B default-bucketed candidates = PIV-015/017/018/019 (+ PIV-016/021).

R2 — Retire candidates are already inactive ⇒ DOT-314 is a no-op

code	is_active	superseded_by	is_matrix
PIV-020	false	NULL	false
MTX-TEST	false	deprecated	true

dot-matrix-retire early-exits ("already retired") on is_active='f'. So it writes nothing for either. Note PIV-020's superseded_by is NULL and cannot be set via DOT-314 (the early-exit precedes the UPDATE) — a cosmetic gap, not a classification issue. Action D: already satisfied; nothing to execute.

R3 — Governed-truth anchor for Action A (entity_species)

species_code	display_name	composition_level
catalog	Danh mục hệ thống	atom
checkpoint_support	Hỗ trợ Checkpoint	molecule
species	Loại	meta

Confirms the 3 target compositions exactly: PIV-001→atom, PIV-021→molecule, PIV-016→meta.

R4 — Retire+recreate workaround is NON-VIABLE

Could Action A be done by dot-matrix-retire (retire the wrong row) + dot-pivot-declare (recreate with correct --level)? No:

source_object	default-filter pivots	codes
meta_catalog	1	PIV-001
entity_species	2	PIV-016, PIV-103
task_comments	1	PIV-021

dot-pivot-declare's duplicate-guard matches on (source_object, filter_spec) without an is_active filter. After retiring PIV-001 it still EXISTS (soft delete), so re-declaring --source meta_catalog --filter '{"filters":[]}' finds PIV-001 → Already exists → creates nothing. Same for entity_species (PIV-103 also blocks) and task_comments. And even if it created, it would mint a new PIV code, breaking identity. ⇒ Retire+recreate is rejected as unbounded, identity-destroying, and self-blocking.

Rehearsal verdict

• ✅ Target state computed and anchored to governed truth.
• ✅ Action A/B confirmed unbindable (no update-DOT) — see doc 02.
• ✅ Action D confirmed already-done / no-op.
• ✅ Retire+recreate confirmed non-viable.
• ✅ Side effects characterized (future UPDATE fires only the refresh trigger).
• ✅ No event/system_issues/UI side effects possible from a read-only rehearsal.

Conclusion: there is no safe, lawful, bounded, reversible DOT mutation to execute in this macro. Proceed to HOLD (doc 04).