The relentless annual increase in top A-level grades has been partly driven by examiners giving pupils the "benefit of the doubt", the head of a major exam board has said.
When faced with borderline decisions about what a paper is worth, examiners tend to favour the pupil and award the higher grade, according to Simon Lebus, chief executive of Cambridge Assessment, parent company of the OCR board.
Because exam boards then use the outcome of that decision as a reference point for the following year's grades, the effect is compounded, gradually pushing down the threshold needed for a particular grade, he said.
"If you are looking at a borderline script and you are trying to decide whether it is an A or B, you are likely to give (the candidate) the benefit of the doubt and give them an A," said Mr Lebus.
"Then the following year the same threshold would be set at a lower level. There is a kind of compound interest effect."
In last month's schools white paper, the Government said exam mark schemes should take greater account of spelling, punctuation and grammar in all subjects. Mr Lebus said he was not suggesting that examiners gave pupils the benefit of the doubt in those areas.
Speaking at the Westminster Education Forum last week, he said examiners' "benefit of the doubt" decisions were one of several factors behind the A-level "rise in attainment". The rise was not confined to school exams and could also be seen in university degrees, he added.
Mr Lebus said others factors behind the A-level improvements included greater transparency in the exam system, the difficulties of maintaining standards at a time of rapid change, and modularisation and resits.
Ministers said last month that they would ask Ofqual, the exam regulator, to change the rules to prevent pupils from resitting large numbers of exam units.
But speaking at the Westminster Education Forum, Isabel Nisbet, Ofqual chief executive, suggested the research on resits quoted by ministers in the white paper was outdated and A-level reforms had since led to a drop in the phenomenon.