Lab
The workhorse behind the research pages. Every sequence ships with its generators, b-file, and run log; every census with the script that produced it. Prose without the scripts is a claim, not a result. Scripts are budgeted to about a minute of wall clock; deep ranges go to fast.c or a stated smaller scale, never to an hour of python.
Sequences
One directory per sequence under sequences/: terms.py (the primary generator), terms2.py (an independent second generator written from the definition), bfile.txt, and RUN.txt with the commands, comparisons, and PASS/FAIL lines. Some entries add a fast.c for deep ranges. The verified summary of every entry is the sequence ledger.
Censuses
census/coprime/ is the coprime census: general.py and its four result files. census/fillclasses/ is the fill-polynomial census: census.py, its csv outputs, and verify_fillclasses.py.
Base 3
base3/ holds the scripts behind the bases page: makefig.py regenerates its figure, precision.py recomputes the constants at high precision, sieves.py runs the structural checks, and claims.md is the per-claim audit.
Visibility
visibility/ covers three arithmetic claims about designs: visible.py for the density of origin-visible lattice points inside a base-q design, gaussian.py for the Gaussian zeta of a 2D design and its 4D quaternion analogue, and pascal.py for the base-2 carpet as the odd entries of Pascal's triangle. A second pass adds connectivity.py, which races self-similar designs against exactly matched random cell sets on components and boundary, baseq.py, which counts base-q designs under the dihedral wreath group and identifies the result in OEIS, and dirichlet.py, which sums one design's fill-ratio fluctuation to pi^4/192. A third pass re-runs those same three claims through independently written generators - matched.py, wreath.py and lambda4.py - which share no code with the second pass and reach larger grids, two more bases and a second pi routine. RUN.txt carries the derivations, the commands, and the PASS/FAIL lines.
Porous
porous/ measures Darcy conductance through a mixed-base carpet sponge: flow.py builds the pore network by Kronecker product and solves it matrix-free, check.py rebuilds it from the digit rule and solves it directly. The finding is that scale ordering changes permeability at fixed porosity.
Millennium
millennium/ recomputes the three measurements a survey of the Clay problems leans on: franel.py for the Franel-Landau discrepancy of the stack's Farey nodes, lattice.py for the log-periodic box counts that place every design in the lattice class, and gap.py with gap2.py for the code-23 Laplacian band gap by two independent routes. A second pass adds stack.py, which re-derives the Farey identity step by step and rebuilds the discrepancy meter from a different generator, and star.py, which pins the upper edge of the band gap by exact integer arithmetic instead of an eigensolver.
Levels
levels/ asks whether mrly fractal Laplacians repel their eigenvalues like a random matrix or cluster them like an integrable system: graphs.py builds the cell graphs, the diagonal slice graph and the two controls, and spacings.py unfolds the spectra two ways and Kolmogorov-Smirnov tests them against both laws.
Lattice
lattice/ works out the complex dimensions of a one-base design three ways: poles.py locates them as roots of the Moran equation, boxcount.py measures the log-periodic oscillation they force in the box count, and tube.py computes the Minkowski content in closed form and shows it never settles. RUN.txt carries the derivation, the commands, the PASS/FAIL lines, and the two places the blanket "no mrly design is Minkowski measurable" statement has to be qualified.
Complex dimensions
complex_dimensions/ is a second, independent pass at the subject of lattice/, reached from a differently worded restatement of the same result. spectrum.py prints the derivation, checks closure under composition cell by cell, computes the tube in exact rationals, proves the box-count identity N(1/(nE)) = k N(1/E) in exact integers, and measures the frequency on two observables over two windows. It agrees with lattice/ on every structural conclusion, disagrees on one measured number, and records both.