VALIDATION_1.8.2.md

ResearchBee 1.8.2 — validation record

Prepared: 10 September 2026.

Executed locally

The complete offline runner finished successfully against the final release files:

python tests/run_offline_release_checks.py

Group	Passed	What ran
Journal/database repairs	7	Actual local index, preferred title parsing, field quartiles and idempotent repair behavior
Chat knowledge	15	Actual Unicode/Arabic retrieval, follow-up context and source review warnings
App request/pipeline boundaries	11	Actual Pydantic models and AST-extracted application function bodies with explicit service/framework stand-ins
Security/admin	14	Actual ASGI limiter/body/stream logic; admin logic with isolated boundaries when FastAPI was absent
OpenAlex/privacy	8	Actual provider/function logic with mocked transport and unavailable HTTP import substituted
Usage lifecycle	5	Actual counter/flush/cancellation logic with mocked transport
Khazna snapshots	12	Actual loader validation, fallback and freshness logic with synthetic snapshots and no network
Policies/repository constraints	31	Actual pure evaluator/registry functions
Policy adapters	12	Actual parser/recommendation bodies; missing httpx import and unused Timeout declaration omitted by the transparent offline loader
Existing OA/repository/frontend assertions	12	Existing assertions executed directly, including eight in the policy runner and four frontend checks
Frontend JavaScript	24	Entire embedded script executed in Node with a minimal test DOM and mocked fetch; asynchronous chat, errors, IDs, controls and exports exercised
Total	151	115 Python regression tests + 12 existing assertion checks + 24 JavaScript checks

Also passed:

• Python compilation across the release.
• Node syntax validation of all embedded frontend scripts.
• HTML parse and 136 unique element IDs, with no duplicate IDs.
• JSON parse checks for included JSON artifacts.
• SQLite integrity check.
• Repaired database SHA-256 matches subject_quartile_repair.json.
• The bundled source CSV SHA-256 matches the repair record.
• Repair rerun changes zero rows and preserves the repaired database hash.
• ZIP CRC verification and byte-for-byte verification of packaged files against the release manifest.
• Standalone frontend TXT is byte-identical to the packaged HTML.

The raw offline runner output is included in validation/offline-validation.log. Test warnings show deliberate synthetic failures and currently supported Python deprecations, not failed checks.

Not executed locally

The preparation environment had Python 3.12.14, Node 24 and numerical/Pydantic dependencies, but could not install FastAPI, httpx, the OpenAI SDK, pytest and other runtime packages. Docker and an installed browser were unavailable.

Accordingly:

• The full pytest suite was not run in this environment.
• The 12 real FastAPI TestClient cases in tests/test_http_release.py were syntax checked and inspected, but not executed here.
• The pinned Python 3.11 Docker image was not built here.
• No live Hugging Face deployment, credentials, billable OpenAI call, upstream API end-to-end response, Cloudflare Worker integration or rendered LibGuides browser/mobile session was verified.
• Keyboard/focus behavior was reviewed and exercised with the test DOM; this is not a browser accessibility certification.
• Provider release metadata checks are not a complete transitive dependency vulnerability scan.

The Dockerfile runs dependency installation, pip check, Python compilation and the actual full Python test suite before creating the production image. A failed dependency or Python test stops that build. The Node regression wrapper skips when Node is unavailable in the Python image; its 24 checks were executed separately here.

Real dependency checks

In a Python 3.11 environment with package access:

python -m pip install -r requirements-dev.txt
python -m pip check
python -m pytest -q
python tests/run_offline_release_checks.py

With Docker:

docker build -t researchbee:1.8.2 .

Use the live smoke checklist in README-DEPLOY.md after Hugging Face reports the build running. Passing local checks reduces identified regression risks; it cannot establish that every live function is error-free.