How a question is checked
“Some drafts are AI-assisted” is a fair thing to be uneasy about, so here is the process rather than a reassuring phrase. Every step below is something our code does today, and each one is written so you can hold us to it. Where we cannot prove something, this page says so instead.
Where the questions come from
Questions come from BrightEleven's own rule-based generators and from authors working to a written brief, and some drafts are AI-assisted. A new question can only be published by a step that refuses to run unless a verifier has independently re-derived that batch's answers and filed a report - the check does not trust the author or the generator, and the record is written against the rows as they go live. Across the bank, automated verifier rules and bias audits run in our build pipeline, and any question a family reports is checked against its source and corrected or removed. We do not claim that a person has individually reviewed each of the questions already live.
That paragraph is the whole of our claim, and it is the same paragraph on our trust page, in our terms and in our privacy notice - one copy, so it cannot quietly get stronger in one corner of the site. The rest of this page is what sits behind it.
The step that refuses to run
A batch of new questions
A batch does not go live because somebody decides it is ready. It goes live through a step that stops with an error unless an independent verifier has re-derived every answer in that batch from scratch - working from the question rather than from the code or the author that produced it - and filed a report. No report on file, nothing published. The record of what was checked, by whom and where the report is gets written against those rows in the same moment they become live, so the claim and the data cannot come apart afterwards.
A single question, published on its own
Since 23 September 2026 the same rule covers one question published by hand from our own admin tools. It is refused outright unless whoever publishes it records, in writing and against that row, that they re-derived the answer themselves. If somebody later edits what the question asks and does not re-derive it, the question takes itself back out of the live pool rather than standing on a record that no longer describes it.
The review is blind
The second check is deliberately made harder than it needs to be. The answer key is hidden from the reviewer, the options are shuffled into a different order first, and the reviewer's answer is compared with ours as text rather than as a letter - so agreeing on “B” proves nothing, and the two have to agree on the actual answer.
Checks that run without anyone asking
On every change, before anything ships
Two gates run with our tests and fail the build rather than filing a warning somebody has to remember to read. One refuses any code path that could put a question straight into the live pool without the checking step above. The other audits every question generator for option bias - the habit of making the right answer stand out, which is how a child learns to pick by shape instead of by reasoning.
Over the live bank, overnight
A separate audit runs against the questions that are actually live and looks for a pattern no single question could show you: a subtopic where the correct answer is systematically the longest of the options, or systematically the shortest. Anything above the ceiling is flagged and repaired by hand. We print no figure from it here on purpose - it re-measures the whole bank every night, so a number on this page would be a snapshot dressed up as a fact.
What we do not claim
No claim of question-by-question human review
The paragraph at the top of this page ends with the limit of what we actually know about the questions already on the shelf, and nothing anywhere on this site goes past it. The checks above are real, and they are recorded against the rows they cover. A promise that a person has personally worked through the whole bank is not one we can substantiate, so we do not make it.
No pass rates, no forecasts
We publish no pass rate and no success figure, because we have none we can verify - and this is the page where inventing one would be easiest. BrightEleven does not produce a standardised score or a pass prediction for anyone: a parent sees a readiness band and a topic breakdown, and a child never sees a mark for a paper at all.
No exam-board affiliation
We describe the practice as GL Assessment-style because that is the format it follows, and for no other reason. We are not connected to any exam board, and the line at the foot of this page says so in full.
When you report a question
Every question your child answers - in practice, in a Discovery Quest and in a mock paper - carries a report button on the same screen. A child picks a reason from a short list and there is nothing for them to type, which is deliberate: no child-facing screen on BrightEleven has a free-text box.
A report reaches us with the question attached. We check that question against its own source - the passage, the rule, the working behind it - and then either correct it or take it out of service. If the correction changes what the question asks, it does not simply carry on: it goes back through review before it can be live again, because the check that passed was a check on different words.
We do not promise a turnaround time, and you will not find one on this page. We are a small team, and a published deadline we cannot hold would be worse than none at all. What we will say is that a person reads every report.
Taking a question out of service
A question we withdraw becomes a draft again, and a draft is invisible to every child from that moment. The code that assembles a practice session, a Daily Quest or a mock paper only ever asks for live questions, so a question we have taken out cannot turn up in one - not in a paper being composed now, and not in Fix-It. Nothing is quietly deleted either: the question and its history stay, so we can see what went wrong and check whether the same mistake is anywhere else.
Read the rest
- What we never do - every commitment we make, gathered in one place.
- How a question goes live - the short version, on our about page, with who runs BrightEleven.
- What we cover, and what we do not - the subjects and test formats we are built for, and the ones we are not.
- Child privacy notice - exactly what we collect about a child, in plain English.
GL Assessment is a trademark of its owner. BrightEleven is independent and is not affiliated with, endorsed by or connected to GL Assessment or any exam board.