Platform · Studio
A desk, not a dashboard.
The people running the work get a queue that clears itself, referrals that open on their evidence, and reports assembled from the record.
Run daily by centre teams across 34 centres in 20 markets
Queue · 5CO01 · Cohort 2026-A
37 of 38 marked01 · The queue
Approve a cohort in one sitting.
Every marked script lands in a queue by state: flagged, up for review, dispatch ready, approved, amends set. Tick the cohort, approve, release.
Each row shows its grading basis: gates passed, judged band, any flags. A referral opens on the quote, the failed check and the written reasoning, so you judge the decision instead of reconstructing it, and an override is recorded like everything else. Ineligible work never reaches this queue at all: it goes back to the learner with its reasons attached, under two minutes on instant plans.
02 · The desk at work
Watch the desk clear itself.
Three screens from the same desk: the queue resolving as it does in a centre, the cohort statistics, and the record answering a question.
Triage happens by itself
Scripts resolve as they are marked. Clean passes release on your rules; only the ones needing judgement reach a human. Your reviewers see referrals, not a pile.
Outcomes you can stand behind
Band distributions per cohort, checked against the calibration set. Drift shows up as a flag on this screen, not as a surprise at moderation.
The record answers for itself
Ask in plain language: why a learner was referred, how a criterion was evidenced, what changed between attempts. The answer is retrieved from stored decisions, with the quote attached.
A cohort that used to take a weekend of clicking now clears before lunch, with the evidence one click away on every row.
“…as our engagement survey shows, absence fell once…04:12
03 · Files & content
Every file, filed against the record.
Whatever a learner submits, one pipeline reads it, and the desk keeps it where you can find it.
A claim at 04:12 of a recorded presentation, a natively parsed chart, a table read as typed fields: non-text evidence runs through the same extraction and gating as prose. And every artefact is stored against the learner, the unit and the attempt: the submission, the marked document, the feedback sheet, each version kept. Nothing lives on a marker's desktop, and nothing needs a folder convention to be found.
04 · Settings you can read
Configuration is a sentence, not a ticket.
Units, brief years and optional modules are configuration. Versions are kept, every change is audited, and each module switches per unit. No engineering ticket required.
A new brief year is a new configuration version. Nothing resets, nothing retrains, and both years stay runnable and auditable side by side. Centre-wide rules read the same way: who approves, what releases automatically, where the data lives, written in words your IQA can take to an EQA visit.
Case study: Meridian BioTech · 8 ACs · 4 gates
2025 brief · v2 remains runnable and auditable
Modules · 5CO01
Most-referred criteria
AC 3.2 referrals doubled under the 2026 brief: question wording flagged
RTD causes this month
word count 41% · missing signature 28% · wrong brief year 17%
Unit health
weakest unit: 5CO02 · pass-first-time down 6 pts vs 2025 cohort
05 · Insights
The analytics your record was already writing.
Every decision the desk stores becomes queryable: most-referred ACs, weakest units, RTD causes, difficulty shifts when a brief year changes.
Programme reviews argue from the record, not anecdote. The assessment question that keeps confusing learners is caught the term it appears, not at the end-of-year board.
06 · The Copilot, in the corner
Every screen can be asked.
Less a chatbot than retrieval over every stored decision, sitting in the corner of the whole desk.
Ask about any learner, AC or trend and the answer comes from the marking record: the quote, the failed gate, the band reasoning, never generated fresh. Forward a learner's email and the reply is drafted from the saved reasoning. The query that used to mean finding which marker graded it becomes a same-day answer from admin or IQA.
From the marking record: 34% of referrals trace to AC 3.2, double the 2025 rate. The 2026 brief reworded task 2, and the learner-confusion flag is up on that question.
Referrals by criterion · Cohort 2026-A
vs 2025 cohort · brief-year difficulty shift flagged
Every figure is retrieved from stored decisions. Ask anything, drill anywhere.
Retrieval, amendments and the drafted replies in full:the Copilot page →
07 · Who sits where
The desk knows who you are.
One record underneath, a different desk per chair, and every action signed.
Approvals per policy
Sampling, review-before-release and batch approval sit exactly where your IQA policy names them. Sample a percentage of marked scripts, hold referrals for sign-off, or hold everything. The desk enforces the policy; it never improvises one.
Role-scoped views
The operations lead opens on the dispatch queue, the lead IQA on samples and moderation-style reports, the student-success team on the Copilot. Each chair sees its own desk, and none of them sees more record than its role allows.
Desk actions on record
Every approval, override and amendment is stored with who made it and why, next to the marking decisions themselves. When the EQA visit opens, the desk's own history is part of the pack, not a separate hunt.
Centre operations · lead IQA · assessors · student-success · tutors · compliance
Put a cohort on the desk.
A pilot runs your own scripts through the full desk: queue, reviews, files, insights and the Copilot.
A 30-minute briefing: one cohort scoped, your own scripts marked in front of you before anything integrates.