Research
DiaCroma™ rests on a formal program: a series of papers that prove, gate by gate, why a decision you can trust, question, and check has to be built a particular way. Each is under peer review with university partners and available on request.
The program · six papers, released
A self-contained arc: the foundation, the gates it builds on, three structural results, and the capstone that composes them.
Proves that a hard limit — something a system must never do — can't be enforced by penalizing violations inside a scoring objective, because a large enough reward always overrides any finite penalty. Safety has to come from removing forbidden actions before scoring, not pricing them against it.
Request access →Builds the four gates — feasibility, safety, law, legitimacy — as yes/no tests on the system's beliefs, and shows they aren't one check repeated four times but three distinct kinds. Proves their conjunction is the only filter that can enforce all four at once.
Request access →Models follow-through as capacity × willingness × coordination, so the weakest of the three caps the whole outcome. The rule that falls out: diagnose the single limiting factor and repair it in its own channel, rather than exhorting more effort.
Request access →Corrects two distortions a learning system must get right: feasibility runs out when the first budget — time, money, energy, attention, others' cooperation — is exhausted, and crediting an action with raw improvement rewards luck. Fixes both, with a first-to-deplete runway rule and a counterfactual baseline that subtracts the improvement that would have happened anyway.
Request access →Defines what it means for a system to be honest about a recommendation. When the right trade-off is genuinely unsettled, the honest output is the frontier of options, not one forced "best" — and any stated confidence must match the frequency the event actually occurs.
Request access →Argues that constraints which must hold unconditionally belong in the mathematics itself — a fixed admissibility projection the optimizer and learner can't reach, reprice, or argue with — and reads the whole series through that one idea.
Request access →The program continues · five papers, held
The six above are part of an eleven-paper program. Five further papers extend it and are held under an active patent filing; they open to request once filed.
How a decision system should read uncertain or low-quality evidence honestly, so its confidence never outruns what the data actually support.
What makes a routine durable: why steady repetition, not intensity, is what lets a behavior become one a person can rely on.
When the right move isn't a single recommendation, what kind of response a system should give instead — and how to make every response, including a refusal, structured and reproducible.
Which parts of a system that learns from experience may safely improve, and which must stay fixed so a hard constraint is never crossed while it learns.
Part of the program, deliberately not described — for this one the subject itself is the protected result. Its content stays sealed until the filing is in place.
Contribute
These papers are under peer review with university partners. Request access to a released paper, propose a review, or tell us where you'd take the work next.
Get in touch →