DATA GOVERNANCE & CONTAMINATION PLAN —
                    MENS-SCCA × APEX
                          Data boundaries, provenance, contamination control, and publication


1. PURPOSE
Define access, provenance, exposure, transformation, use, and publication boundaries before confirmatory execution.

2. FOUR-ZONE SEPARATION
Zone A — MERCOR CONFIDENTIAL: reserved tasks, data, environments, executions, rubrics, or materials designated
confidential/restricted.
Zone B — EXPERT / REFERENCE: expert rubrics, annotations, professional criteria, and reference materials.
Zone C — MENS-SCCA INFRASTRUCTURE: research code, specifications, documentation, fixtures, and methodological
components owned or controlled by the researcher.
Zone D — AUTHORIZED DERIVATIVES: aggregate tables, metrics, results, fixtures, or artifacts whose extraction/publication
is explicitly authorized.

3. PRIMARY DATA RULE
Mercor primary data remain in authorized environments. Public repositories will not contain reserved tasks, confidential
documents, credentials, unnecessary personal information, proprietary content, or trajectories that permit reconstruction of
protected materials.

4. PROVENANCE
Where possible, each relevant artifact should carry: artifact_id, source_class, creation_date, access_group, task_id, run_id,
benchmark_split, training_exposure, publication_status, license_status, and transformation_history.
The objective is a sufficient provenance chain to reconstruct where an observation came from and what transformations it
received.

5. CONTAMINATION CONTROL
DEVELOPMENT — permitted for design/calibration.
PILOT — permitted for testing.
HELDOUT — reserved for confirmatory evaluation.
RESTRICTED — not usable without specific authorization.
Held-out tasks will not be used retrospectively to tune confirmatory instruments, rubrics, or hypotheses.

6. TRAINING EXPOSURE
When information is available, exposure of models, agents, judges, or evaluation components to tasks, documentation, or
derivatives will be recorded. Lack of evidence will not be treated as proof of non-exposure; it will be recorded as unknown when
appropriate.

7. DEVELOPMENT VS. CONFIRMATION
DEV/PILOT may adjust instrumentation format, fixtures, ingestion procedures, rubric clarity, and technical error handling.
Before confirmation, hypotheses, primary variables, partitions, success criteria, statistical plan, exclusion procedures, and
instrument/rubric versions must be fixed. Later changes that affect confirmatory hypotheses are treated as exploratory and
versioned.

8. ACCESS AND MINIMUM NECESSITY
Access will be limited to what is necessary for the authorized research task. The preferred pattern is authorized environment →
minimal extraction → instrumentation → metrics → analysis → authorized derivative.

9. PUBLICATION
Artifacts will be classified as PUBLICABLE, PUBLICABLE WITH TRANSFORMATION, or CONFIDENTIAL / NOT
PUBLICABLE.
Public outputs should favor aggregates, synthetic examples, researcher-owned fixtures, and methodological documentation
when those forms support reproducibility without exposing protected materials.

10. CODE AND LICENSING
An eventual public package may contain only researcher-owned or appropriately licensed code, authorized fixtures,
documentation, schemas, synthetic/derived data, and reproducible instructions.

11. GOVERNANCE PRINCIPLE
The existence of an artifact in the research environment does not by itself establish permission to publish, redistribute, train on,
or disclose it. Authorization status remains explicit and versioned.