dot-iu-cutter v0.5 — Deferred Source Family: Recommendation + Ruling Request

Phase: v0_5_deferred_source_family_grammar_binding_gate · Nature: design_plus_authoring__no_execution · Date: 2026-05-18 Decision authority: GPT / User ONLY. Agent recommends; does not decide.

1. Agent recommendation (summary)

recommended_overall_route: HYBRID  (C operating basis + D design-only ruling request)
seed_now: NONE  (0 of 6 families)  -> file 3 = NO-SEED-DRAFT
reject: Option A (fake per-family profiles) — QG1 violation
discourage_strongly: Option B (one generic structured-artifact profile) — QG1 +
  defeats fail-closed grammar detection
adopt_as_operating_basis: Option C (defer seed; recognise non-document sources)
open_as_ruling_request_only: Option D (schema revision design; NOT executed)

per_group:
  GROUP_ENTITY {sql_entity, code_artifact}:
    route: serve via the ALREADY-LIVE entity dimension
           (entity_kind_registry + entity_reference_registry + iu_entity_binding);
           do NOT create source_family rows for them.
    ruling_needed: confirm "non-document sources are NOT source_family rows" OR
           authorise an Option-D schema-revision design cycle (nullable
           grammar_profile_ref / add source_kind|parser_strategy).
  GROUP_DOC {internal_process, architecture_note}:
    route: defer seed; open a SEPARATE grammar-profile ratification sub-gate to
           author + sovereign-ratify a REAL process / architecture_note profile
           (WS-2 D3-style: level_definitions + matcher config + address_template).
           No generic profile.
  GROUP_AMBIGUOUS {report, lesson}:
    route: defer seed; ruling on dimension first — evidence DOCUMENT (light
           grammar profile) vs evidence ARTIFACT (entity via report_path).

Rationale in one line: two of the six are not documents at all (entities), two need real ratified grammars that do not exist yet, and two have an unresolved document-vs-entity identity — so nothing can be seeded now without breaking QG1 or the still-forbidden list.

2. Explicit ruling request for GPT/User

Choose one disposition per group (the four headline options the task enumerated, applied per group rather than globally, because the 6 are not homogeneous):

RULING-1  GROUP_ENTITY (sql_entity, code_artifact):
  option_A_seed_all_now_with_new_profiles : NOT RECOMMENDED (QG1 — fake grammar)
  option_B_seed_subset_only               : N/A (nothing safe in this group)
  option_C_defer_all                      : ACCEPTABLE interim — they are already
                                            covered by the entity dimension
  option_D_open_schema_revision_design    : RECOMMENDED (design-only) — make
                                            grammar_profile_ref nullable for
                                            non-document families OR add a
                                            source_kind/parser_strategy discriminator
  agent_recommendation: C now + D as a design-only follow-up cycle

RULING-2  GROUP_DOC (internal_process, architecture_note):
  option_A_seed_now_with_new_profiles : NOT RECOMMENDED now (profiles not ratified;
                                        "generic profile" forbidden by closeout)
  option_B_seed_subset_only           : NOT YET (no ratified profile for either)
  option_C_defer_all                  : RECOMMENDED until a dedicated
                                        grammar-profile ratification sub-gate
                                        authors + ratifies REAL profiles
  option_D_open_schema_revision       : NOT NEEDED (these genuinely are documents
                                        and legitimately require a grammar profile)
  agent_recommendation: C (defer) + schedule a profile-ratification sub-gate

RULING-3  GROUP_AMBIGUOUS (report, lesson):
  dimension_decision_required:
    (a) evidence DOCUMENT  -> needs a light report/lesson grammar profile
                              (then route like GROUP_DOC)
    (b) evidence ARTIFACT  -> model via entity dimension (report_path entity_kind
                              already live) + iu_entity_binding; NOT a
                              grammar-profile source_family
  option_C_defer_all : RECOMMENDED until (a)/(b) is ruled
  agent_recommendation: defer; request GPT/User to pick (a) or (b) per family

Headline mapping to the task's A/B/C/D for the package as a whole: A. seed all 6 now with new profiles → NO. B. seed subset only → NO (no subset is safe). C. defer all 6 → YES (operating basis, all 6). D. open schema revision → YES, but design-only ruling request (for GROUP_ENTITY).

3. Tradeoff table

route	satisfies NOT NULL FK w/o schema change	semantic correctness	QG1 (no fake profile)	unblocks WS-3 production	risk / debt	execution needed now
A — 6 fake profiles	yes	low (entities ≠ documents)	FAIL	yes (but on a false model)	high permanent semantic debt; breaks fail-closed detection	DML (forbidden)
B — 1 generic profile	yes	low	FAIL	partially	over-generic; hard to split later	DML (forbidden)
C — defer all, recognise non-doc	n/a (no seed)	high	PASS	not yet (deferred)	low; assembly of process/report/lesson delayed	none (design-only) ✓
D — schema revision (design-only)	yes (long-term, correctly)	highest	later (after schema cycle)	later	needs a new sovereign schema cycle	none now (design-only) ✓

4. Blocking level per family (task-mandated taxonomy)

must_resolve_before_constitution_dryrun:
  - NONE of the 6.
    reason: Constitution hardtest uses internal_incomex_constitution (seeded) +
    grammar profile incomex-architecture-constitution-v4 (seeded). The 6 deferred
    families do NOT gate the Constitution dry-run. (Separate item OD-G1 — actual
    constitution fixture grammar ratification — is defer_post_pilot and is NOT
    part of OD-SF1.)

must_resolve_before_cross_source_assembly_production:
  - internal_process   # WS-3 core node N3 of the collection_structure thread
  - report             # WS-3 node N6
  - lesson             # WS-3 node N7
  - sql_entity         # WS-3 node N4 — but resolved via iu_entity_binding
                       #   (needs the binding physical, a separate gate)
  - code_artifact      # WS-3 node N5 — same as sql_entity
  note: physical cross-source thread production for collection_structure cannot
        run until these are modeled (doc families seeded with REAL profiles;
        entity families served by physical iu_entity_binding).

can_defer_post_pilot:
  - architecture_note  # NOT in the WS-3 sample chain; needed later for
                       #   design/architecture-doc ingestion
  - the Option-D schema-revision execution itself (design-only now)
  - iu_entity_binding physicalization (WS-1 OD-FA2, defer_post_pilot)

5. Readiness for command-review

ready_for_GPT_command_review: NO
needs_design_ruling_first: YES
reason:
  - no safe seed exists -> there is no command package to review yet
  - RULING-1/2/3 are schema/long-term-semantic decisions (QG7) that must be
    taken by GPT/User before any seed-extension authoring can begin
on_ruling:
  - if RULING-2/3 -> "author real profiles": open a grammar-profile
    ratification sub-gate (authoring-only), THEN a seed-extension command package
  - if RULING-1 -> Option D: open a design-only schema-revision cycle
  - none of the above is self-advanced; each is a separate sovereign gate

6. Statements

• Recommendation = defer all 6 (Option C basis) + open Option D as a design-only ruling request for the entity group; reject A, strongly discourage B (QG1).
• Explicit ruling request stated; no modeling/schema/long-term decision taken silently (QG7). Blocking levels classified against the task taxonomy (QG4 feeds the report).
• No DML/DDL/GRANT/schema-alter/nullability-change/generic-profile/Directus/CUT/VERIFY/deploy/git. No execution (QG8).
• Self-advance PROHIBITED — doc 2 of 5; STOP after package → route GPT/User.

Companion: classification-and-options, seed-extension-draft-if-safe (NO-SEED-DRAFT), verification-and-rollback-plan, gate-report.