1. Ofqual has released information on accuracy of grading: roughly 45-65% at A level depending on subject. See graph - each blob is a subject: middle top is biology- most accurate at ~67%. Even here a third of grades likely to be inaccurate.

2. Method used here: model applied to 2019 results and compared to actual grade distribution. Assumes centre rank order perfect i.e. measures their level of success in dealing with the centre-level uncertainties highlighted in this thread. https://twitter.com/JonColes01/status/1292787888401403912?s=20

3. Ofqual have released a similar graph showing accuracy if centres below 10 entries removed. (This bears some relation to what they have done - they've replaced model with CAGs below 5.) Most accurate subjects then at 70% accuracy - still many with 40-50% inaccuracy.

4. I said they'd do well to get better than 75% accuracy. GCSE accuracy graph shows this perfectly. Top right are Eng and Ma - biggest entries, large cohorts in most centres - close to 75%. Lots of GCSEs below 50% accuracy.

5. Predictions get less accurate as entry size gets smaller, unsurprisingly. This is biology A level where model performs best - still under 50% accurate for centres under 10.