dcl-core v0.2.1: exact integer A=1, and the hunt for a probability floor
dcl-core v0.2.1 is out — deposited on Zenodo (doi:10.5281/zenodo.20615410) and tagged on GitHub. It bundles the v0.2.0 feature release with a v0.2.1 patch.
What changed: probability you can count
The engine now does A=1 by exact integer arithmetic. A session is a fixed budget of N indistinguishable probability tokens spread over the bipartite octahedral lattice, and \mathcal{A}=1 means \sum_x N(x) = N exactly — no floating-point drift, no renormalisation fudge. The fractional bits that used to live in floating point are carried by a Bresenham-style residual accumulator, so the books balance to the integer. This integer-token engine sits alongside the original continuous-amplitude engine (a verbatim port of Paper I), under one import.
The question it lets us ask: is there a minimum Δp?
This is the release that makes the δp_min investigation testable head-on. The question is simple to state: is there a minimum delta-probability — a floor? With integer tokens there is a natural candidate, because the smallest change the lattice can make is one token out of N.
A floor would mean probability is pixelated, and that is falsifiable with teeth. It says interference and coherence are pixelated — fringe contrast cannot fall continuously to zero, and coherence carries a grain. And it predicts that higher-dimensional lattices pixelate more, because probability spreads thinner across more directions and hits the one-token floor sooner — turning the grain into a measurement of the lattice’s dimension. v0.2.1 is the instrument: run discrete A=1 against the continuum, and at different dimensions, and watch. The new essay Is probability pixelated? develops this.
Why discrete is more than a reformatting
Once probability is integer tokens, a great deal of the lattice becomes exact combinatorics — counting tokens and paths — even after the lattice is calibrated to physical units. Much of what was floating-point becomes integer calculation: reproducible, drift-free, and decidable by counting rather than by tolerance. That is a different epistemic footing for the framework’s claims, and it is what the discrete-vs-continuous comparison is built to exploit.
What’s next
The next release brings GPU and parallel processing to bear: the integer/combinatorial workload parallelises well, and scaling it up is the path to the larger lattices and longer horizons the open audit rows need.
Citation
- Version: v0.2.1
- DOI: 10.5281/zenodo.20615410
- Prior version: 10.5281/zenodo.20350952 (v0.1.0)
- Repository:
dcl-core - Pin:
dcl_core @ git+https://github.com/JackDMenendez/dcl-core@v0.2.1
See the papers page for how the engine underlies the series, and Paper I for the continuous-amplitude implementation it reimplements.