Changelog

2026-08

Research Tool Comparison Guides

• Expanded the 15-page research-tool decision cluster to all 14 supported languages: one comparison hub, five direct comparisons, four alternatives guides, and five task-specific best-of guides (210 localized URLs).
• Each locale now has native search terminology, UI copy, metadata, FAQ, CTA, canonical and hreflang output. Rankings, source URLs, review dates, and published price facts continue to share one verified factual baseline.

Open-Access Paper PDFs

• Expanded original PDF coverage: when PubMed Central has no full text, LitSource now uses the DOI to query Unpaywall and look for a verifiable open-access PDF.
• PubMed Central remains the preferred source; an Unpaywall link is shown only when the record is explicitly open access and provides a direct PDF URL.
• The PDF endpoint now returns the source, open-access status, version, and license for easier verification and traceability.
• If no open full text or valid PDF URL is available, or an upstream service fails, LitSource fails closed and never guesses a download URL from metadata.

MCP Support for AI Agents

• Launched MCP (Model Context Protocol) endpoint at /mcp with Streamable HTTP transport.
• Available tools: litsource_search_evidence (1 credit with results), litsource_verify_references (existing Verify formula), litsource_get_run (free), litsource_get_usage (free).
• Compatible with Codex, Claude Code, and Cursor — see the MCP connection guide for setup instructions.
• Paid tools require a stable request_id for idempotent retries.

REST API v1

• Launched public REST API v1 with scoped Agent Token authentication.
• Endpoints: POST /api/v1/evidence/search, POST /api/v1/references/verify, GET /api/v1/runs/{id}, GET /api/v1/usage.
• OpenAPI 3.1 specification available at /api/v1/openapi.json.
• Paid POST endpoints support Idempotency-Key for safe retries; errors and zero-result searches charge 0 credits.
• Interactive API explorer and code examples (curl, JavaScript, Python) at /docs/api.

Agent Token Management

• New settings page at /settings/agents for creating and managing scoped Agent Tokens (lsa_…).
• Per-token scope control: evidence:search, references:verify, runs:read, usage:read.
• Token revocation and expiration support.
• Per-token credit usage tracking with UTC-day boundary.

2026-04

Broader Subject Support, Faster Responses, Still Biomedical-First

• Retrieval and verification flows now support more non-biomedical academic use cases, including interdisciplinary writing, methodology references, and general scholarly claims.
• Core response speed has been improved again, so common searches surface the first usable results earlier and complete faster overall.
• Multi-source orchestration and fallback retrieval have been streamlined, making complex inputs feel more responsive.
• LitSource remains primarily focused on biomedicine, and ranking quality, terminology handling, evidence judgment, and coverage in that domain will continue to get priority.
• For broader subject support, the current goal is better usability and wider coverage first; for biomedicine, we are still optimizing for higher accuracy and stronger evidence quality.

Biomedical Evidence Search Is More Accurate, Faster, and More Decisive

• Evidence search now performs better on complex biomedical claims, especially mechanism-heavy, risk-association, and comparison-style queries.
• Support, partial support, against, and uncertain classifications are now more stable, so results do more than return related papers and better reflect evidence direction.
• Core conclusions are now surfaced earlier in the results view, with support/against analysis treated as a primary output rather than a secondary layer.
• The fallback retrieval path has been optimized for speed, reducing wait time when extra literature expansion is needed.
• Real-user high-frequency biomedical searches are now included in focused regression testing, making future quality improvements more representative of real usage.

More Accurate Reverse Search for Complex Biomedical Inputs

• Reverse search now works better for long biomedical claims pasted directly from drafts, notes, or research outlines.
• Results are now less likely to drift toward loosely related papers when a biomedical claim contains multiple mechanisms or domain-specific terms.
• Complex biomedical inputs now return evidence that is closer to the user’s actual claim, not just the broader topic.
• Long-form biomedical paragraphs are now handled more cleanly, with better focus on the parts that matter most for evidence matching.
• Biomedical mechanism-heavy searches now surface more useful papers for claims involving fibrosis, microcirculation, extracellular matrix changes, and functional decline.
• We also expanded testing with harder real-user biomedical searches to improve quality in practical use.

2026-03

Reverse Literature Search Improvements

• Improved long-claim handling so complex biomedical inputs are checked in more meaningful sentence-level units.
• Strengthened Chinese-to-English claim matching for literature retrieval.
• Expanded evidence recall to surface more candidate papers before ranking.
• Improved ranking to favor papers that are closer to direct support rather than loose topical similarity.
• Added evidence-type filtering in results: Support, Partial support, Against, and Uncertain.
• Improved sitemap coverage for blog content so new posts can be discovered more consistently.
• Published two new blog posts on evidence-focused search improvements in English and Chinese.

Verify Improvements

• Expanded support for more Chinese-language journal and local-language reference records in LitSource Verify.
• Improved matching accuracy for English-language citations with DOI-based verification.
• Reduced cases where valid English references were marked as suspicious because metadata alignment was weak.
• Improved reference recovery when a citation text is incomplete but a trustworthy DOI-backed record can still be identified.

PubMed Search Assistant Lite (Chrome Extension)

• Launched the Chrome extension: PubMed Search Assistant Lite, now integrated with LitSource.
• Added PubMed query generation with direct in-page query viewing.
• Added journal impact factor lookup to support faster paper screening.
• Added query export for easier strategy sharing and reproducibility.
• Added batch export for papers and query spreadsheets.

2026-02

LitSource Verify

• Launched LitSource Verify at /verify for reference authenticity checks.
• Added detection/classification pipeline: Verified / Suspicious / Fake with DOI + Crossref metadata matching.
• Added credit-based billing for verify runs (shared credit pool, no separate quota plan).
• Improved multi-reference parsing for pasted content (supports inline numbered references like 1) ... 2) ...).
• Added paste format example block to guide first-time users.
• Tuned title-match thresholds to reduce false Suspicious cases when Crossref match quality is high.
• Added Suggested citation output with one-click copy.
• Added citation style selector in Verify results: APA, MLA, Chicago, BibTeX, Vancouver, IEEE, Harvard, GB/T 7714.

Sharing

• Added one-click share links for analysis results.
• Added token-based shared result pages with expiration support.
• Shared visitors can view evidence without login, while AI summary and personal actions remain login-gated.

Paper Access

• Added original PDF open/download flow when a source PDF is available.

AI Summary

• AI paper summaries now consume credits and return clearer error messages when credits are insufficient.

Payments

• Fixed checkout callback origin resolution to use configured app URL (avoids localhost callback issues).
• Added admin reconciliation for pending Creem orders to sync paid/expired/canceled outcomes.
• Improved Creem status mapping to safely handle pending/expired/canceled/failed states.

Pricing

• Updated plan details (Pro monthly/yearly credits, AI summary capability, plugin-in-progress messaging).
• One-time package is now displayed as One-Time, with clearer scope and naming consistency.
• One-Time now includes Pro export entitlement during its validity window.

Localization

• Replaced native browser tooltips with site-styled tooltips.
• Localized results UI, history, filters, and paper cards.
• Expanded supported locales: Korean, Japanese, German, Spanish, Italian, Russian.
• Added locale normalization/fallback improvements and language-aware locale switch prompts.

SEO

• Added sitemap generation with multilingual alternates for core pages.

2026-01

Results experience

• Improved sentence display with original sentence + hover translation.
• Added caching for search runs and improved history replay.