S02.3 — Integrate make data and enforce the data contract

Migration identity#

• Canonical identifier: S02.3 (GitHub issue type: subtask).
• Legacy identifier: C02.3 (retained as old_identifier).
• Parent epic: E02.
• Milestone: M1 — IntentGuard Weekend MVP.

Objective#

Wire the data script to the Make contract and add acceptance tests that fail on provenance, split, schema, or label drift.

Rationale#

The data boundary needs an executable, regression-resistant acceptance gate.

Parent identifier#

E02.

Source task#

T-002.

Traceability#

Primary: AC-001. Secondary: FR-001, FR-010.

Prerequisites#

C02.1 and C02.2 complete with stable outputs.

Likely files#

Makefile, scripts/prepare_data.py, tests/contract/test_data_contract.py, data/README.md.

Implementation boundary#

Command orchestration and contract tests; do not train a tokenizer or model.

MUST scope#

Idempotent command, actionable errors, output validation, and exact acceptance assertions.

Explicit non-goals#

Network retries platform, data version server, synthetic success path, or benchmarks.

Acceptance criteria#

make data exits zero only for a validated pinned dataset and the contract test catches wrong counts, labels, overlap, revision, or fingerprints.

Validation commands#

make data && uv run pytest tests/contract/test_data_contract.py -q.

Expected evidence#

Command output, provenance file, passing positive tests, and controlled negative-test failures.

Fallback and status consequence#

Cached execution is valid only with verified provenance; a dummy success makes AC-001 fail.

Stop condition#

Stop if the command can pass without validated outputs or if tests depend on uncontrolled network state.

Definition of ready#

Data output schema and paths are frozen.

Definition of done#

The command and contract suite establish AC-001 with reproducible evidence.

Labels#

type:subtask, priority:MUST, area:data

Estimate#

0.75 hour.