2026-01-31-MAMMOCHAT-TO-MAGIC

MammoChat to MAGIC: How One Mammogram Changed Everything

In October 2025, we published a whitepaper about governing mammograms. Four months later, the governance framework it introduced is worth more than the mammograms.

Figures

Context Type Data
post flow-chain nodes: MammoChat → Hadley Lab → MAGIC

The Patient

The patient was 54. Screening mammogram. BI-RADS 4. The AI system said one thing. The clinical evidence said another. And between those two answers was a woman waiting for a phone call that would change her afternoon, or her life.

Nobody could trace why the AI gave the answer it gave.

I’m Dexter Hadley. MD/PhD. 23 years in academia, 37 years writing code. 65 peer-reviewed publications, 6 patent families, 4 clinical trials as PI, $38M+ in funded research across NIH, NCI, and state grants (VITAE). This is the story of how I built a system that proves.

The Paper

On Halloween 2025, we published the MammoChat OPTS–EGO Ledger (PAPERS/OPTS-EGO PDF) — a whitepaper that closed Breast Cancer Awareness Month with open data. MammoChat was funded by a $2M Florida Department of Health grant and enrolled as a clinical trial (NCT06604078), validated through 80+ customer discovery interviews via NSF I-Corps (VITAE → GRANTS). It described a provenance-driven architecture where every mammogram, biopsy, and clinical note becomes a verifiable, patient-consented data token. 
graph LR
    MAMMOGRAM["🔬 Mammogram"] --> HASH["SHA-3 Hash"]
    BIOPSY["🧬 Biopsy"] --> HASH
    NOTE["📋 Clinical Note"] --> HASH

    HASH --> OPTS["OPTS Token<br/>(Dᵢ, Mᵢ, σᵢ, τᵢ)"]
    OPTS --> LEDGER["EGO Ledger<br/>Keccak256 chain"]
    LEDGER --> PROOF["Zero-Knowledge<br/>Proof of Compliance"]

    style OPTS fill:#4ecdc4,color:#fff
    style LEDGER fill:#1a1a2e,color:#fff
    style PROOF fill:#f7931a,color:#fff

The mathematics were elegant. Every data object gets four things: a content hash (Dᵢ), mCODE metadata (Mᵢ), the patient’s digital signature (σᵢ), and a timestamp of consent (τᵢ). The ledger is append-only. Compliance is proven by zero-knowledge proof — verifying policy without revealing personal health information.

We proved three lemmas and one theorem:

Theorem (Constructive Compliance). If Access Minimality, Immutable Auditability, and Confidentiality/Integrity hold, then the OPTS–EGO Ledger enforces HIPAA safeguards by design.

The paper had 128 references. It described Maria, a 47-year-old Colombian schoolteacher with moderate anxiety and a BI-RADS finding lost in translation. It described Zaida, a 52-year-old Pakistani engineer “watched but not understood” by her own monitoring devices. It proposed a $12M Series A to reach 20,000 women across partner health systems in Florida and California.

It was the best paper I’d ever written. And it wasn’t enough.

The Gap

graph TB
    subgraph "What OPTS–EGO Governed"
        G1["✅ Data provenance"]
        G2["✅ Patient consent"]
        G3["✅ Record immutability"]
        G4["✅ HIPAA compliance"]
    end

    subgraph "What OPTS–EGO Couldn't Govern"
        R1["❌ Why the AI recommended what it did"]
        R2["❌ Whether the clinician was credentialed"]
        R3["❌ Whether the system learned from errors"]
        R4["❌ Whether the billing code matched the service"]
        R5["❌ Whether the vocabulary was consistent"]
        R6["❌ Whether the architecture was sound"]
    end

    style G1 fill:#4ecdc4,color:#fff
    style G2 fill:#4ecdc4,color:#fff
    style G3 fill:#4ecdc4,color:#fff
    style G4 fill:#4ecdc4,color:#fff
    style R1 fill:#e94560,color:#fff
    style R2 fill:#e94560,color:#fff
    style R3 fill:#e94560,color:#fff
    style R4 fill:#e94560,color:#fff
    style R5 fill:#e94560,color:#fff
    style R6 fill:#e94560,color:#fff

OPTS–EGO governed the data. It didn’t govern the system.

It could prove a mammogram was hashed and consented. It couldn’t prove the AI recommendation was based on current evidence. It couldn’t prove the radiologist was board-certified. It couldn’t prove the system learned from the last time it was wrong. It couldn’t prove the billing code matched the service rendered.

Four dimensions out of eight. Half the governance. Half the proof.

The realization hit me in December 2025, on a Sunday, the way important things hit — not like lightning, but like gravity. Slowly, then all at once.

The Generalization

OPTS–EGO was a 4-tuple: (Dᵢ, Mᵢ, σᵢ, τᵢ). Four variables. Four dimensions of governance. But healthcare doesn’t fail in four dimensions. It fails in eight.

graph TB
    subgraph "OPTS–EGO (Oct 2025)"
        direction TB
        O_E["Evidence<br/>Dᵢ = SHA-3 hash"]
        O_H["History<br/>τᵢ = timestamp"]
        O_C["Community<br/>σᵢ = signature"]
        O_S["Structure<br/>Mᵢ = mCODE metadata"]
    end

    ARROW["December 2025<br/>The Sunday Insight"]

    subgraph "MAGIC 255 (Feb 2026)"
        direction TB
        M_D["D₀ Declaration"]
        M_E["D₁ Evidence"]
        M_H["D₂ History"]
        M_C["D₃ Community"]
        M_P["D₄ Practice"]
        M_S["D₅ Structure"]
        M_L["D₆ Learning"]
        M_V["D₇ Language"]
    end

    O_E --> ARROW
    O_H --> ARROW
    O_C --> ARROW
    O_S --> ARROW

    ARROW --> M_D
    ARROW --> M_E
    ARROW --> M_H
    ARROW --> M_C
    ARROW --> M_P
    ARROW --> M_S
    ARROW --> M_L
    ARROW --> M_V

    style ARROW fill:#f7931a,color:#fff,font-weight:bold
    style M_D fill:#e94560,color:#fff
    style M_P fill:#e94560,color:#fff
    style M_L fill:#e94560,color:#fff
    style M_V fill:#e94560,color:#fff
    style M_E fill:#4ecdc4,color:#fff
    style M_H fill:#4ecdc4,color:#fff
    style M_C fill:#4ecdc4,color:#fff
    style M_S fill:#4ecdc4,color:#fff

The four new dimensions (red) are exactly the ones missing from healthcare’s worst failures:

The Theory

The generalization didn’t come from engineering. It came from biology.

I spent a decade at Penn — MSE in Systems Engineering (2001-2003), PhD in Genomics & Computational Biology under Junhyong Kim (2003-2007), MD in Precision Medicine (1999-2009) (VITAE → EPOCH 0). Kim taught me that a genome is a program and evolution is its compiler. Warren Ewens — father of mathematical population genetics — sat on my thesis committee. Ewens proved that drift, not selection, governs genetic variation.

In December 2025, those threads converged. I wrote four papers in thirty days — all hosted in the CANONIC gov tree:

timeline
    title The Theoretical Foundation (December 2025 – January 2026)
    December 2025 : Code Evolution Theory
                  : "Drift wins. Code evolves through neutral drift."
                  : Mapped Kimura's neutral theory to software governance
    January 2026 : The Neutral Theory of CANONIC Evolution
                 : Proved that at 255-bit equilibrium, all change is drift
                 : Mathematical proof using Ewens's framework
    January 2026 : Evolutionary Phylogenetics
                 : All domains share common ancestor
                 : 9 runtime clades diverge simultaneously
    February 2026 : The CANONIC CANON
                  : Master specification — 7 parts, 5 stages
                  : "Structure is a side effect of intelligence"

The code evolution theory (PAPERS/THEORY) proved that software governance follows the same mathematics as biological evolution. The neutral theory (PAPERS/NEUTRAL) proved that at fitness equilibrium (255 bits), all change is drift — the system is optimally governed. The phylogenetics (PAPERS/PHYLO) proved that governance frameworks can diversify across languages and platforms while maintaining a common ancestor. The master specification is the CANONIC CANON (PAPERS/CANONIC-CANON).

Twenty years of biological training (VITAE → EPOCHS 0-7), applied to the governance problem that started with one mammogram.

From One Mammogram to Twenty Health Systems

The OPTS–EGO paper governed a mammogram. MAGIC governs an industry.

In February 2026 — four months after the OPTS–EGO publication — we compiled every publicly documented violation against the twenty largest U.S. health systems. The findings are published in our companion paper, The $255 Billion Wound.

graph LR
    subgraph "October 2025"
        MC["MAMMOCHAT<br/>OPTS–EGO<br/>1 application<br/>1 institution<br/>4 dimensions"]
    end

    subgraph "February 2026"
        MAGIC["MAGIC 255<br/>20 health systems<br/>$6.8B in violations<br/>8 dimensions<br/>82% preventable"]
    end

    MC -->|"4 months"| MAGIC

    style MC fill:#4ecdc4,color:#fff
    style MAGIC fill:#1a1a2e,color:#fff,stroke:#f7931a,stroke-width:3px

The math that proved one mammogram is governed now proves that $7.5 billion in healthcare violations were preventable. The framework that started with Maria’s BI-RADS 4 finding now addresses the structural failure of the entire U.S. healthcare compliance apparatus.

The Reference Deployment

MAMMOCHAT is still running. Supported by AdventHealth (DEALS/ADVENTHEALTH) — 550+ facilities across nine states, $14B system. Clinical trial recruiting toward 20,000 patients (NCT06604078). FHIR-native. mCODE-compliant. Evidence traced to NCCN guidelines. Clinician credentials verified. Every encounter on the ledger. Free to the patient.

graph TB
    subgraph "MAMMOCHAT at AdventHealth"
        PATIENT["Patient<br/>Mammography question"] --> CHAT["CHAT Primitive<br/>Empathy-first response"]
        CHAT --> INTEL["INTEL Primitive<br/>NCCN evidence chain"]
        INTEL --> COIN["COIN Primitive<br/>Work receipt minted"]
        COIN --> LEDGER["TRANSCRIPT Ledger<br/>Immutable record"]
    end

    LEDGER --> GOVERNANCE["MAGIC 255<br/>All 8 dimensions validated"]

    subgraph "What This Proves"
        P1["Clinical AI can be governed"]
        P2["Governance doesn't slow workflows"]
        P3["Compliance is continuous, not periodic"]
        P4["The audit trail is complete"]
    end

    GOVERNANCE --> P1
    GOVERNANCE --> P2
    GOVERNANCE --> P3
    GOVERNANCE --> P4

    style GOVERNANCE fill:#f7931a,color:#fff,font-weight:bold
    style PATIENT fill:#4ecdc4,color:#fff

AdventHealth paid $118.7 million in 2015 for governance failure — Stark Law violations and false coding. The system that now governs MAMMOCHAT’s clinical AI recommendations is the same MAGIC framework that would have prevented that settlement. Same institution. Same math. Different outcome.

The Three Primitives

MAMMOCHAT was the first service composed from CANONIC’s three primitives:

graph TB
    INTEL["INTEL<br/>What you KNOW<br/>━━━━━━━━━<br/>Clinical evidence<br/>mCODE data<br/>NCCN guidelines"]

    CHAT["CHAT<br/>What you SAY<br/>━━━━━━━━━<br/>Patient conversation<br/>Empathy-first design<br/>Multilingual"]

    COIN["COIN<br/>What you DO<br/>━━━━━━━━━<br/>Work receipts<br/>Immutable ledger<br/>OPTS tokens"]

    INTEL <--> CHAT
    CHAT <--> COIN
    COIN <--> INTEL

    CENTER["MAMMOCHAT<br/>INTEL + CHAT + COIN<br/>= Governed Clinical AI"]

    INTEL --> CENTER
    CHAT --> CENTER
    COIN --> CENTER

    style INTEL fill:#0f3460,color:#fff
    style CHAT fill:#533483,color:#fff
    style COIN fill:#e94560,color:#fff
    style CENTER fill:#f7931a,color:#fff,font-weight:bold

Remove any primitive and the system fails:

All three together: the stool stands.

What Comes Next

This blog post is the first in a series that launches February 28, 2026.

Post 1 (this post): How MAMMOCHAT led to MAGIC — the origin story.

Post 2: The $255 Billion Wound — the full whitepaper. Every publicly documented violation against the 20 largest U.S. health systems, mapped to missing MAGIC dimensions, with formal proofs that 255-bit governance prevents 82% of documented losses. The business case for CANONIC.

The MAMMOCHAT OPTS–EGO paper asked: Can you govern a mammogram? Yes.

This series asks: Can you govern an industry?

The math says yes. 255 bits of yes.

Internal Sources — CANONIC Gov Tree:

External Sources:

All claims verified against VITAE — canonical CV, source of truth.

Dexter Hadley, MD/PhD — Founder & CEO, CANONIC Origin date: October 31, 2025 — Publication date: February 28, 2026