A useful AI paper trading prompt specifies the market, interval, entries, exits, position rules and observation plan, then asks the assistant to resolve missing assumptions before testing. In Premiss, inspect the generated strategy and historical evidence before creating a versioned paper run. A prompt is a starting specification; it does not prove that a strategy works or start a bot from this page.
A Premiss-published reference. Sources and the ownership relationship are linked below.
Replace an ambition with an observable rule
“Find a winning strategy” supplies a desired outcome rather than a reproducible instruction. Begin with an observable condition and the information needed to calculate it. Describe the market and decision interval, then state what would make the strategy enter, exit or do nothing. If a term such as dip, momentum or reversal is still subjective, ask the assistant to identify that ambiguity before it writes code.
Ask for missing choices before a test
A useful assistant should identify unresolved sizing, timing, cost and position-state choices. It should also confirm whether the chosen market, interval and history are supported. An invented default can make a report appear more complete while changing the actual research question. Record any agreed defaults in the specification and review the implementation against them.
Request evidence you can inspect
Ask for readable strategy code and the actual historical trade record. Select an entry, exit and no-trade period to audit manually. If the strategy depends on a completed candle, identify when its final value became available and when a simulated fill could occur. A narrative about the idea should not stand in for a recorded engine result.
Separate the historical test from the paper plan
The historical test and the subsequent forward observation window answer different questions. Freeze the version you intend to observe, record its start time and define an initial review horizon. Also define operational pauses, such as an unexplained data interruption. Those pauses are different from the strategy’s own exit rule. Preserve changes as new versions.
Use a review prompt after observations accumulate
At review, ask what decisions actually occurred and whether they matched the saved rule. Include quiet periods and interruptions. Explain whether a difference came from a new market sequence, changed assumptions, a revised strategy or an operational problem. Do not ask the assistant to rewrite the strategy until the existing record looks attractive; first preserve what the original experiment showed.
Weak prompts and more inspectable alternatives
| Unresolved request | Clarification to add | Evidence to ask for |
|---|---|---|
| Buy dips | Define the dip, decision time and eligible market | A sampled signal and its inputs. |
| Use sensible sizing | Specify simulated allocation and position-state rules | The sizing implementation and an example position. |
| Make it realistic | Name fill timing and cost assumptions | What the engine modeled and what it omitted. |
| Tell me if it works | Define the research question and review window | Historical trades and separate forward observations. |
A prompt is a specification, not a result
For an illustrative daily rule, “use only completed candles” removes one ambiguity but leaves several others. You still need the exact entry condition, exit condition, position allocation, warm-up and permitted fill. Resolve those choices and confirm engine support before interpreting a report.
The template below intentionally contains placeholders. Replace them with your own research specification and ask Premiss to identify unsupported inputs. Nothing is submitted or executed when you copy it from this page.
Copy the complete paper-only research prompt
Copy this starting request, replace the placeholders and review it in Premiss.
Help me design a paper-only experiment for [supported market] on [interval]. My proposed observable entry is [condition], exit is [condition], and simulated position allocation is [rule]. Before testing, identify ambiguous terms, required history, warm-up, decision timing, assumed fills, costs and existing-position handling. Confirm what the engine supports and ask me to resolve missing choices. Show readable code and actual historical trades; audit an entry, exit and no-trade interval. Keep that report separate from a frozen paper run with [review horizon] and explicit operational pause conditions. Preserve revisions as new versions. Do not invent returns or place real-money orders.
Copying stays on your device. Opening Premiss does not submit this prompt, run a test or start a bot.
Questions about this workflow
Can I copy this prompt directly into Premiss?
Yes, as a starting research request. Replace placeholders and review the missing choices in the application. Copying from this page does not automatically create or start anything.
Should I ask the AI to optimize the backtest return?
First define how the experiment will be evaluated and preserve the trial history. Selecting among many variants on the same historical data can make the chosen result hard to interpret.
Do I need to share exchange keys to use this template?
No. This is a paper-only research specification. Premiss documents paper simulations separately from approved live exchange access.
Sources and next steps
Product sources establish the published scope. Worked examples, checklists and prompts are original editorial material. External technical references do not certify the behavior of the Premiss engine.
- Premiss product and workflow ↗Official product scope: AI-assisted research, readable strategies, historical tests and paper bots.
- Premiss plans and execution boundaries ↗Current research allowances, simulated paper bots and separate approval for live access.
- Alpaca: paper and live trading differences ↗A broker’s explanation of simulation limits. Its platform details do not establish how Premiss models fills.