Three things worth doing first

These are complete sessions, not prompt snippets. Each one shows what you say, what the assistant does, and — most importantly — where to push back, because the difference between a useful answer and a confident one is whether it can show you the receipt.

1. Build and test a strategy without opening the app

The shortest path from an idea to a number you can trust.

You say

Build me an EMA(20)/EMA(50) crossover on BTCUSDT 1h.
Long when the fast crosses above the slow, flat otherwise, 2% stop loss.
Then backtest it over 2024 and tell me what happened.

What happens. The assistant reads the node catalog so it uses real node types rather than invented ones, creates the strategy, and compiles it. A compile that fails comes back with diagnostics and it repairs in place — you may see a couple of rounds of this. Then it binds a backtest to that exact revision and reports trades, return, and drawdown.

Push back with

Which revision did you test, and did it compile clean?

A real answer names a revision and a compile receipt. If it cannot, the numbers are describing something other than the strategy you are looking at.

2. Turn a disappointing result into a tested variant

The part people skip. Do not edit the original — branch it, so you can compare.

You say

That drawdown is too deep. Branch it and try a 1% stop
and an ATR filter so we skip the chop, then backtest both
over the same window and show them side by side.

What happens. It creates an experiment branch, edits only the branch, compiles, and runs the second backtest over the same window so the comparison means something. You end up with two revisions and two result sets, both still openable in the Spera app.

Push back with

Same symbol, timeframe, and date range on both? Show me the losing trades from the worse one.

Changing the window between runs is the easiest way to make a variant look better than it is. Ask.

3. Hand it over with the evidence attached

A strategy nobody can evaluate six months from now is not finished.

You say

Document this strategy: what it trades, why the filter is there,
and what the backtest actually showed. Cite the run.
Then record in the project notes why we moved to the 1h timeframe.

What happens. It writes documentation attached to that exact artifact revision, grounded in the real compile and backtest receipts rather than a summary of the conversation, and adds a project note you can cite later.

If the strategy has a protected STRATEGY.md, changes to it arrive as a proposal you accept explicitly — the assistant cannot write that file directly, by design.

Push back with

Is every performance claim in there backed by a run you actually did?

What it will not do, however you ask

Deploy a bot, start live trading, read your exchange keys, or move money. Those tools are not on the server, so this is not a refusal you can rephrase past — see the full tool list. Deploy from the Spera app when you are ready.

Next

• Every tool, and what it does
• What to ask for
• When it doesn't connect