Deep Dive

Teleporting to the Right Neighborhood: Expert-Defined Similarity in QIS

Precision-first routing when it matters, discovery-first routing when you want it — and a protocol that supports both without breaking the scaling law.

By Christopher Thomas Trevethan · January 16, 2026

In the flagship QIS Scaling Law article, I explained the core scaling law: as a network grows to N agents, the opportunity space for pattern synthesis grows as Θ(N²) while per-agent routing stays efficient at O(log N).

This article is the practical follow-up. It answers the question engineers (and skeptical reviewers) immediately ask:

How do agents actually find the right peers? And how do we make that routing precise enough for safety-critical domains like healthcare — without turning the network into an expensive broadcast system? (Every component already exists—this article shows how they fit together.)

Routing Is Where the Magic Happens

A network can have infinite potential and still be useless if discovery is random. QIS solves discovery through semantic fingerprints—representations of the problem itself that route queries directly to where relevant outcomes already exist.

But implementations come in flavors—structured templates, approximate nearest neighbors, vector similarity, and more—and that choice matters. Not because it changes the scaling law, but because it changes the purity of each neighborhood. In other words: how often you land near peers who matter.

Expert-Defined Criteria (Precision-First)

In many domains, you don’t want fuzzy similarity. You want structured identity. For example: a specialist-defined clinical fingerprint for colorectal cancer metastasis, built from exactly the fields that govern treatment selection and outcomes.

The routing key represents the problem itself—defined by the expert best positioned to determine what makes two cases "similar." The key becomes an address. Outcome packets from agents with matching problems collect at that address. When you query, you're not searching—you're going directly to where relevant insights already live.

// Expert-curated template (illustrative) E(patient) = [ age_bucket, stage_code, biomarker_flags, genomic_signature_id, comorbidity_bits, treatment_history_vector, response_outcome_encoding ] // One approach: structured template → deterministic address // Expert defines similarity criteria → matching problems route to same location

When similarity criteria are consistent across a network, you can route with high confidence: matching problems map to the same address, and outcome packets from all those cases collect there. Query that address and you retrieve real insights—not predictions.

The “teleportation” principle

When your fingerprint is deterministic and your routing preserves its structure, an agent does not wander the network. It lands where it should.

Diagram: Teleporting to the Right Neighborhood

Deterministic → Addressable
Expert Template stage_code, biomarker_flags, genomic_signature_id, history... embedding vector E(x) deterministic encoding interpretable dimensions precision-first similarity Locality-Preserving Address H(E(x)) → key / bucket 0xA7F3...9C2B stable mapping DHT Neighborhood route cost: O(log N) relevant peers cluster here “Teleportation” = engineered locality: the representation determines the neighborhood, so routing lands on purpose.

This is not marketing language — it’s a practical engineering target: the problem becomes the address, and outcomes collect there. In safety-critical domains, this mapping can be deterministic (exact addresses), not merely approximate.

DHT is just one option. There are many ways to implement this. The innovation is in the architecture—not any single routing method.

But QIS isn’t locked to curated templates

Precision is not always what you want. If you’re exploring a new domain, or looking for weak signals, or you don’t know which features matter yet, then approximate methods can outperform curated templates. (This is distinct from federated learning—QIS routes to outcomes, not model weights.)

Precision-first (curated): stable, interpretable, consistent neighborhoods

Discovery-first (approximate methods): adaptive, exploratory, fuzzy matching

QIS supports both — and can blend them as the network matures.

Two-lane architecture: run both at once

The most powerful deployment strategy is not choosing one. It’s running a two-lane system:

Lane Representation Routing style When to use
Lane A Expert-defined similarity criteria Deterministic / exact addresses Clinical decisions, compliance, safety-critical outcomes
Lane B Approximate similarity methods Fuzzy / exploratory matching Discovery, drift adaptation, early-stage domains

Diagram: Parallel Tiers (Precision + Discovery)

Hybrid Routing
Agent local state Lane A: Curated deterministic → exact buckets Lane B: Approximate approximate → fuzzy matching Match → Synthesize outcomes propagate through relevant peers Hybrid deployments let you keep safety-critical precision while still discovering new patterns as the world changes.

This is how you avoid false binaries. Curated criteria can route problems to exact addresses where proven outcomes collect, while approximate methods explore adjacent space for emerging patterns — both feeding the same synthesis primitive.

What Can Go Wrong

Teleportation is powerful, but it comes with constraints. If you define similarity too narrowly, you can miss cross-cutting insights. If you define it too loosely, you can drown in irrelevant matches. The right strategy is tiered:

Key insight: QIS guarantees scaling of opportunity. How you define similarity—what makes two problems "the same"—determines the yield of that opportunity in a given domain.

This is why The Three Elections matter: networks compete to define similarity best, and the winners are the ones whose pattern curation produces the highest-quality outcomes.

Why This Matters Beyond Healthcare

Once you can route by engineered locality, you can build addresses for any problem type: machinery failure signatures, crop disease patterns, aviation anomalies, supply chain disruptions, early warning signals. QIS is not the domain. QIS is the protocol that lets domains share real outcomes without centralizing data.

Define the problem. Route to the address. Retrieve the outcomes. That's it. The experts define similarity, the network does the rest.

Subscribe on Substack The Three Elections 11 Flips Every Component Exists All Articles