Safeguarding & casework

The operational job: triage incoming concerns - across forms, emails, transcribed phone calls - into a structured incident record with a severity classification and a mandatory routing recommendation. Never replace the human decision; just ensure that the human sees what they need to see, in the order they need to see it, faster than the volume otherwise allows.

The bar here is different from the other sector offerings. A false negative in safeguarding can be catastrophic. The deployment posture is conservative by design.

The shape Loop takes here:

• High-stakes governance profile - mandatory DPIA, mandatory HITL approval before any external action, trustee-board briefing required pre-deployment
• Conservative eval threshold - the system is tuned for the cost of missing a real concern being much higher than the cost of a false alarm. The model flags. It doesn't decide
• A queue ordered by severity - what the designated lead sees on a Monday morning is everything the system thinks needs their attention, in the order it thinks it needs attention
• Full audit on every decision - every line in every triage record is sourced; nothing the system says is unattributed

How we'd shape it

Safeguarding deployments take longer to start than the others. We need a conversation with the designated safeguarding lead, the head of services, and ideally a trustee. We need to read your current safeguarding policy and understand where decisions get made and by whom. The discovery output is the threshold conversation: what does the system flag, what does it ignore, who sees what when.

We won't proceed without trustee sign-off on the deployment plan.

What gets built and stays running

We'd build:

• The triage pipeline taking concerns from whatever channels they arrive in
• Severity classification + routing recommendation with confidence levels surfaced clearly
• A review queue for the designated lead ordered by the system's confidence × severity score
• DPIA + trustee briefing pack generated alongside the system itself

Edge cases compound here. We hold the calibration with you and review the conservative thresholds quarterly. Every "missed" case (one the system didn't flag but the team felt it should) feeds back into the eval set.

Where this stands today

A catalogue Loop with the governance profile and eval pattern fully designed; we are looking for the right partner to land it. The conservative deployment posture is non-negotiable - if a charity wants speed over rigour here, this isn't the right Loop to start with.