# Pre-flight

> The publish gate. A draft that fails does not reach the approval queue.

Canonical: https://clawli.ai/concepts/preflight

When the cost of producing a competent post drops to zero, competence stops being a signal. Pre-flight exists so the system cannot ship the median.

## The five checks

- Proprietary detail: at least one number, name, or outcome only the author could know.
- Named cost: what the approach broke, who complained, or what it took to undo.
- A position: a claim a reasonable person could disagree with.
- Opening line: a sentence, not a manufactured curiosity gap.
- Slop markers: the recognizable phrasing patterns of unedited model output.

## How it runs

Pre-flight runs after drafting and before the approval queue. Each check returns a pass, a fail, and the specific lines that triggered it, so the fix is a rewrite of two sentences rather than a regeneration of the whole post.

A failing draft is returned to the drafting agent with the failed checks as constraints. Two failed attempts escalate to a human with the diagnosis attached, rather than looping until the token budget runs out.

## Why it is a gate and not a score

Scores get ignored. A number in the corner of a screen becomes wallpaper within a week. A gate that refuses to advance the draft changes what gets published, which is the only measurable outcome that matters.

## Related

- [Pre-flight demo](https://clawli.ai/preflight)
- [Anti-slop standard](https://clawli.ai/concepts/anti-slop)
- [Evals or it didn't ship](https://clawli.ai/blog/evals-or-it-didnt-ship)