Required inputs
Locked astrology model
Ordinary baseline
Identical eligible tasks
Scoring rule
Minimum sample
Ordered execution
Work the protocol
- 1
Choose the baseline before outcomes.
- 2
Give both models identical non-outcome information.
- 3
Score all tasks including abstentions.
- 4
Calculate paired differences.
- 5
Report the baseline even when both models fail.
Outputs
- Paired score table
- Coverage and abstention rates
- Baseline comparison verdict
Done only when
- Task sets are identical.
- The comparison is prospective.
- Uncertainty and null results are reported.
Fail closed
Refusal conditions
- Only astrology successes are sampled.
- The baseline gets less information.
- The comparator is changed after scoring.