DEV Community

Cover image for A school trust wants web filtering. I thought it was a blocklist. It isn't.
Gavriliuc Vlada
Gavriliuc Vlada

Posted on AI-assisted

A school trust wants web filtering. I thought it was a blocklist. It isn't.

This week I found a small notice on Find a Tender that I almost skipped. The title is just "Web Filtering and Monitoring". The buyer is The David Ross Education Trust, a multi-academy trust in England (a group of schools run by one organisation).

My first thought was that this is boring. Web filtering is a blocklist, right? You have a list of bad websites, the school network says no, done. I have seen this as a student myself. Half of the internet was blocked and we all knew which proxy sites still worked.

But I read the notice and then the government guidance it points to, and I changed my mind a bit. It is more like an access control problem with a safeguarding team at the end of it.

You can read the notice here.

What the notice says

First the facts, then my guesses.

It is a preliminary market engagement notice, so it is not a tender yet. The trust wants to talk to suppliers before it writes the real procurement. It was published on 26 October 2026 and edited the next day. The deadline to register interest was 6 November 2026, so that part is already closed. The estimated date for the real tender is 11 January 2027. The contract would run from 1 July 2027 to 1 July 2030, with a possible extension to 1 July 2032. I did not see a value in the notice.

The things they are interested in include:

  • staff and student devices
  • filtering and monitoring on-site and off-site
  • working across a multi-academy trust
  • "Centralised safeguarding, alerting and reporting functions"
  • "Role-based administration and access controls"
  • following the Department for Education filtering and monitoring expectations

They also ask suppliers if one company should do both filtering and monitoring, or if it is better to split them into separate contracts. I found that question interesting, and I come back to it at the end.

What the DfE guidance adds

The notice mentions DfE expectations, so I opened the filtering and monitoring standards on GOV.UK. The page says it was updated on 16 September 2026.

A few lines surprised me as a developer:

  • Filtering should block harmful content "without unreasonably impacting teaching and learning". So blocking too much is also a failure.
  • The filtering system should not have "a blanket filtering profile for all users". So one list for everybody is not enough.
  • Filtering should also apply to school devices taken off-site, and to personal devices under a bring your own device (BYOD) scheme.
  • Some blocklists, from the IWF and the CTIRU, "cannot be disabled, overridden, or altered by any user". Not even an admin.
  • Monitoring should give "immediate reports when an incident is classed as high-risk" and also weekly reports.
  • The designated safeguarding lead (DSL) is the person responsible for safeguarding issues found through monitoring.

So now I have users with different profiles, different devices, two kinds of rules (locked ones and normal ones), and two speeds of reporting. This is the point where I wanted to write code.

My tiny model

Everything in this example is invented by me. The profile names, categories and rules are not from the notice. I just wanted to see the shape of it.

from dataclasses import dataclass

# lists that nobody is allowed to switch off
LOCKED = {"child_abuse", "terrorism"}

# who may see which category (my own invented rules)
PROFILES = {
    "year7":  {"social_media": False, "gaming": False, "adult": False},
    "year12": {"social_media": True,  "gaming": False, "adult": False},
    "staff":  {"social_media": True,  "gaming": True,  "adult": False},
}

@dataclass
class Request:
    user: str
    profile: str
    device: str   # "school" or "byod"
    on_site: bool
    category: str

def decide(r):
    if r.category in LOCKED:
        return "block", "high"
    allowed = PROFILES[r.profile].get(r.category, True)
    if allowed:
        return "allow", None
    return "block", "low"

def route(r, risk):
    if risk == "high":
        return "alert DSL now"
    if risk == "low":
        return "weekly report"
    return "-"

requests = [
    Request("amir", "year7", "school", True, "education"),
    Request("amir", "year7", "school", False, "gaming"),
    Request("lena", "year12", "byod", True, "social_media"),
    Request("mr_k", "staff", "school", True, "social_media"),
    Request("lena", "year12", "byod", True, "terrorism"),
]

for r in requests:
    action, risk = decide(r)
    where = "on-site" if r.on_site else "off-site"
    print(f"{r.user:5} {r.profile:6} {r.device:6} {where:8} {r.category:13} -> {action:5} | {route(r, risk)}")
Enter fullscreen mode Exit fullscreen mode

When I run it I get:

amir  year7  school on-site  education     -> allow | -
amir  year7  school off-site gaming        -> block | weekly report
lena  year12 byod   on-site  social_media  -> allow | -
mr_k  staff  school on-site  social_media  -> allow | -
lena  year12 byod   on-site  terrorism     -> block | alert DSL now
Enter fullscreen mode Exit fullscreen mode

The important line for me is the first check in decide. The locked list is checked before the profile, so no profile can open it. If I put that check after the profile lookup, a badly configured staff profile could allow something that should never be allowed. In my first version I did exactly that, and only noticed when I wrote the comment above it.

Where my model is too simple

In quite a lot of places.

My model sends every normal block to the weekly report. But a Year 7 student trying a gaming site at lunch is probably not a safeguarding concern. If the DSL gets 4,000 lines like that every week, the one real problem gets lost. I think a big part of the real product is deciding what is noise.

I store device and on_site but my rules don't use them. In real life I guess they matter a lot. How do you even filter a school laptop at home? My guess is some agent software on the device, or traffic going through the school's system, but I don't know which is more common.

Monitoring is more than URLs. The guidance talks about alerts and observations, and some products watch what people type. That is a big privacy question for students and I don't feel ready to model it.

And the trust has many schools. Who can change the profiles? One central team, or each school? "Role-based administration" probably means that a teacher in one school should not see alerts from another school. That would be one more check in my code.

The one supplier or two question

Now I understand why the trust asks this. Filtering and monitoring share the same data: who the user is and which profile they have. If two suppliers do them, they both need the same user list and they need to agree on who is who. That sounds like a sync problem to me. But one supplier for everything means you depend on one company for five years or more.

I don't know the right answer. I am curious what people think.

If you have worked on school IT or on any content filtering system, did you split filtering and monitoring between tools, or keep them in one? And how did you stop the alerts from turning into noise?

Top comments (0)