fix: validate prediction feature dimensions

[CoreyLeath-code]

Summary

Makes SentinelAI's /predict endpoint enforce the 16-feature input contract of the shipped default SentinelModel. Invalid-width vectors are now rejected by FastAPI/Pydantic validation instead of reaching PyTorch and failing as internal server errors.

Architecture Changes

• The deployed default MLP has input_dim=16; the public API now exposes that existing constraint at its boundary.
• This does not add a new model, alter weights, claim model quality, or change valid 16-feature inference behavior.
• Deployments that deliberately use a differently sized model must update their model artifact and API contract together.

Files Changed

• api/main.py — defines the 16-feature contract with Pydantic list-length validation.
• api/core/model.py — corrects the source-level model contract documentation; the previous “arbitrary-length” claim did not match its linear input layer.
• tests/test_predict.py — verifies one valid 16-element request returns its mocked model result and one 15-element request is rejected with HTTP 422.

Validation Performed

The following workflows completed successfully on this pull request:

Workflow	Result	Evidence
Continuous Integration & Telemetry	Passed	run #40: 8 passed, 1 warning
Security Analysis	Passed	run #38
Ingestion Schema Validation	Passed	run #75
CodeQL	Passed	run #57
Performance Benchmarking	Passed	run #36
Code Quality Assurance	Passed	run #63
SAST Code Scan	Passed	run #37

Commands and Real Output

pip install -r requirements-dev.txt
PYTHONPATH=. pytest --cov=. --cov-report=term-missing --cov-report=xml tests/ --junitxml=pytest-results.xml

8 passed, 1 warning in 0.24s
Coverage XML written to file coverage.xml

Risks

• Clients currently sending a feature vector of any length other than 16 will now receive the explicit HTTP 422 validation response rather than an internal inference error.
• A non-default model dimension is not auto-discovered; that requires an intentional model/versioning design rather than an inferred fallback.

Follow-up Recommendations

• Version the feature schema alongside model artifacts before supporting multiple input dimensions.
• Add representative model-output tests when a trained, versioned artifact and evaluation data are committed.

NEEDS HUMAN DECISION

None.

Summary by CodeRabbit

• New Features

  • Prediction requests now require exactly 16 numeric features.
  • Added clearer input-field guidance for prediction requests.

• Bug Fixes

  • Invalid feature vectors are now rejected with a validation error instead of proceeding to inference.

• Tests

  • Added coverage for valid 16-feature requests and invalid 15-feature requests.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 26cb362d-93a7-41dc-ad72-6f79ca956dbd

📥 Commits

Reviewing files that changed from the base of the PR and between ff6aa8c and e46d617.

📒 Files selected for processing (3)

• api/core/model.py
• api/main.py
• tests/test_predict.py

---

📝 Walkthrough

Walkthrough

The API now enforces exactly 16 numeric features for prediction requests. Model documentation describes the fixed-width contract. Tests cover successful inference and rejection of 15-feature vectors.

Changes

Feature Dimension Contract

Layer / File(s)	Summary
Define fixed feature dimension api/core/model.py, api/main.py	The model documentation and request schema define a fixed 16-feature input contract. RequestModel.features uses Pydantic validation and field metadata.
Validate prediction requests tests/test_predict.py	Prediction tests cover valid 16-feature input and HTTP 422 validation for 15-feature input.

Estimated code review effort: 2 (Simple) | ~10 minutes

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/validate-predict-feature-dimensions

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}