This document defines the shipped runtime contract for lpp_seed and lpp_refined_predictor.
It is separate from the benchmark-leader summary in CANDIDATE_CATEGORIES.md.
The Lorentz Prime Predictor defines two related objects, and this repository keeps them separate in both prose and measurement.
lpp_seed is now the repository's official deterministic inversion seed. It returns an integer estimate for the $n$th prime for
For the main regime, this is the fixed-step inverse construction defined in FORMULA.md. For the narrow low-index compatibility window below 100, the runtime uses the legacy closed-form path so the public API stays defined on the full declared domain.
This is the correct object for the repo's primary seed contract. The older closed-form path and the other comparison formulas remain available as alternates, not as the shipped default.
lpp_refined_predictor starts at the official seed and moves forward deterministically until a prime output is reached, again for
This answers a different question. It measures whether the shipped seed is a useful launch point for a practical prime-returning method, not only whether the seed lands near
A strong refined predictor does not by itself prove a stronger seed formula. A strong seed formula does not by itself prove the best prime-output stack. The two layers support different claims and must not be merged into one headline result.
For this repository, that means:
- seed tables report seed accuracy only
- refined tables report prime-output accuracy only
- summary language must name which estimand is being discussed
- refined comparisons are valid only when the refinement rule is shared across predictors
The codebase also exposes:
legacy_lpp_seedcipolla_log5_repacked_seedli_inverse_seedr_inverse_seed
r_inverse_seed is the explicit method name for the same construction that now ships as lpp_seed. The other three names are alternate formulas retained for benchmark and research work.