Text written by a model

anomalY uses language models to write short summaries. Numbers never come from a model: every figure in a summary comes from data that anomalY has stored.

Where models write text

• summaries of KAP disclosures,
• the weekly bulletin,
• the macro outlook, including summaries of brokerage reports,
• finding management guidance in filings for the guidance tracker.

Which models

The models run on Cloudflare Workers AI. Today they are Meta Llama 3.3 70B and Meta Llama 4 Scout.

How numbers are checked

Each summary goes through a citation audit. The audit looks for every number in the text in the stored data the summary was written from. A summary with a number the audit cannot find is marked as flagged.

What the models see

The models see only public data: filings, reports and stored market data. They never see your account, your lists or your alarms.

Limits

A summary can still leave out something important or describe it badly. Read the filing at its source before you act. Report a bad summary to hello@useanomaly.com.