Docs / Rule language

Rule language

Common mistakes

The rules that look right and are not, with what they really answered and what to write instead. Every number is a real answer.

1. Using max() for a breakout

C > max(C, 20) can never be true, because the window includes today. On XLE, which closed at a 20-session high, it is 0.5 while C > highest(C, 20) is 0.6462.

Instead: Use highest/lowest for breakouts and max/min for "where it stands now".

2. Expecting a tie to count as true

>= and <= answer 0.50 on an exact tie, not a high belief. C >= max(C, 20) can at best tie, because the window contains today: on XLE, at its 20-day high, it is 0.5, and on SPY C >= max(C, 1) (today against itself) is 0.5.

Instead: Give the line some room: C >= 0.98 * max(H, 20), or use the excluding form.

3. Putting an offset on a function

sma(C, 20)[-1] is not valid: Unexpected token '[' at column 11

Instead: ref(sma(C, 20), -1).

4. Looking forward by accident

C[1] is the next bar, which does not exist: C > C[1] on XLE is 0.5.

Instead: Offsets that look back are negative: C[-1].

5. A window longer than the history

About 1,500 sessions are loaded; C > sma(C, 2000) is 0.5 ("not known") on XLE.

Instead: Keep windows under about 1,000 sessions; a year is 252.

6. Comparing today with an average that includes today

On a spike day, today's volume is part of avg(V, 20), which dampens the comparison.

Instead: V > 2 * ref(avg(V, 20), -1) compares with the 20 sessions before today.

7. Mixing and with or without parentheses

and binds tighter. On AAPL, a or b and c is 0.9119 and (a or b) and c is 0.7437.

Instead: Always bracket a mix of and and or.

8. Mixing units

Returns written by hand are fractions (C / C[-1] - 1 > 0.02 is 2%), but roc() and atr_pct() are percents (roc(C, 20) > 5 is 5%).

Instead: Check the unit on the function's entry in the reference.

9. Writing a state field

rsi > 70 reads rsi as a field, and rules on Ora have none: No field "rsi" — this request has no named state, only price bars (O, H, L, C, V and the functions).

Instead: rsi(C, 14) > 70.

10. Trusting that an English question became a rule

"Is it at a 52-week high?" was answered by the forecast model on TSLA (0.533, no rule in the answer), because the compiler did not recognise it as a "now" question.

Instead: Check for rule in the answer, or write it: H > highest(H, 252).

11. Reading an event as a state

crossup(C, sma(C, 50)) is 0.95 only on the day of the cross and 0.05 on every day after, even while price stays above.

Instead: Use C > sma(C, 50) for the state, or barssince(crossup(…)) <= 5 for "recently".

12. Reading a belief as a probability

A rule's 0.70 does not mean a 70% chance of anything; the condition already holds or not. It says how clearly.

Instead: Use rule beliefs as filters and strength, forecasts as odds: Reading beliefs.

Information, not investment advice. Beliefs and probabilities on this page are real responses for the session shown and change with every session.