Screens
Say it, then readwhat it wrote
"Create a screen on mexc with RSI below 35 and ADX above 20." The assistant produces a real specification — venue, market type, conditions, interval — and shows it to you before anything is created.
Why the draft is shown rather than applied
Because a screen that scans nothing looks like a quiet market
The failures in a screen specification are silent. A wrong market type scans zero symbols and reports a clean run. A condition on an indicator the screen never asked for produces no error and no result. Showing the specification means those are visible before the screen exists rather than discoverable weeks later by wondering why it never fires.
Because the drafting has known traps
Two in particular are written into the assistant’s own instructions, because they cost us real runs: a perpetual is market type "futures" here, and every condition on a feature must have that feature declared. The guidance carries the full schema so the draft is a valid specification rather than plausible-looking text.
Because creating is confirmable and deleting is not offered
The assistant can create a screen, add a version, and pause or resume scanning — each through a confirmation you press. It cannot delete one, deliberately: deletion cascades away versions, runs, schedules and alerts, and that is not an action to make available to a sentence that might have been ambiguous.
What happens after the sentence
- 01
You describe it
In whatever words you use for it, including the venue and the timeframe if you have a preference.
- 02
It drafts a spec
Venue, market type and quote come from the registry rather than from guessing, so a market that does not exist cannot be named.
- 03
The service validates it
The specification is checked by the screener before it is saved, not after it has run and found nothing.
- 04
You confirm
You read the spec and press create. Nothing scans, schedules or alerts until you do.
Questions about describing screens
What does it cost?
A drafting turn is a few tenths of a cent — creating a screen from a sentence has been measured at under a penny, and pausing one at half that. Turns are budgeted per account so the number stays predictable.
Can I edit the spec afterwards?
Yes, and editing produces a new version rather than changing the old one in place. Your alert history therefore stays attributable to the rule that produced it.
Can it build the screen presets for me?
The presets exist as starting points with an honest note on each — what it finds, when it helps, what it misses. Describing a variation of one is usually a better prompt than describing a screen from scratch.
Does it turn the screen on?
Only if you say so and confirm. Creating, scheduling and alerting are separate steps, and the console can do all three in one dialog when that is what you want.
Describe one and read what comes back
Free paper accounts get the assistant and the whole screener.