Context Signals & Risk Factors
Formerly The Context Tensor; the catalogue (Context Signals) and the scoring layer (Risk Factors) it conflated are now named separately, and the page moved with the object it describes —
learn/context-tensoris retired, this page islearn/risk-factors. The six weighted inputs are named by their keys; the physics words they replaced are recorded in the specification's Analogy lines.
The Geometry of Trust in a Digital Universe
Trust isn't a yes/no flag. It's a field. Context Signals capture the real-time state of trust for any interaction—its weight, direction, and volatility—so authorization can respond structurally, not by policy. Instead of static rules, KTP evaluates a living vector space that changes as conditions change.
Live Trust Geometry
Ebase → Risk Deflation → Etrust
For the interactive source illustration, follow “View on GitHub” below.
The Seven Named Inputs
The Risk Factors are expressed across seven named inputs. Click any input to explore its mechanics.
For the full signal breakdown (1,644 signals), see KTP-Signals RFC.
Incoming Signals
Context Signals don't operate in a vacuum—they're fed by a continuous stream of live telemetry from across your infrastructure. These signals are normalized, classified through ARQ, and projected into the Risk Factor inputs.
Signal Ingestion Pipeline
From raw telemetry to trust geometry
Signal Categories
Every signal entering the system falls into one of four categories. Each category has distinct sources, update frequencies, and trust implications.
Network and infrastructure telemetry—the pulse of connectivity
Security events and behavioral patterns—the narrative of activity
Performance and saturation indicators—the stress gauge
Cryptographic proofs and human approvals—the trust anchors
Category Deep Dive
Packets
What It Measures
Packet telemetry captures the physical and logical reality of network connectivity. Every TCP handshake, every dropped packet, every millisecond of latency tells a story about reachability, stability, and infrastructure health.
This is the most fundamental signal category—if packets aren't flowing, nothing else matters. Packet signals form the bedrock of Accessibility scoring and provide early warning of infrastructure degradation.
Why It Matters
Network telemetry is impossible to fake at scale. An attacker can forge credentials, but they can't hide the latency of routing through Tor. They can compromise an endpoint, but the C2 beaconing pattern shows up in flow data. Packet signals provide ground truth that higher-layer deception can't mask.
Key Signals
Signal Sources
ARQ Mapping
Risk Factor Projection
Full Signal Catalog (~250 signals)
Layer 4: Transport
Signal Description Derived Metrics src_portClient ephemeral port Port exhaustion, fixed port detection, scan patterns dest_portService port Service distribution, dark port access, scan detection tcp_flagsSYN/ACK/FIN/RST SYN flood, RST rate, null/xmas scans, handshake completion window_sizeTCP receive window Zero window, window scaling, retransmission correlation retransmission_ratePackets resent Spike detection, high-retransmit hosts, global trends Layer 3.5: Flow
Signal Description Derived Metrics flow_bytes_in/outByte counts Volume spikes, exfiltration detection, asymmetric flows flow_packetsPacket count per flow Small packet floods, scan detection, avg packet size flow_durationConnection length Long-lived flows, C2 beaconing, tunnel detection flow_start_timeInitiation timestamp Off-hours activity, burst detection, time correlation application_idDPI-identified app Shadow IT, new application alerts, usage trends Layer 3: Network
Signal Description Derived Metrics src_ip/dest_ipSource/destination Top talkers, reputation, beaconing, new host detection latencyRound-trip time P50/P95/P99, anomaly detection, trend analysis packet_lossDrop indication Loss rate/spikes, path-specific loss, P99 loss hop_countTTL-derived Path length, routing changes, excessive hops bgp_peer_stateRouting health Flap detection, idle alerts, prefix changes Layer 2: Data Link
Signal Description Derived Metrics src_macAsset identifier New/rogue MAC detection, spoofing, OUI distribution vlan_idBroadcast domain VLAN hopping, unused VLAN detection interfacePhysical port Utilization, flapping, broadcast storms, errors link_statusUP/DOWN state Availability, flapping, MTTR Layer 1: Physical
Signal Description Derived Metrics rssi/snrWiFi signal quality Coverage holes, low-signal clients, interference channelWiFi channel Utilization, co-channel interference, DFS events optical_rx_powerFiber light levels Power degradation, link margin, asymmetry transceiver_tempSFP temperature Overheating alerts, thermal trends poe_power_drawPoE consumption Budget usage, power anomalies
Logs & Events
What It Measures
Log telemetry captures the semantic story of what's happening across identities, applications, and endpoints. Authentication attempts, policy violations, configuration changes, and security events all flow through this category.
Unlike packet data (which shows that something happened), logs show what happened and who did it. This category is essential for behavioral analysis and detecting patterns that span multiple sessions.
Why It Matters
Logs are where intent becomes visible. A failed login is just noise—but 50 failed logins from 50 different IPs against the same account is credential stuffing. Logs provide the context to distinguish between accidents and attacks, mistakes and malice.
Security events in this category directly influence adversarial_pressure (anomalies increase friction) and attestation_coverage (audit coverage affects confidence).
Key Signals
Signal Sources
ARQ Mapping
Risk Factor Projection
Full Signal Catalog (~120 signals)
Layer 8: Identity
Signal Description Derived Metrics user_idHuman/machine identity Auth volume, failed login rate, concurrent sessions roleRBAC assignment Privilege escalation, toxic combinations, dormant usage geo_locationAccess location Impossible travel, new country, high-risk country device_idEndpoint identifier Device trust score, new device rate, jailbreak detection Layer 7: Application
Signal Description Derived Metrics http_methodGET/POST/PUT/DELETE Method distribution, unusual usage, high-volume POST http_statusResponse codes 5xx error rate, 403 denials, 404 spikes url_pathResource accessed Path traversal, admin access, sensitive files user_agentClient string Rare agents, bot detection, spoofing Layer 6.5: API
Signal Description Derived Metrics api_endpointAPI route Usage patterns, deprecated endpoints, shadow APIs api_key_idClient identifier Invalid key rate, concurrent usage, key rotation rate_limit_statusThrottling events Quota consumption, abusive clients Layer 6: Presentation
Signal Description Derived Metrics tls_versionProtocol version Legacy protocol usage, downgrade attacks cipher_suiteEncryption algo Weak cipher usage, PFS adoption Layer 5: Session
Signal Description Derived Metrics session_idSession identifier Fixation, churn, concurrent sessions session_durationActive length Short/long sessions, timeout rate login_statusAuth result Brute force, credential stuffing Layer 0: Endpoint/Host
Signal Description Derived Metrics process_nameExecutable Rare process, living-off-the-land techniques process_hashFile hash Malware match, unknown hashes, first-seen process_cmd_lineFull command Encoded commands, suspicious patterns registry_keyWindows registry Run key mods, persistence detection file_operationFile changes evidence_density changes, sensitive file access
Metrics
What It Measures
Metrics telemetry captures the quantitative health of infrastructure. CPU utilization, memory pressure, queue depths, error rates, and saturation levels all reveal whether systems are operating within safe bounds or approaching failure.
These signals are periodic snapshots rather than event-driven—they show the state of the system at regular intervals, enabling trend analysis and capacity planning.
Why It Matters
Infrastructure stress directly correlates with trust risk. A service running at 95% CPU has less capacity to validate requests properly. A queue backing up might indicate an attack or a failing dependency. Metrics provide the early warning system before events occur.
Metrics heavily influence adversarial_pressure (saturation increases friction) and trust_trend (degradation trends signal declining trust trajectory).
Key Signals
Signal Sources
ARQ Mapping
Risk Factor Projection
Full Signal Catalog (~50 signals)
Compute Resources
Signal Description Derived Metrics cpu_utilizationProcessor load Average, peak, per-core distribution memory_usageRAM consumption Heap/stack, swap activity, OOM proximity disk_ioStorage throughput IOPS, latency, queue depth network_ioNIC throughput Bytes in/out, packet rate Application Health
Signal Description Derived Metrics request_rateIncoming traffic Requests/sec, burst detection error_rateFailure ratio 5xx rate, exception frequency response_timeLatency P50/P95/P99, anomaly detection queue_depthBacklog Size, growth rate, drain time Trust Dynamics
Signal Description Derived Metrics throughputSession velocity Sessions/sec processing rate trust_massFriction capacity 100 - Accumulated Risk env_frictionInstantaneous risk Current environmental resistance (0-100) accumulated_riskSession risk Total risk accrued during session Infrastructure
Signal Description Derived Metrics connection_poolPool status Active/idle/waiting counts thread_countConcurrency Active threads, pool exhaustion gc_pressureGarbage collection Pause time, frequency, heap churn
Attestations
What It Measures
Attestation telemetry captures cryptographic proofs and human validations. Digital signatures, audit certifications, provenance trails, and explicit approvals all provide external verification that grounds trust in something more than self-reported behavior.
This is the only signal category that can directly influence the Soul veto—attestations carry the weight of constitutional constraints, jurisdictional requirements, and consent verification.
Why It Matters
Attestations provide trust that cannot be fabricated by the subject being evaluated. An agent can manipulate its own logs, but it can't forge a signature from a trusted third party. Attestations anchor the trust calculation in external reality.
When multiple independent attesters agree, confidence compounds. When attestations are missing or stale, attestation_coverage drops. When constitutional constraints are attested (GDPR consent, TK Labels), Soul gates engage.
Key Signals
Signal Sources
ARQ Mapping
Risk Factor Projection
Full Signal Catalog (~30 signals)
Cryptographic Proofs
Signal Description Derived Metrics signature_validDigital signature verification Signer identity, algorithm strength, timestamp certificate_chainPKI chain validation Chain completeness, revocation status hash_verificationContent integrity Match status, algorithm used Audit & Compliance
Signal Description Derived Metrics audit_certificationThird-party audit status Certification type, expiry, scope compliance_attestationRegulatory compliance Standard (SOC2, ISO, etc.), last audit date policy_attestationInternal policy compliance Policy version, attestation freshness Consent & Sovereignty
Signal Description Derived Metrics consent_stateUser consent record Consent type, timestamp, scope data_sovereigntyJurisdictional constraints TK Labels, OCAP/CARE markers, geofence status opt_out_flagsExplicit denials Scope of denial, timestamp Peer & Human Validation
Signal Description Derived Metrics peer_attestationCross-agent vouching Attester trust level, attestation count human_approvalManual authorization Approver identity, approval scope, expiry break_glass_eventEmergency override Override reason, authorizer, audit trail
The Signal Pipeline: From Telemetry to Trust
Before signals reach the Risk Factor inputs, they pass through a classification layer called ARQ — the foundational grammar of digital experience.
The Trust Pipeline
Raw signals → Classification → Projection → Score
Metrics, Attestations
Retainability, Quality
trust geometry
Action permitted?
The Three Foundational Questions
Every signal entering the system is classified by three fundamental questions about the digital experience it represents:
ARQ-to-Risk-Factor Projection Matrix
Each ARQ classification feeds specific Risk Factor inputs. The mapping is not 1:1 — signals often influence multiple inputs with different weights.
| ARQ Input | Primary Inputs | Secondary | Signal Examples |
|---|---|---|---|
| Availability | evidence_density moment_criticality | trust_trend | Latency, packet loss, reachability, handshake success |
| Retainability | update_resistance trust_trend | evidence_density | Session stability, connection drops, retry rates |
| Quality | adversarial_pressure attestation_coverage | update_resistance | Error rates, throughput, security signals, attestations |
Dimension Lens
Each dimension has a distinct signal signature and behavioral effect. Click any dimension to explore its mechanics.
evidence_density
How It Works
evidence_density measures the density and volume of telemetry flowing through the system. Think of it as gravitational weight—agents with more observed, consistent behavior carry more mass and are harder to perturb.
Unlike a simple counter, evidence_density rewards sustained activity over spikes. An agent processing 1,000 requests over an hour gains more mass than one processing 1,000 requests in a burst. The tensor values consistency and predictability.
| Signal Source | Example Signals | Influence |
|---|---|---|
| Packets | Request volume, throughput, bytes transferred | Primary |
| Logs | Event density, audit trail depth | Secondary |
| Metrics | Concurrent sessions, queue depth, active connections | Secondary |
Behavioral Pattern
Consistency compounds. Steady behavior over time builds mass more effectively than volume spikes. The tensor rewards agents who show up reliably, not those who flood the system episodically.
Failure Mode
Noisy traffic inflation. High-volume garbage requests can artificially inflate evidence_density without genuine trust improvement. Anti-Goodhart measures detect and penalize this pattern.
trust_trend
How It Works
trust_trend captures the rate and direction of change in an agent's trust posture. It's not about where you are—it's about where you're heading and how fast.
Positive momentum (improving trust trajectory) can accelerate permission grants. Negative momentum (declining trust) triggers early intervention before thresholds are crossed. The system watches the derivative, not just the value.
| Signal Source | Example Signals | Influence |
|---|---|---|
| Trust Deltas | Score changes per window, trend direction | Primary |
| Volatility | Swing magnitude, oscillation frequency | Primary |
| Step Changes | Sudden jumps, threshold crossings | Secondary |
Behavioral Pattern
Direction matters more than position. An agent at E=60 with positive momentum may be granted more than one at E=75 with negative momentum. The system anticipates where you'll be, not just where you are.
Failure Mode
Oscillation whiplash. Rapidly swinging between states—even if averaging to a good score—signals instability. The tensor penalizes agents that can't hold a steady course.
update_resistance
How It Works
update_resistance measures an agent's resistance to rapid fluctuation. High inertia means the agent has established stable patterns that dampen noise. Low inertia indicates volatility and fragility.
Think of inertia as behavioral ballast. Agents with deep, consistent histories are harder to knock off course by a single bad event—but also slower to recover from genuine compromise. It's a stabilizing force that cuts both ways.
| Signal Source | Example Signals | Influence |
|---|---|---|
| Configuration | Config stability, drift detection | Primary |
| Dependencies | Dependency consistency, version churn | Primary |
| Behavior | Pattern consistency, routine adherence | Secondary |
Behavioral Pattern
Stability is earned over time. New agents have low inertia by design—they haven't proven themselves yet. update_resistance builds through sustained, predictable operation. It cannot be rushed.
Failure Mode
Rapid drift signals fragility. Frequent configuration changes, dependency updates, or behavioral shifts indicate an agent that hasn't settled. The tensor interprets this as risk.
adversarial_pressure
How It Works
adversarial_pressure measures environmental stress, anomaly load, and attack indicators. It's the friction coefficient of the trust environment—the resistance that actions must overcome.
adversarial_pressure rises fast but cools slowly. A burst of errors, a spike in blocked requests, or detected attack patterns all generate heat. Recovery requires sustained periods of clean operation—there are no shortcuts to cooling down.
| Signal Source | Example Signals | Influence |
|---|---|---|
| Errors | Error rates, exception frequency, 5xx responses | Primary |
| Security | WAF blocks, anomaly detections, threat indicators | Primary |
| Load | CPU pressure, memory saturation, queue depth | Secondary |
Behavioral Pattern
adversarial_pressure is asymmetric. It spikes immediately on stress events but decays on a longer curve. This prevents agents from rapidly alternating between attack and recovery to game the system.
Failure Mode
Sustained heat collapse. Prolonged stress doesn't just reduce trust—it can trigger adaptive dormancy, forcing the agent into Hibernation mode until the environment stabilizes.
moment_criticality
How It Works
moment_criticality captures temporal context, signal freshness, and decay functions. Not all history is equal—recent signals carry more weight than stale ones, and the scoring layer applies exponential decay to age out old data.
moment_criticality also encodes session context. Long-lived sessions shift baselines differently than ephemeral requests. The tensor understands that an agent's behavior at hour 1 of a session may differ from hour 8—and accounts for it.
| Signal Source | Example Signals | Influence |
|---|---|---|
| Freshness | Signal age, last-seen timestamps | Primary |
| Session | Session duration, activity windows | Primary |
| History | Historical baselines, trend windows | Secondary |
Behavioral Pattern
Recency dominates. A clean last hour matters more than a clean last month. The tensor implements exponential decay—old signals fade, giving agents the ability to recover from past mistakes.
Failure Mode
Stale signal poisoning. If the Risk Factors rely on outdated data, it may grant trust based on conditions that no longer exist. moment_criticality decay prevents this, but gaps in telemetry can still mislead.
attestation_coverage
How It Works
attestation_coverage measures attestation coverage, audit visibility, and peer corroboration. Trust confidence depends not just on what happened, but on how well it was witnessed.
Multiple independent observers raise confidence. A single observer—or gaps in observation—lower it. The tensor implements observer-weighted trust: signals from high-trust attesters carry more weight than those from unknown sources.
| Signal Source | Example Signals | Influence |
|---|---|---|
| Attestations | Signed proofs, audit logs, peer validations | Primary |
| Coverage | Observation gaps, blind spot detection | Primary |
| Peer Visibility | Cross-agent corroboration, mesh attestation | Secondary |
Behavioral Pattern
Independent witnesses compound trust. Three independent observers seeing the same behavior creates higher confidence than one observer seeing it three times. The tensor rewards attestation diversity.
Failure Mode
Blind spot exploitation. Attackers may target gaps in observation coverage. The tensor flags low-visibility zones and may require additional attestation before granting high-privilege actions.
soul
How It Works
Soul is the constitutional veto layer. Unlike the six weighted inputs (which contribute weighted values to trust calculations), Soul operates as a binary gate. If a Soul constraint is violated, the action is forbidden—regardless of how high the trust score is.
Soul encodes immutable constraints: jurisdictional requirements, consent state, Indigenous data sovereignty (TK Labels, OCAP/CARE), sacred land protections, and governance vetoes. These are the laws of the digital universe that cannot be negotiated away for operational convenience.
| Signal Source | Example Signals | Influence |
|---|---|---|
| Jurisdiction | Regulatory flags, data residency requirements | Veto |
| Consent | User consent state, opt-out flags | Veto |
| Sovereignty | TK Labels, OCAP/CARE markers, sacred geofences | Veto |
Behavioral Pattern
Soul is not a score—it's a gate. An agent with E=99 is still blocked if Soul=1. This ensures that human rights, cultural sovereignty, and constitutional constraints are never traded away for performance or efficiency.
Failure Mode
Soul violations are absolute. There is no gradual degradation, no warning tier—just immediate denial. Recovery requires explicit governance action to change the underlying constraint, not just improved behavior.
Risk Deflation
Trust isn't just built—it's also deflated by risk. The tensor doesn't ignore threats; it applies them as friction that reduces your effective trust score.
Your base performance is deflated by risk. Higher risk means lower effective trust.
How Risk Deflates Trust
Even a perfect operational score can be crushed by unaddressed risk:
Unpatched vulnerabilities, exposed secrets, insecure configurations, active threat indicators.
40% security risk drops an excellent score to marginal.
Regulatory violations, audit failures, policy breaches, expired certifications.
25% compliance risk turns excellent into merely acceptable.
Anomalous patterns, deviation from baseline, sudden capability changes.
50% behavioral risk triggers adaptive dormancy.
Risk Compounds
Multiple risk factors multiply together. Security risk of 20% plus compliance risk of 15% doesn't equal 35%—it equals
0.8 × 0.85 = 0.68, or 32% total deflation. Risk stacks against you.