IU Core MVP Sandbox — 04 Sandbox Probe Seed

Date: 2026-05-21 · Branch 3. Harmless sandbox-only probe data.

New repo artifacts

• sql/iu-core/sandbox/010_sandbox_probe_seed.sql — additive probe seed.
• sql/iu-core/sandbox/010_sandbox_probe_seed.rollback.sql — exact cleanup.

Design

• Additive only. Inserts into the 7 iu-core tables + creates one sandbox-only table public.iu_core_probe_source. It only reads information_unit / unit_version as FK anchors — never INSERT/UPDATE/ DELETE on them (so the "dependency rows unchanged" invariant holds).
• Every probe row tagged created_by='sandbox-probe' or route_code/ idempotency_key prefixed probe_* / iu_op:…:probe, so the rollback removes exactly the seed.
• Re-runnable: the DO block is a no-op if iu_op:add_new_piece:probe exists.
• Routes seeded enabled=false, dry_run=true — incapable of delivery.

Seeded rows (restore_test)

Anchors picked deterministically by a single row_number() ranking over information_unit (canonical_address, id) — ranks 1/2/3: parent a67cb4bc… (D38-DIEU28-ROOT), child a419f320… (D38-DIEU28-S0), sibling 1601c9b3… (D38-DIEU28-S1).

Table	Rows	Probe content
iu_core_probe_source	1	sandbox-only source table + 1 row
iu_sql_link	1	parent IU represents table iu_core_probe_source
iu_sql_event_route	1	inbound, probe_inbound_disabled, disabled+dry_run
iu_outbound_route	1	outbound, probe_outbound_disabled, disabled+dry_run
iu_route_attempt	1	inbound attempt, status dry_run
iu_structure_operation	1	add_new_piece, status draft
iu_tree_path	2	depth-0 root + depth-1 child (synthetic sidecar path)
iu_relation	2	contains parent→child + derived_from sibling→parent

dot_config key iu_core.routes_master_enabled seeded = 'false' (explicitly OFF; only if absent).

Bug found + fixed during seeding (H1)

First seed run: the child/sibling anchor selects combined WHERE id<>v_parent with OFFSET 1/2, double-skipping rows — it picked the 1st, 3rd and ~5th IU instead of 1st/2nd/3rd. Worse, the rollback recomputed anchors with a different ORDER BY … OFFSET, so it would have orphaned the depth-1 iu_tree_path row (no created_by column to match on).

Fixed:

• Seed now uses one explicit row_number() ranking (ranks 1/2/3).
• Rollback now derives the exact probe anchor IUs from the probe's own iu_structure_operation row (parent_unit_id + target_unit_ids) — no re-derivation of an ordering, no coupling to data order.

The fixed rollback was run against the buggy seed and cleaned it perfectly (DELETE 2 relation / 1 attempt / 1 route / 1 outbound / 1 link / 2 tree_path / 1 op / 1 config, DROP TABLE), then the fixed seed was re-applied. Post re-seed counts match the table above. Re-running the seed a 2nd time correctly logged sandbox probe already seeded — DO block is a no-op.

Cleanup

010_sandbox_probe_seed.rollback.sql returns all 7 iu-core tables to 0 rows and drops iu_core_probe_source. Proven twice (buggy-seed cleanup + the rollback drill).