Candidate set
The metric requires a history of comparable variants rather than one published winner.
VALIDATE / OVERFIT RISK
The in-sample winner can become weak after selection.
PBO studies candidate behaviour across partitions and estimates whether the selection procedure rewards a random peak.
EVIDENCE / VERDICTRESILIENT
UNSEEN
HOLDOUTPASSDSR0.93PBO18%
SYSTEM DOMAIN03 / VALIDATE
Challenge the candidate before risk
- DISCOVER
- OPTIMIZE
- VALIDATE
- RUN
- OBSERVE
A result is a record, not a green badge.
The page follows the same order as the application: declare the question, inspect the run, then state what the data cannot prove.
- 01
Candidate set
The metric requires a history of comparable variants rather than one published winner.
- 02
Partitions
In-sample ranks are compared with out-of-sample behaviour for the same variants.
- 03
Selection risk
High PBO signals that the space or economic mechanism needs revision.
- 04
Not loss probability
PBO estimates procedure overfitting, not the probability of a future losing trade.
Partitions
In-sample ranks are compared with out-of-sample behaviour for the same variants.
Selection risk
High PBO signals that the space or economic mechanism needs revision.
Not loss probability
PBO estimates procedure overfitting, not the probability of a future losing trade.
DECISION FRAMEWORK
Which independent test could still reject the candidate?
Treat holdout, time stability, multiple-testing correction and stress paths as separate evidence. Agreement matters more than one green metric.
Protect unseen data
Record every trial
Accept FAIL and insufficient data
Validation verdict
Important
Historical, statistical and model results do not promise future returns. kquant provides computational tools and technical execution, but does not assess whether a specific trade is suitable for a user.