A second reader for human marking
Every mark, read twice.
Your markers do the marking. OutFigure lines up every first mark with its second reading and hands the ones that don’t add up to a senior marker before results go out. Nothing changes behind anyone’s back, and every call is written down.
Mock paper, accountancy
Flagged 61 a gap over the allowed
Changed 49 by a senior marker
Kept 12 the second reading was wrong
- marker-05/014 marked 1, read again as 5 mark changed
- marker-06/032 marked 4, read again as 0 mark changed
- marker-07/013 marked 4, read again as 0 mark changed
- marker-07/021 marked 7, read again as 3 mark changed
- marker-08/041 marked 5, read again as 1 mark changed
A wrong mark is usually found by the candidate.
A senior marker can only check a handful of scripts from each marker. The rest goes out on trust.
Right now a wrong mark gets caught one of two ways. A senior marker happens to pick that script, or a candidate who knows they did better asks for a review, weeks after results, when fixing it means a re-mark, a reissued result and a little less faith in you.
Markers are good people doing a long, tiring job. At eleven at night anyone can miss a mark given on the candidate’s own figure, go soft on the last batch, or read one question tougher than everyone else. You can’t see it until someone looks at every script.
Every script gets a second reading. People decide the rest.
OutFigure reads what your marking platform exports, or what your system sends through the API. You keep your tools and your people.
-
Start with a past session. Send a session you’ve already marked and moderated, with whatever second marking you did and the marks that changed later. You’ll see which of those changes a flag would have caught before results.
-
Bring the marks. While marking is going on, or right before results. Today the second reading is your own second marker’s, on every script or on the sample you already check.
-
Flag the gaps. When the two marks are further apart than you allow for that question, the script goes to a senior marker. A marker who runs hard or soft, is all over the place, or drifts through their pile is called out beside their name.
-
A senior marker decides. A person looks at every flag and keeps the mark or changes it, and can say why. Nothing changes a mark on its own.
-
Sign it off. An owner or admin signs the session off once every flag is decided. The record and the final marks are ready for results, with a fingerprint that shows whether the file has changed since.
What works today, on made-up scripts.
All of it runs in the console and the API, tested on a made-up paper. None of it has run on a real organisation’s marking yet.
-
Line up both marks
worksUpload your platform’s export or send scripts through the API: each first mark beside its second reading, by marker and question. A script nobody read twice stands as marked and is counted, so a sample is never mistaken for the whole.
-
Flag the gaps
worksLargest gap first, against the allowed gap you set for each question.
-
Call out markers
worksHarsh, lenient, inconsistent or drifting, measured against the second reading.
-
A senior marker decides
worksFlags are shared out among your senior markers, who decide each one and can say why. A flag the senior marker keeps means the second reading was the one that was wrong, and that’s counted too.
-
Audit a past sitting
worksSet what a second reading would have flagged beside the marks that changed later, at moderation, a review or an appeal.
-
Read the scripts itself
being builtA model reading the scripts nobody second-marked, against your mark scheme, only where your agreement allows it and through a provider whose terms rule out training on your candidates’ answers. It will flag. People will still decide.
Proof, for whoever asks.
Each session keeps a record you can hand to a moderator, a regulator or a candidate’s review panel. It names no candidate.
-
Every flag, with both marks and the gap you allowed.
-
Every decision, with who made it, when and why. Decisions are added, never written over.
-
Each marker’s pattern across the session, and feedback written to them.
-
How much was read twice, so a sample is never mistaken for the whole paper.
-
The final marks, one row per script, with a fingerprint taken when the session was signed off.
For anyone who marks at volume and answers for it.
No research team needed. OutFigure isn’t tied to any exam body or marking platform.
-
Awarding bodies
Qualifications marked by teams of examiners, with reviews of marking to answer for.
-
Professional bodies
Their own exams, and their tuition partners’ mock exams.
-
Universities
Coursework that is second-marked and moderated, against a deadline.
-
Test-series providers
Mock exams marked in a rush before the real thing.
What we won’t do.
-
No machine changes a mark. A flag only ever hands a script to a person. Your markers mark and your senior markers decide.
-
Your candidates’ answers never train a model for anyone else. Today OutFigure needs marks, not answers. When it reads scripts, it will do so only where your agreement allows it.
-
No accuracy we haven’t measured on your scripts. The only number we’ll show you is the one from your own marked session.
-
No replacing your marking platform. OutFigure reads what it exports. Nothing about how your markers work has to change.
Send us a session you’ve already marked.
We’ll show you what a second reading would have flagged, and which of the marks that changed later it would have caught first.
Run it on a past session See a sample report
Every figure on this page comes from a made-up paper with made-up markers, run through the same code as the API. It says nothing about any real marker or organisation.