Evaluating an AI writing tool without getting burned by invented claims

In Lekha, no — there is deliberately no action meaning 'make this pass review.' The guard re-runs on every rewrite regardless of how the request is phrased.

A price, a number, or a proper noun that the rewrite dropped compared to the original — the specific categories of fact most likely to matter and least likely to be noticed missing from a smooth-reading rewrite.

Not for every rule. Forbidden-term and required-phrase rules are pattern matching against the workspace's own guideline rows and run with nothing configured. Tone and style rules — the ones that need judgment rather than a word list — do run, but need an LLM key; without one they are skipped and reported as skipped rather than counted as passed.

No — editing an approved document takes it back out of approved status automatically, so an approval can never quietly attach to text that was never reviewed.

No — Hinglish is named as its own explicit option because a model told only 'Hindi' tends to produce formal transliteration most of the intended audience does not actually speak or read casually.

Not yet — its status is Coming soon, the same honest label used next to every agent still in build on this site.