Section overview · guide
ML system
A guide explains; a reference enumerates. This guide walks the lifecycle in the order the repository executes it. The exhaustive inventory lives in the Repository Reference.
The headline result, unretouched
TF-IDF baseline: Macro-F1 = 0.8654 (measured on the untouched test split).
DistilBERT: Macro-F1 = 0.6620 (measured on the untouched test split).
The fine-tuned transformer loses to the lexical baseline, and nothing was retuned after that number was seen — see Limitations for what these numbers do not support.
Read the system
ML system designspecificationData contract, both models, threshold selection, evaluation design.ArchitecturespecificationModule boundaries: data, baseline, training, threshold, evaluation, API.Requirements & acceptancespecificationFR-001…FR-010, NFR-001…NFR-010, AC-001…AC-014 — the contract.Interface contractspecificationPOST /v1/predict and GET /health, typed and versioned.Implementation statusmeasuredPer-capability status with run-level measured evidence.LimitationslimitsWhat the measured numbers do not support. Read before quoting.Requirement traceabilitygraphUS → FR → AC → test, as a navigable graph preset.Evidence chaingraphWHY → WHAT → HOW VERIFIED → PROOF → CLAIM.