E03 — TF-IDF logistic-regression baseline

Migration identity#

• Canonical identifier: E03 (GitHub issue type: epic).
• Legacy identifier: U03 (retained as old_identifier).
• Parent umbrella: W02.
• Milestone: M1 — IntentGuard Weekend MVP.

Objective#

Train, persist, reload, and measure one deterministic scikit-learn TF-IDF logistic-regression baseline using the U02 contract.

Rationale#

A credible lexical baseline makes transformer improvement measurable and provides an early working checkpoint.

Parent identifier#

Umbrella issue W02 in milestone M1 — IntentGuard Weekend MVP.

Source task#

T-003.

Traceability#

Primary: T-003, FR-002, AC-002.

Prerequisites#

U01 complete and U02 pinned data outputs available.

Likely files#

src/intentguard/baseline.py, src/intentguard/metrics.py, src/intentguard/artifacts.py, scripts/train_baseline.py, tests/unit/test_baseline.py, tests/unit/test_metrics.py, baseline artifact/report paths.

Implementation boundary#

One explicitly configured TF-IDF plus logistic-regression pipeline; shared splits, labels, seeds, and metric code; no baseline tuning on test.

MUST scope#

S03.1 pipeline and fit; S03.2 artifact reload parity; S03.3 measured report and regression checks.

Explicit non-goals#

Additional classical models, ensembles, hyperparameter search, transformer embeddings, or fabricated performance thresholds.

Acceptance criteria#

make baseline fits on training data, uses validation only for allowed choices, evaluates the canonical test split at the declared checkpoint, writes accuracy/macro-F1 and provenance, and reloads with prediction parity.

Validation commands#

make baseline && uv run pytest tests/unit/test_metrics.py tests/unit/test_baseline.py tests/unit/test_artifacts.py -q.

Expected evidence#

Serialized pipeline, configuration, data fingerprint, dependency versions, class mapping, deterministic predictions, and measured report.

Fallback and status consequence#

Reduced safe iteration settings may remain Implemented if they preserve the specified estimator and are recorded. A different or unmeasured model is Partial and fails strict MVP.

Stop condition#

Stop for data mismatch, convergence/error instability, artifact reload divergence, test-driven tuning, or missing required metrics.

Definition of ready#

U02 evidence is complete and baseline hyperparameters are explicitly resolved in configuration before fitting.

Definition of done#

S03.1–S03.3 pass and the working-baseline checkpoint is recorded as Implemented and Measured.

Labels#

type:epic, priority:MUST, area:baseline

Estimate#

3.0 hours.