dot-iu-cutter v0.5 Schema Q5 — Command Sequencing & Gates

Phase: v0_5_schema_Q5_and_evidenced_by_command_review_planning · Nature: design_only / command_review_preparation · Date: 2026-05-18 Authority (consumed, NOT reopened — QG1): GPT review next-phase mandate (phase_should_produce: implementation sequence plan + decision gates separating design, dry-run, command approval, production execution); handoff 2026-05-17 §2.2 accepted sequencing; pre-scale index DDL GPT review (D-5 separate sovereign cycle); fabric-addendum open decisions.

⚠️ GATING BANNER

phase: design_only
sequencing_is: a PROPOSED gated plan — NOT an authorization to run any step
executable_sql: none            # QG2
any_gate_executed: false        # G0..G6 are described, none entered
execution_authorized: false
self_advance: PROHIBITED
decision_authority: GPT / User ONLY (sequence + every gate transition)

This file proposes order and gates only. It does not approve the order, does not enter any gate, runs nothing. Where any ordering depends on an unresolved decision, it is flagged as an Open Decision (OD) — NOT self-resolved (QG1).

---

1. Gate model (applies to every workstream below)

G0  design_review            : design package authored + GPT/User PASS (prerequisite to G1)
G1  isolated_dry_run_plan    : authored dry-run plan on a restored isolated env (NOT prod)
G2  command_review           : exact command package + gates + verification + rollback authored
G3  sovereign_approval       : explicit GPT/User authorization of the command package
G4  production_apply         : execute the reviewed commands on production
G5  structural_verification  : catalog-level post-apply checks (no rendered-string compare)
G6  rollback / forward_compensation : invoked ONLY on G5 fail or defect (no hard-delete default)
properties:
  - each gate is GPT/User-gated; the Agent NEVER self-transitions a gate (QG7).
  - G1 (dry-run) is itself a separately authorized cycle; design PASS at G0 does
    NOT authorize G1.
  - "command-review planning" (this package) is PRE-G2 input: it prepares the
    G2 material; it is NOT G2 and not an approval.
gate_distinction (QG7 — explicit):
  planning            = this package (no run, no command authored to execute)
  authoring           = produce exact commands/DDL (e.g. index DDL already done)
  dry_run             = G1 run on isolated restored env
  production_execution= G4 run on prod after G3
  these are FOUR distinct states — never collapsed.

2. Workstream decomposition

The implementation surface is four independent workstreams that must NOT be merged into one migration:

WS-Q5  schema_registry_objects : OBJ-01..06 (+ matcher_config + address_template)
                                  — the registry / fabric substrate
WS-VA  evidenced_by_vocab_amend : OBJ-09 — §3.3 step-4 logical vocab entry made physical
WS-CK  cap4_checker_extension   : OBJ-10 — §3.3 step-5 R-EP/AU/NO/PV/LC/AD enforcement
WS-IX  pre_scale_indexes        : OBJ-11 — the 7 GPT-PASSed indexes (own D-5 cycle)
each WS carries its OWN G0..G6 lane; they do not share a transaction or a gate.

3. Proposed cross-workstream order (PROPOSAL — GPT/User ratifies)

proposed_order:
  1: WS-Q5  schema registries / fabric substrate
       why_first: GPT review explicit — "running vocab/checker implementation
         alone before schema Q5 may create an edge type without the surrounding
         registry/fabric substrate". Substrate is the no-hardcode resolution
         base every later layer reads.
  2: WS-VA  evidenced_by vocab amend
       why_after_1: vocab entry references owner_law/authority context; the
         Cap-4 checker that will enforce it reads authority_semantics from the
         WS-Q5 substrate. Vocab precedes any edge row (additive, no rows yet).
  3: WS-CK  Cap-4 checker extension
       why_after_2: R-AD3 requires evidenced_by to be live in the vocab store
         (read-from-vocab, never hardcoded); checker enforcement is meaningless
         before the vocab entry exists.
  4: WS-IX  index dry-run / execution
       why_here: indexes are additive and independent; sequence after substrate
         so indexed tables/columns are stable. Runs its OWN D-5 dry-run →
         command-review → sovereign cycle (NOT folded into WS-Q5 apply).
  5: source / grammar ratification (OD-G1/OD-S1 + Constitution fixture)
       why_after: ratify the real Constitution source authority + grammar
         profile only once the registry substrate that stores them is real.
  6: Constitution dry-run
       why_last: end-to-end hardtest needs substrate + vocab + checker + indexes
         + ratified source/grammar all in place; never before.

4. Ordering conflicts — FLAGGED, not self-resolved (QG1/QG7)

OD-SEQ1 source_document_registry_precedence:
  conflict: WS-Q5 OBJ-03 grammar_profile.address_template depends on
    source_document registry (address_docprefix, OD-A1/BR-A1), but
    source_document registry is an ingestion-layer object NOT yet logically
    specified (see inventory PRE-REQ).
  options: (a) specify+gate source_document registry as WS-Q5 step 0;
           (b) author address_template without docprefix binding, defer
               docprefix to source-ingestion cycle; (c) reorder: source
               ingestion design before WS-Q5.
  recommendation (advisory): (a) — add source_document registry as the first
    WS-Q5 sub-step so address namespacing is not orphaned.
  decision_owner: GPT / User. NOT self-resolved.

OD-SEQ2 vocab_amend_vs_substrate_minimum:
  conflict: does WS-VA need the FULL WS-Q5 substrate, or only the
    authority_semantics-relevant subset (source_family_registry +
    authority_override)?
  options: (a) full WS-Q5 G4 before WS-VA G2; (b) partial — only
    source_family_registry + authority_override live before WS-VA.
  recommendation (advisory): (a) conservative — full substrate first;
    revisit only if timeline pressure, via GPT/User.
  decision_owner: GPT / User.

OD-SEQ3 index_cycle_independence:
  note: WS-IX D-5 is already ruled "separate sovereign cycle". It MAY run in
    parallel with WS-Q5 (additive, disjoint objects) — but ONLY if GPT/User
    explicitly authorizes parallel lanes. Default = serial per the order above.
  decision_owner: GPT / User.

5. Per-workstream gate state (current)

WS-Q5: G0 = design PASS (WS-2 GPT cleanup PASS + this planning) ; next = G1 plan
        authoring (NOT entered). G2..G6 not reached.
WS-VA: G0 = design PASS (evidenced_by vocab-cap4 GPT PASS 2026-05-18) ;
        next = G2 command-review authoring AFTER OD-VC2 resolved. NOT entered.
WS-CK: G0 = design PASS (same GPT PASS) ; next = G2 AFTER OD-VC3 resolved +
        WS-VA live. NOT entered.
WS-IX: authoring DONE + GPT PASS ; D-2 APPROVED_PARTIAL ; D-5 = separate
        dry-run→command-review→sovereign cycle. G1 NOT entered.
nothing in any lane is at G3/G4 — no execution authorized anywhere.

6. Statements

• QG1: authority consumed, not reopened; every ordering dependency that is uncertain is raised as OD-SEQ1..3 for GPT/User, not decided here.
• QG2: no executable SQL/code. QG4: sequencing + explicit G0..G6 gate model present. QG7: planning / authoring / dry-run / production explicitly separated (§1). QG8: smallest next step is named in the report file, not a production jump.
• No repo/VPS access in this session. code_changed: false, commit_made: false.
• Self-advance PROHIBITED — doc 2 of 5; STOP after package complete → route GPT/User.

Companion files: scope-and-object-inventory, verification-and-rollback-master-plan, open-decisions-and-risk-register, command-review-planning-report.