Canonical: https://botbento.com/usecases/verification/
Format: Markdown representation of the public HTML page.

BOTBENTO / USE CASES

# How use cases are verified

798 of 798 published use cases currently carry the badge.

The Verified mark on a use case means it passed every automated gate below on the date shown. It is an automated process with AI review. No person reviews each entry before it is published, and the badge does not claim one did.

## Automated gates

- Schema. Required fields, lengths, known categories, tools and workflow steps, and consistent dates.
- Uniqueness. One entry per source URL, title and tagline across the whole directory.
- HTTPS and live source. Every link is HTTPS, with no URL shorteners, raw IP or punycode hosts, lookalike domains or links to installers. The source page answered a rate-limited HEAD/GET request.
- Injection and safety scan. No instructions aimed at AI agents (for example “ignore previous instructions”), hidden or zero-width characters, encoded payloads, shell commands piped from the internet, or prompts that send secrets or credentials anywhere.
- Approval-step rule. A prompt that sends, posts, pays, deletes, books or contacts anyone must tell the bot to wait for your approval first.
- Second-model review. A different model from the one that drafted the entry checks it against the source for accuracy, unsupported “works with” claims, missing guardrails, thin content and copying. Any critical or high finding blocks publishing.

## Models

- Drafting: the pipeline drafts with google/gemini-3.8-flash via OpenRouter by default, in our own words from the public source. A badge names the drafting model only when it was recorded at drafting time; entries drafted before recording started say “not recorded”.
- Review: a strong model that differs from the drafter, tried in order: deepseek-v4-pro (DeepSeek API), then Grok and DeepSeek through other providers or OpenRouter, with a per-run spend cap. Lightweight model tiers are never used for review. Each entry’s badge names the model that actually reviewed it.
- Arabic: translated by a strong model (currently anthropic/claude-sonnet-5.5 via OpenRouter) and checked by a reviewer model. The badge on an Arabic page names the translator and reviewer recorded for that translation. The English text is the reference.

## What it does not mean

Verification checks the entry, not the tools it mentions. Results quoted from a source are the source authors’ own. If anything is wrong, [tell us](/contact/); an edited entry loses its badge until it passes again.

[Use cases](/usecases/)