Error carried forward.
The rule that stops one slipped decimal in step two costing a student the whole question. Every board expects it. Almost nobody applies it consistently at eleven at night on the fortieth paper.
What it actually says
If a student makes an error early in a multi-step answer, and then carries that wrong value correctly through the remaining steps, the later steps still earn their marks. The student is penalised once, for the mistake they made, and not repeatedly for the consequences of it.
A three-mark numerical is a mark for the correct approach, a mark for correct substitution, a mark for the arithmetic. A student who writes the right formula, substitutes correctly, and then divides wrongly has earned two of the three. Marking that answer zero because the final number is wrong is not strictness — it is a misapplication of the scheme.
Where it applies
- Numericals in Physics, Chemistry and Mathematics, which is the obvious case.
- Derivations, where a wrong intermediate expression is carried through correctly to the end.
- Accountancy and Economics, where a wrong figure in one statement flows into the next and the treatment is still right.
- Multi-part questions where part (b) uses the answer from part (a).
Where it does not
It is not a general amnesty. A wrong formula is a wrong method, not a carried error. If the mistake is conceptual rather than arithmetical, the marks for the concept are gone — the student has not demonstrated the thing the mark was for. The distinction is between “knew what to do and miscalculated” and “did not know what to do”.
Why it is so inconsistently applied
Because applying it properly means reading every step of every answer, including the answers that reached the wrong total. The temptation, on a large pile, is to check the final value against the key and move on. That is faster, and it systematically punishes the students who understood the method.
It is also the single most common reason a parent arrives at a centre holding a paper and asking why their child got nothing for an answer that is visibly almost right. They are usually correct.
The consistency problem is the real one. Most teachers apply error carried forward correctly on paper one and progressively less carefully by paper forty — not through carelessness, but because it is genuinely tiring to trace a wrong value through four steps, two hundred times.
What to do about it
If you are marking by hand: mark question-wise rather than paper-wise. Doing all forty copies of question 3 in a row holds the standard far better than doing all thirty-three questions of one paper before starting the next.
If you are using software to help, this is the specific thing to test it on. Take a paper where a student got an early value wrong and carried it through. If the tool zeroes the question, it is matching final answers against a key, and it will do that to your students every week. Our note on where AI checking should and should not be trusted has the other tests worth running.
See it on your own paper.
Send us one you have already marked, including the answers where a student slipped early. Compare what comes back. The first paper is free.