Maths · Predicted Papers
Maths
29
Papers to sit
3
Exam boards
2027
Forecast year
These are forecasts, not certainties. They are trained on Edexcel, AQA and OCR series up to 2023 and 2024, depending on the board. The questions are ours: every paper here is an original built to the shape and topic mix its board's forecast expects, not a copy of a real paper and not a leak.
How far past the data this reaches
The series sat since each board's own cutoff are not in the corpus yet, so predicting 2027 means extrapolating up to 4 years past the last real data point rather than the usual one year. Each board's figures are its own first forecast carried forward, and are refreshed as later series are tagged; its exact cutoff is on its section below.
01
AQA GCSE
Forecast for 2027, trained on AQA series to 2023.
6 papers · 480 marks
F1
Foundation Paper 1
FoundationF2
Foundation Paper 2
FoundationF3
Foundation Paper 3
FoundationH1
Higher Paper 1
HigherH2
Higher Paper 2
HigherH3
Higher Paper 3
Higher02
AQA A Level
Forecast for 2027, trained on AQA series to 2023.
3 papers · 300 marks
03
Edexcel GCSE
Forecast for 2027, trained on Edexcel series to 2024.
6 papers · 480 marks
F1
Foundation Paper 1
FoundationF2
Foundation Paper 2
FoundationF3
Foundation Paper 3
FoundationH1
Higher Paper 1
HigherH2
Higher Paper 2
HigherH3
Higher Paper 3
Higher04
Edexcel A Level
Forecast for 2027, trained on Edexcel series to 2024.
3 papers · 300 marks
05
Edexcel AS Level
Forecast for 2027, trained on Edexcel series to 2024.
2 papers · 160 marks
P1
Paper 1
P2
Paper 2
06
OCR GCSE
Forecast for 2027, trained on OCR series to 2024.
6 papers · 600 marks
F1
Foundation Paper 1
FoundationF2
Foundation Paper 2
FoundationF3
Foundation Paper 3
FoundationH1
Higher Paper 1
HigherH2
Higher Paper 2
HigherH3
Higher Paper 3
Higher07
OCR A Level
Forecast for 2027, trained on OCR series to 2024.
3 papers · 300 marks
How the Edexcel forecast is built, and where it is weak
- These figures are Edexcel Maths and nothing else. The same model is run separately over each board this site holds question-level topic tagging for (Edexcel, AQA and OCR), and each board is tuned on its own history, so none of these numbers carries over to the Edexcel papers' neighbours in the list above. GCSE Foundation/Higher Papers 1-3 (2017-2024), A-Level and AS-Level (2018-2024). The only board in this corpus carrying AS-Level papers.
- Built from every question in 8 years of this site's own tagged Edexcel past papers (2017 to 2024 for GCSE, 2018 to 2024 for A Level and 2018 to 2024 for AS Level), not an opinion about the specification.
- The 2020 and 2021 papers behind this model were pandemic-disrupted resit series sat by small, non-representative cohorts, not a normal summer exam - they are kept in the data (dropping them would leave a hole) but are a genuinely weaker data point than the other years.
- Validated by predicting each of 2022, 2023 and 2024 from only the years before it. Over those folds it placed 70-75% of the topics that really came up in its top-ranked list of the same length. That clearly beats simply repeating last year's paper, but only modestly beats a plain historical-average baseline (a few percentage points of hit rate, at a small cost to how well the raw probabilities are calibrated). It is a real, if modest, edge - not a breakthrough.
- Probabilities below about 10% are poorly calibrated in this data, so the model's raw numbers are published nowhere on this site: they decide how many marks of each topic the papers above are built from, and nothing else. Treat a paper's topic mix as a direction of travel, not as a reading of the real one.
How the AQA forecast is built, and where it is weak
- These figures are AQA Maths and nothing else. The same model is run separately over each board this site holds question-level topic tagging for (Edexcel, AQA and OCR), and each board is tuned on its own history, so none of these numbers carries over to the AQA papers' neighbours in the list above. GCSE Foundation/Higher Papers 1-3 and A-Level Papers 1-3, 2017-2023. No 2024 sitting in the corpus and no AS-Level papers at all.
- Built from every question in 7 years of this site's own tagged AQA past papers (2017 to 2023 for GCSE and 2018 to 2023 for A Level), not an opinion about the specification.
- The 2020 and 2021 papers behind this model were pandemic-disrupted resit series sat by small, non-representative cohorts, not a normal summer exam - they are kept in the data (dropping them would leave a hole) but are a genuinely weaker data point than the other years.
- Validated by predicting each of 2022 and 2023 from only the years before it. It correctly ranked the right topic 51-57% of the time, close to a coin flip. Its "recency-weighted" ranking loses to a plain historical-average count in 2 of its 3 own validated years, and its expected-marks figures ARE that plain historical average - AQA's own past-paper history showed no benefit from weighting recent papers more heavily. This is not a validated improvement over simply counting how often a topic has come up before.
- Probabilities below about 10% are poorly calibrated in this data, so the model's raw numbers are published nowhere on this site: they decide how many marks of each topic the papers above are built from, and nothing else. Treat a paper's topic mix as a direction of travel, not as a reading of the real one.
How the OCR forecast is built, and where it is weak
- These figures are OCR Maths and nothing else. The same model is run separately over each board this site holds question-level topic tagging for (Edexcel, AQA and OCR), and each board is tuned on its own history, so none of these numbers carries over to the OCR papers' neighbours in the list above. GCSE Foundation/Higher Papers 1-3 and A-Level Papers 1-3, 2017-2024. A-Level Papers 2 and 3 are OCR's genuinely combined Pure+Statistics and Pure+Mechanics components, modelled as such rather than split.
- Built from every question in 8 years of this site's own tagged OCR past papers (2017 to 2024 for GCSE and 2018 to 2024 for A Level), not an opinion about the specification.
- The 2020 and 2021 papers behind this model were pandemic-disrupted resit series sat by small, non-representative cohorts, not a normal summer exam - they are kept in the data (dropping them would leave a hole) but are a genuinely weaker data point than the other years.
- Validated by predicting each of 2022, 2023 and 2024 from only the years before it. It correctly ranked the right topic 45-53% of the time, close to a coin flip. Its "recency-weighted" ranking loses to a plain historical-average count in 2 of its 4 own validated years, and its expected-marks figures ARE that plain historical average - OCR's own past-paper history showed no benefit from weighting recent papers more heavily. This is not a validated improvement over simply counting how often a topic has come up before.
- Probabilities below about 10% are poorly calibrated in this data, so the model's raw numbers are published nowhere on this site: they decide how many marks of each topic the papers above are built from, and nothing else. Treat a paper's topic mix as a direction of travel, not as a reading of the real one.