CR019 development verification

This is a development increment, not a release qualification. Candidate source and test SHA-256 identities and actual tool versions are retained in candidate.json alongside this page. Input baseline is ae0c0b5a5af6a03e13fd3ef4911888bfaa66442e plus the identified worktree. The runs occurred on 2026-10-28 UTC using Linux x86_64, Python 3.12, Firefox 159.0, geckodriver 0.37.1 and Node 24.19.0. Native browser sandbox process creation is restricted in this container; Firefox content sandboxing was disabled for this isolated test environment. Product/browser defaults are unchanged. Browser runtime networking was blocked except localhost by the existing fixture policy.

Actual results

  • feature.xml: six final feature checks passed, including direct-file/HTTP reload, stable post-reload navigation, ordered cards/FIFO, opt-out, isolation, stale/corrupt state, blocked/quota storage, layout races, overview paging, separate pages, concurrent tabs and a full browser restart with the same profile.

  • browser-initial.xml: 42 cases passed before a Firefox NoSuchWindow error during setup of the next existing navigation test. No assertion failed in that run. browser-retest.xml: all four remaining navigation cases passed on fresh browser startup. Together these cover 46 distinct selected cases: six feature and 40 existing regression cases. This is not a single all-green release execution. The final feature rerun includes added anchor assertions.

  • node.tap: 48 JavaScript tests passed, including six new snapshot/storage cases and current-graph authority, 200/201-node limits and geometry reuse.

  • python.xml: all 107 non-browser Python tests passed. This includes clean offline wheel installation/build without Node and source-distribution rebuild. The Linux development artifact retains version 1.1.0; it is not a new release.

  • Strict Sphinx/publication, complete graph/evidence validation with test coverage and both trace CSV exports passed. Executable mapping reports 86 SYS/SWR requirements and no unmapped active requirement or unknown ID.

  • Draw.io 32.6.1 exported the editable remembered-view source to its adjacent SVG. Link checks and Firefox standalone/embedded visual inspection passed.

Commands and reproduction

Use the repository virtual environment and pinned dependencies. Provision STV_GECKODRIVER and a Firefox executable on PATH as in the browser fixture. Run from the repository root unless a working directory is stated:

# frontend/ working directory
npm ci
node build.js
node --test --test-isolation=none --test-reporter=tap src/*/tests/*.test.js

# repository root
python -m build --no-isolation
python -m pytest -m "not browser" -q
python -m pytest docs/test-scripts/test_remembered_view.py -q
python -m pytest docs/test-scripts/test_persistent_details.py docs/test-scripts/test_extended_controls.py docs/test-scripts/test_position_stability.py -q
python scripts/check-coverage.py --strict
python scripts/build_docs.py
python tools/method/tooling/validate_traceability.py docs/_build/html/needs.json --version 1.1 --require-test-coverage --repo-root . --csv docs/_build/cr019-trace.csv --test-csv docs/_build/cr019-tests.csv
python scripts/normalize-diagram-links.py --diagram-dir docs/architecture/diagrams --published-dir docs/_build/html

The clean-wheel test needs runtime wheels under docs/_build/wheelhouse; prepare them using pip download before offline installation. Node’s process isolation in this environment reported only file-level wrappers; explicit --test-isolation=none executes and reports all 48 named assertions/cases. The extended reset harness uses the same option; its two cases both execute.

Corrections and limits

Initial integration attempts exposed incorrect event-listener wiring and an unchecked-view pagehide regression; both were fixed and the affected tests rerun. Linux setup needed hatchling and an offline wheelhouse. Existing test harnesses assumed a Windows venv path and a JavaScript esbuild launcher; they now use a platform-aware interpreter and esbuild’s JavaScript API. Generated license newlines are normalized without changing audited vendor bytes. The publication audit rejected a private tooling name in the review prose; the public record now uses portable workflow wording, and the audit passed.

Formal Validate and store remember... (STV_TC_SW_056) and Remember and restore a real... (STV_TC_SY_070) remain accepted test designs with last_result not_run: these developer runs do not invent formal execution Needs. Store opt-in snapshots inde... (STV_SWR_073), Validate and reconstruct re... (STV_SWR_074), Restore presentation withou... (STV_SWR_075) and the amended stakeholder obligations have partial verification. Windows desktop qualification, additional browsers, release publication and later occurrence-mode behavior remain outside this increment. The previous release and historical acceptance/results are unchanged.