Background tasks (ingestion, PDF resolution, scheduler) now acquire an
asyncio.Semaphore before checking out a DB connection, guaranteeing at
least N connections remain available for API request handlers. Removes
duplicate polling loop from PublicationsPage, debounces publication
reload on run-status changes, and increases poll interval from 5s to 15s
since SSE handles live discovery.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Remove unused type:ignore comment in scholars.py
- Fix 51 mypy errors across test files (Any return types, cast(), protocol stubs)
- Fix arxiv unit test failures caused by session isolation (patch_session_factory fixture)
- Fix integration test rate-limit bleed between search and create
- Add scripts/premerge.sh for local CI verification
- Add docs/website/.docusaurus/ to .gitignore
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Replace fixed batch_size=25 with URL-byte-length-aware chunking for
OpenAlex title.search queries. Long/Unicode titles caused URLs to
exceed server limits, resulting in 500 errors.
- Always hydrate scholar profile metadata (name, image) on creation,
even when an initial scrape job is queued, so the UI shows profile
info immediately.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The upsert rolls back the entire transaction on partial failure,
so new_pub_count and link count should be consistent (both 0),
not assuming the first publication survived.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Extract _load_unenriched_publications, _enrich_batch, _flush_and_sweep_duplicates from enrich_pending_publications
- Extract _sanitize_titles shared helper for title cleaning
- Extract _run_iteration_and_complete and _inline_enrich_and_finalize from run_for_user
- Extract _run_first_pass and _run_depth_pass from run_scholar_iteration
- Extract _finalize_successful_queue_job and _finalize_failed_queue_job from _finalize_queue_job_after_run
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Rename _int_or_default → int_or_default (public API, was using private convention)
- Extract shared pdf_queue utilities (utcnow, event_row, queued_job, status constants) to pdf_queue_common.py to avoid circular imports
- Consolidate _effective_request_delay_seconds into queue_runner.py with explicit floor parameter
- Update tests to match new function locations and names
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Split app/services/ingestion/application.py (2,955 lines) into three
focused modules:
- page_fetch.py (219 lines): PageFetcher class for single-page fetch,
parse-or-layout-error handling, and retry with backoff
- pagination.py (486 lines): PaginationEngine class for multi-page
orchestration, loop state management, and short-circuit detection
- publication_upsert.py (239 lines): module-level functions for
resolve_publication, upsert_profile_publications, and all
find/create/update helpers
Also fixes two external import paths in portability/ that incorrectly
sourced normalize_title and build_publication_url from application.py
instead of fingerprints.py.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add white logo variant at .github/logo-dark.png using <picture> tag
with prefers-color-scheme media query
- Update mermaid diagram to show Vue<->FastAPI REST+SSE connection
and bidirectional API<->DB relationship
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
