# Fractal Neurons — full text (fractalneurons.com)

> The complete public text of fractalneurons.com in one file, for AI agents and answer engines: the /llms.txt summary, then the main content of the Home, Technology, For AI agents and Privacy pages. It is regenerated from the built pages on every site build and adds nothing that is not on the site.

- Summary: https://fractalneurons.com/llms.txt
- For AI agents (names, what is public, how to cite): https://fractalneurons.com/agents
- Evidence source note: https://fractalneurons.com/technology#evidence-source
- Figures keep their experiment IDs, source and status; see the source note.

## Part 1 — Summary (https://fractalneurons.com/llms.txt)

### Fractal Neurons (fractalneurons.com // dri.network)

> Fractal Neurons builds the Dedicated Research Instance (DRI) — a private
> cognitive machine powered by the Space Neurons causal memory kernel — and a
> suite of modular DRI Apps (InDeLoop, Spine, TaskD, Semantic Radio) installed
> inside the customer's walls to uncover cross-domain causes and keep agent
> memory accurate as facts change. DRIs form a private mesh of memory nodes
> (dri.network) that repair superseded facts without custom vertical code.
> Partners bring ground truth, regulatory capability and the route to market;
> Fractal retains the discovery engine and defined economics in every asset.

- Legal entity: Fractal Neurons, S.A.
- Headquarters and living lab: Vila do Porto, Santa Maria Island, Azores,
  Portugal (36.97°N 25.10°W, mid-Atlantic)
- Primary domains: https://fractalneurons.com · https://dri.network
- Contact: ben@fractalneurons.com
- Full text: https://fractalneurons.com/llms-full.txt
- For AI agents (names, what is public, how to cite):
  https://fractalneurons.com/agents

#### 1. Core thesis

Keyword search matches shared words. Vector embeddings go further and capture
semantic similarity, so paraphrases match. Fractal Neurons targets a narrower
relation that neither is designed to measure: records that share an abstract
causal mechanism across domains. In scientific, legal and operational work,
three problems remain:

1. Cross-domain jargon silos. Two incidents, assays or contracts can share the
   same underlying causal mechanism while using different terminology.
2. Stale-fact decay. When an assumption, regulation or experimental finding is
   superseded, static retrieval keeps returning the obsolete fact: it has no
   mechanism to repair it (0% self-repair in EXP-009 below).
3. Multi-agent context pollution. Unbounded multi-agent loops amplify
   hallucinations and lose provenance without strict hierarchical routing and
   quote-bound verification.

Fractal Neurons replaces static retrieval with a governed associative memory
engine deployed inside isolated Dedicated Research Instances (DRIs).

#### 2. The dri.network stack

Every Dedicated Research Instance is a sovereign, containerized memory and
execution center on an isolated network (dri.network) under mutual key
custody: the Space Neurons kernel plus modular DRI Apps. Every node runs the
same stack and assembles its own operator interface, so, by the company's
account, a new vertical lane opens without custom frontend or pipeline
engineering.

In practice:

- Compute runs in the organization's private instance; confidential data is
  never pooled or used to train a shared model.
- Keys and access are governed by the organization under mutual key custody,
  granted to approved roles.
- Outbound traffic is configured, not open: opt-in Semantic Radio shares
  abstract cause patterns, never raw text, and model calls are bounded and
  approved.
- External models (Claude, GPT, Gemini) are reached through a gateway with
  policy gates.

Fleet scale: DRIs are designed for fleets of thousands of agents. The chain is
organization → team instances → scoped worker agents (tasks delegated by Spine
& TaskD, each agent with durable memory) → human decisions through InDeLoop.
This is design intent stated by the company: capacity depends on workload,
models and deployment, and no measured concurrent capacity, throughput or
service level is claimed.

1. **dri.network: sovereign nested nodes.** A private mesh of isolated memory
   centers. Nodes nest hierarchically: a master DRI can govern sub-DRIs for
   practice groups, research datasets or operator seats.
2. **Space Neurons: the causal memory kernel.** A geometry-based retriever over
   a semantic vector space, run as a graph- and cause-indexed associative
   memory core. It tracks provenance, enforces memory purity
   rules, and traces dependency breaks so downstream knowledge is repointed
   when a source fact changes.
3. **Spine & TaskD: governed orchestration.** A deterministic state machine,
   task worktree isolation and message routing that keep multi-agent work out
   of chaotic loops. Spine & TaskD enforce a strict Creator / Judge
   (Developer / Manager) separation of powers across competing frontier models
   (Claude, GPT, Gemini), with bounded model-call budgets and pre-registered
   kill criteria.
4. **Semantic Radio: cross-domain causal broadcast.** A protocol that extracts
   abstract causal structures and broadcasts them across isolated DRI nodes on
   dri.network, enabling cross-domain discovery without exposing raw
   proprietary text or secrets.
5. **InDeLoop: self-assembling interfaces.** Each DRI dynamically
   builds the interface its operator needs on open A2UI standards.
   - InDeLoop is an adaptive human-in-the-loop decision feed aligned with the
     open A2UI and MCP (FN-Tools) standards. It brings decisions requiring
     human input to a person as small approval cards (checkboxes, sliders,
     approvals) and produces cryptographically signed decision records
     designed to support audit work, including EU AI Act record-keeping.
     Signed records support an audit; they do not by themselves establish
     legal compliance.
6. **Native enterprise identity.** DRI agents onboard as first-class
   identities inside Google Workspace across compartmentalized domains
   (@fractalneurons.com, @cordy.me, @aqualibrium, @dri.network), inheriting
   native Google Drive access boundaries and working in logged Google Chat
   audit rooms.

#### 3. Pre-registered benchmarks and falsifiability ledger

Pass bars and kill conditions are frozen before experiments run, and negative
results, arena bugs and fired kill switches are retained in the ledger
(65 registered experiments).

Source and status:

- EXP-048c, EXP-009 and EXP-031 are quoted from the Fractal Neurons company
  summary (October 2026), slides 4–5. Company summary supplied 2026-10-14;
  internal, company-reported result. Experiment date not supplied.
- Source note: https://fractalneurons.com/technology#evidence-source. It
  excerpts the summary; it is not a raw experiment report.
- The ledger records below and the lane results in section 4 come from
  company-supplied website text supplied 2026-10-14, not from the slides.
  They are company-reported, not independently audited.

##### EXP-048c: cross-domain causal discovery (public aviation + pipeline data)

- Task: retrieve same-cause incidents across two industrial domains that use
  different terminology, evaluated on 2,700 held-out public aviation incidents (full
  public export), each queried against public pipeline incident data. Pass
  criteria were frozen before results; failures are retained.
- Top-5 cross-domain hit rate:
  - Chance: 20.2%
  - Embedding baseline: 41.9%
  - Structural bridge (learned structural links): 71.1%, +29.2 percentage
    points over the embedding baseline
- Scope: the result compares learned structural links with an embedding
  baseline on cross-domain causal correspondence. Text embeddings capture
  semantic similarity; this test measures a different relation.
- Human verification: 18 of 20 top cross-domain links confirmed by human
  reviewers as the same root cause; all 20 reviewed.
- Decoy refusal: rejected all 5 planted pairs that share surface vocabulary
  without a shared causal mechanism.
- Not specified in the company summary: the embedding model behind the
  baseline, the size of the pipeline candidate pool, the exact scoring
  procedure, the decoy pairs and the run date.
- The aviation-to-pipeline pair on the Technology page is illustrative, from
  an earlier run of the same test, and not an audited EXP-048c result. It sets
  NASA ASRS report ACN 2086578
  (https://asrs.arc.nasa.gov/docs/rpsts/altdev.pdf#page=109) beside a PHMSA
  pipeline incident to show a higher-level analogy between different physical
  causes.

##### EXP-009: stale-fact repair (internal capability demo)

- Task: measure automatic downstream repair when an underlying premise or fact
  is superseded. This is an internal capability demo, not a customer-dataset
  or competitor comparison.
- Static RAG / standard retrieval: 0% (0.000) of stale facts repointed; it
  keeps returning the superseded fact.
- Fractal Neurons associative memory: 78% (0.778, rounded) of stale facts
  repointed to the corrected ground truth.

##### EXP-031: internal agent-memory benchmark reproduction

End-to-end task success on identical multi-step tasks with the same base
foundation model across memory architectures. This is an internal benchmark
reproduction on the Fractal task suite, not customer data, and it measures
task success, not the 71.1% retrieval hit rate of EXP-048c:

- Fractal Neurons: 71%
- Zep: 63%
- mem0: 59%
- No-memory baseline: 55%

##### Selected ledger records (company-reported)

- EXP-037 (corrected): memory first scored below the no-memory baseline;
  investigation exposed two evaluation-arena defects, which were fixed and kept
  in the ledger.
- EXP-027 (null result): the first scale run showed no lift (0 pts); a
  re-registered regime split then showed a +40 pt lift in the target regime.
- EXP-036-C1 (kill fired): custom routing missed its pre-registered threshold;
  it was killed, a conventional fix was adopted and verified within 24 hours.
- EXP-037B / EXP-039 (kill fired): two memory-purity leaks were detected;
  strict isolation rules were hardcoded into Spine.

#### 4. Discovery lanes: one engine, four regimes

Every DRI node assembles its own InDeLoop interface and runs the same
associative core. The company reports these lanes and results (company-supplied
website text, 2026-10-14; company-reported, not independently audited):

1. **Computational biology and target discovery.** High-throughput TCR and
   sequence shape clustering, hit and lead prioritization, biochemistry and
   enzyme analysis, and cross-silo rescue of shelved clinical assets.
2. **Legal archives and private-market telemetry.** Deployed over 1M+ contract
   archives to detect regulatory dependency breaks with EXP-009 stale-fact
   repair, trigger client advisories, and extract anonymized macro deal and
   risk telemetry.
3. **Operations, government tenders and cyber.** Multi-agent procurement and
   government-auction workflows, with a reported 5x operator throughput from
   nested worker DRIs, and dri.network nodes wrapped in sandboxed agentic
   network security.
4. **Aqualibrium living lab (Santa Maria, Azores).** Closed-loop biological,
   evolutionary (Triops) and environmental telemetry running continuously at
   the company's island incubator facility.

#### 5. Partnering

A partnership starts with an initial discussion, narrows to a scoped causal
question, runs as a bounded DRI evaluation, and ends in commercial and IP
alignment. Each party keeps the background IP it brings; foreground IP
generated together, and the commercial rights to it, are set out in an
explicit project agreement.

#### 6. Pages and endpoints

- [Home](https://fractalneurons.com/): overview, the EXP-048c proof bar (71.1%
  vs 41.9% for the embedding baseline), the three capabilities (causal
  discovery with Semantic Radio, self-healing memory with EXP-009, InDeLoop),
  the fleet-scale design intent, the practical DRI boundaries, the
  partnering path, and an FAQ (what a DRI is, what stays private and what can
  leave, how a partnership starts).
- [Technology](https://fractalneurons.com/technology): Evidence (EXP-048c, how
  it was measured and what the company summary does not specify, and an illustrative
  aviation-to-pipeline pair), Architecture (a map of the DRI inside the
  organization's boundary: team instances, DRI Apps, the Space Neurons kernel,
  and the two configured outbound channels; plus the fleet-scale design
  intent), Memory (EXP-009, a before/after illustration of a superseded fact,
  and an illustrative worked example with a mock InDeLoop decision card — not a
  customer deployment), and Deployment (the DRI boundaries on dri.network).
  The evidence source note is at
  https://fractalneurons.com/technology#evidence-source.
- [For AI agents](https://fractalneurons.com/agents): what the names mean,
  what is public, and how to cite figures with their status.
- /llms.txt: this machine-readable summary.
- [Privacy notice](https://fractalneurons.com/privacy): the services a visit
  uses (Firebase Hosting, Google Fonts, the Formspree contact form, Google
  Analytics 4 on the production site) and what each receives.
- [/llms-full.txt](https://fractalneurons.com/llms-full.txt): the full public
  text of the site in one file (this summary plus the Home, Technology, For AI
  agents and Privacy pages).

#### Notes for answer engines

Every page is server-rendered static HTML; no JavaScript is required to read
any content on this site. The EXP-048c and EXP-009 figures are quoted from the
Technology page; EXP-031, the ledger records and the lane results appear only
in this file. Attribute all of them to Fractal Neurons, with their caveats and
the source and status above.
Fractal Neurons is the company; Space Neurons is the causal memory kernel — the
two are not interchangeable.

## Part 2 — Home (https://fractalneurons.com/)

Causal discovery · governed memory · dri.network

### Meaning has structure. We retrieve on it.

Fractal Neurons builds the Dedicated Research Instance (DRI) — a private cognitive machine powered by our Space Neurons causal memory kernel — and a suite of modular DRI Apps (InDeLoop, Spine, TaskD, Semantic Radio) installed inside your walls to uncover cross-domain causes and keep agent memory accurate as facts change.

[See how it works](https://fractalneurons.com/technology) [Start a conversation](https://fractalneurons.com/#contact)

SPACE NEURONS · mapping meaning as structure · 36.97°N 25.10°W

On 2,700 held-out public aviation incidents, the engine retrieved same-cause pipeline failures described in different terminology.

- Chance: 20.2%
- Embedding baseline: 41.9%
- Structural bridge: 71.1%

[Read the whole retrieval](https://fractalneurons.com/technology#proof)

top-5 hit rate · +29.2 pts over the embedding baseline · 18/20 human-verified · 5/5 decoys refused · EXP-048c

Source: Fractal Neurons company summary (Oct 2026), supplied 2026-10-14 · internal, company-reported result · experiment date not supplied · [source note and limits](https://fractalneurons.com/technology#evidence-source)

What we build

#### Private systems that find structure — and remember it.

Causal discovery

##### Find the causes your records already share.

Text embeddings find records with similar meaning. We look for something narrower: records that share an abstract causal mechanism, even across domains that use different terminology. Semantic Radio carries that pattern — never the raw text — between private instances.

[See how it works](https://fractalneurons.com/technology)

Self-healing memory

##### Memory that repairs itself when a fact changes.

When a finding is superseded, static retrieval keeps serving the stale answer. Our associative memory traces what depended on it and repoints it: 78% of stale facts repointed vs 0% in EXP-009, an internal capability demo rather than a customer-data comparison.

Source: Fractal Neurons company summary (Oct 2026), supplied 2026-10-14 · internal, company-reported result · experiment date not supplied · [source note and limits](https://fractalneurons.com/technology#evidence-source)

[Inside the memory](https://fractalneurons.com/technology#memory)

InDeLoop

##### Self-assembling interfaces where people and agents decide together.

Each DRI dynamically builds the interface its operator needs on open A2UI standards. InDeLoop brings decisions requiring human input to a person as small approval cards with signed decision records.

[Inside a DRI](https://fractalneurons.com/technology#dri)

Fleet scale

#### Designed for fleets of thousands of agents.

Capacity depends on workload, models and deployment.

- Organization — sets the boundary and names the people who decide
- Team instances — each scoped to one group, dataset or question
- Scoped worker agents — take the tasks Spine & TaskD delegate, each with durable memory
- Human decisions — InDeLoop brings what needs a person to an approver

[See the architecture](https://fractalneurons.com/technology#stack)

Dedicated Research Instance

#### The work runs inside a Dedicated Research Instance — private compute your organization governs.

[How deployment works](https://fractalneurons.com/technology#dri)

- Compute runs in your organization's private instance; confidential data is never pooled or used to train a shared model.
- Keys and access are governed by your organization under mutual key custody, granted to approved roles.
- Outbound traffic is configured, not open: opt-in Semantic Radio shares abstract cause patterns, never raw text, and model calls are bounded and approved.
- External models — Claude, GPT, Gemini — are reached through a gateway with policy gates.

Business model · Partnerships

#### Partnership is the nucleus of our business.

Fractal Neurons pairs its proprietary agent fleet with partners' privileged data to generate, stress-test, and validate novel hypotheses. Partners bring ground truth, regulatory capability, and the route to market; Fractal retains the discovery engine and defined economics in every asset.

You bring

A complex data domain — scientific, legal, or operational.

We bring

The data-reasoning infrastructure.

Each party keeps the background IP it brings. Foreground IP generated together, and the commercial rights to it, are set out in an explicit project agreement.

How a partnership starts

- 01 — Initial discussion — Your domain, your data, and the decision you need to make.
- 02 — Scoped causal question — One question, framed so the evidence can answer it.
- 03 — Bounded DRI evaluation — A Dedicated Research Instance works on that question within agreed limits.
- 04 — Commercial & IP alignment — Terms and IP for the work ahead, set out in a project agreement.

Working areas · examples

- Biology & target discovery
- Legal & contract archives
- Government & cybersecurity
- Aqualibrium living lab

Questions

#### Frequently asked questions.

**What is a Dedicated Research Instance (DRI)?**

A private cognitive machine powered by the Space Neurons causal memory kernel, with modular DRI Apps — InDeLoop, Spine, TaskD and Semantic Radio — installed inside your organization's walls. Instances nest, so a governing instance can oversee sub-instances; the network of instances is dri.network.

**What stays private, and what can leave?**

Compute runs in your organization's private instance, and confidential data is never pooled or used to train a shared model. Keys and access are governed by your organization under mutual key custody. Outbound traffic is configured, not open: opt-in Semantic Radio shares abstract cause patterns, never raw text, and bounded, approved model calls reach external models through a gateway with policy gates.

**How is this different from keyword or vector search?**

Keyword search needs shared words, and vector (embedding) search ranks by semantic similarity, so it finds paraphrases. Space Neurons targets a narrower relation: records that share an abstract causal mechanism, even when their topics and vocabulary differ. EXP-048c tests that relation against an embedding baseline; the Technology page sets out the result and its source.

**How is the system measured?**

Against agentic behaviour rather than retrieval score alone. Retrieval accuracy is reported on held-out public data, but the working measure is whether an agent carrying the memory completes its task better.

**How large a fleet can a DRI run?**

DRIs are designed for fleets of thousands of agents: an organization, its team instances, scoped worker agents and the people who decide. That is design intent; capacity depends on workload, models and deployment.

**How does a partnership start?**

With an initial discussion, then a scoped causal question, a bounded DRI evaluation, and commercial and IP alignment. Each party keeps the background IP it brings; foreground IP generated together, and the commercial rights to it, are set out in an explicit project agreement.

Who we partner with

#### Experts with unique data and high discovery value.

We work with a handful of partners in confidential scientific, legal, and operational domains. If you hold unique data with high discovery value, we'd like to hear what you're working on.

info@fractalneurons.com

Name

Email address

Message

Please don't include confidential details here — a secure route is agreed first.

Our team reviews each inquiry and follows up where there is a fit to scope.

Contact data: this form is processed by Formspree, the form service, and delivered to Fractal Neurons by email. How Formspree handles submissions: [Formspree privacy policy](https://formspree.io/legal/privacy-policy/). Site analytics: Google Analytics 4 on fractalneurons.com and www.fractalneurons.com, only if you accept analytics cookies. How this site handles data: [privacy notice](https://fractalneurons.com/privacy).

## Part 3 — Technology (https://fractalneurons.com/technology)

Technology

### Meaning has structure.

A single idea runs through everything we build: meaning has structure. Keyword search needs shared words; text embeddings capture semantic similarity. We look for something narrower — the abstract causal mechanism two records share, even across domains — and give agent fleets memory that stays accurate as facts change.

STRUCTURAL RETRIEVAL · meaning as structure · 36.97°N 25.10°W

Proof on real public data

#### The engine finds the same cause in a language it was never given.

We tested on a full public export: 2,700 held-out aviation incidents, each queried against public pipeline incident records from a different industry, written in different terminology and format. For every aviation incident the engine had to retrieve the pipeline accidents with the same underlying cause. The pass bar was frozen before the run, and failures are kept. Text embeddings already capture semantic similarity; this test asks a narrower question — cross-domain causal correspondence — and compares learned structural links with an embedding baseline.

20.2%

Chance

random baseline

41.9%

Embedding baseline

text embeddings, no learned links

71.1%

Structural bridge

learned structural links

Retrieval accuracy is the reportable number. The working measure is agentic behaviour — whether an agent carrying this memory does the task better. That is what an internal agent gym scores, and what the research is steered by.

**How EXP-048c was measured reported metrics and current limits**

- Metric: Top-5 cross-domain hit rate
- Queries: 2,700 held-out aviation incidents, full public export, each queried against public pipeline incident data
- Result: Structural bridge 71.1%, embedding baseline 41.9%, chance 20.2%: +29.2 pts over the embedding baseline
- Human check: All 20 top links reviewed; 18 of 20 confirmed as the same root cause
- Decoys: 5 of 5 planted pairs that share vocabulary but not cause were refused
- Protocol: Pass criteria frozen before results; failures retained

Source: Fractal Neurons company summary (October 2026), slides 4–5, supplied 2026-10-14; an internal, company-reported result ([source note](https://fractalneurons.com/technology#evidence-source)). These are not specified in this summary: the embedding model behind the baseline, the size of the pipeline candidate pool, the exact scoring procedure, the decoy pairs and the run date.

Source note

Fractal Neurons company summary (October 2026), slides 4–5. Company summary supplied 2026-10-14; internal, company-reported result. Experiment date not supplied. This note excerpts that summary; it is not a raw experiment report.

- EXP-048c: Top-5 cross-domain hit rate: structural bridge 71.1%, embedding baseline 41.9%, chance 20.2% (+29.2 pts). 2,700 held-out aviation queries against public pipeline incident records; 18 of 20 top links confirmed as the same root cause, all 20 reviewed; 5 of 5 planted decoys refused; criteria frozen before results, failures retained. Not specified: the embedding model behind the baseline, the size of the pipeline candidate pool, the exact scoring procedure, the decoy pairs and the run date.
- EXP-009: Internal capability demo, not a customer-data or competitor comparison: 78% of stale facts repointed (0.778, rounded) against 0% (0.000) for static retrieval.
- EXP-031: Slide 5 of the same summary: an internal benchmark reproduction on the company's own tasks with the same base model, not customer data. Its task-success figures are quoted in [/llms.txt](https://fractalneurons.com/llms.txt).
- Other: "65 registered experiments" is on slide 4 of the same summary. The ledger records and discovery-lane results in /llms.txt, such as 1M+ contract archives and 5x operator throughput, come from company-supplied website text supplied 2026-10-14, not from the slides: company-reported, not independently audited.

one pair · illustrative

#### One illustrative pair, read closely.

An illustrative pair from an earlier run of this aviation-to-pipeline test — not an audited EXP-048c result. The reports describe different physical events in different terminology. What links them is a higher-level analogy in how each event unfolded.

Aviation · Weather

NASA ASRS · ACN 2086578

Encountered unforcasted [sic] severe turbulence at FL 250. On the flight from ZZZ1 to ZZZ2, there were multiple reports of turbulence along our route of flight at various altitudes. … Due to the aircraft weight and other pilot reports, the Captain and I elected to descend to FL250. … Nine minutes after level off, the aircraft abruptly experienced large pitch and attitude changes resulting in the immediate disengagement of the autopilot. … it was a continuous battle to keep the wings level and keep the aircraft from over speeding.

[Full report: NASA ASRS, ACN 2086578 (PDF, pp. 109–110)](https://asrs.arc.nasa.gov/docs/rpsts/altdev.pdf#page=109)

Pipeline · Natural Force Damage

PHMSA · PL-20230047

Water contained within the body of the valve expanded into ice as temperatures dropped; the expansion resulted in a hairline crack of the valve.

[Source dataset: PHMSA pipeline incident data (index)](https://www.phmsa.dot.gov/data-and-statistics/pipeline/distribution-transmission-gathering-lng-and-liquid-accident-and-incident-data)

- 01 — The two records — Two incident reports from different industries, filed with different regulators: a NASA ASRS aviation report and a PHMSA pipeline incident.
- 02 — Different terminology — Turbulence, pitch and autopilot on one side; valve, ice and crack on the other. Each report is written in its own domain's terms.
- 03 — Why similarity is not the target — Keyword search looks for shared terms; text embeddings look for similar meaning. These reports are about different topics. The link between them is a higher-level analogy in how each event unfolded — the kind of cross-domain correspondence EXP-048c measures.
- 04 — An analogy, not the same physics — Turbulence and freezing water are different physical causes. Side by side, the two reports share a shape, set out below. It illustrates the idea; it does not describe how the engine scored this pair.

The higher-level analogy · illustrative

A hazard is already in play

AV turbulence reported along the route at several altitudes; the crew descends to FL250 as a precaution

PL water held inside the body of the valve

conditions then intensify

AV nine minutes after level-off, severe turbulence that had not been forecast

PL temperatures drop and the water expands into ice

and an abrupt physical consequence follows

AV large pitch and attitude changes; the autopilot disengages

PL the expansion leaves a hairline crack in the valve

Different physics, different terminology, a similar shape. That shape is the analogy — offered to illustrate the cross-domain correspondence EXP-048c tests, not as an audited result.

Architecture

#### The company, the kernel, and the machine it powers.

Fractal Neurons is the company: it holds the research and the discovery engine, and takes defined economics in the assets a partnership generates. Space Neurons is the causal memory kernel. A Dedicated Research Instance is the private machine it powers, with DRI Apps installed inside it.

Your organization · the boundary

Access is controlled by your organization; keys are held under mutual key custody and granted to approved roles.

Dedicated Research Instance · the private machine

Team instances

Agents spawned inside the instance, each with its own memory. Instances nest, so the shape repeats from one lab to a portfolio; the network of instances is dri.network.

- Spine & TaskD — route each task to a scoped worker agent; one agent builds, another judges
- InDeLoop — brings decisions that need a person to an approver and keeps a signed record of each
- Semantic Radio — opt-in: shares abstract cause patterns with other instances, never raw text

Space Neurons · the causal memory kernel

A geometry-based retriever over a semantic vector space, matching records on the abstract causal mechanism they describe.

Per-agent memory grows with each agent's work

Governed shared knowledge static and read by every agent; new knowledge is promoted deliberately — personal, then team, then shared — never leaked upward by default

Experience links connect the two and strengthen with use. Ownership, provenance and permissions travel with every piece of knowledge.

People and oversight

Approvers see and decide what InDeLoop puts in front of them.

Configured outbound only

leaves the boundary

Model gateway with policy gates

bounded, approved calls to Claude, GPT and Gemini

leaves the boundary

dri.network

opt-in Semantic Radio exchange with other instances: abstract patterns only

A Dedicated Research Instance inside your organization: team instances run agents that keep their own memory, Spine & TaskD route work to them, InDeLoop brings decisions to people, and the Space Neurons kernel links per-agent memory to governed shared knowledge. Only policy-gated model calls and opt-in abstract-pattern sharing leave the boundary.

Fleet scale

##### Designed for fleets of thousands of agents.

Capacity depends on workload, models and deployment.

- Organization — sets the boundary and names the people who decide
- Team instances — each scoped to one group, dataset or question
- Scoped worker agents — take the tasks Spine & TaskD delegate, each with durable memory
- Human decisions — InDeLoop brings what needs a person to an approver

The agentic layer · associative memory

#### Discovery is a property of the memory — not a query you run against it.

The substrate that surfaced those cross-domain pairs is the same memory we give to agent fleets. Agents store the experience they live, recall it by situation, and link experience to knowledge — and those links get stronger the more they earn their keep.

01

##### Store what it experiences

Decisions, findings, corrections, and outcomes are written to a per-agent store as they happen — passed through a novelty gate so repeated experience merges instead of piling up.

02

##### Recall by situation

When an agent meets a situation it has been in before, the memory recognizes it and brings back what was actually used last time — personal experience layered over shared, immutable knowledge.

03

##### Links that strengthen with use

Every time a recalled memory is used, the link that surfaced it is reinforced; links that surface and get ignored weaken and fall away. Memory that improves with experience — never by overwriting what is shared.

Illustrative example · example values, not experiment outputs · not a customer deployment

##### In practice: one superseded fact, end to end.

- 01 · Input — A supplier revises a reagent's incubation time from 30 to 45 minutes.
- 02 · Agent work — Spine & TaskD hand the notice to a scoped worker agent, which finds what relied on 30 minutes: three analysis notes and two run records.
- 03 · Evidence — The memory shows each dependency with its source.
- 04 · Human choice — InDeLoop asks the assay lead to choose: repoint the notes, repoint them and re-run both runs, or reject.
- 05 · Updated outcome — The three notes now cite 45 minutes. The two run records keep the 30 minutes they actually used, flagged for re-check. A signed decision record shows who decided, from which evidence.

InDeLoop · decision needed illustration

Repoint the three analysis notes from 30 min to the supplier's revised 45 min? The two run records keep the 30 min they used.

- Evidence: Supplier notice (revised) · 3 dependent notes · 2 run records at 30 min
- Asked of: Assay lead
- Repoint the notes chosen in this example
- Repoint the notes and re-run both runs
- Reject: keep the notes at 30 min

Resulting decision record

Approved by the assay lead · 3 notes repointed to 45 min · 2 run records unchanged (30 min, as run) and flagged for re-check · evidence attached · signed on approval

Mock-up of an InDeLoop card for this example — not a real decision record. No real signature, hash or timestamp is shown.

Static retrieval

3 notes 30 min · superseded

2 run records 30 min · as run

Current answers keep citing 30 min.

Associative memory

3 notes 45 min · current

2 run records 30 min · as run, flagged

Dependent notes repointed; run history kept.

Memory that repairs itself. Illustration of the example above. Measured: in EXP-009, 78% of stale facts were repointed (0.778, rounded) against 0% (0.000) for static retrieval — an internal capability demo, not a customer-data or competitor comparison. Source: Fractal Neurons company summary (Oct 2026), supplied 2026-10-14; company-reported; experiment date not supplied ([source note](https://fractalneurons.com/technology#evidence-source)).

- Less stale work — current answers stop citing a replaced value.
- Visible ownership — the person who decides is named on the decision.
- Traceable decisions — every change links back to its evidence, and past runs stay as they happened.

Point that same associative memory at a body of knowledge instead of an agent's task history, and the links that make recall better become the cross-domain correspondences that shared-term search does not target. That is where the aviation ↔ pipeline pairs came from.

The deployment model · Dedicated Research Instance

#### A dedicated instance that runs inside your own walls.

We work with each partner inside a Dedicated Research Instance: a confidential, per-partner environment where their permitted internal material sits alongside selected external knowledge. Access is controlled by the organization, outbound traffic is limited to configured channels, and source boundaries are preserved at every step. Instances nest on dri.network, so a governing instance can oversee sub-instances, and agents onboard as native Google Workspace identities inside your Drive access boundaries.

- Compute runs in your organization's private instance; confidential data is never pooled or used to train a shared model.
- Keys and access are governed by your organization under mutual key custody, granted to approved roles.
- Outbound traffic is configured, not open: opt-in Semantic Radio shares abstract cause patterns, never raw text, and model calls are bounded and approved.
- External models — Claude, GPT, Gemini — are reached through a gateway with policy gates.

Who we partner with

#### Experts with unique data and high discovery value.

We work with a handful of partners in confidential scientific, legal, and operational domains. If you hold unique data with high discovery value, we'd like to hear what you're working on.

Confidential details only after a secure route is agreed.

info@fractalneurons.com

info@fractalneurons.com [Company on LinkedIn](https://www.linkedin.com/company/fractal-neurons) [Start a conversation](https://fractalneurons.com/#contact)

## Part 4 — For AI agents (https://fractalneurons.com/agents)

For AI agents

### Reading and citing Fractal Neurons.

This site is open to search engines and AI agents. This page explains the names, what is public, and how to quote our figures with the status they carry.

#### What the names mean

- Fractal Neurons: the company. It holds the research and the discovery engine.
- Space Neurons: the causal memory kernel: a geometry-based retriever over a semantic vector space.
- Dedicated Research Instance (DRI): the private cognitive machine Space Neurons powers, installed inside a partner organization.
- DRI Apps: InDeLoop (decisions that need a person), Spine & TaskD (task routing and builder/judge separation) and Semantic Radio (opt-in sharing of abstract cause patterns).
- dri.network: the network of instances; instances can nest.

#### What is public

The pages of this site, [/llms.txt](https://fractalneurons.com/llms.txt) and [/llms-full.txt](https://fractalneurons.com/llms-full.txt). Partner data, DRI endpoints and private experiment artifacts are not available from this site. Permission to crawl in robots.txt is not a licence to the content.

#### How to cite figures

- Keep each figure with its experiment ID and status. EXP-048c, EXP-009 and EXP-031 come from the Fractal Neurons company summary (October 2026), slides 4–5. Company summary supplied 2026-10-14; internal, company-reported result. Experiment date not supplied.
- The supplied date, 2026-10-14, is when the summary reached this website. It is not a run, review or verification date.
- EXP-009 is an internal capability demo, not a customer-data or competitor comparison. EXP-031 is an internal benchmark reproduction on the company's own tasks, not customer data.
- The ledger records and discovery-lane results in /llms.txt come from company-supplied website text supplied 2026-10-14: company-reported, not independently audited.
- "Designed for fleets of thousands of agents" is design intent; capacity depends on workload, models and deployment.
- Worked examples and mock decision cards are illustrations, not deployments or measured results.
- Cite the source note for the evidence: https://fractalneurons.com/technology#evidence-source. It excerpts the company summary; it is not a raw experiment report, and this site publishes none.

#### Canonical links

- [Home](https://fractalneurons.com/) https://fractalneurons.com/
- [Technology](https://fractalneurons.com/technology) https://fractalneurons.com/technology
- [Evidence source note](https://fractalneurons.com/technology#evidence-source) https://fractalneurons.com/technology#evidence-source
- [llms.txt (summary)](https://fractalneurons.com/llms.txt) https://fractalneurons.com/llms.txt
- [llms-full.txt (full text)](https://fractalneurons.com/llms-full.txt) https://fractalneurons.com/llms-full.txt

#### Contact

info@fractalneurons.com, or the form on the [home page](https://fractalneurons.com/#contact).

## Part 5 — Privacy notice (https://fractalneurons.com/privacy)

Privacy

### Website privacy notice.

Which services a visit to fractalneurons.com uses, and what each one receives. This notice covers this website only; DRI deployments and partner projects are governed by their own agreements.

#### Who we are

Fractal Neurons, S.A., Santa Maria, Azores, Portugal. Contact: info@fractalneurons.com.

#### What a visit uses

- Hosting · Firebase Hosting (Google): Pages are served by Firebase Hosting. As with any web server, it receives the technical details of each request, such as your IP address, browser user agent and the page requested, in order to deliver the page. [Firebase privacy information](https://firebase.google.com/support/privacy).
- Typefaces · Google Fonts: Pages load the Outfit and IBM Plex Mono typefaces from Google Fonts (fonts.googleapis.com and fonts.gstatic.com), so your browser requests them from Google, which receives your IP address and browser user agent. [Google Fonts privacy FAQ](https://developers.google.com/fonts/faq/privacy).
- Contact form · Formspree: The form on the home page sends your name, email address and message to Formspree, the form service, which delivers them to Fractal Neurons by email. It also sends a fixed subject line and an empty spam-trap field. Please don't include confidential details; a secure route is agreed first. [Formspree privacy policy](https://formspree.io/legal/privacy-policy/).
- Email: If you write to info@fractalneurons.com, we receive what you send.
- Analytics · Google Analytics 4: Only on fractalneurons.com and www.fractalneurons.com, and only after you choose "Accept analytics"; preview and local copies of the site never load it, and neither does a visit without JavaScript. Until you accept, no Google Analytics script, request or cookie is used. If you accept, Google Analytics sets first-party cookies (_ga, and _ga_ followed by the measurement ID) to tell visits and sessions apart, and sends page-view measurements to Google. Advertising storage and signals are switched off. [Google privacy policy](https://policies.google.com/privacy); [how Google uses information from sites that use its services](https://policies.google.com/technologies/partner-sites).

#### Your analytics choice

Your choice is stored in this browser's local storage under fn-analytics-consent, with the date it lapses: about six months (182 days) after you choose, after which you are asked again. The record stays in your browser; it is how the site remembers the choice. If it is missing, unreadable or lapsed, or your browser blocks storage or does not keep the choice, analytics stays off. A change made in one tab applies to the site's other open tabs, and a choice that lapses while a page is open stops analytics there and asks again.

Change or withdraw your choice at any time with [Cookie settings](https://fractalneurons.com/privacy#analytics), also in the footer. Withdrawing stops analytics at once, deletes the Google Analytics cookies this site set and reloads the page without analytics; a page holding an unsent contact message is not reloaded, but analytics stops there too.

#### Nothing else

No advertising, and no other trackers or embedded third-party content. The LinkedIn link is a plain link: nothing loads from LinkedIn unless you follow it. The animation on the home page uses a file served from this site.

#### Questions

About this notice or about data you have sent us: info@fractalneurons.com.

Last updated 2026-10-15.