04 — Seed & Readiness Plan

Minimal seeds required before a dry-run, classified by whether each may be a transient rehearsal seed (rolled back) or must stay empty. No seed is created by this mission.

4.1 Seed inventory

Seed	Needed?	Mode for first dry-run	Persist?
Ruleset row (governance_ruleset)	Yes — scan must resolve a ruleset_version	one draft ruleset, rehearsal-seeded inside BEGIN..ROLLBACK	No (rolled back)
Snapshot row (evolution_snapshots)	Reusable — 1 already exists (id present)	reference the existing snapshot; do not create	n/a (already 1)
Cursor row (gov_worker_cursor)	Yes for object-grain; not for collection-grain	collection-grain dry-run needs none; if seeded, rehearsal-only	No
Scope seed (governance_responsibility_scope)	Already seeded (6, reference data)	reuse as-is	already present
Owner seed (governance_object_ownership)	Intentionally empty	leave empty — the dry-run must honestly show owner gaps	No
Issue-type seed (system_issues buckets)	No new bucket	issues ride existing buckets; none created in dry-run	No
Event-type activation (SB-11)	Stays active=false	unchanged	No
Seam wiring (doc 02)	Yes for a non-trivial run	CREATE OR REPLACE inside BEGIN..ROLLBACK (collection-grain)	No
Axis seeds (axis_registry/axis_assignment)	Out of scope — substrate unbuilt	not seeded	No

4.2 What stays empty (hard invariants for the dry-run)

• governance_object_ownership = 0 (proves honest under-coverage reporting).
• governance_candidate_state / candidate_scan_run / governance_candidate_object = 0 persisted.
• event_outbox governance emit = 0.
• os_proposal_approvals = 0, approval_requests/apr_approvals unchanged.
• All 5 governance event types active=false.
• governance_relations=8, dot_tools=309.

4.3 The honest-result principle

The first dry-run must not seed owners. An empty ownership table means every governed collection shows as an owner gap — and that is the correct result to verify: it proves the path reports reality (under-coverage) rather than a manufactured PASS. Seeding owners to "make it green" would be the exact anti-pattern (a checked-forever / declared-coverage lie) the substrate was designed to prevent.

4.4 Seed classification vs the forbidden list

Every seed above is either (a) reference data already present (scopes, snapshot), or (b) a transient rehearsal seed inside a rolled-back transaction (ruleset, optional cursor, seam wiring). None is a persistent production candidate/owner population. This satisfies the mission's "no production candidate/owner data unless explicitly classified as inert test seed and justified" — here the classification is: no persistent seed at all; only rolled-back rehearsal seeds, justified as the minimum to exercise the path.

4.5 Readiness gate before any dry-run

A dry-run may proceed only when all hold (re-verified read-only immediately prior):

1. os_proposal_approvals=0 (gate intact) and dry-run is explicitly non-activating.
2. Substrate census matches doc 01 (8 tables + 6 views, all inert).
3. Seam wiring spec (doc 02) chosen as collection-grain (object-grain deferred).
4. Operator has the rollback script staged (doc 02 §2.5 inert-restore is the SB-2 rollback already on disk).
5. idle_in_transaction=0 at start.

4.6 Seed/readiness verdict

CLEAR. Exactly one transient seed (a draft ruleset) plus inside-transaction seam wiring are required; everything else is reused-as-present or stays empty. The plan forbids persistent candidate/owner seeding and mandates an empty-ownership honest-gap result.