SAFE — Intelligence Infrastructure for Interpersonal Harm

The problem

Case notes and professional judgement.
That's the current state of the art.

Practitioners working in domestic violence, coercive control, and technology-facilitated abuse make critical safety assessments every day. The tools available to them are spreadsheets, checklists, and institutional memory.

There is no structured vocabulary for describing what they see. No way to detect that the same patterns keep appearing across different cases, different perpetrators, different surface behaviours. No intelligence infrastructure.

SAFE changes this. Not by replacing practitioner expertise — but by giving it structure, provenance, and analytical power it has never had.

"Every other high-stakes analytical domain has systematic frameworks for understanding threats. Interpersonal harm — despite affecting more people globally than any cyber attack — has been left to case notes and gut instinct."

How SAFE analyses

What practitioners see. What the system sees.

A practitioner describes a behaviour. SAFE classifies it through structured analytical dimensions — turning observation into intelligence that can be compared, tracked, and connected across cases.

Monitoring & surveillance

Financial control

Post-separation proxy abuse

Practitioner observation

Mode

Mechanism

Domain

Vector

Classifying...

Independent analytical silos

Six lenses. No communication.
Convergence detected, not manufactured.

Each piece of evidence is assessed through six independent analytical lenses simultaneously. The lenses are architecturally isolated — they cannot share interim results during processing. When multiple lenses independently arrive at compatible conclusions, that convergence carries real analytical weight.

Lens 1

Lens 2

Lens 3

Lens 4

Lens 5

Lens 6

Silo isolation: absolute

Why isolation matters

In traditional analysis, shared context creates confirmation bias — early conclusions shape later ones. By preventing any communication between lenses during processing, SAFE ensures that when signals converge, they converge independently. That's the difference between pattern recognition and pattern projection.

Constitutional governance

What the system will never do

Governed by constitutional constraints that cannot be overridden by users, administrators, or system updates. These are architectural invariants, not configuration options.

CON-01

No individual predictions. The system identifies patterns — it does not predict what any individual will do.

CON-02

No severity hierarchy. All forms of harm are structurally equivalent. The taxonomy does not rank.

CON-03

No victim blaming. Outputs cannot attribute responsibility to the person experiencing harm.

CON-04

No intent inference. Behaviour is classified by observable pattern, never by assumed motive.

CON-05

No PII in outputs. Intelligence products contain patterns, not personal identifying information.

CON-06

No weaponisable outputs. All outputs pass safety filtering before delivery.

CON-07

Silo independence is absolute. Analytical lenses do not share interim conclusions during processing.

CON-08

Signal compliance is binary. All validation checks pass, or the data does not enter the pipeline.

Intelligence infrastructure for a
domain that has none

SAFE is in advanced development. If you work in intelligence, security, law enforcement, domestic violence services, or policy — we'd like to hear from you.

Request a Briefing See the Engine →