Native Windows development is accepted with Podman Compose and Dagger image builds. Start with the development entry point and native migration closeout. Run commands from the repository root; keep Windows disposable outputs under D:/BuildArtifacts/QOpenBB.
Separate financial browser acceptance passed Income, company selection, failure recovery and retained recreation followed by fresh login. The accepted run records 58 FMP and 6 FRED sends. This establishes the stated local image journey, not a new published release or private-server deployment.
Portable Linux CI passed with 459 tests passed, 12 skipped and one live test excluded; CodeQL succeeded. Skipped or excluded tests do not establish live-provider behavior.
./pipeline/dev.ps1 -Action verify -SessionPath <owned-session-path> requires healthy owned Workspace/data services, locked verification dependencies and an explicit browser cache. It runs synthetic Python/browser checks. Its financial-acceptance field remains unverified because financial acceptance is a separate recorded journey.
The following commands and results preserve the earlier WSL checkpoint. Its limitations describe that fixture run; current native and Linux CI evidence is linked above.
For the separately accepted backend test and query layers, use the SQLite repair verification. With the frozen backend Python environment restored, run from the repository root:
/home/mc/qopenbb-work/sqlite-runtime/venv/bin/python \
vendor/workspace/backend-api/backend/tests_sqlite/run.pyThat interpreter path identifies the exercised WSL environment. The maintained launcher creates an isolated child with synthetic configuration, blocks operator env/account reads and network access, and removes its file databases after exit. Its 60 tests cover six read paths, sharing/creator semantics, actual bulk save/update and SQLite engine recreation. They do not prove browser save/reopen, container restart or deployment. No provider credentials are needed.
The historical fixture checks display a synthetic observation of 123.45 USD, verify fresh cookie/local-storage state, and deliberately produce console/network failures through the owned browser's CDP session. A green result establishes verification infrastructure only. That fixture run did not establish Workspace, live-provider, Anthropic, Excel, Compose or deployment acceptance.
Restore the pinned development tools. From the repository root:
export PATH="$PWD/artifacts/task-1.2/node-v24.22.1-linux-x64/bin:$PATH"
PLAYWRIGHT_BROWSERS_PATH="$PWD/artifacts/playwright-browsers" \
node verification/node_modules/@playwright/test/cli.js install chromium
# A prepared runner also needs Playwright's Chromium OS dependencies.
node verification/node_modules/@playwright/test/cli.js test --config verification/playwright.config.ts
pipeline/.venv/bin/python verification/runner.py tests/delivery/test_verification.pyPlaywright 1.64.0 selects Chromium revision 1248 (156.0.8078.4). The historical runner launched that project-local Linux executable; no operator browser discovery or CDP attachment was available. On the observed Debian host, the first launch failed because libglib was absent. Installing the pinned Playwright CLI's install-deps chromium OS dependency list enabled actual browser execution. Current Linux CI and native Windows execution have separate evidence linked above.
The canonical browser command uses one worker and rejects CLI worker overrides and debug environment settings. The Python wrapper permits zero to two xdist workers, removes inherited pytest/plugin options, uses explicit synthetic fixture markers, and isolates output and temporary directories. live is excluded by default and recorded as unverified. No live credential is loaded. A separately authorized live check must report missing credentials as skipped/unverified, never successful acceptance.
Reviewed evidence is synchronized into the committed artifact folder before commit and push. Its manifest maps each published file to its local source and records both hashes and any transformation. Raw runtime state, credentials and caches remain local. The index states which completed runs are included; a missing historical artifact is not newly verified evidence.
Each browser run writes one artifacts/task-1.7/runs/run-*/evidence.json with source revision, tool/browser versions, timestamp, command identity, result, diagnostic categories and artifact root. Browser profiles are disposable and removed after success and fault paths. The fixture listener binds a dynamic loopback port and refuses an existing listener. Tests prove closure after success, startup failure, artifact failure and diagnostic failure. The standalone node verification/runner.ts --failure command exits nonzero on unexpected console/network errors.
CDP uses Playwright context.newCDPSession on the browser this run owns. Runtime and Network events retain category and HTTP status only. They discard URLs, headers, cookies, request/response bodies, exception text and console arguments before writing evidence. This protects even secret-bearing error strings; it is deliberately less detailed than an application debug log. Synthetic sentinel regressions check retained evidence for leakage. Raw screenshots, video and traces remain off; requesting capture for credential entry is refused before creating a run. Credential scenarios must never enable capture.
Browser children receive an environment allowlist; provider/model/registry variables, dotenv files and inherited Node options are unavailable. The runner invokes no external server helper: the HTTP fixture is an owned in-process server. Playwright launches hidden browser processes; the Git metadata helper uses windowsHide, and the Python subprocess uses CREATE_NO_WINDOW where supported. This WSL record remains distinct from the accepted native Windows run linked above.
Python evidence lives under artifacts/task-1.7/pytest/run-*, with bounded worker count, exact command, exit and numeric result counts. Raw pytest stdout and JUnit are never retained, because failure assertions can expose credentials. xdist temporary roots and test-local paths separate workers; the wrapper removes its temporary root in finally so failing-test sources and bytecode are not retained. These are disposable synthetic records. Do not reuse operator account state or interpret skipped live tests as verified provider behavior.
Task-local RED, flag-enabled GREEN, flag-removed GREEN, type checks, dependency setup and complexity reports are retained under artifacts/task-1.7. Re-run the combined complexity gate after changes. The acceptance disproof is a missing observation/unit, shared browser state, retained synthetic secret, leaked owned process/listener, unbounded worker override, or a failure reported as passed.