UNDER THE HOOD

Inside K9 Proxy Guard: the trust architecture.

K9 Proxy Guard is the server-side Behavior Trust Layer: a 42-layer architecture fed by independently deployed, cross-platform K9 Proxy Agents. The Technical Analysis Engine is one layer within that server-side platform. Its research foundation includes 100 TB of combined ground truth (35 TB Android). 19M+ signals contribute to a behavioral signature, distinct from 15,000+ incoming telemetry signals per second. Watch the working product demonstrations.

19M+
Signals defining the behavioral signature
15,000+ signals per second supply ongoing telemetry to the 42-layer trust architecture.
01 · INSIDE THE BEHAVIOR TRUST LAYER

42 layers between raw behavior and a decision.

Evidence flows down from devices to decisions; intelligence flows back up, because every evaluation refines the baselines. The 42 layers are grouped below into seven stages, in the order the data travels.

01ACQUIRE

Signal Collection Unit

Harvests thousands of high-volume signal types at three timescales — keyboard dwell & flight dynamics, file create/delete/rename rates, extension entropy, ransomware-like burst patterns, process creation trees, DNS resolution behavior, domain entropy & novelty, and session-level browsing abstractions — with integrity context (missingness ratios, agent health) attached to every batch.

120s / 900s / 3600s multi-timescale windows
2,100+ signals per agent, six signal families
metadata-only network side — privacy by design
02REPRESENT

Feature Engineering Unit

Converts noisy real-world activity into a mathematical representation of behavior: aggregation, extraction of behavioral, statistical and temporal patterns (cadence, entropy, sequence, distributional moments, trends, bursts, periodicity), then missing-resilient normalization so degraded telemetry never poisons a score.

1,024+ feature vector per entity per window
+100–250 integrity / meta signals
MIDAS-based missing-aware normalization
03BASELINE

Trust Intelligence Server

Learns what normal looks like — per entity, per session, per host, per tenant, per fleet. Exponentially-weighted baselines adapt as the fleet drifts; a neural autoencoder learns each entity's behavioral manifold; embedding space is monitored so concept drift is caught before it becomes blindness.

z₁ = αx₁ + (1−α)z₀ EWMA adaptive baseline
Enc(X)=Z · Dec(Z)=X̂ autoencoder manifold
CSV core registry versioned by registry hash
04MEASURE

Behavior Divergence Detection

Quantifies how far current behavior sits from the learned signature, five ways at once: statistical residual deviation, change-point detection for abrupt shifts and gradual drift, reconstruction error against the learned manifold, KL divergence between embeddings, and distance from the peer/fleet baseline — with EVT/GPD adaptive thresholds for genuinely novel behavior.

Z₁ = (x₁ − μ)/σ residual deviation
Page-Hinkley · CUSUM change-point & drift
E₁ = ‖X − X̂‖² · Dₖₗ(P‖Q) manifold & embedding drift
05FUSE

Fusion Intelligence Layer

Synthesizes divergence signals, graph data (inter-sensor linkages, similarity and causality, peer baselines) and contextual evidence into one coherent risk view — probabilistic fusion weighs each signal by significance and influence, and temporal-graph relevance weighs graph changes by recent drift history.

dynamic Bayes probabilistic signal fusion
outlier rank similarity + meta-signal ensemble
Jaccard · FCD fleet clustering & campaign likelihood
06DECIDE

Continuous Trust Scoring

One score in [0,1], recomputed continuously, with a priority level, a transparent decomposition of exactly which signals moved it, and suppression/policy gates so 150 suppression rules keep noise from becoming alert fatigue. Thresholds trigger alerting, blocking and contextually-aware response — detect, monitor, explain, act.

S₁ ∈ [0,1] descriptive confidence attached
Low / Med / High / Critical priority levels
explainable transparent model decomposition
07ACT

Policy Gates & Response

Turns scores into safe action: suppression rules and policy gates keep noise from becoming alert fatigue, trust-score thresholds trigger alerting, blocking and contextually-aware response — and every decision feeds back to retrain baselines, closing the loop between detection and defense.

150 suppression rules noise kept below alert fatigue
alert · block · respond fusion-based action on thresholds
feedback loop every decision retrains the baselines
02 · CONTINUITY OF TRUST

Trust is a state, with decay and recovery.

Rather than a binary verdict, each entity holds a modelled trust state that decays as anomalies accumulate and recovers as behavior returns to baseline — every transition attributable to understandable contributors.

Stablewithin bounds
Watchmild anomaly
Degradingaccumulating
Challengedstep-up / gate
Disruptedcontinuity broken
Recoveringreturning
// State transitions are asymmetric — recovery demands stronger temporal consistency than decay requires to raise concern.
~1,274
High-dimensional behavioural variables per entity
120·900·3600s
Multi-timescale windows — burst to long-term drift
Hybrid
Statistical + neural + temporal + fleet context
Linux·Win·macOS
Cross-platform agents (macOS on the roadmap)
03 · RESEARCH PRINCIPLES

Research-grade by construction.

K9 Proxy Guard is not just a server dashboard: it is a behavioral trust platform grounded in applied research. These are the principles every layer is built to honor.

R · 01

Evidence over testimony

Verdicts anchor in hardware attestation verified off the device and network metadata observed from outside — signals a resident implant cannot forge, no matter how deep it sits.

R · 02

Corpus-tuned models

Deterministic rules beside deep neural detectors, trained and continuously retrained on 1.5M+ labeled samples with a matched goodware set — driving false positives and false negatives down together, not one at the other’s expense.

R · 03

Fusion, mapped & explained

A 15-dimension fusion engine produces one calibrated score, decomposed into contributors and mapped to MITRE ATT&CK Mobile — so every verdict arrives with its reasoning attached.

R · 04

Honesty as architecture

Confidence stated, limits declared, blind spots documented. Decision support, not oracle theatre — the property that governments, journalists and executive-protection teams actually buy.

04 · XAION CORPUS × DEEP LEARNING

From raw samples to a learning verdict engine.

The pipeline that turns 100 TB of combined ground truth — 35 TB of it Android — into deployed intelligence. Abstracted by design; specifics live under NDA.

L1
AcquireContinuous collection — crawlers, feeds, partner intake, consent-based scan write-back
1.5M+ samples
L2
CurateDeduplication, family labeling, fuzzy-hash indexing, time-stratified ground truth
100 TB curated
L3
RepresentStatic, structural & behavioral feature forging — the mathematical shape of intent
high-dim vectors
L4
LearnDeterministic rules beside deep neural models — drift-aware retraining, analyst label-capture loop
hybrid models
L5
ServeScoring APIs and on-device verdicts — every decision writes back into the corpus
closed loop
05 · THE INTELLIGENCE CORE

Acquisition is commodity. The corpus is the company.

Any competent engineer can rebuild collection. What cannot be rebuilt is a decade of labeled ground truth — and a scoring stack trained against it.

Fewer false positives. A matched known-good corpus proves "this exact binary was checked and found clean" — killing a false positive instantly, offline, with no third-party lookup.
Fewer false negatives. 1.5M known-bad hashes catch unmodified samples; TLSH fuzzy hashing catches repacked variants; deterministic rules and deep-learning risk models retrained on the corpus catch the novel.
A compounding flywheel. Every scan and every score writes back into the shared intelligence database — baselines, labels, fuzzy indices — so the moat deepens with use.
0
Labeled Android malware samples · 10 years
0
Raw APK corpus with detection metadata
0
Behavioral features per entity per window
0
Live rules — critical · suppression · fleet
0
Agent signals across six families
0
Combined ground truth across platforms
06 · PROVENANCE

Where the data comes from.

Intelligence is only as trustworthy as its sources. Every signal in our stack traces to one of six documented origins — collected read-only, consent-based, and metadata-only on the network side.

01 · CORPUS

Decade-scale malware collection

1.5M+ Android malware samples gathered continuously since 2016 — ~35 TB of APKs, each carrying multi-engine detection metadata, family evolution history and signing-certificate lineage.

02 · ALLOWLIST

Verified goodware hashes

A matched known-good corpus built from CIRCL's public hashlookup service — the negative signal that proves a binary clean, offline, and kills false positives instantly.

03 · FEEDS

Live open-source threat intel

Continuously refreshed open-source IOC feeds — hashes, domains, IPs — merged into the intelligence database, so today's campaigns are matched against today's indicators.

04 · ROOTS OF TRUST

Hardware attestation anchors

Google's current attestation roots, tracked through the 2026 root rotation, so both established and newest devices verify. Evidence signed by secure hardware — not vouched for by software.

05 · FIELD TELEMETRY

Consent-based scan write-back

Where consented and enabled, endpoint assessments and K9 Proxy Agent telemetry can inform the shared corpus — baselines, fuzzy indices, verdicts — under explicit consent, read-only on the device.

06 · ANALYST LABELS

Human-verified ground truth

A label-capture loop records analyst verdicts, retroactive corrections and survival labels on every reviewed finding — the training signal that keeps deterministic rules and ML models honest.

SELECTED RESEARCH

Working papers from the lab.

WORKING PAPER

Behavioural Trust Modelling: A High-Dimensional Approach to Security

Introduces continuous behavioural trust modelling — integrating baseline statistics, change-point detection, neural autoencoders and embedding-drift analysis into a single continuous trust score — and shows earlier detection of credential abuse, slow insider threats and AI-assisted attacks than event-based approaches.

STATISTICAL + NEURAL · MULTI-TIMESCALE
SPECIFICATION

Behaviour Trust Layer — Ontology v1

A formal, privacy-aware, cross-platform semantic model for trust-relevant signals: a clean signal universe, entity model, canonical naming, causal relationships and trust-state definitions with decay and recovery contributors — the foundation for explainable trust scoring.

CROSS-PLATFORM · EXPLAINABILITY-FIRST

Research access on request — write to the lab →

WORK WITH US

Bring the trust layer to your fleet.

Compromise assessments, design partnerships for the Behavior Trust Layer, corpus and API licensing, or an investor briefing — tell us which, and we'll come prepared.

◈ PICK A THEME