> ## Documentation Index
> Fetch the complete documentation index at: https://docs.portir.xyz/llms.txt
> Use this file to discover all available pages before exploring further.

# News check

> An LLM reads recent headlines before every buy, and may only hold the buy back.

After the price and market-hours checks pass, the agent asks a language model one question: *should this buy go ahead now, or wait up to a day?*

## What it reads

The latest headlines for each holding (Yahoo Finance, plus CoinDesk crypto/RWA context when configured).

## When it says wait

Only for a **concrete near-term risk** in the headlines:

* earnings or guidance within about 24 hours
* a trading halt
* fraud or accounting issues
* a regulator or court action
* a sharp news-driven move happening right now

Ordinary news, analyst opinions and general market chatter mean *buy*.

## Guarantees

| Property | Behavior |
| - | - |
| Power | It can only hold a buy back. It can never trigger, enlarge or redirect one |
| Transparency | The one-sentence reason (naming the headline) is written on-chain with the run |
| Fail-open | If no model answers, the buy follows the Guard alone |
| Window | Once the plan's window is over, the schedule wins over the news |
| Cost | One opinion per ticker per hour, cached across 15-minute scans |

Example recorded reason: *"Price is fair, but holding back for the news: NVIDIA reports earnings after the close today."*


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.