Klarbefund: running the evidence on a Mac

The Swift package targets macOS 26 and iOS 26, and the scanning and benchmark tests need Apple's Vision framework, so they run only on a Mac. This is the checklist for producing the evidence in evidence.md and self-assessment.md on one.

Prerequisites

What	Why	Check
macOS 26	`Package.swift` declares `.macOS("26.0")`	`sw_vers -productVersion`
Xcode 26, selected	Swift 6 toolchain and the macOS 26 SDK with Vision	`xcode-select -p`, `swift --version`
XcodeGen (app only)	generates `MeinBefund.xcodeproj` from `project.yml`	`brew install xcodegen`
CodeQL (optional)	the static analysis the pre-push hook and CI run	`brew install --cask codeql`

Nothing else: the package has no third-party dependencies, so there is nothing to resolve or download.

1. Build and test

git fetch origin && git switch feat/186-klarbefund-accuracy-evidence
cd clients/ios

swift build                       # Shared, ScanReplay, LabFile, AnalyteParity
swift run AnalyteParity           # the Swift dictionary matches its TypeScript source
KLARBEFUND_BENCHMARK_OUT=$HOME/Desktop/klarbefund-accuracy-benchmark.json swift test

swift test runs the whole suite, 319 tests in 54 suites. The scanning and benchmark suites render A4 sheets and read them with Vision, serially; on an Apple-silicon Mac the whole suite took about ten seconds after the build. Two suites are new on this branch:

ExtractionAccuracyTests: the accuracy measure itself, no Vision needed;
AccuracyBenchmarkTests: nine scan conditions, zero silent errors required under each, counts written to the JSON file named above.

Only that JSON is evidence to publish: synthetic sheets, counts and LOINC codes, no values. To run one suite: swift test --filter AccuracyBenchmarkTests.

2. Pilot on real reports (outside the repository)

In the app: menu → Export diagnostics, AirDrop the zip to the Mac.
Unzip it outside the repository, for example ~/Pilot/klarbefund-diagnostics-….
Next to each diagnostics.json, write a truth.json from the paper (format in evidence.md).

Run:

swift run ScanReplay ~/Pilot/klarbefund-diagnostics-2026-10-20 --truth          # per row, no values
swift run ScanReplay ~/Pilot/klarbefund-diagnostics-2026-10-20 --truth --json   # counts only

Both tools refuse a path inside clients/ios/. Get each participant's written consent before step 1, and delete the export when the counts are taken.

3. Optional checks

Scripts/codeql-swift.sh           # CodeQL on the app, ~15 minutes, exits 1 on a new finding
Scripts/run-ui-tests.sh           # XCUITest on a simulator, demo data only
xcrun simctl launch <udid> red.mabu.meinbefund -MBSelfTest   # 27-check self-test, see clients/ios/README.md

4. Put the Swift suite into CI

CI today builds only the Shared target (ios-analyte-parity in .github/workflows/test.yml); it never runs swift test. Add this job after ios-analyte-parity and push it with an account allowed to change workflows:

ios-unit-tests:
  name: iOS Unit Tests + Accuracy Benchmark (Swift)
  runs-on: macos-26
  defaults:
    run:
      working-directory: ./clients/ios
  steps:
    - uses: actions/checkout@v7
    - name: Swift version
      run: swift --version
    - name: Run unit tests
      env:
        KLARBEFUND_BENCHMARK_OUT: ${{ runner.temp }}/klarbefund-accuracy-benchmark.json
      run: swift test
    - name: Upload accuracy benchmark
      if: always()
      uses: actions/upload-artifact@v7
      with:
        name: klarbefund-accuracy-benchmark
        path: ${{ runner.temp }}/klarbefund-accuracy-benchmark.json
        if-no-files-found: warn

The benchmark JSON then becomes a downloadable artifact of every run that touches clients/ios/, which is the reproducible form an outside reviewer can check.

5. What to bring back

the swift test summary line, and any failure in full;
klarbefund-accuracy-benchmark.json;
for the pilot, the --truth --json output and the number of participants and sheets.