# Preprints.ai — extended machine-readable notes

The canonical, current product and API description is `/llms.txt`. This file no
longer publishes the retired two-axis grade or journal-anchored “Publication
Fit” scale as the primary signal.

## Screening contract

A compact profile can contain three independent axes, for example `AC4`:

- Evidence `A–E`: rubric-guided model screening of methods and claim support.
- Trust `A–E`: applicable transparency and reporting markers.
- Novelty `1–10`: a machine-generated distinctness signal; lower is more
  novel. The API labels its basis as literature-grounded, reconstructed from
  retained literature artifacts, LLM-derived, or unknown.

The API may retain `publication_tier` and `tier_label` for backwards
compatibility. They are experimental derived fields, not validated predictions
of journal placement or editorial decisions.

## Pipeline

Full text is required. Automated Layer-1 checks run before an eleven-role model
panel. Structured findings are synthesized by a rubric-guided judge, with
deterministic hard gates for safety-critical findings. Literature retrieval
grounds Novelty when available. Insufficient panels are deferred; degraded
checks and fallback grading are recorded.

`screening_signal.status` is `complete`, `limited`, or `withheld`; each axis is
separately `available`, `limited`, or `unavailable`. `provenance_complete: true`
requires full-text basis, a full recorded panel, non-empty and non-degraded
safety checks, non-empty literature grounding, the rubric judge, and valid
score contracts. It
does not mean that the paper or its conclusions are scientifically verified.
Legacy Novelty values with retained model findings or literature artifacts are
shown as limited signals with an explicit basis label rather than withheld.

## Limitations

Most indexed papers are not yet assessed. Context windows are finite, model
errors can be correlated, Trust applicability is partly heuristic, Novelty
depends on retrieval coverage, and repeat model runs can differ. All outputs are
machine-generated screening signals requiring expert interpretation.

See `/llms.txt`, `/methodology`, `/evidence`, `/v1/stats`, and `/pipeline` for
the current public description and live state.