How to Read a Backtest Card

A Stingray backtest card is a rejection tool. It does not say a rule will work in the future. It shows how a typed rule behaved over a stated historical window, with enough detail to decide whether the idea is worth monitoring, tightening, or discarding.

Read the card from top to bottom: scope, data coverage, trigger quality, forward returns, then activation risk.

1. Check the scope first

The header tells you what was tested: the market, symbol, interval, and date range. If the tested window is too short, too regime-specific, or mismatched to the thesis, do not over-read the result.

Useful questions:

• Was the range user-selected or automatically chosen?
• Does the interval match the signal speed?
• Does the market match the asset or venue you meant to test?
• Did the test include the exact data source your live rule will monitor?

2. Read the data coverage note

The coverage note explains whether the replay had continuous data, sampled data, filled gaps, skipped gaps, or trimmed start/end ranges. This matters more than the headline number. A clean result on incomplete data is not clean.

For example, a one-year price backtest can be continuous, while a data source such as open interest may use sampled or segmented history. That does not make the test useless, but it changes how much confidence you should put in timing-sensitive triggers.

3. Compare raw triggers with cooldown triggers

Raw triggers are every moment the rule was true. Cooldown triggers are the notifications that would survive after Stingray applies the alert cooldown.

That distinction is important:

• Many raw triggers in a tight cluster usually mean one market episode, not many independent signals.
• Cooldown removes duplicate notifications so the card reflects what a live alert would actually send.
• If cooldown retention is very low, the rule may be detecting persistent conditions rather than discrete events.

For period-style rules, the card can show active periods and estimated notifications instead of one row per event.

4. Treat forward returns as base rates

Forward returns show what happened after the rule fired at standard horizons such as 1h, 4h, and 24h. They are not a forecast. They are the base rates of the rule you described.

Read them with sample size:

• A positive 24h return on five fires is weak evidence.
• A small 1h return can be noise if the signal is built for a slower move.
• A strong 4h result with poor 24h follow-through may describe a short-lived setup, not a swing thesis.
• Direction matters: the opposite version of the rule can be worth testing.

5. Use signal density to catch alert spam

The signal-density strip shows whether triggers are evenly distributed or clustered. A useful alert should not wake you up repeatedly for the same market state.

Dense clusters often mean the rule needs:

• A longer cooldown
• A stricter threshold
• A trend or regime filter
• A cleaner definition of the event boundary

Sparse signals are not automatically better. A rule that fired twice in a year may be too rare to evaluate.

6. Inspect the table before trusting the summary

The table gives the exact trigger time, indicator values, and forward returns for each event or period. This is where you find whether the rule is doing what you thought it was doing.

Look for:

• Indicator values that barely cross the threshold
• Triggers during obviously unusual regimes
• Events with missing follow-up samples
• Best and worst events that dominate the average
• A trigger pattern that contradicts the original thesis

7. Decide the next action

After reading the card, choose one of three outcomes:

• Reject: the rule is too noisy, too rare, too fragile, or dependent on one regime.
• Tighten: adjust the threshold, cooldown, data source, or time window and run it again.
• Monitor: turn it into an alert when the rule is specific enough and the historical behavior is understandable.

Execution should remain behind the trust ladder: notify first, preview-confirm where supported, and only use opt-in execution when the rule and assumptions are explicit.

What not to do

Do not treat a backtest card as a profit promise. Do not compare rules without checking sample size and data coverage. Do not activate a rule just because the average return is positive. The card is there to make a strategy easier to reject before it touches capital.

For concrete examples, see Strategy Examples or start with the funding-rate rule walkthrough.