How the check decides
A chat box will answer anything you ask it, including things it has no way of knowing. This is built the other way round: a set list of questions, rules written against named guidance, and an answer that shows its working.
Everything below describes how the check behaves today. Anything still to come is marked as such.
What happens to what you tell it
Ask a general chatbot about your child’s support and it will produce something that sounds right, cannot say where it came from, and answers differently if you ask again. This is the other shape: a short path with a named step at each point.
What you tell it
Your answers are choices from a set list. They’re kept against a reference number that’s picked at random and means nothing on its own. Nothing about your child goes into the web address.
Turned into labelled facts
Each one is marked Confirmed, You told us, Not known, or Does not match. The label travels with the fact all the way onto the page you read.
Matched against written rules
The rules are checked in a set order and the first one that fits your answers is the one you get. Nothing is scored, weighted or predicted.
Tied to a source
A finding that says something is the case has to name the guidance it comes from. One that cannot name it never reaches this site at all.
Shown to you to act on
One next step, in plain words, for you to take or to ignore.
Kept and tracked over timeComing soon
Following what you sent and what came back is the account side of Best Chance, and it’s still to come.
Where a finding gets its authority
A finding that states something names the guidance behind it: the document, who published it, and the date a person last checked it was still current. A rule that claims something without naming a source never reaches this site, so a finding cannot quietly lose its footing. The one result that cites nothing is the one that found nothing.
Who is behind the words
No expert has signed off on the words on this site, and nothing here pretends one has. When a named reviewer does check a page, you’ll see their name and what they checked, on the page.
What it will not do
These are not promises written beside the product. They’re things it cannot do.
It will not chat
There’s no open text box anywhere in a check. Every question has a set list of answers, so there’s no way to ask it something it was not built to answer.
It will not guess
If your answers match none of its rules you get the "we cannot say" outcome, which cites nothing, claims nothing, and sends you to the help pages.
It will not take orders from a document
Text inside a letter you paste is content to report on, never an instruction to follow. Where a document contains instruction-like wording, the summary says so rather than acting on it.
It will not act on its own
Nothing is sent, submitted, published or deleted unless you press something. Nothing on this site contacts a school, a council or any other organisation.
If something here is wrong
Tell us. Every finding and every guidance page carries a report link, and the form behind it sends. A report is recorded against the page or the source it came from, and held in this build’s queue until the moderation process that reads it arrives with the product. Nothing you write in one is published, shown back on the site, or treated as an allegation about a named person.
How little it needs to know
The check does not call a language model at all. That is why it needs so little from you: no name, no documents, no history. Only the answers, and only while the check is open.
Judge it by what comes back
The quickest way to test any of this is to run a check and look at what it gives you.