Manual checking vs a criterion-checking tool: the honest comparison
Some things manual marking does better than any tool, and pretending otherwise sells teachers short. Some things tools do better than any human at 10pm, and pretending otherwise wastes teachers' evenings. The honest comparison, from the company that builds the tool — believe the critique or don't, the mechanics are checkable.
What manual checking genuinely does better
- Knowing the learner. "This is thin for her" — a judgement no model can make and every assessor makes daily.
- Novel evidence shapes. Unusual submissions, creative formats, evidence that doesn't sit still — human flexibility wins.
- The viva instinct. Noticing that something's off and asking the right question in the corridor — authentication at its best is a human skill.
- Accountability. The decision is the teacher's; manually-derived verdicts come with built-in ownership.
What a criterion-checking tool genuinely does better
- Locating evidence across 30 scripts without fatigue. The locate-judge-record cycle's mechanical two-thirds — the workload maths.
- Verb discipline at scale. Described work recorded as met against analyse criteria is script-40 fatigue; a checker reads the verb the same way on every script.
- Cited evidence on every verdict. The records write themselves.
- Determinism. Same evidence, same verdict, every time — the property that makes re-checks and appeals boring (in the good way).
- Brief-version awareness — reading evidence against the right scenario's data, V1 or V2, every time.
Where each fails
| Failure mode | Manual | Tool |
|---|---|---|
| Fatigue drift | Yes — the 10pm problem | No — but see next row |
| Misreading evidence | Occasionally, tired | Occasionally, confidently — which is why verdicts must be overridable in one click and every flag a question, not a verdict |
| Learner context | Blind to nothing you know | Blind to everything you know |
| Paperwork | Late, reconstructed, incomplete | Instant, but only as good as the verdicts behind it |
That last tool-failure column is the design brief for checkb.tech: advisory verdicts with cited evidence, triage-first review (the queue), teacher override on everything, and no grade emission anywhere — the vendor questions applied to ourselves.
The actual answer to "which should I use"
Both, in the order the workload wants: tool for the locate-and-check pass across the batch, human for the judgement layer — borderline verdicts, learner context, authentication instincts, and the final decision on every flag. Departments that hand the whole job to either extreme lose something: all-manual loses the evenings and the records discipline; all-tool loses the professional judgement that the assessment is legally built on. The checkb.tech workflow is built around that split by design.
FAQ
Is using a tool "cheating" on my assessment duties?
No more than an IV is — checking decisions against criteria with cited evidence is verification, and the decision layer stays with the teacher, on the record.
What if the tool disagrees with me?
You win — one click, and the record shows your decision. The disagreement was free calibration.