# Stage 143: Room-B Synthetic Kaggle Twin (aq143_synthesize_cleanroom_kaggle_twin / DB-UNIT-18)

Rebuilds realistic Kaggle-style competition datasets and evaluation splits solely from Room-A mathematical specs while refusing any raw source payload.

- Atomic micro-queue: `aq143_synthesize_cleanroom_kaggle_twin` (visibility: `public_commons`)
- Backing database unit: `DB-UNIT-18` (`Causal Scenario Foundry, Simulation Federation & Opportunity Core`)
- Database cell: `entities-db.teleon.dev` (`teleon_core.scenario_ir_manifests`, `teleon_core.world_hidden_truth_vault`, `teleon_core.actor_observation_views`, `teleon_core.opportunity_hypotheses`)
- Claim row hydration: `single_self_contained_flat_task_row` (100% context inline, normalized writeback)
- Track 2 Sister queue (`SISTER-18`): `sq18-improve-scenarios-queue.teleon.dev` (Enforce POET minimal-criterion solvability bands, audit SPDX simulator licenses, and turn production failures into regression worlds.)
- Lanes: Cholesky Gaussian Copula, Refuse Raw Source, Realistic Train/Test

## Execution engines & AI Agent freedom contract

- **Deterministic (0 model calls)**: `mesa_amlsim_and_sql_state_world_compiler_v1` (Seeded Cholesky normal-score copula & inverse-CDF twin generator (0 calls))
- **Hybrid (<=1 model call)**: `autoplay_explore_first_scenario_synthesizer_v1`
- **AI Agent & Bot Fleet (1..6 calls)**: `multi_engine_concordia_oasis_and_foresight_author_v1` (Realistic edge-case, noise-regime & metric-alignment designer)
  - Contract: **Instructions + Resources + Freedom to search for new resources and new ways not bound by instructions**

## Database invariants

- Investigator and solver views never access world_hidden_truth_vault or derive test fixtures from detector rules.
- Every suspicious or failure scenario pairs with a matched benign counterfactual and abstention control.
- Opportunity and problem generation stays decoupled from pre-selected solution promotion.

## External & internal bot fleet actions

- Refuse immediately if raw_source_text or raw_source_rows cross into Room B
- Sample correlated normal scores via Cholesky decomposition and map back to target marginals
- Preserve realistic missingness, skew, and evaluation signal without copying source rows

## Sister self-tuning areas

- Enforce POET solvability bands and family-level train/holdout lineage splits (DIM-01, DIM-02)
- Wrap permissive simulation adapters (MIT/Apache-2.0/BSD-3) while blocking copyleft or BSL traps (DIM-03)
- Convert verified production traces and failed experiments into permanent regression scenarios (DIM-06, DIM-07)