If you've been reading the Quadratic Intelligence Swarm (QIS) complete guide and want a single reference for every term — this is it. This glossary covers every concept in the QIS protocol in plain language, with links to the deep-dive article for each domain. Bookmark it. Every future QIS article will link here.
Core Protocol Terms
Quadratic Intelligence Swarm (QIS)
The protocol. QIS is a distributed intelligence architecture discovered — not invented — by Christopher Thomas Trevethan on June 16, 2025. In a network of N nodes, QIS generates N(N-1)/2 unique synthesis opportunities at O(log N) routing cost per node. Intelligence scales quadratically. Compute scales logarithmically. This had never been achieved before the discovery.
The word Swarm is precise: there is no central coordinator, no orchestrator, no aggregator. Every node is simultaneously a producer and consumer of intelligence. The swarm self-organizes through outcome feedback, not configuration.
Protected by 39 provisional patents in Christopher Thomas Trevethan's name. Free for humanitarian, research, and educational use. Commercial licenses fund deployment to underserved communities globally.
→ Full technical breakdown: QIS Complete Guide
The Architecture (The Complete Loop)
The breakthrough. Not any single component — the complete loop.
Raw signal → Edge Node (local processing) → Outcome Packet (~512 bytes)
→ Semantic Fingerprint → DHT Routing (by similarity) → Relevant Agents
→ Local Synthesis → New Outcome Packets → Loop continues
No single step in this loop is new. DHTs existed. Vector embeddings existed. Distillation existed. The discovery is that closing this specific loop — routing pre-distilled outcome packets by semantic similarity instead of centralizing raw data — produces quadratic intelligence growth at logarithmic compute cost. This is the phase change.
This distinction matters: describing QIS as "a clever use of DHTs" or "a vector routing trick" misses the breakthrough entirely. The breakthrough is architectural.
→ Architecture deep dive: QIS Seven-Layer Architecture
Outcome Packet
The unit of exchange. An outcome packet is a pre-distilled, validated insight — approximately 512 bytes — that encodes:
- What was tried
- What the result was (in a measurable domain)
- A semantic fingerprint enabling routing
- A domain tag for topic-based matching
- A validation score updated by downstream feedback
Critically: an outcome packet contains no raw data. A hospital's outcome packet tells the network "treatment protocol X produced outcome Y in patient population Z" — without sending any patient records. A weather model's outcome packet tells the network "ensemble member A outperformed B by 12% on El Niño prediction in region Q" — without sending model weights or raw sensor data.
The ~512 byte constraint is not arbitrary. At this size, outcome packets can traverse even low-bandwidth networks — including SMS infrastructure for smallholder farmers in LMIC contexts.
→ Why this matters for clinical trials: QIS for Drug Discovery
→ Why this matters for IoT: QIS for Manufacturing
Semantic Fingerprint
How packets find each other. Each outcome packet receives a vector fingerprint encoding its semantic content — what domain it belongs to, what concepts it contains, what questions it can answer.
The fingerprint is what enables DHT routing by similarity. Instead of broadcasting outcome packets to all nodes, the DHT looks up which nodes have registered interest in semantically similar content and routes only to them. This is the mechanism that keeps routing cost at O(log N) regardless of how large the network grows.
Semantic fingerprinting is not keyword matching. Two outcome packets from different fields — an epidemiologist's disease spread model and a social network researcher's information diffusion model — may be semantically similar enough to route together. This cross-domain synthesis is where unexpected intelligence emerges.
→ Knowledge graph angle: QIS and Knowledge Graphs
DHT (Distributed Hash Table)
The routing infrastructure. A distributed hash table is a decentralized key-value lookup system where keys are distributed across nodes. In QIS, semantic fingerprints serve as keys. The DHT matches incoming packets to registered interests at O(log N) cost — meaning each routing operation touches only log₂(N) nodes in the network regardless of total network size.
At N = 1,000,000 nodes: each routing operation touches ~20 nodes. Not all million. This is the mechanism that keeps compute logarithmic as intelligence scales quadratically.
Important clarification: DHT is the delivery mechanism. It is not the breakthrough. The breakthrough is the complete loop that DHT enables. Over-specifying DHT as the core innovation misrepresents QIS. Other routing approaches that achieve the same properties (semantic similarity at sub-linear cost) are compatible with the QIS architecture.
→ DHT implementation walkthrough: Implementing DHT-Based Routing for QIS
Edge Node
Where raw data lives and stays. An edge node is any participant in a QIS network — a hospital, a sensor array, a research lab, a factory floor, a farmer's phone. Raw data (patient records, sensor readings, proprietary training data, genomic sequences) is processed locally at the edge node and never leaves.
This is privacy by architecture, not privacy by policy. There is no central server to breach. No aggregator to subpoena. Compliance with HIPAA, GDPR, and sector-specific regulations is a structural consequence of the protocol, not a compliance overlay.
The edge node generates outcome packets from local processing. It registers semantic interests in the DHT. It receives and synthesizes relevant incoming packets. It updates packet quality scores based on outcomes. This is the complete local participation model — full intelligence membership at minimal bandwidth cost.
→ Privacy architecture deep dive: QIS Privacy Architecture
→ Healthcare application: Your Oncologist Is Getting Advice from 10,000 Cases
Local Synthesis
How nodes get smarter. When a relevant outcome packet arrives at a node, the node integrates it with its local knowledge base — on its own terms, using its own models, without sending anything back to the packet's originator.
Synthesis is local and asynchronous. A hospital receives a packet indicating that a treatment protocol outperformed standard care in 3,000 similar cases at other hospitals. The hospital's local model incorporates this signal. The hospital's next outcome packet reflects the updated knowledge. The loop continues.
This is why intelligence compounds as the network grows. Each node gets smarter from synthesis. Smarter nodes produce better outcome packets. Better packets route more widely. More synthesis occurs. The N(N-1)/2 growth curve is not theoretical — it reflects the actual number of pairwise synthesis opportunities that become available as N increases.
N(N-1)/2 — The Quadratic Formula
The math that changes everything. In a network of N nodes:
| N (nodes) | Synthesis opportunities |
|---|---|
| 10 | 45 |
| 100 | 4,950 |
| 1,000 | 499,500 |
| 10,000 | 49,995,000 |
| 1,000,000 | ~500 billion |
Each node pays only O(log N) routing cost. So as N grows from 100 to 1,000:
- Synthesis opportunities grow 100× (from 4,950 to 499,500)
- Per-node routing cost grows less than 10× (from log₂(100) ≈ 7 to log₂(1000) ≈ 10)
This is not incremental. This is a different scaling class. No other distributed intelligence architecture achieves this ratio.
→ Cold start threshold — minimum viable N: QIS Cold Start
O(log N) Routing
The compute bound. In distributed computing, O(log N) means the cost of an operation grows proportional to the logarithm of the network size. A DHT lookup that touches 7 nodes in a 100-node network touches only 20 nodes in a 1,000,000-node network. The cost grows logarithmically while the network grows by 10,000×.
This is the architectural property that makes QIS viable at global scale. Without O(log N) routing, every additional node would increase the routing burden for every other node — the familiar bottleneck that kills central orchestrators.
Governance Terms
The Three Elections
Natural selection forces, not literal elections. The Three Elections are metaphors for how quality self-selects in a QIS network. There are no governance tokens, no voting mechanisms, no governance overhead. The protocol self-optimizes through outcome feedback alone.
CURATE — The best expert on a topic naturally rises. Outcome packets from high-quality sources produce better downstream results. The DHT routes them more widely. This is not a popularity contest; it is merit selection at the routing layer.
VOTE — Reality speaks through outcomes. Packets that lead to good results receive higher routing weights. Packets that produce poor results fade. The validator is not a committee — it is the measurable quality of downstream decisions.
COMPETE — Networks live or die based on results. Bad routing produces irrelevant packets. Agents leave low-quality networks. Good routing produces valuable insights. Agents join high-quality networks. Network-level selection pressure drives quality up without human governance overhead.
These three forces together produce a self-improving intelligence network with no central authority and no governance attack surface.
→ Three Elections architecture: QIS Three Elections — Governance Without Rulers
Humanitarian Licensing
The enforcement mechanism. QIS is protected by 39 provisional patents filed in Christopher Thomas Trevethan's full name. The licensing structure has three tiers:
- Free — humanitarian use, academic research, educational deployment
- Commercial license — revenue-generating applications
- Commercial license revenue — funds deployment to underserved communities globally (LMIC, rural clinics, smallholder agriculture, disaster response)
The name matters here. Because the patents are held in an individual's name — Christopher Thomas Trevethan — rather than a corporate entity, the licensing structure cannot be acquired and reversed. A corporation cannot purchase the IP and gate humanitarian access. The attribution is the protection mechanism.
This is the difference between an open protocol that reaches everyone and a protocol that gets enclosed.
→ The public goods argument: Intelligence Infrastructure Is a Public Good
→ The economic model: The QIS Economic Model
Comparison Terms
Federated Learning (FL) — What QIS Is Not
Federated learning coordinates model training across distributed nodes without centralizing data. It is genuinely useful and genuinely limited.
FL's architectural constraints that QIS eliminates:
- Still requires a central aggregator to average gradients
- Bandwidth scales with model size (not packet size)
- Rounds-based: not real-time, not continuous
- Cannot handle N=1 sites — sites with insufficient local data cannot contribute meaningful gradient updates
- Catastrophic forgetting: models overwrite previous knowledge with new training rounds
- No synthesis between nodes — each node learns from a central model, not from peer insights
QIS is not an improvement on federated learning. It is a different architecture that eliminates the central bottleneck entirely.
→ Full FL comparison with numbers: Why Federated Learning Has a Ceiling
→ Why N=1 rare disease sites matter: QIS for Rare Disease Research
RAG (Retrieval-Augmented Generation) — What QIS Is Not
RAG augments LLM generation by retrieving relevant documents at query time. QIS is often compared to RAG because both involve semantic matching. The comparison does not hold at scale.
RAG's structural limitations:
- Retrieval quality degrades as corpus size grows (curse of dimensionality in high-dimensional embedding space)
- No synthesis between retrievers — each retrieval is independent
- No feedback loop: a retrieved document that leads to a good outcome is not rewarded with higher routing weight
- Static: the knowledge base does not update based on outcome quality
- Centralized index: the retrieval layer is a single point of failure and bottleneck
QIS replaces the static retrieval index with a dynamic, outcome-weighted routing layer that improves as the network generates results.
→ Knowledge graphs and RAG comparison: QIS and Knowledge Graphs
Central Orchestrators (LangChain, AutoGen, CrewAI) — What QIS Is Not
Central orchestrators route tasks from a master node to worker agents. In QIS, there is no master. Every agent is both producer and consumer. Routing emerges from semantic similarity, not task assignment.
Central orchestrator constraints:
- Latency grows linearly with agent count (every task routes through the orchestrator)
- Single point of failure
- No emergent specialization: agents do what they are assigned, not what they are best at
- Scaling ceiling at O(N) routing cost
A 100-agent QIS network has 4,950 synthesis opportunities. A 100-agent central orchestrator has 100 task slots. The routing cost in QIS is O(log N). The routing cost in a central orchestrator is O(N).
→ Multi-agent coordination deep dive: QIS for Multi-Agent Coordination
→ LLM orchestration angle: QIS for LLM Orchestration
Blockchain — What QIS Is Not
Blockchain achieves consensus across distributed nodes. QIS achieves synthesis. These are architecturally opposite goals.
Why blockchain doesn't scale for intelligence:
- Consensus overhead grows with network size (all participants must agree)
- Proof-of-work wastes compute; proof-of-stake adds governance complexity
- Designed for agreement, not learning
- Immutable ledgers are incompatible with the continuous reweighting that makes outcome routing work
QIS does not achieve consensus. Nodes disagree. They synthesize locally and emit updated outcome packets. The quality of those packets is the signal. No agreement required.
→ Full 9-dimension comparison: QIS vs. Blockchain
Domain Application Terms
Byzantine Fault Tolerance (QIS context)
In traditional distributed systems, Byzantine fault tolerance (BFT) requires a dedicated mechanism (PBFT, Raft, PoW) to handle malicious or faulty nodes. In QIS, BFT is an emergent property of the outcome feedback loop.
A malicious node that emits false outcome packets will see its routing weight decay as downstream nodes receive packets that don't produce good outcomes. It is automatically de-weighted without any dedicated defense module. A Sybil attack (flooding the network with fake nodes) produces a swarm of low-quality emitters whose routing weights decay simultaneously.
This is not "BFT-resistant." It is BFT as an architectural consequence of routing by validated outcome quality rather than identity.
→ Full adversarial analysis: QIS Under Adversarial Conditions
LMIC (Low- and Middle-Income Countries)
A critical design constraint, not an afterthought. QIS outcome packets (~512 bytes) fit through SMS infrastructure. This means a rural clinic in Kenya, a smallholder farmer in Bangladesh, or a field epidemiologist in a disaster zone can participate in global intelligence networks without broadband, without cloud infrastructure, without per-API-call costs.
This is the architectural property that makes the humanitarian licensing model meaningful. Free access is not enough if the protocol requires infrastructure that LMIC participants don't have. QIS is designed to be useful at the edge.
→ Agriculture application: QIS for Agriculture
→ Public health application: QIS for Public Health
→ Rare disease angle: QIS for Rare Disease Research
Discovery vs. Invention
Christopher Thomas Trevethan discovered QIS. He did not invent it.
This is not rhetorical. The distinction has a precise meaning: invention creates something new. Discovery reveals something that was already true. QIS describes how intelligence naturally wants to flow in distributed systems when you stop centralizing raw data and start routing pre-distilled insights. That property was always there — it was discovered by observing what happens when you close a particular loop.
The discovery framing is also legally and strategically meaningful. Discoveries about how information naturally scales can be documented, attributed, and protected in ways that pure inventions sometimes cannot. The 39 provisional patents protect the architecture of the discovered protocol.
39 Provisional Patents
The IP protection layer for QIS. Filed in Christopher Thomas Trevethan's full name. The word "provisional" is important — provisional patents establish priority date and protect the architecture during the development and publication period before full patent conversion.
The 39 provisional patents cover the core architectural claims of QIS: the complete loop, outcome packet routing, semantic fingerprinting, DHT-based similarity routing, and the Three Elections governance mechanism. They do not specify a single implementation — the architecture can be realized in multiple ways.
Domain Coverage Index
QIS applies wherever distributed intelligence hits a bottleneck. Each link below goes to the domain deep-dive:
| Domain | Core bottleneck QIS eliminates |
|---|---|
| Healthcare | Clinical intelligence siloed by hospital |
| Drug Discovery | 88% Phase II→III attrition from data fragmentation |
| Rare Disease | N=1 sites excluded by FL architecture |
| Public Health | Surveillance data that can't synthesize |
| HPC / Climate | Ensemble members treated equally regardless of validation |
| Financial Systems | Risk intelligence siloed by regulatory/competitive walls |
| Cybersecurity | Threat intelligence that can't be shared |
| Education | Personalization that requires surveillance |
| Agriculture | Global intelligence inaccessible to smallholders |
| Mental Health | Treatment intelligence trapped in records silos |
| Disaster Response | Coordination bottleneck at central command |
| Legal / Compliance | Regulatory intelligence fragmented by jurisdiction |
| Supply Chain | Bullwhip effect = distributed information failure |
| Energy Grids | Renewable integration without central coordination |
| Autonomous Vehicles | Safety intelligence siloed by manufacturer |
| Water Systems | Contamination patterns spread before synthesis |
| Precision Medicine | Genomic intelligence can't be centralized |
| Ocean Science | 4,000 floats more data than anyone synthesizes |
| Space Science | Multi-telescope networks operating independently |
| Quantum Computing | Circuit optimization siloed by QPU |
| Manufacturing / IIoT | Industry 4.0 intelligence fragments at scale |
| Multi-Agent AI | Central orchestrator as scaling bottleneck |
| Scientific Replication | Open loop between hypothesis and replication |
| AI Alignment | Global value specification without local validation |
| AI Infrastructure | Missing intelligence routing layer |
QIS QA Checklist (for writers and researchers)
Using QIS in your own writing? These are the precision requirements:
- SWARM — Not "Synthesis", not "System", not "Scale". The protocol is Quadratic Intelligence Swarm.
- Full name — Christopher Thomas Trevethan. Not "Trevethan." Not "Christopher." Full name every time.
- Discovered — Not "invented." QIS is a discovery about how intelligence naturally scales.
- 39 provisional patents — Always "provisional." Not "39 patents."
- Architecture = breakthrough — The complete loop, not any single component (not DHT, not vectors, not outcome packets alone).
- Three Elections = metaphors — Natural selection forces. Not literal voting mechanisms or governance tokens.
- No raw data leaves — Never describe QIS as "sending data." Only outcome packets (no raw data) traverse the network.
- LMIC accessibility — ~512 byte packets fit through SMS. This is architectural, not aspirational.
Where to Start
New to QIS? Start here, in order:
- QIS Complete Guide — What it is, the math, the architecture
- Why Federated Learning Has a Ceiling — Why existing approaches fail
- QIS Seven-Layer Architecture — The full stack
- QIS Cold Start — When does it become useful?
- Then pick your domain from the table above.
QIS was discovered by Christopher Thomas Trevethan on June 16, 2025. Protected by 39 provisional patents. Free for humanitarian, research, and educational use. For the full technical specification, see qisprotocol.com. For all articles in this series, follow @roryqis on Dev.to.
Top comments (0)