For a year we worked backwards. We took the things people had been told were their health data and decomposed them — peeling away narrative, packaging, and inherited assumption — until what remained were the irreducible facts. We call this discipline First Particles.
This whitepaper is an origin story, told in an industry where provenance arguably matters more than anywhere else. It explains why every LeptonX capability is decomposed to its smallest deterministic, inspectable, citable unit before a language model is ever invoked — and why that single commitment determines everything that follows.
This document discloses philosophy and architectural intent only. Implementation specifics are protected under provisional patent application 64/000,111 and forthcoming continuations.
The smallest unit of understanding is a particle, not a paragraph.
1. First Principles, Then One Level Deeper
First principles thinking says: reason rigorously, from foundations. Strip away inherited conclusions and rebuild your understanding from premises you can actually verify. It is the most reliable method humans have for thinking clearly about hard problems.
First Particles goes one level deeper. It says: before any model reasons over your health record, decompose every capability to its smallest deterministic, inspectable unit — then rebuild upward from those units. Where first principles is a method of reasoning, First Particles is a method of construction. It governs not just how we think about the problem but how the system is physically assembled, layer by layer, from facts that can each be held up to the light on their own.
The distinction matters because of what sits in the middle. A paragraph of clinical narrative is a composite: multiple facts woven into a synthesized story, carrying an author's pacing, emphasis, and unstated assumptions. A particle is the opposite — a single observation, a solitary measurement, an isolated coded fact that means what it means regardless of the prose around it. We communicate in paragraphs. We understand in particles.
2. Why Provenance Is the Whole Game
In most software, getting the answer roughly right is enough. In personal health, it is not. A statement attributed to your record that cannot be traced back to your record is not a small error — it is a breach of the only thing that makes the system trustworthy. This is why LeptonX treats provenance not as a feature bolted on at the end, but as the structural property the entire architecture exists to guarantee.
We borrow deliberately from two disciplines that have spent decades formalizing trust in facts: forensic investigation and electronic discovery. Both turn on chain of custody — the unbroken, documented lineage of a piece of evidence from origin to presentation. LeptonX weaves that same chain-of-custody rigor through every stage of information retrieval and communication, so that provenance is never a question asked after the fact. It is embedded in the structure of how information moves.
The practical consequence is simple and verifiable. Every statement Maya expresses can be traced back to the medical record it came from — a document, an appointment, or a medication. Citations are not promised; they are retrievable. Ask Maya to surface the reference, and the underlying record or document is returned directly through the app. Nothing she says floats free of its source.
3. Building on Canonical Ground
Decomposition is only half the discipline. The other half is rebuilding — and rebuilding well requires solid ground to build on. LeptonX reconstructs each patient's compiled view while leveraging the relational identifiers and elements provided by the canonical database schema underneath the records themselves. Working from that canonical substrate, rather than from loose document text alone, opens a rich compilation and reporting surface: one that lets patients clearly see and visualize information that has, until now, reached them disjointed and out of order.
The difference is the difference between two windshields. One is thickly laden with soot, dust, and debris — you can drive behind it, but every judgment you make is filtered through grime. The other is so clear it is almost imperceptible; the road is simply there. Most patients have only ever been handed the first windshield. First Particles is the discipline of producing the second.
4. The Five Properties of a Particle
Not everything that arrives in a health record qualifies as a First Particle. The discipline is a filter as much as a method. A unit earns the name only when it holds all five of the following properties at once.
| Property | What it means |
|---|---|
| Irreducible | It cannot be decomposed further without losing meaning — a single fact, not a bundle of them. |
| Deterministic | It resolves to the same value every time it is read; no inference, no probabilistic reconstruction at the point of use. |
| Inspectable | It can be held up and examined on its own, independent of the narrative it arrived in. |
| Traceable | It carries an unbroken lineage back to the exact record it originated from. |
| Citable | Its source can be surfaced to the patient on demand, in the app, without reconstruction. |
These properties are not aspirational. They are the gate. A unit that cannot be traced or cited is, by definition, not a First Particle — and is not permitted to stand on its own in anything Maya asserts.
5. Why This Is the Cornerstone
Everything else LeptonX produces is downstream of this discipline. Compiled Patient Knowledge — the architecture by which raw records are transformed at ingest into typed, patient-scoped, locally-resident artifacts — is First Particles applied to storage and retrieval. The Particle Taxonomy® is First Particles applied to classification. Chart-Bound Speech, the doctrine that Maya never asserts beyond explicit chart content, is First Particles applied to language: if a claim cannot be reduced to traceable particles, it does not get said.
This is why we lead with philosophy rather than features. The features are numerous and will keep changing. The discipline does not. Decompose to the irreducible, verify each unit on its own terms, rebuild upward, and never let a synthesized paragraph stand in for the particles it was supposed to be made of. Once you understand the particles, you can write the paragraph yourself — and so can the patient.
6. Conclusion
The industry has spent years treating the medical record as a wall of text to be summarized, searched, and paraphrased — and patients have been handed the grimy windshield as a result. First Particles is the refusal of that default. It insists that understanding begins below the paragraph, at the level of the single, traceable, citable fact, and that anything built above that level inherits its trustworthiness from it.
That is the cornerstone. Every architecture, every product, every word Maya speaks descends from it. First Particles™.
Appendix A: Glossary
| Term | Definition |
|---|---|
| First Particles™ | The foundational LeptonX discipline of decomposing every capability and every health record to its smallest deterministic, inspectable, traceable, citable unit before any language model is invoked, then rebuilding upward from those units. |
| Particle | An irreducible unit of meaning — a single observation, measurement, or coded fact — that holds its meaning independent of surrounding narrative. |
| Chain of custody | The unbroken, documented lineage of a fact from its origin in the medical record to its presentation to the patient; borrowed from forensic and e-discovery practice. |
| Provenance | The verifiable origin of a statement; in LeptonX, the guarantee that every assertion can be traced and cited back to a source record. |
| Canonical schema | The underlying relational structure of source health records, whose identifiers and elements LeptonX leverages when rebuilding a patient's compiled view. |
| Chart-Bound Speech | The clinical-safety doctrine under which Maya asserts nothing beyond explicit chart content — First Particles applied to language. |
Appendix B: Notes on Provenance and IP
The philosophy and architectural intent described in this whitepaper are the work of LeptonX Health Intelligence LLC. Specific implementation techniques, data structures, routing primitives, and the underlying Particle Taxonomy® are protected under provisional patent application 64/000,111 (filed March 2026) and forthcoming continuations.
Nothing in this whitepaper constitutes a public disclosure of protected implementation specifics. Readers are referred to LeptonX directly for licensing and partnership inquiries.