Coverage scope: decide on full-lib measurement vs. current activated-surface convention

[TaprootFreak]

Context

Today the coverage floor gate measures the activated surface only:

• lib/packages/** (services, repositories, signers, utils)
• lib/screens/**/cubit(s)/** and lib/screens/**/bloc/**

*.g.dart (Drift codegen) is explicitly removed from the scoped tracefile. Widget files (lib/screens/**/*_page.dart, lib/widgets/**) are exercised via testWidgets specs but not counted in line coverage — see README.md "Coverage scope" and docs/testing.md "Tier 0".

Current scoped coverage: 90.8 % (post wave 1 uplift, PR #538/#539).

Question

Is the activated-surface scope the final long-term convention, or should it eventually be widened to cover the rest of lib/ (widgets, routes, themes, bootstrap, generated localizations)?

Trade-offs

Pro full-lib measurement

• Single number to point at
• Catches UI files that have zero render-level test (today these are invisible in the gate)
• Forces every new screen to ship at least a smoke testWidgets spec

Contra

• ~50 % of widget code is layout (padding, sizing, theming) — line coverage there says "rendered once", not "behaviour verified". Mixing it with cubit/service coverage dilutes the signal.
• lib/main.dart and routing entrypoints never execute under flutter test
• lib/generated/ (localization codegen) would need the same coverage:ignore-style exclusion as *.g.dart
• Headline % would drop ~15-20 pp on the first measurement, so the floor would need a one-off lowering — easy to misread as a regression

Options

1. Keep activated surface (status quo) — accept that "98 %" means "98 % of behaviour-bearing code", not "98 % of lib/". Tracks well with the Tier model in docs/testing.md.
2. Add a second, advisory metric — measure full lib/ (with *.g.dart + lib/generated/** excluded) and surface it in the CI summary, but keep the floor gate on the activated surface. Best of both worlds, more YAML.
3. Switch the gate to full lib/ — broaden scope, accept the dilution, raise widget-test discipline.

Recommendation: option 2 if anyone wants the wider visibility, otherwise option 1.

Related

• README.md — "Coverage scope" / "Coverage infrastructure roadmap"
• docs/testing.md — "## CI"
• .github/workflows/pull-request.yaml — lcov --extract + lcov --remove '*.g.dart'
• PR test: cover the four Drift repositories + tighten coverage scope #538 (initial scope + first ratchet), PR test: coverage uplift batch (storage + dfx services + config + sell_bitbox + kyc cubits) #539 (wave 1 uplift)

[TaprootFreak]

Closing — superseded by the "vollständige Testabdeckung" push tracked in the issues below.

The scope-narrowing trade-off becomes moot once every layer (cubits, pages, widgets, integration, e2e, visual-regression) is in fact covered, since the activated-surface filter exists to hide measurement noise where no test exists. With full coverage there is no noise to hide.

Concrete numbers re-checked against coverage/lcov.info: full-lib coverage today (post #539 wave 1, *.g.dart + lib/generated/** stripped) is 83.6 %, not the 70-75 % suggested by the trade-off section. Out-of-scope code (widgets, models, pages) sits at 76.4 % already — testWidgets specs are producing line coverage that is currently being thrown away by the lcov --extract step. The path to a full-lib gate is therefore much shorter than this issue assumed.

Follow-up issues will replace this discussion with concrete work items per tier / surface.