Sheet 4 ·Is every line tested?
Every line and every branch of the code runs under a test, and CI rejects a change that drops below 100%.
The shared core
What both products share: the Claude and Voyage clients, pricing, the budget ledger and rate limits, trace events and server-sent events. Clients are faked in tests.
core/src/highnet_core · measured by pytest + coverage.py
- Lines100%518 of 518
- Branches100%46 of 46
The RAG API
The FastAPI app, every pipeline stage, the providers, the storage and the recorder. Tests swap Claude and Voyage for fakes, so no test calls a paid API.
api/src/highnet_rag · measured by pytest + coverage.py
- Lines100%1,260 of 1,260
- Branches100%240 of 240
Scope
The rules, the nearest-neighbour vote, the judge and its validation, the decision model, fusion, thresholds, the twelve steps, the recorder and the routes, run on fake models and a fake-embedded dataset.
scope/src/highnet_scope · measured by pytest + coverage.py
- Lines100%1,144 of 1,144
- Branches100%194 of 194
RAG’s eval runner
The golden sets, the runner and its judge, tested the same way with fake models.
evals/src/highnet_rag_evals · measured by pytest + coverage.py
- Lines100%550 of 550
- Branches100%52 of 52
Scope’s eval runner
Scoring rules, kNN, the judge and the fused pipeline, the threshold sweep and the cost per method, on fake models.
evals/src/highnet_scope_evals · measured by pytest + coverage.py
- Lines100%333 of 333
- Branches100%26 of 26
This website
Every page, component and helper, rendered in a simulated browser. Generated files (trace types, code excerpts) and stylesheets are not counted.
web · measured by Vitest + V8
- Lines100%2,476 of 2,476
- Branches100%1,934 of 1,934