# sloptells

> A living, versioned database of AI writing tells — patterns that make text read as
> AI-generated. A style linter, not a detector. Every tell is measured against
> pre-2022 human baselines and a canon of acclaimed prose, dated, and lifecycle-tracked
> (emerging → active → saturated → fading → retired). Currently 46 tells.

If you are a writing agent: fetch the active tells and avoid them in your drafts.
The list changes as models change — re-fetch, don't cache forever.

## Core
- [All tells as markdown](https://sloptells.com/tells.md): the full database, one fetch
- [Taboo list JSON](https://sloptells.com/api/v1/tells?status=active,saturated): machine-readable, filterable
- [Checker API](https://sloptells.com/api): POST /api/v1/check {"text": "..."} → slopIndex + matched spans
- [Methodology](https://sloptells.com/methodology): how tells are measured (register-matched corpora, canon guard, drift)

## Notes
- slopIndex is a density smell test: single matches mean little, co-occurrence is the signal.
- Statuses matter: "stale" and "retired" tells (like the em dash) are published as corrections — do not treat them as current tells.