Level 7 · Lesson 3 of 5

Prompting for interpretation

7 min · Goal: Write a prompt that produces checkable interpretation rather than confident narrative.

Give it the positions

The single biggest improvement: compute the chart yourself and hand the model the positions, rather than giving it birth details and asking it to work them out. That removes the arithmetic failure mode entirely and leaves the model doing the thing it is actually good at.

Include the Lagna, the planetary positions with signs and degrees, the house each occupies, and the current dasha. That is enough context for genuinely useful interpretation.

Ask for reasoning, not conclusions

A prompt asking "what does this chart say about career" gets a confident paragraph. A prompt asking "which factors bear on the 10th house here, and do they agree or conflict" gets something you can check.

The second form also surfaces disagreement rather than hiding it, which is what Level 5 taught you to look for.

Four instructions worth including

Ask it to state how common each configuration is. This directly counteracts the training-data bias toward describing common yogas as remarkable.

Ask it to report where schools disagree rather than choosing. This surfaces the smoothing failure.

Ask it to distinguish what the tradition holds from what it is inferring. Models blur this constantly.

Forbid prediction and outcome claims explicitly. Ask for emphasis and tendency, in the language Level 6 established.

And then check it

A well-prompted model still fabricates, just less. Verify any classical claim, any base-rate figure, and any calculation. The prompt improves the draft; it does not remove the need to read it critically.

Try it with your own chart

Write and test a prompt

Take your computed chart and write a prompt that supplies the positions rather than the birth details.
Ask for the factors bearing on one house, whether they agree, how common each is, and where schools differ.
Explicitly forbid predictions and outcome claims.
Run it, then check three claims from the answer against what you learned in Levels 2 to 6.
Note how many survive checking. That ratio is the useful thing to know.

Try our astrologer

Key points

Supply computed positions rather than birth details — it removes the arithmetic failure mode.
Ask for factors and whether they agree, not for conclusions.
Explicitly request base rates, reported disagreement, and tradition-versus-inference.
Forbid prediction and outcome claims in the prompt.
A better prompt improves the draft; it does not remove the need to check it.

Check yourself

Why is supplying computed positions better than supplying birth details?

← Verifying the computation Where human judgment is irreplaceable →