Technical Reference

One Diagram: The Complete QIS Architecture

From raw data sources through edge processing to real-time, scalable, distributed intelligence. Every layer. Every component. One diagram.

By Christopher Thomas Trevethan · January 17, 2026

QIS Protocol: Full System Architecture

DATA SOURCES Raw inputs from anywhere IoT Sensors Temperature, motion, biometric APIs FHIR, REST, GraphQL, HL7 Databases EHR, SQL, NoSQL, Graph Wearables Apple Watch, Fitbit, Oura Voice Input Speech-to-text, commands Manual Entry Forms, surveys, symptoms Medical Devices CGM, BP monitor, ECG Lab Results Blood work, imaging, pathology Genomics 23andMe, WGS, panels Vehicle Telemetry CAN bus, OBD-II, sensors Industrial SCADA PLCs, HMIs, historians Environmental Weather, soil, air quality ANYTHING THAT PRODUCES DATA INGEST EDGE NODES Any compute that can process Smartphone iOS / Android app LOCAL PROCESSING Edge AI Device NVIDIA Jetson, Coral TPU ON-PREM ML Cloud AI Model GPT, Claude, Gemini, Grok API-BASED AGENT Local LLM Llama, Mistral, Phi PRIVATE INFERENCE Smart Sensor ESP32, Raspberry Pi EMBEDDED LOGIC Hospital System Epic, Cerner integration INSTITUTION NODE Vehicle Computer Tesla FSD, AV stack FLEET NODE Industrial PLC Factory automation OPERATIONAL NODE Browser / PWA WebAssembly, WASM ZERO INSTALL AGGREGATE → DISTILL → ROUTE → SYNTHESIZE determine SEMANTIC FINGERPRINT Any method of defining similar Similarity = your address Similarity Definition Expert template AI-determined Doctor-assigned Network-inferred ANALYSIS IS HERE Your Situation = Your Address However determined Routing Key Hash, Vector, ID, any key 0x7f3a9c2b... Experts compete to define templates (Three Elections) Same key = same bucket post DEPOSIT Insight distilled at edge node → Distill into outcome packet → Post to your address in routing Baseline rises for this situation Propagates to all who need it route ROUTING LAYER Any infrastructure that routes by similarity DHT (Kademlia) BitTorrent: 28M concurrent Vector Database Pinecone, FAISS, Weaviate Service Registry Consul, etcd, Zookeeper Gossip Protocol SWIM, Serf, Memberlist IPFS / libp2p Content-addressed P2P Skip Graph O(log N) range queries Pub/Sub (MQTT) 100M+ connections proven Chord / Pastry Ring-based DHTs Central Vector DB Managed option (simpler) Hybrid / Custom Mix methods as needed O(log N) hops Same fingerprint = same bucket Everyone comparable retrieve OUTCOME PACKETS The insight itself, any domain ~Bytes to KB treatment: "FOLFOX" outcome: "responded" duration: "24 months" confidence: 0.94, etc... PRE-DISTILLED INSIGHT WHAT MOVES Distilled outcome Ready to route Privacy-safe NEVER MOVES PII / PHI Medical records Raw sensor data The insight was created when the outcome occurred Human, machine, or system Now it routes to you synth LOCAL SYNTHESIS Any method of combining outcomes Vote / Tally Weighted Median Bayesian Update Ensemble Confidence Filter Outlier Detection Meta-Learning Custom Logic YOUR INSIGHT Personalized O(K) local ~2ms / 1K packets ↩ BACK AT EDGE NODE Insights return to your Edge Node for local synthesis EXTERNAL AUGMENTATION (Optional Network Layer) Supercomputer, cloud AI, or human analyst — any observer of the network Monitors outcomes • Spots correlations • Generates & tests hypotheses in real-time • Refines templates ingest refine query QUADRATIC SCALING: Same situation = access to all who share it 3 nodes 3 connections → 5 nodes 10 connections → 7 nodes 21 connections → 10 nodes 45 connections → SCALE TO ANY SIZE 1,000 → 499K 1M → BILLIONS Scales to billions — any method Within each bucket: • Every node connects to every other • Routing did the filtering — all comparable • N² synthesis fully realized Add as many nodes as you want → Nodes (N) N² INSIGHT O(log N) cost Insight quadratic • Compute flat
Data Sources — Raw inputs from anywhere
Edge Nodes — Any compute that can process
Semantic Fingerprint — Any method of defining similar
Routing Layer — Any infrastructure that routes
Outcome Packets — Bytes to kilobytes of distilled insight
Local Synthesis — Back at your Edge Node
External — Any observer of the network

Example: One Edge Node, Many Data Sources

Apple Watch HealthKit API Hospital FHIR Lab Results Symptom Entry → Your AI Health Agent → Outcome Packet

Multiple data sources feed into ONE edge node. The node aggregates, distills the outcome locally, the fingerprint/address is determined, and routes to the matching neighborhood for the insight needed. Synthesis happens locally. Raw data never leaves.

Any Domain. Any Situation. Same Architecture.

The pattern is always the same: a node with a situation needs real-time insight on what's working from all similar situations.

🏥 ONCOLOGY PATIENT
Stage III colon, KRAS+, 58yo → What treatments working for similar?
🚗 AUTONOMOUS VEHICLE
Rain, highway, sensor degraded → What worked in similar conditions?
🧬 CRISPR LAB
Target gene, off-target risk profile → What guide RNAs worked for similar?
🌾 PRECISION AGRICULTURE
Soil pH 6.2, pest detected, corn V6 → What interventions working nearby?
⚙️ CNC MACHINE
Titanium, 0.02mm tolerance, tool wear → What parameters working for similar?
⚡ POWER GRID
Peak demand, renewable fluctuation → What load balancing working now?
🚑 EMERGENCY RESPONSE
Multi-car pileup, rural, limited resources → What triage protocols working?
🛰️ SATELLITE CONSTELLATION
Orbital debris detected, maneuver needed → What avoidance worked for similar?
♾️ PRECISION EVERYTHING
Real-time optimization • Precision coordination • Continuous adaptation • Any node that needs insight on what's working

Same architecture. Same protocol. Any domain where "what's working for similar" matters.

Layer-by-Layer Breakdown

📡

0. Data Sources — Anything That Produces Data

The raw inputs. IoT sensors streaming temperature. FHIR APIs pulling medical records. Wearables tracking heart rate. Voice commands. Manual symptom entry. Lab results arriving as HL7 messages. Genomic data from 23andMe. Vehicle telemetry from CAN bus. Industrial SCADA systems. Environmental sensors. Multiple sources can feed into a single edge node.

IoT Sensors APIs Databases Wearables Voice Manual Lab Results Genomics Vehicle SCADA Environmental FHIR HL7
🔌

1. Edge Nodes — Any Compute That Can Process

The processing unit — and the synthesis unit. Ingests from multiple data sources. Aggregates locally. The semantic fingerprint is determined — created locally, pre-assigned, or network-inferred. Routes to peers. Then receives outcome packets back and synthesizes locally. Could be your phone, an edge AI device, a cloud LLM acting as an agent, a smart sensor, a hospital system, a vehicle computer, or a browser running WebAssembly. The same node that asks the question synthesizes the answer. Full loop.

Smartphone Edge AI Cloud LLM Local LLM Smart Sensor Hospital System Vehicle Computer Industrial PLC Browser/PWA
🎯

2. Semantic Fingerprint — Any Method of Defining Similar

What makes two cases comparable? That's the fingerprint. Your situation IS your address — however that gets determined. Experts curate a template. An AI learns the mapping. A doctor assigns you. The network infers it. You self-select. A simple hash. A complex embedding. Doesn't matter WHO defines it or HOW. What matters: same situation = same address = same bucket = everyone comparable. The similarity logic IS the analysis. The network just routes.

Expert Template AI-Determined Doctor-Assigned Network-Inferred Self-Selected Learned Embedding Hash Function Domain-Specific
🔀

3. Routing Layer — Any Infrastructure That Routes by Similarity

Decentralized or centralized. P2P or managed. Battle-tested or custom-built. DHT like BitTorrent (28M concurrent nodes). Vector databases like Pinecone (1.5T vectors). Service registries. Gossip protocols. IPFS. Skip graphs. MQTT pub/sub. Chord/Pastry. A simple central vector DB for quick deployments. A hybrid for scale. The protocol doesn't care which infrastructure — just that similar keys land in the same neighborhood.

Kademlia DHT Pinecone FAISS Consul MQTT IPFS Skip Graph Chord Gossip Hybrid
📦

4. Outcome Packets — The Insight Itself

Bytes to kilobytes of distilled insight — the outcome packet IS the insight. "Treatment A worked. 24 months progression-free. Confidence 0.94." Or "Configuration B reduced latency by 40%." Or "Route C avoided the hazard." NOT raw data. NOT full records. The outcome IS the insight, created when it occurred — human, machine, or system. It already exists. It routes to you.

treatment: FOLFOX outcome: responded duration: 24mo confidence: 0.94 Bytes to KB
⚡

5. Local Synthesis — Back at Your Edge Node

The loop completes. Outcome packets return to the SAME edge node that initiated the query. Your device does the final synthesis — any method of combining outcomes: vote, tally, weighted median, Bayesian update, confidence filter, outlier detection, ensemble, meta-learning, custom logic. Trivial compute — ~2ms for 1,000 packets. The edge node that sent the query is the edge node that synthesizes the answer. Privacy preserved. Full circle.

Vote/Tally Weighted Median Bayesian Ensemble Outlier Filter jStat TensorFlow.js
🧠

6. External Augmentation — Any Observer of the Network

Could be a supercomputer processing millions of packets. Could be a cloud AI model spotting correlations. Could be a data analyst with a dashboard and spreadsheet. The mechanism is the same: ingest outcome packets across the network, spot patterns no individual node would see, generate and test hypotheses in real-time, query edge nodes to validate, refine similarity templates based on what actually works. The network learns from itself. Individual nodes serve users. The external layer serves the network.

Supercomputer Cloud AI Model Human Analyst Pattern Discovery Real-Time Hypothesis Testing Template Refinement

The Paradigm Shift in One Sentence

The insight already exists. The outcome already occurred. Route to it.

Data sources feed edge nodes. Your situation becomes your address — however determined. Routes to all who share it. Outcome packets return. Local synthesis produces YOUR insight. No central compute. No data movement. The similarity logic was defined upfront. The insight was pre-deposited by every entity — human, machine, or system — that experienced the outcome. The network is a post office, not a computer.

One architectural decision inverts everything: private, cheap, real-time, transparent, adaptive, universal — not trade-offs, but consequences. See the 11 Flips →

Go Deeper

The Paradigm Shift in Three Words
Subscribe on Substack The Three Elections The 11 Flips The QIS Scaling Law The Inevitability The Cure Already Exists Every Component Exists All Articles