A person in a psychiatric crisis enters the justice system. A defence needs a forensic psychiatric assessment. There is a waiting list measured in months, a shortage of qualified assessors, and a hearing date that does not move.
That gap is the entire reason this system exists. Not to replace the forensic psychiatrist, which it cannot do and does not attempt. To produce the structured evidentiary groundwork the psychiatrist would otherwise spend weeks assembling, so that scarce expert time is spent on judgement rather than on compilation.
The intended readers were named at design time and the whole architecture follows from them: psychiatric tribunals and courts, magistrates and public prosecutors, emergency and medical personnel who need to know how to help someone quickly, mental health professionals deciding where to focus, and forensic or defence legal teams trying to understand a client.
Every one of those readers is either adversarial or operating under time pressure. Design for that reader and you get a very different machine than the one you get from designing for a curious consumer.
Most systems in this category are built to produce an answer. This one is built to decline to produce one, wherever the data does not support it.
That inversion is the whole contribution, and every constraint below exists because a hostile cross-examination would otherwise find it.
An instrument that cannot say "not enough data" is not an instrument. It is a generator.
Nothing in the output originates with the system. Every marker used is drawn from published, peer-reviewed association work, and the report names the source for each one so that a reviewer can check it independently.
Polygenic risk is reported as a percentile against a reference population and as a relative multiplier against baseline. It is never reported as a diagnosis, because it is not one, and the report says so on its own first page.
It cannot diagnose. Polygenic liability is probabilistic. A high percentile is a statement about a population, not about a person's condition, and certainly not about their conduct.
It cannot establish intent. No sequence of base pairs speaks to what somebody meant to do. A report that is read as though it did would be a misuse of it, and the cover letter exists to prevent exactly that reading.
It cannot separate gene from environment. Expression depends on circumstances the genome has no record of. Trauma, poverty, injury, and the ordinary accidents of a life are not in the file.
It cannot cover what was never measured. Consumer arrays sample the genome, they do not read it. The coverage figure printed beside every domain is there so nobody mistakes a partial panel for a complete one.
It has not been externally validated. This is a method paper describing an instrument in use, not a clinical validation study. That study has not been run, and until it is, the honest status of this work is promising and unproven. Stated plainly rather than buried
DNAGenomicsGPT v7.10 was designed and built from scratch by Rio Ezra Kho, and predates the research programme it now sits inside. It is the origin instrument of the blackpaper line, which concerns suicide reduction, and it is the technical basis of the accelerated postgraduate research proposal that line grew out of.
The full instruction set and the 105 operator prompts are held privately. They are the working apparatus rather than the argument, they carry commercial value, and any real evaluation contains a specific living person's psychiatric and genomic profile. Subject material is never published, in whole or in part, under any circumstances. Access to the method for research or clinical collaboration is available on request under agreement.
Somebody in a cell tonight is being assessed by a system with no waiting list, no notes, and no genome. The instrument does not fix that. It shortens the part a machine can honestly shorten, and refuses the part it cannot.
Complement the psychiatrist. Never replace them. Never pretend to.