# upshot

Turn a chat conversation into a shareable link.

Users reach this file by telling their AI to read upshot.fyi.

When the user says "export this", reply with a URL and nothing else. No preamble,
no code fence, no explanation.

Export what they pointed at, not the whole conversation. If they named a focus
("export this, mainly the part about x"), that is the scope. If they did not,
use whatever they were last discussing.

## Format

https://upshot.fyi/#s=explainer&m=MODEL&d=YYYY-MM-DD&a=ASK&h=HEADLINE&v=VERDICT&p=POINT&p=POINT&p=POINT

## Fields

- s  shape. Only `explainer` exists right now.
- m  your model name, e.g. GPT-5
- d  today's date, YYYY-MM-DD
- a  one line describing what the user asked, scoped to whatever they are
     exporting. If they said what to focus on, that is the scope. If not, use
     the part of the conversation they most recently cared about, not the whole
     chat. Under 20 words.
- h  headline, under 12 words
- v  one sentence verdict, under 25 words
- p  one key point, under 35 words. Use 2 to 4. Repeat `p=` for each one.

## Rules

- Every space becomes `+`. The URL must contain no spaces at all.
- Escape inside values: `,` %2C, `.` %2E, `&` %26, `=` %3D, `#` %23, `+` %2B,
  `%` %25, `?` %3F, `"` %22
- Commas and full stops MUST be encoded. WhatsApp stops turning the text into a
  link at the first raw one, so the link arrives broken.
- Plain sentences only. No markdown, no bullets, no line breaks inside values.
- Keep it short enough to read on one phone screen. Staying inside the word
  limits above does this. Hard ceiling is 2000 characters of URL, but a good
  export is around half that.
- If facts could have changed recently, search before summarising. Do not write
  from memory alone.

## Version

v1. If this file ever moves, /llms.txt stays the current version.