01 / THE PRACTICAL QUESTION
A decision, before a model.
A forecast interval looks reassuring until volatility changes and observations begin escaping it more often.
Monitor how often intervals contain the observed value, alongside how wide they must become.
02 / DATA & COMPARISON
The idea in plain language.
Conformal prediction calibrates uncertainty using past prediction errors. Its usual coverage guarantee depends on exchangeability; a regime shift can break that condition.
- Coverage
- The fraction of observed outcomes inside their forecast intervals.
- Exchangeability
- The assumption that observations can be reordered without changing their joint distribution.
Data. Separate training, calibration, and evaluation samples.
Baseline. Equivalent-target conventional intervals on held-out observations.
03 / THE EXPERIMENT
What the saved experiment shows.
The saved finite-sample check selects the seventh ordered calibration score from nine observations at 20% miscoverage, matching the expected statistic. This checks the implementation; the interactive story separately demonstrates coverage under changing conditions.
Inspect the supporting resultEvidence record: research-validation.json#numerical_checks/11
Explore the related lab04 / RESULTS & LIMITATIONS
Evidence with its boundaries attached.
The related lab is a cross-project demonstration. Read this investigation’s evidence and limits before transferring its conclusions.
Inspect numerical checks and validation records
Loading validation evidence…
Exchangeability can fail in time series. Observed coverage is not a guarantee under shift.
05 / REPRODUCE
Reproduce and challenge the result.
Code, configuration, and reproduction
The project contains its implementation, configuration, tests, and walkthrough. Download the lab configuration to record the exact parameters used in an interactive run.
Project code and walkthrough