# Results

Analysis status: **blocked**.

This is a reproduction and robustness audit of the pinned paper-release standard-PE samples. It is not a discovery analysis, population analysis, or event classification.

## Reproduction ledger

- Derived-quantity comparisons completed: 48
- Comparisons with any sample outside preregistered tolerance: 0
- Largest sample-by-sample absolute algebraic difference: 8.882e-15
- Rounded paper/portal comparisons completed: 176
- Paper comparisons outside rounding tolerance: 7
- All such paper differences are confined to the HPD-in-cosine tilt bounds. Ordinary central-interval table quantities agree at their printed precision; the tilt estimator difference is retained and diagnosed rather than absorbed into wider tolerances.

## One-dimensional marginal information gain

- KL estimates completed: 0
- KL missing/block records: 6
- The release's six matching `priors/samples` groups are empty and its `config_file` groups do not contain a complete machine-readable prior specification. KL is therefore blocked and no prose prior is substituted.
- Values are marginal Bayesian-surprise estimates in bits. They are prior dependent, non-additive, and are not the event's total information.
- Bootstrap intervals quantify Monte Carlo estimator variation, not physical credible intervals.

## Pairwise posterior-analysis divergence

- JS comparisons completed: 42
- Every value is scipy's base-2 Jensen-Shannon distance squared, so it is a divergence in bits.
- Combined samples are excluded. Prior-vs-prior check status, bin sensitivity, and same-model split-sample noise-floor distributions are recorded in metrics.json. The prior-vs-prior checks are unavailable here because the released prior groups are empty.
- Marginal JS does not measure changes in joint correlations and has no universal scientific-significance threshold.

Descriptive maxima within the audited marginal set (not significance rankings):

- GW241011_233834: largest observed audited marginal was `chi_eff_infinity` for IMRPhenomXO4a vs IMRPhenomXPHM: 0.587667 bits; 90% bootstrap Monte Carlo interval [0.581258, 0.595403] bits.
- GW241110_124123: largest observed audited marginal was `chi_eff_infinity` for IMRPhenomXO4a vs IMRPhenomXPHM: 0.016942 bits; 90% bootstrap Monte Carlo interval [0.015594, 0.019612] bits.

## Guardrails

No hierarchical-merger classification, population inference, formation-channel Bayes factor, ancestral inference, eccentricity inference, or qualitative divergence grading is performed.

The complete machine-readable result is [outputs/metrics.json](outputs/metrics.json).