OPT-GOV Prompt 0 — Mapping Table + Collision Report
Session: 7pjhmg · Date: 2026-08-22 ~16:3xZ · Task: read-and-map per ROADMAP_REVIEW_OPT-GOV_v1.1_20260821.md (Drive prompt board), lane claimed on the ledger (2ee05277). No feature code written.
Ground truth (Step 0): main synced (96b926f4+); CELL_REGISTRY = 39 cells ✓; OFFERING_MAP v2.3 ✓; claude/growth-control-completion-v1-1-w7vg1c LANDED ✓ (and the superseding GC ACTIVATION d58bd79c landed 2026-08-21 ~18:45Z — two hours after this spec was written, which drives the collision report below); .claude/rules/ current (00/01 additions ingested).
1. Mapping table (roadmap name → real path + owning surface)
| # | Roadmap name | Real path(s) | Notes |
|---|---|---|---|
| 1 | Ingress/SENSE serving path + JourneyEvent ingest gate | Gate: src/shared/virtuoso_models/ (MAPPERS → ingest_gate; wiring truth src/shared/virtuoso_models/WIRING.md; byte-identical twin in services/virtuoso-models-service/). Existing edge module: miz-oki-adk-agents/boss/boss_agent_core.py (ENABLE_EDGE_INFERENCE default false; pins tests/test_edge_inference_hardening.py) |
The landed edge seat is boss-core, evidence-only — not a SENSE-cell ONNX evaluator. A distiller must reconcile with it, never open a second edge door |
| 2 | Experiment/holdout registry | src/cells/cell36/causal_cell/ — POST /v1/causal/holdouts:assign (write-once deterministic-hash arms), POST …:preview, GET /v1/causal/holdouts/{kind}/{id}; holdouts.py (+71/-… extended and deployed 2026-08-22, run 32578630680) |
The geo scheduler is a client of this; method=geo, holdout 10–20%, MDE up front |
| 3 | X-Learner / DR-Learner / DoWhy battery | src/cells/cell26/Cell26.py (X-Learner), src/cells/cell36/causal_cell/{estimators.py, refutation.py} (DoWhy chain), src/cells/cell25/causal_journey_integration.py; continuous dosage: services/lift-engine/src/core/continuous_dosage.py (shim at services/lift-engine/continuous_dosage.py) |
New estimator classes register beside cell36's estimators.py / the lift-engine core |
| 4 | PLAN bid/budget valuation (net-yield/NCM) | services/measurement-rails/{ncm_v1.py, causal_credit.py, flags.py, bq.py}; services/net-yield/ (incl. writeback/, vault_costs.py); contracts: contracts/mizoki_contracts/metric_contracts.py |
Tenant margins/costs come from the onboarding vault (services/service-marketing-connectors/{tenant_costs.py, tenant_economics.py}), never constants |
| 5 | DEL (Decision Eligibility Layer, PR #656) | src/shared/mizoki_core/{objects.py, control_plane.py} + services/service-decision-control-plane/; treasury gate already live in code: services/service-policy-engine/main.py load_treasury() (F5 v1 fail-closed, ruling 2026-08-19-B) reading contracts/mizoki_contracts/treasury.py; veto reasons treasury_liquidity_floor_breached / …tightened_cap… / …covenant_cap…; TBR review queue on audit-replay |
Threshold floors raise-only — untouched by this lane |
| 6 | Decision trace / credit ledger (passport assembly) | contracts/mizoki_contracts/passport_package.py (assemble/persist/verify_package_seal; joins audit-chain bodies) + audit-replay routes (services/service-audit-replay/main.py: assemble/read + KMS sign_passport_package/verify_passport_signature since the activation); credit: services/measurement-rails/causal_credit.py |
No per-tenant prior-passport hash chaining exists — that is net-new |
| 7 | Cell 37 external-data door (supply feeds) | src/cells/cell37/ — /inject/batch|gcs|synthetic + /inject/canonical/* (dark, register row 26); F3 inventory: src/cells/cell37/market_cell/{f3_inventory.py, f3_config.py} (deployed 2026-08-22, run 32578920031), writes f3_recommendations, actuator "f3-recommendation-only" registered in NO adapter registry (two-key rule) |
The sanctioned supply-data door; no parallel ingestion |
| 8 | Feature-flag conventions | services/net-yield/{main.py, test_flags.py} + writeback/ and services/measurement-rails/flags.py — env-read flags with fail-if-flipped source-literal tests |
None of the four new flag names exists anywhere yet (repo-wide grep = 0) ✓ |
Adjacent existing frontier libraries (the spec's own homes for "extend, don't rebuild"): F4 geo src/shared/growth_control/f4_calibration/ (candidate pools, approval-gated "f4-geo-reservation" proposals, null-cap refusal, clean-cycle eligibility; its docstring already names the fuller Meridian/Robyn MMM as a labeled future lane) · F2 LTV src/shared/growth_control/f2_ltv/ (MIN_OBSERVED_QUARTERS=2 pinned, data_insufficient refusals) · F1 creative src/shared/creative_aesthetic/{elements.py, dr_effects.py, store.py} (honest simplified AIPW over creative-exposure aggregates; provisional=true double-locked).
2. Collision report (pre-approval condition 1: a collision voids the approval for that feature until the owner rules)
The GC ACTIVATION (d58bd79c, landed 2026-08-21 ~18:45Z, deployed GATE-2 2026-08-22) post-dates this spec and built into its territory:
| Phase feature | Verdict | Evidence |
|---|---|---|
| P2 BUILD 1 — ContinuousDosageRLearner | HARD COLLISION — already built to spec. Cross-fitted (_cross_fit, group-split folds), R-learner (METHOD="r_learner_local_marginal_moment_wls_v1"), bootstrap/CI machinery present, refusal statuses (STATUS_INSUFFICIENT_*, invalid-input) instead of spend-to-max. 18 tests landed with it. |
services/lift-engine/src/core/continuous_dosage.py |
| P4 BUILD 1 — treasury gate | HARD COLLISION on the liquidity-floor gate (exists, fail-closed, veto reasons + TBR queue + /reload) and on reservations (DCP transactional reservations, activation Workstream). Genuinely absent: a DEL_TREASURY_GATE flag (F5 arms by config-presence, not flag — adding a second arm is a design decision, not a gap) and the stockout/supply veto. |
services/service-policy-engine/main.py:23-51, contracts/mizoki_contracts/treasury.py, runbook §2 |
| P4 BUILD 2 — ValidationPassport | PARTIAL. Assemble/persist/sha256 seal/KMS sign/verify/fixtures exist. Net-new: per-tenant prior-passport hash chaining + chain-walk verification utility (and model_version in the hashed payload if absent). |
contracts/mizoki_contracts/passport_package.py, audit-replay main.py |
| P1 BUILD 1 — edge distiller | PARTIAL. Edge inference exists in boss-core (evidence-only, ENABLE_EDGE_INFERENCE default false, EDGE_WRITE_TOKEN tiering still an owner item). The ONNX distillation job, fidelity floor, registry-provenance demotion, and a SENSE-side evaluator are net-new — but must reconcile with the existing module (one edge concept, not two; whether EDGE_INFERENCE_SERVE should exist beside ENABLE_EDGE_INFERENCE is an owner design call). |
miz-oki-adk-agents/boss/boss_agent_core.py, tests/test_edge_inference_hardening.py |
| P1 BUILD 2 — geo-SCM scheduler | PARTIAL. F4 calibration library + cell36 registry client surface exist. Net-new: automated DMA selection, real SCM weight fitting + diagnostics, MMM prior proposal records + MMM_PRIOR_WRITEBACK flag. Build INTO f4_calibration, not beside it. Ships dark until F4 geo list + caps (owner item). |
src/shared/growth_control/f4_calibration/ |
| P2 BUILD 2 — creative unbundler | PARTIAL. dr_effects.py is an honest simplified AIPW per-element estimator over aggregates. The spec's cross-fitted joint residual regression + collinearity diagnostics would extend/supersede its estimator core — extend in place, keep the provisional=true double-lock and the aggregates-not-identities unit. |
src/shared/creative_aesthetic/dr_effects.py |
| P3 BUILD 1 — LTV governor | PARTIAL. f2_ltv + tenant cost/economics vault exist and are the mandated sources. Net-new: retention-discounted blend (governed config, no literals), shadow-mode comparison records, LTV_BID_WEIGHTING flag — extend f2_ltv. |
src/shared/growth_control/f2_ltv/, onboarding vault |
| P3 BUILD 2 — stockout DEL gate | PARTIAL. F3 inventory feeds exist (cell37, observe-only, two-key rule). Net-new: the DEL-side SUPPLY_CHAIN_VETO. A veto is a reduction, not a dispatch, so it appears compatible with F3's recommend-only invariant — flagged for explicit owner confirmation anyway. |
src/cells/cell37/market_cell/f3_inventory.py |
No competing session claims on any of the seven features (ledger + board + branch greps clean; the GC-activation lane is CLOSED — its row is the overlap's provenance, not a live claim).
3. Corrections to the spec's assumptions (Step 4c)
All five held: 39 cells ✓ · DEL exists (mizoki_core + DCP) ✓ · holdout registry exists (cell36) ✓ · Neo4j retired ✓ · canon gate = scripts/skill_sync.py --audit + pytest ✓. Additions: (a) scripts/gen_canon_status.py --check is a real CI drift gate for docs/CANON_STATUS.md — treat it as part of the gate set for anything touching canon-status inputs; (b) the spec's "sub-10ms" was already relabeled aspirational by the activation (binding SLO p95<100ms) — matches A1(c); (c) treasury_constraints.yaml template lives at config/treasury_constraints.yaml, read via TREASURY_CONSTRAINTS_PATH on policy-engine.
4. Recommended branches (Step 4d) — re-scoped by the collisions
As named in the spec, with contents narrowed to the net-new remainders above: claude/opt-gov-phase1-edge-geo (distiller reconciled with boss-edge + SCM/DMA/MMM-proposals into f4_calibration) · claude/opt-gov-phase2-dosage-creative (dosage: no build — at most CI/refutation deltas if the owner rules any remain; creative: joint-DML extension of dr_effects) · claude/opt-gov-phase3-ltv-supply (f2_ltv extension + DEL supply veto) · claude/opt-gov-phase4-treasury-passport (treasury gate: no rebuild — flag-wrapper question to owner; passport chaining + chain-walk verifier + SIG-042 e2e harness).
STOP per the spec. Phases 1–4 do not start until the owner rules on the collisions above (pre-approval condition 1). Status file posted to the Drive board at this boundary.