The KCSIE evidence gap: what DSLs are actually accountable for

Ask a designated safeguarding lead what takes the most time, and very few will say "identifying concerns." They will say recording them, chasing them, and being able to demonstrate afterwards that the right thing happened.

That gap — between doing the work and being able to evidence the work — is where most safeguarding pressure actually sits.

What KCSIE asks for

Keeping Children Safe in Education places a specific set of obligations on schools: a designated lead, clear reporting routes, records of concerns, timely action, and the ability to show all of it to an inspector.

The obligations themselves are rarely controversial. Schools agree with them. The difficulty is that most of the machinery for meeting them is manual:

• A member of staff notices something and writes it up, later, from memory
• A concern is logged in one system, the response in another, the outcome in a third
• Pulling a coherent account of a single incident means reconstructing it from three places and several people's recollections
• The reconstruction happens under time pressure, often in the week before an inspection

None of that is a policy failure. It is an information-architecture failure, and it produces two distinct problems.

Problem one: the record is thinner than the reality

Schools generally act well and record partially. An incident is dealt with properly in the moment — staff intervene, a parent is called, a follow-up happens — and what survives in writing is three lines entered two days later.

When that record is examined months afterwards, it does not represent what the school did. It represents what someone had time to type. The gap between those two things is invisible to the school and highly visible to an inspector.

Problem two: patterns disappear into individual entries

This one matters more, and it is specific to bullying.

Bullying is defined by repetition. A single incident of unkindness is not bullying; the same child being targeted by the same group over six weeks is. That means the safeguarding-relevant unit is not the incident — it is the pattern across incidents.

Now consider how the pattern is usually stored. Three separate log entries, written by three different staff members, using three different descriptions, filed at three different times. Each entry is individually unremarkable. The pattern exists in the data and is visible to nobody, because seeing it requires someone to read all three and make the connection.

Schools frequently discover the pattern only when a parent complains, and then find, on review, that the evidence was there all along. That is not a failure of concern. It is a failure of aggregation.

What "audit-ready" should actually mean

The phrase gets used loosely. Applied strictly, an audit-ready safeguarding record has four properties:

Contemporaneous. Created at the time of the event, not reconstructed later. A timestamp generated by the system carries weight that a recalled time does not.

Complete. The record contains what happened, where, when, and what followed — in one place rather than assembled across systems.

Connected. Related incidents involving the same student are linked, so a pattern is visible as a pattern rather than as a set of unrelated entries.

Exportable. It can be handed over — to an inspector, to a local authority, to a parent exercising a subject-access request — without a preparation exercise.

Most school safeguarding processes reliably achieve one of these four. Contemporaneous and connected are the ones that almost always fail, and they are the two that determine whether a pattern is caught early.

The role of automatic detection

This is where detection technology earns its place, and it is worth being precise about why. The value is not that a machine spots things a human would miss — although in a corridor with nobody in it, that is true.

The value is that a detected event produces a record without anyone choosing to write one.

When a physical altercation is detected, a timestamped entry exists with location, category and footage attached, whether or not a staff member later has time to log it. The contemporaneous property is satisfied automatically. When the same student appears in several such records, the connection is available immediately rather than after someone notices.

The DSL still exercises judgement — the system does not decide what is significant, and it should not. But the judgement is applied to a complete record rather than to a partial reconstruction.

What this does not solve

Worth stating plainly, because safeguarding technology is routinely oversold.

Detection covers what happens in view of a camera. A great deal of what matters to safeguarding does not: disclosures, home circumstances, online harassment, coercion in private spaces. No video system addresses those, and any vendor implying otherwise should be treated with suspicion.

What detection does is remove one specific category of failure — the incident that occurred, was recorded by a camera, and was never seen by an adult who could act. That is a narrower claim than "we make your school safe." It is also a claim that can be substantiated, which is the more useful property.

The rest of safeguarding remains what it has always been: attentive adults with enough time to notice, and a culture where children believe telling someone will help. Technology's job is to stop consuming the time those adults need.