Whether actual trials met the conditional target; this is conditional on the assumed sampling/support model.
Actual final-support reestimations, separate from sampled hypotheses.
Last final-support refinement report, or null when no refinement was admitted.
Local refinements refused by numerical or support admission.
Number of local consensus refinements actually attempted.
Row-major 3 by 3 epipolar model in the declared image charts; homogeneous scale is arbitrary.
Attempted hypotheses rejected for numerical admission or degeneracy.
Conditional trial requirement computed from final observed support, or null when it is not representable.
Residual-gated original correspondence indices and their pixel quantiles for the current model.
Actual random-sample hypotheses attempted under the caller's explicit work ceiling.
Estimate a calibrated pixel epipolar consensus with bounded local refinement. Every proposal is projected to the supplied-calibration essential manifold before scoring. The first bounded proposals with at least eight calibrated inliers receive local refinement before comparison with the current winner; a raw score cannot bound the result of local optimization. Final re-estimation invalidates an older sampling target; any larger target resumes proposals within the same total trial budget.