How Departments Triangulate AI Marking to Self-Moderate and Close the Feedback Loop

A Pattern We Didn't Design, That Several Departments Built Anyway

We've written before about the real ways schools used GradeDrive over summer 2026 (see AI Marking in Summer 2026 for the full picture). One pattern from that summer is worth its own, closer look, because it's not a feature anyone shipped, it's a process several departments in London built on top of GradeDrive themselves, for a reason that has nothing to do with saving time: moderation.

The pattern: every script in a department goes through AI marking first. Every teacher's own marking of those same scripts is then checked against that AI-marked first pass, before results go out. Because GradeDrive applies the same mark scheme the same way to every script regardless of which class or which teacher it came from, that first pass gives a department a consistent second read across an entire cohort, the kind of check that's normally expensive in teacher time to do properly at the end of a busy term.

One school went a step further and added a third layer: students marked their own papers against the mark scheme before the AI-marked and teacher-reviewed results came back. That gives a department three independent reads on the same scripts: the student's own assessment, the AI's, and the teacher's. It also means students engage directly with the mark scheme as part of getting their results back, not just the final number (we've covered that side of it from a student's perspective in What AI Marking Actually Gives Students).

What This Isn't

Worth being precise about this, because it's easy to read "self-moderation" and assume it's a mode you switch on. It isn't. GradeDrive doesn't have a dedicated multi-pass, triangulation or moderation mode. What these departments built is their own process, layered on top of the same AI-marks-then-teacher-reviews workflow every GradeDrive submission gets by default, just applied deliberately across a whole department's papers rather than one teacher's set, and in one case with a student self-assessment step added in front of it. If you want this, you're not configuring a setting, you're designing a department process and using GradeDrive as one (or two) of the independent reads inside it.

Why Three Reads Catches More Than One

A single mark, from any source, tells you what one marker (human or AI) thought a script was worth. It doesn't tell you whether that's reliable. Three independent reads on the same script, each produced without seeing the others first, tell you something a single mark can't: where they agree, you have real confidence; where they disagree, you've found exactly the scripts worth a second look, rather than discovering it blind in an exam board remark months later.

This is the same logic behind ordinary moderation meetings, just applied earlier and at a different scale. A department moderation meeting typically samples a handful of scripts and compares a few teachers' judgement on them, after marking is already finished, which means problems get caught after students already have their results. Running the AI pass first, and treating it as one of the three reads rather than the final word, means the same kind of check happens on every script, not a sample, and before feedback goes out rather than after.

Building This in Your Own Department

Decide what "disagreement" means before you start. Three reads are only useful if you've agreed in advance what counts as a meaningful gap, a mark or two on a six-mark question is different from a mark or two on a two-mark question, and what you do when you find one (default to the teacher's professional judgement, flag it for a second teacher, or something else).

Run the AI pass across the whole set, not a sample. The value of this pattern is that every script gets the same check, not just the ones that would have been sampled anyway. Every mark GradeDrive proposes still goes through a teacher's review before anything is finalised; see How GradeDrive Helps Teachers Cut Marking Time Without Losing Control for exactly how that review step works.

Decide whether a student self-assessment layer is worth adding. It's the least essential layer and the one requiring the most setup (students need the actual mark scheme, or a version of it appropriate to share), but it's also the one with a benefit beyond moderation: it changes how a student engages with their own result, not just how reliable the result is.

Treat this as a department decision, not an individual teacher's workflow. The whole point is a consistent check across every class and every teacher in the department, which means it needs a head of department coordinating it, not each teacher deciding independently whether to run it on their own set.

Closing the Feedback Loop

The phrase "closing the feedback loop" gets used loosely, but what it means concretely here is this: feedback is only as useful as it is trustworthy and timely, and a process that catches marking drift before results go out, rather than after, is feedback a student and a parent can actually rely on the first time, not a mark that might get quietly revised after a department moderation meeting weeks later. Combined with GradeDrive's own turnaround (see How AI Exam Marking Feedback Transforms Student Progress for the numbers), the loop closes faster and with fewer corrections needed after the fact, not because any single part of it is infallible, but because three independent checks on the same script catch more than one check does.

Frequently Asked Questions

Does GradeDrive have a built-in self-moderation or triangulation feature?

No. GradeDrive doesn't have a dedicated multi-pass or moderation mode. The pattern described in this post is a process some departments have built themselves on top of GradeDrive's standard AI-marks-then-teacher-reviews workflow, not a setting you switch on.

What is the three-layer moderation pattern, exactly?

Every script goes through AI marking first, as one independent read. Each teacher's own marking is then checked against that first pass, as a second read, giving the department a consistent benchmark across the whole cohort. Some departments add a third layer: students marking their own papers against the mark scheme before either result comes back.

Why would a department do this instead of a normal moderation meeting?

A normal moderation meeting usually samples a handful of scripts after marking is finished. Running an AI pass across every script, and comparing it to every teacher's marking, surfaces disagreement across the whole cohort rather than a sample, and does it before feedback goes out rather than after.

Is this pattern specific to London schools, or can any department use it?

The real examples we're aware of, described in AI Marking in Summer 2026, happened at several schools in London. Nothing about the pattern is specific to London; any department using GradeDrive can build the same process.

Does adding a student self-assessment layer require a separate tool?

No separate tool is required. It requires sharing the mark scheme (or an appropriate version of it) with students before AI marking and teacher review happen, which is a process decision for the department, not a product feature to configure.

Internal Links

For the real usage patterns this post expands on, see AI Marking in Summer 2026. For the student-side benefit of self-assessment, see What AI Marking Actually Gives Students. For the consistency case for departments generally, see Why Teachers Choose GradeDrive: Consistency Is the Real Decision Factor. For how the teacher-review step works day to day, see How GradeDrive Helps Teachers Cut Marking Time Without Losing Control.