Opcode Registry · 31 May 2026
What the model is allowed to invoke
Each opcode is a deterministic function the model calls instead of generating a clinical answer in free text. The function reads a citable canon table and returns a verified value. The model never writes the medical answer. It only decides which opcode to invoke. The opcode produces the answer, and the citation chain travels with it. Below: the registry, organized by tissue and domain. Opcode names and implementations are withheld for IP integrity; the function each performs and the evidence class each invokes is shown.
27Clinical opcodes
6Domains
—Registered opcodes (live)
—Cardiac cells (live)
The discipline
An opcode is bound to one or more citation classes. A clinical opcode invoked at runtime must produce a value whose provenance traces to at least one source in the bound class. The SHA-256 hash of each bound citation will live with the opcode at compile time, so any verifier can confirm the citation chain without trusting Nexus Concordat or any third party. The hash slots are reserved below; population is in progress as the citation registry is finalized.
Opcode names and source code are not published. The function each opcode performs and the evidence it relies on are. That is the contract. Trust the evidence, not the operator.
Registry by tissue and domain
Cardiac System
Opcode #1Drug-drug interaction screen
Given two or more drug agents and patient context, returns the documented interaction risk class, mechanism category, and recommended management posture (avoid, monitor, dose-adjust, no action). Does not generate the interaction; looks it up.
36 training pairscitation class: cardiac pharmacology
Bound citation classes
- FDA approved drug labels (DailyMed)
- RxNorm + DrugBank curated interaction tables
- AHA/ACC/HRS clinical practice guidelines
- PubMed cardiology pharmacology literature
SHA-256 binding: pending citation registry compilation
Opcode #2Contraindication flag
Given a proposed agent and patient profile, returns whether the agent is contraindicated and the specific contraindication category (absolute, relative, dose-modify) plus the underlying physiologic rationale class.
19 training pairscitation class: regulatory + clinical guidelines
Bound citation classes
- FDA drug labels (boxed warnings, contraindication sections)
- NHLBI clinical practice guidelines (JNC8, ATP3, Million Hearts)
- AHA/ACC guidelines for specific drug classes
- PubMed contraindication case literature
SHA-256 binding: pending citation registry compilation
Opcode #3Dose titration
Given a drug, current dose, and clinical response parameters (vitals, symptoms, lab markers), returns the next-step dose decision (hold, decrease, maintain, increase) per guideline-documented titration ladders.
14 training pairscitation class: guideline-directed medical therapy
Bound citation classes
- AHA/ACC heart failure guideline-directed titration ladders
- ESC heart failure dose escalation protocols
- NHLBI ATP3 antihypertensive titration recommendations
- FDA-approved dosing escalation language per agent
SHA-256 binding: pending citation registry compilation
Opcode #4Electrocardiographic interpretation
Given ECG features (rate, axis, intervals, segments, waveforms), returns the documented finding category (rhythm, conduction, ischemic, structural) and the criteria that produced the call. Pattern recognition only; does not invent.
18 training pairscitation class: ECG criteria registry
Bound citation classes
- AHA/ACCF/HRS standardized ECG recommendations (Wagner et al.)
- Sgarbossa criteria for ischemic patterns
- Brugada criteria (wide-complex tachycardia)
- ESC chamber enlargement / hypertrophy criteria
- PubMed cardiac electrophysiology literature
SHA-256 binding: pending citation registry compilation
Opcode #5Hemodynamic estimation
Given measured or invasively-derived hemodynamic parameters, returns the calculated derived value (cardiac output, systemic vascular resistance, pulmonary vascular resistance, cardiac index, etc.) using documented formulas. Closed-form computation only; bounded by physical equations.
16 training pairscitation class: physiologic computation
Bound citation classes
- Fick principle (cardiac output from oxygen consumption)
- Stewart-Hamilton thermodilution equations
- SVR / PVR calculation per AHA/SCAI consensus
- Berne & Levy Physiology, 7th edition
SHA-256 binding: pending citation registry compilation
Opcode #6Differential diagnosis
Given a presenting symptom complex and key features, returns the ordered list of plausible diagnoses with the discriminating evidence class for each. Probabilistic over a closed catalog; does not invent diagnoses.
10 training pairscitation class: diagnostic criteria + epidemiology
Bound citation classes
- DSM-5 / ICD-11 cardiac-relevant codes
- AHA/ACC chest pain / dyspnea / syncope clinical pathways
- NHLBI Framingham Heart Study epidemiologic priors
- UpToDate / Harrison's Principles of Internal Medicine (cardiac sections)
SHA-256 binding: pending citation registry compilation
Opcode #7Referral tier classification
Given a clinical presentation and resource setting, returns the appropriate level-of-care tier (outpatient follow-up, urgent same-day, emergency department, cardiology consult, cath lab activation). Mapped to documented escalation criteria.
16 training pairscitation class: care escalation protocols
Bound citation classes
- AHA/ACC STEMI activation criteria
- HEART / GRACE / TIMI risk score thresholds for escalation
- NHLBI heart failure hospitalization criteria
- ACEP chest pain emergency pathways
SHA-256 binding: pending citation registry compilation
Opcode #8Risk stratification
Given a patient feature set, returns the documented risk score (HEART, TIMI, GRACE, CHA2DS2-VASc, HAS-BLED, etc.) with the component contributions. Deterministic score calculation; numbers come from published rubrics.
10 training pairscitation class: validated risk scores
Bound citation classes
- HEART score (Six et al., Crit Pathways Cardiol)
- TIMI risk score (Antman et al., JAMA)
- GRACE risk model (Eagle et al., JAMA)
- CHA2DS2-VASc (Lip et al., Chest) and HAS-BLED (Pisters et al., Chest)
SHA-256 binding: pending citation registry compilation
Cellular Metabolism
Opcode #9Glycolysis pathway yield
Given substrate quantity, returns the deterministic ATP, NADH, and pyruvate yield through the documented ten-step cytoplasmic pathway. Pure stoichiometry.
citation class: biochemistry canon
Bound citation classes
- Lehninger Principles of Biochemistry (current edition)
- KEGG Pathway Database (map00010)
- Voet & Voet Biochemistry
SHA-256 binding: pending citation registry compilation
Opcode #10Krebs cycle yield
Given pyruvate or acetyl-CoA input, returns the documented yields of NADH, FADH2, GTP, and CO2 through the tricarboxylic acid cycle. Stoichiometric.
citation class: biochemistry canon
Bound citation classes
- Lehninger Principles of Biochemistry
- KEGG Pathway Database (map00020)
- Stryer Biochemistry
SHA-256 binding: pending citation registry compilation
Opcode #11Oxidative phosphorylation yield
Given electron carrier input (NADH, FADH2), returns the documented ATP yield via the electron transport chain coupled to chemiosmotic gradient. P/O ratios from canonical biochemistry.
citation class: biochemistry canon
Bound citation classes
- Lehninger Principles of Biochemistry
- Hinkle P/O ratio measurements (FEBS Letters)
- Mitchell chemiosmotic hypothesis (Nature, Nobel 1978)
SHA-256 binding: pending citation registry compilation
Opcode #12Beta-oxidation yield
Given a fatty acid chain, returns the ATP yield from sequential two-carbon shortening through mitochondrial beta-oxidation. Chain-length dependent stoichiometric formula.
citation class: biochemistry canon
Bound citation classes
- Lehninger Principles of Biochemistry
- KEGG Pathway Database (map00071)
- Voet & Voet Biochemistry
SHA-256 binding: pending citation registry compilation
Opcode #13Gluconeogenesis yield
Given precursor input, returns the deterministic glucose production cost (ATP and GTP expended) through the documented gluconeogenic pathway. Reverse of glycolysis with documented bypasses.
citation class: biochemistry canon
Bound citation classes
- Lehninger Principles of Biochemistry
- KEGG Pathway Database (map00010 reverse)
SHA-256 binding: pending citation registry compilation
Opcode #14Total ATP yield from substrate
Composes the four canonical pathways above to return the total ATP yield from a given starting substrate quantity through complete aerobic metabolism. Pure composition of stoichiometric primitives.
citation class: composed pathway sum
Bound citation classes
- Composes opcodes 9, 10, 11 (glycolysis + Krebs + oxphos)
- Lehninger Principles of Biochemistry, complete oxidation chapter
SHA-256 binding: pending citation registry compilation
Molecular Biology
Opcode #15Codon to amino acid translation
Given a three-base RNA codon, returns the encoded amino acid per the universal genetic code. Pure lookup against the canonical table.
Bound citation classes
- The Universal Genetic Code (NCBI Taxonomy translation table 1)
- Crick et al., Nature (1961) — adapter hypothesis
- Nirenberg / Khorana / Holley Nobel work (1968)
SHA-256 binding: pending citation registry compilation
Opcode #16Start codon flag
Given a codon, returns whether it is the canonical start codon (or one of the alternate initiators by organism class). Boolean against documented translation initiation sites.
Bound citation classes
- NCBI translation tables (per-organism initiation codons)
- Kozak consensus sequence (Kozak, Nucleic Acids Research)
SHA-256 binding: pending citation registry compilation
Opcode #17Stop codon flag
Given a codon, returns whether it is one of the three documented termination codons (or the per-organism class equivalent). Boolean against canonical translation tables.
Bound citation classes
- NCBI translation tables
- Universal Genetic Code stop signals
SHA-256 binding: pending citation registry compilation
Amino Acid Chemistry
Opcode #18Amino acid full name
Given the one-letter symbol, returns the canonical amino acid name per IUPAC nomenclature.
Bound citation classes
- IUPAC-IUB nomenclature (Eur J Biochem, 1984)
- Lehninger Principles of Biochemistry, amino acid appendix
SHA-256 binding: pending citation registry compilation
Opcode #19Amino acid three-letter code
Given the one-letter symbol, returns the corresponding three-letter abbreviation per IUPAC nomenclature.
Bound citation classes
- IUPAC-IUB nomenclature (Eur J Biochem, 1984)
SHA-256 binding: pending citation registry compilation
Opcode #20Amino acid monoisotopic mass
Given an amino acid identifier, returns the documented monoisotopic mass in daltons. Pure physical constant lookup.
Bound citation classes
- ExPASy ProtParam reference masses
- NIST atomic mass data
SHA-256 binding: pending citation registry compilation
Opcode #21Amino acid hydrophobicity index
Given an amino acid, returns the documented hydrophobicity index per the chosen scale (Kyte-Doolittle, Eisenberg, etc.).
Bound citation classes
- Kyte & Doolittle (J Mol Biol, 1982)
- Eisenberg hydrophobicity scale (Proc Natl Acad Sci, 1984)
SHA-256 binding: pending citation registry compilation
Linguistic Structure
Opcode #22Part-of-speech open-class membership
Given a word, returns whether it belongs to the open class (noun, verb, adjective, adverb) or closed class (determiner, preposition, conjunction, pronoun) per documented syntactic taxonomy.
Bound citation classes
- Penn Treebank POS tag set (Marcus et al., Computational Linguistics, 1993)
- Universal Dependencies POS tag set v2
SHA-256 binding: pending citation registry compilation
Opcode #23Part-of-speech slot taxonomy
Given a word, returns the candidate set of POS slots it can fill (e.g., "run" can fill {verb, noun}). Closed enumeration over documented dictionary entries.
Bound citation classes
- WordNet lexical database (Princeton)
- Penn Treebank POS tag set
SHA-256 binding: pending citation registry compilation
Opcode #24Phoneme category
Given a phoneme (in IPA or English mapped notation), returns its documented articulatory category (vowel, stop, fricative, nasal, etc.).
Bound citation classes
- International Phonetic Alphabet (IPA), 2020 revision
- Ladefoged & Maddieson, Sounds of the World's Languages
SHA-256 binding: pending citation registry compilation
Opcode #25Phoneme voicing flag
Given a phoneme, returns whether it is voiced or voiceless per the documented articulatory feature matrix.
Bound citation classes
- International Phonetic Alphabet feature matrix
- Chomsky & Halle, The Sound Pattern of English
SHA-256 binding: pending citation registry compilation
Opcode #26Phonogram-to-sound mapping
Given a written phonogram, returns the documented set of sounds it represents (e.g., "ough" maps to multiple documented pronunciations). Pure lookup against an explicit phonogram dictionary.
1,440 training pairs (largest single opcode)
Bound citation classes
- Spalding, The Writing Road to Reading (70 phonogram set)
- Orton-Gillingham phonogram corpus
- Merriam-Webster pronunciation database
SHA-256 binding: pending citation registry compilation
What the registry guarantees
The model does not generate the answer. The model decides which opcode to invoke. The opcode reads its bound citation class and returns a value whose provenance traces to a documented source.
Names and implementations are withheld; bindings are not. The numbered registry above tells you what each opcode does and what evidence it is constrained to. The opcode names and source code are proprietary. The citation chain is public.
SHA-256 hashes will travel with each opcode. The hash slots above will be populated as the citation registry is compiled. After compilation, an external verifier can confirm the chain by querying the registry endpoint and comparing hashes against the bound citation set. The model itself cannot return a value whose hash is not in the registry.
This is what makes the architecture verifiable. Standard generative medical AI emits free text and then attempts post-hoc citation. We emit citation first, value second. The hash chain is built in.