This post explains our peer-citable research paper, published on Zenodo: OmniSynx: A Provenance-Gated Deterministic Framework for Stage-Wise Optimisation of Oncology Treatment Decisions (Jain & Rai, 2026, DOI: 10.5281/zenodo.22586517).
What is OmniSynx, in one sentence?
OmniSynx is a computational platform that reads a cancer patient’s complete medical records and answers one question honestly: at each decision point in the treatment journey, what did the evidence support — and what actually happened?
What makes it different from “AI medical advice”?
Two design rules, both unusual, both in the paper’s title:
- Provenance-gated means every single fact the system uses must carry a citation to the exact page of the patient’s own records it came from. If a claim cannot be traced to a source page, the system is not allowed to use it. No guessing, no filling in blanks.
- Deterministic means the reasoning engine is pure mathematics — the same records always produce the same analysis. There is no “creative” AI in the decision layer that could hallucinate a finding. The engine’s correctness is checked by eight machine-verified invariants — automated tests that must pass before any result is reported.
In plain words: the part that reads is careful, and the part that thinks is checkable.
What did the research actually test?
The framework was applied to three real, fully anonymized cancer journeys:
| Cancer analysed | What the engine did |
|---|---|
| Bile-duct cancer of the liver (intrahepatic cholangiocarcinoma) | Stage-wise counterfactuals: what each alternative decision would have changed |
| Aggressive B-cell lymphoma (DLBCL) | Dose-intensity modelling and a search over admissible drug combinations |
| Thyroid carcinoma | Survival simulation under actual versus evidence-supported treatment paths |
The headline finding
Across all three cases, the information gap was zero at every decision point. The knowledge needed for a better decision was already inside the patient’s own records, at the time, every time. Treatment losses came from process and access barriers — delays, unexecuted plans, drugs that were hard to obtain — not from anything unknowable.
The simulated survival difference between what happened and what the same records supported ranged from 4.26 to 16.40 months per patient.
Frequently asked questions
Is OmniSynx a diagnostic tool? No. The paper is explicit: the platform is designed for molecular tumour boards and retrospective analysis — education and decision support, never diagnosis or treatment. Every decision belongs to patients and their doctors.
Can it analyse my records? That is exactly what our helpdesk does: you open a ticket, attach your reports (any format, any language, even handwriting), and receive a plain-language analysis grounded in your own documents.
Where can I read the full research? Freely, on Zenodo: zenodo.org/records/22586517. It is open-access under a Creative Commons licence.