Skip to content

Research

Design decisions in the Naughtian ecosystem are argued in public before they are implemented. This section holds the long-form work: the mathematics, the benchmarks, and the reasoning that the tool documentation only summarises.

These are papers, not product documentation. They are deliberately kept whole rather than split into task-oriented pages — an argument loses its force when you chop it into how-to guides.

Reformulates Kubernetes pod scheduling as a Multi-Dimensional Bin Packing Problem by drawing a structural isomorphism with the Container Stowage Planning Problem used by mega-vessel shipping lines.

Presents a three-phase hybrid pipeline — FFD warm start, genetic algorithm with gang-aware repair, and CSP enforcement with forward checking — and evaluates it on a synthetic benchmark of 186 nodes, 2,714 pods and 5,128 affinity constraints.

Headline results: 100% scheduling success with zero constraint violations, consolidation onto 152 of 186 nodes, 88.7% average CPU utilisation, and an approximation ratio of α = 1.34 against the LP lower bound, in under 44 seconds. Monte Carlo testing puts significance at p < 10⁻⁴.

The tool it describes is Kuberina. For the shorter version of the argument, see the maritime isomorphism.

Argues that Operator Experience — the cost, borne by a human, of understanding and correcting a system’s behaviour under time pressure — is decided by architecture rather than by documentation, and can therefore be treated as a design input.

Applies the constraint in two directions: reflexive, whether a tool is cheap to operate itself, and assistive, whether it lowers the operating cost of the incumbent platform it sits alongside. The second is what distinguishes the position from a critique.

This is the theoretical umbrella over the day-2 problem, which states the same argument in the form the rest of the ecosystem answers.

The framework used to write every paper here. Rather than starting with prose, it treats a paper as a list of interview questions — answer all of them and you have roughly 80% of the content already.

The Kuberina paper is archived on Zenodo with a DOI: 10.5281/zenodo.21582492.