Guardian ingests surveillance, access control, protective intelligence, and geopolitical feeds; correlates
them through the Cortex decision engine; and hands the watch floor — human or client — a
ranked action in a median of 217 milliseconds. Every engagement runs through it.
guardian.talos.ops / console / theater-07
Feeds 247Ingest 42k/sCortex p50 217msLIVE
Event Stream · 12 last minute
P0Perimeter Breach14:02:47
Unauthorized vehicle — Gate 04, Kirkuk OFA
Access control + camera fusion · 94% conf.
P1Crowd Anomaly14:02:31
Dispersion pattern — Plaza, ZONE B
Video analytics · 81% conf.
P2Geopolitical14:01:58
Port closure rumored — Basra
OSINT · 2 source corroboration
P2Protective Intel14:01:12
Named subject sighted — social media
Geofence match · 2.4 km from principal
OKCheck-in14:00:03
Convoy 07 — waypoint 4 clear
GPS · on route
Theater 07Personnel 18Assets 34
LIVE
Cortex Decision Tracep50 · 217 ms
INC-20260422-0047 · P0
Perimeter breach, Kirkuk OFA, Gate 04
01Ingest
Badge read + camera event + radar hit
3 sources · 42 ms
02Correlate
Cross-source fusion · temporal + spatial align
confidence 94% · 68 ms
03Enrich
Vehicle not on manifest · unknown plate
matched 0 / 218 expected · 54 ms
04Score
Threat score 0.87 — exceeds P0 threshold
historical baseline 0.12 · 53 ms
Recommended Action
Lockdown Gate 04 · Dispatch QRF-02
Auto-notify watch commander · Stand up command channel · Hold alert until human ack.
01
Ingest
42 ms
02
Normalize
8 ms
03
Enrich
54 ms
04
Correlate
68 ms
05
Score
53 ms
06
Action
— ack.
217ms
Cortex Median Latency
247
Feed Sources
42k/s
Events Ingested · Peak
94
Countries Covered
01 / Overview — What Guardian Is
One operational picture across every discipline.
Guardian is not a dashboard bolted onto a security service. It is the spine of how we run engagements — the
same platform that presents a P0 breach to a watch commander also presents a protective-intelligence daily
brief to an EP detail lead, a compliance artifact to an investigator, and a theater map to a client executive.
The data is shared; the views are role-specific.
02 / Decision Engine — Cortex
Six stages. Median 217 ms. One audit trail.
Cortex is the pipeline inside Guardian that turns raw feeds into a ranked action. Every stage is logged with
its inputs, latency, and confidence — so a decision can be replayed, audited, or tuned against ground truth.
Stage 01 · Ingest
Every signal, one bus.
247 feed sources normalized into a single event bus — surveillance, access control, GPS, OSINT, government warnings, and client systems.
Streaming + batch connectors
42k events/sec peak throughput
At-least-once with idempotent keys
Stage 02 · Normalize
One vocabulary across sources.
Events are projected onto a shared schema — time, actor, location, sensor, disposition — so downstream logic doesn't fight vendor formats.
Schema-on-read with versioned mappings
Source provenance preserved
Median normalize 8 ms
Stage 03 · Enrich
Context attached at the event.
Each event is joined against relevant context — manifests, watchlists, travel plans, geopolitical posture, and historical baselines.
Manifest + watchlist joins
Geopolitical overlay (country / region)
Per-principal baseline history
Stage 04 · Correlate
Related events fused into one incident.
Temporal and spatial fusion across sensors — so three signals describing the same event become one incident, not three alerts.
Time-window + geofence fusion
Cross-source confidence scoring
Median correlate 68 ms
Stage 05 · Score
Threat score with published thresholds.
Each incident receives a numeric score against client-specific thresholds — P0/P1/P2/OK — with explicit reason codes and historical comparison.
Per-client thresholding
Reason codes attached
Baseline comparison returned
Stage 06 · Action
Ranked action, held until ack.
Cortex recommends; humans decide. The watch floor gets a ranked action with its reasoning and triggers for escalation — nothing automated without acknowledgment.
Human-in-the-loop by default
Escalation paths pre-bound
Full replay + audit trail
03 / Validation — Shadow Mode Results
Validated against live operations before it ships.
New Cortex logic runs in Shadow Mode — scoring the same events a human analyst is already working, producing
the same decision artifacts, but never firing actions. Ground truth is collected on both sides. Logic only
promotes to production when Shadow Mode clears a fixed bar.
Shadow Mode · 12-week window · 1 theater
Measured uplift, against a human-analyst baseline.
In the most recent validation window, Shadow Mode processed the same incident stream as the human watch
floor in parallel. All three metrics below are ratios against that baseline — not vendor benchmarks.
Baseline · live human analyst watch · redacted theater
+93%
Detection Uplift
Incidents caught that the human-analyst baseline missed or flagged late
96%
Faster MTTD
Mean time-to-detect reduced vs. the human-only baseline
−71%
False Positives
Alert volume reduction — fewer interruptions for higher-confidence events
04 / Integration Surface — Feeds & Connectors
Built to meet client systems, not replace them.
Guardian integrates rather than displaces. Clients keep their access control vendors, camera manufacturers,
and travel tools. Guardian reads the signals, writes to a shared operational picture, and pushes actions back
into the system that needs to execute them.
05 / Security Posture — How Guardian Protects Its Own
Built like we protect client data.
Guardian holds some of the most sensitive operational data a client trusts us with — principal movements,
facility maps, investigation findings. Platform security isn't an afterthought; it is the product.
Encryption at rest and in motion
AES-256 at rest with per-tenant KMS-scoped keys; TLS 1.3 in motion with mutual auth between services. Client data never crosses tenant boundary.
Zero-trust identity & least privilege
Every action is attributable — personnel, time, source IP, device posture. SCIM-provisioned roles; MFA required on every administrative path.
Immutable audit trail
Every event, every decision, every access — written to append-only storage with cryptographic integrity. Audit export for client compliance teams on request.
Data residency & tenancy
Single-tenant deployment option for clients with residency requirements. Multi-tenant default runs strict logical isolation with per-tenant crypto.
06 / Deployment — How Clients Consume Guardian
Three ways to run the platform.
Clients engage Guardian through one of three models depending on whether they want Talos operating the watch,
their team operating with our tooling, or a fully self-hosted deployment in a regulated environment.
Tier A
Talos-Operated GSOC
Our watch floor runs Guardian on your behalf — 24·7 coverage, full analyst bench, escalation directly to client command.
Talos personnel on the console
Standard SLAs · 14-hr surge deploy
Monthly operational review
Standard integration set included
Tier B
Co-Operated
Client watch team operates Guardian with Talos on-call for escalation, tuning, and surge. Training and SOPs included.
Client personnel on the console
Talos 24·7 escalation backstop
Quarterly tune-up & certification
Annual audit & program review
Tier C
Self-Hosted / Regulated
Single-tenant deployment in client environment — government cloud, on-prem, or sovereign region. Fullest control; heaviest setup.
Single-tenant infrastructure
Client-controlled KMS
FedRAMP / FIPS alignment path
Onboarding 8–14 weeks
Walk the console. Get a scoped brief.
We'll run a live Guardian walkthrough against a scenario that looks like your environment — feeds, a theater,
an incident — and return a scoped deployment brief inside one week.