Skip to content

CHANGELOG.md

Latest commit

 

History

History
141 lines (128 loc) · 8.17 KB

CHANGELOG.md

File metadata and controls

141 lines (128 loc) · 8.17 KB
Raw

Changelog

1.4.0 (2026-07-13)

New output formats

  • CSV bundle (--format bundle): writes citing_papers.csv, authors.csv (with country + match confidence), venues.csv, and timeline.csv into one directory for spreadsheet analysis.
  • Citation tree: interactive terminal tree (menu option 8, groupable by year / venue / institution) and a plain-text export (--format tree).
  • HTML report (--format html): a single self-contained HTML file with stat tiles, citation timeline chart, world map of citing institutions, citation-intent chart with context quotes, scholar/venue tables, and a collapsible citation tree. Works offline, light/dark aware.

International reach

  • Citing authors' institution countries are now captured from OpenAlex (ISO codes on every author, a per-country breakdown in results, a "cited across N countries" grant statement, country column in exports).

Reliability

  • Live end-to-end testing caught and fixed a crash: OpenAlex returns explicit null for summary_stats / last_known_institutions / affiliations on some records; all parsing sites are now null-safe (regression-tested).
  • max_citations is clamped to Semantic Scholar's 1000-limit in API mode instead of failing with HTTP 400.
  • Full-diff regression audit confirmed and fixed 10 defects introduced during the sprint, including: Google Scholar citation IDs leaking from one analysis into the next; OpenAlex title lookups breaking on commas (filter injection); author dedup creating "chimera" entries (one person's identity with another's h-index); Rich-markup crashes on citation text containing bracket tokens (e.g. [/INST], [sic]); scripted -o - output polluted by progress prints (now on stderr); explicit --h-index-threshold 0 being ignored; 12-letter surnames misread as Google Scholar IDs.
  • Data-quality safeguards: correct Semantic Scholar request pacing with an API key (was 40 req/s against a ~1 req/s allowance, causing silent mass failures), Retry-After honored on 429s, per-run API-failure tracking with a visible "data may be incomplete" banner in the UI and all report formats, and degraded results are no longer written to the 7-day cache. The alarm is proportional: a few transient failures produce a note, not the banner.
  • Batch author fetching: citing-author profiles are now fetched from Semantic Scholar's batch endpoint (one request for up to 500 authors instead of one per author), then enriched via OpenAlex as before — analyses make an order of magnitude fewer requests and survive strict rate limits.
  • Adaptive throttle: consecutive rate-limit failures automatically slow request pacing (up to 8×) and recover on success. OpenAlex now runs at half its polite-pool allowance by default; set your email in Settings to join the polite pool (dramatically more reliable than anonymous access).

1.3.1 (2026-07-12)

Author disambiguation overhaul (fixes reported same-name mix-ups)

Addresses the reported issue where authors sharing a name could be mistaken for each other. Author resolution is now ID-first with evidence-verified fallbacks:

  • Every resolved author profile carries a match_confidence provenance: id (unique Semantic Scholar / Google Scholar identifier), verified (name search corroborated by the candidate's publication list), or name (unverified name-only match).
  • Name-keyed author-cache hits are now gated: a cached profile that merely shares an author's name is only reused when its stored publications contain the citing paper. Unverifiable name hits are rejected rather than trusted.
  • Browse Other Authors by name now shows a disambiguation picker (name, affiliation, h-index, paper count from Semantic Scholar author search) instead of silently using the first same-named hit.
  • The UI marks each author with ✓ (ID-matched), ≈ (verified), or ? (name-only, may be a different person), with a legend; the summary shows "Author profiles: N ID-matched, N verified, N name-only", and the Markdown export gains a Match column.
  • Exports now use the configured h-index threshold in their labels instead of a hardcoded "≥ 20".

1.3.0 (2026-07-12)

New features

  • Field-normalized impact (FWCI): analyses now fetch the Field-Weighted Citation Impact and field citation percentile from OpenAlex (1.0 = world average for the same field/year), shown in the overview, grant summary, impact statements, and exports. Degrades gracefully when unavailable.
  • Self-citation detection: every analysis now reports how many citations are independent of the original authors (matched by Semantic Scholar author ID, falling back to name compatibility) — the number grant reviewers ask for.
  • "How Your Work Is Used": new drill-down (option 7) showing the citation intent distribution (methodology / background / result) and sample in-text context quotes from citing papers.
  • Report exporter (citationimpact/export.py): export analyses as grant-ready Markdown, LaTeX (appendix-ready section), CSV (all citing papers), BibTeX, or JSON — from the results screen (e) or the CLI.
  • Non-interactive CLI (citationimpact/cli.py): script the tool without menus — citation-impact analyze "Paper title" --format markdown -o report.md, citation-impact cache list|clear, citation-impact --version. Running ./citation-impact with no arguments still opens the interactive UI.
  • Test suite: 200+ pytest tests covering exporters, models, categorization, rankings, caches, analyzer logic, clients, UI helpers, and the CLI.
  • CI: GitHub Actions workflow running the suite on Python 3.9–3.12.

Bug fixes (72 verified defects, found by multi-agent audit)

Highlights — full details in git history:

  • Institution misclassification (critical): substring matching classified Princeton/Cincinnati as Industry ('inc'), and any "Department of …", Newcastle, or NIST-like affiliation as Government ('epa', 'cas', 'nist' substrings). Now word-boundary matched.
  • Wrong university credit: fuzzy matchers gave "National University" NUS's Top-10 rank and matched arbitrary superstring venues; QS range ranks (601-610 etc., 60% of the file) were silently dropped; ICORE junk strings ("Unranked", "TBR") were surfaced as ranks.
  • Dead features revived: methodological citations (S2 returns lowercase intents), Crossref citation-count/venue merging (key mismatch/unreachable branch), S2 DOI enrichment (externalIds never requested), ORCID fallback (wrong class name + list/dict mismatch), DBLP author publications (nonexistent endpoint), GS-only papers yielding 0 citations in comprehensive mode.
  • Crashes fixed: google_scholar mode signature mismatches, unified search_paper on zero-score results, null citationCounts, ORCID null-name records, QS range-rank parsing, None affiliations.
  • Correctness: author dedup no longer merges distinct authors (first-initial guard) and no longer loses repeat citers' papers; publication-overlap cache matching no longer attributes co-authors' profiles to each other; author-profile cache no longer merges same-name researchers; grant statements no longer conflate author counts with university counts and respect the configured h-index threshold; citation URLs no longer dropped by an operator-precedence bug; "recent citations" now really spans 2 years.
  • Robustness: config writes are atomic (API keys can't be destroyed by a failed save); connection errors are retried; cache expiry handles unlink races; UTF-8 enforced for profile files; shared client no longer reused across data-source switches (stale-mode analyses); browse-author-by-name works in API mode; Selenium clients are tracked and closed.

Packaging & hygiene

  • setup.py console entry point referenced a nonexistent module; now installs working citation-impact / citationimpact-ui commands; version single-sourced from citationimpact/__init__.py (was 0.1.0 vs 1.2.0).
  • Added pyproject.toml, .gitignore (the .citationimpact/ folder holding plaintext API keys was previously trackable), and untracked committed __pycache__ files.

1.2.0

  • Prior release (see git history).