Docs
REST API for retrieving assessments and paper data. All endpoints are read-only and unauthenticated.
Base URL
https://preprints.aiAll endpoints below are relative to this base.
Endpoints
GET /v1/stats
Aggregate counts for the live pipeline — total papers indexed, assessed, grade distribution, queue status.
curl https://preprints.ai/v1/stats{
"papers": { "total": 318526, "assessed": 1934, "fulltext": 316459 },
"queue": { "pending": 338, "processing": 6, "failed": 4 },
"grades": { "A": 51, "B": 637, "C": 982, "D": 278, "E": 66 },
"novelty": { "1": 120, "2": 340, "3": 890, "4": 300, "5": 60 }
}GET /v1/pipeline
Recent assessments + grade heatmap. Useful for live feeds or custom dashboards.
curl https://preprints.ai/v1/pipelineGET /reviews/{doi}
Full assessment view for a paper. Supports any DOI that has been assessed.
curl https://preprints.ai/reviews/10.1101/2025.01.15.633214POST /v1/assess
Request an assessment for a specific paper. Returns an existing assessment if one is cached, otherwise triggers a new pipeline run.
curl -X POST https://preprints.ai/v1/assess \
-H "content-type: application/json" \
-d '{"doi": "10.1101/2025.01.15.633214"}'Grading scale
Every paper receives a 10-tier Publication Fit score (the primary public signal) plus a legacy two-character A–E × 1–5 grade (kept for backwards compatibility). The tier is derived deterministically from the grade — see methodology.
| Tier | Label | Journal anchor |
|---|---|---|
1 | Landmark | Nature, Science, Cell |
2 | Broad-impact | Nature subjournals, eLife, Cell Reports |
3 | Field-flagship | PLOS Biology, society flagship journals |
4 | Specialty solid | Nature Communications, JBC, NAR |
5 | PLOS One bar (corpus median) | Methodologically adequate, narrower scope |
6 | Specialty mid-tier | MDPI, Frontiers field journals |
7 | Major revisions needed | Not publishable as-is |
8 | Reject — methodology broken | Don't publish |
9 | Inadequate evidence | Panel verdict on deep-E papers |
10 | Not for publication | Paper-mill / fabrication (Layer 1 deterministic only) |
Legacy 2-axis grade (A1–E5)
| Letter | Meaning | Integrity score range |
|---|---|---|
| A | Exceptional evidence — rigorous methods, comprehensive controls | ≥ 0.85 |
| B | Compelling — strong methods, minor gaps | 0.70–0.84 |
| C | Solid — appropriate methods, some limitations | 0.55–0.69 |
| D | Incomplete — significant methodological gaps | 0.40–0.54 |
| E | Inadequate — fundamental flaws, unreliable conclusions | < 0.40 |
Significance digit — lower number = higher novelty. A1 is the best possible grade; E5 is the worst. (Inverted 2026-05-12; the prior convention had 5 as Landmark.)
| Digit | Significance |
|---|---|
| 1 | Landmark — field-defining, transformative |
| 2 | Fundamental — major conceptual advance |
| 3 | Important — significant advance, broad interest |
| 4 | Valuable — addresses a real gap, moderate novelty |
| 5 | Useful — incremental advance, narrow scope |
Rate limits
Public API is open; no auth required. Fair-use throttle of 120 requests per minute per IP. If you need bulk access for research, get in touch — we can share a data dump rather than have you hammer the API.
Webhooks
Not yet public. If you're building an integration that needs push-style delivery of new assessments, contact us — we have a private beta.
Machine-readable specs
Full OpenAPI 3.1 spec at /openapi.json. LLM-friendly summary at /llms.txt.
Methodology
For the full pipeline architecture — four layers, eleven agents, rule-based grade derivation, Opus arbitration — see the methodology page.
Stability
The /v1 API is stable. New fields will be added additively; existing fields will not change shape within v1. Breaking changes go to /v2.
Source & transparency
Every assessment is reproducible — the paper text, model versions, prompts, and agent responses are logged. Grades are deterministic given the text labels (evidence strength, significance) agents emit — so two runs differ only where model sampling differs.