October 2026 CBLE
Import your Talogy PDF or paste 80 answers. Compare your answers with the survey’s provisional key and see where respondents disagree.
Your provisional estimate
The passing estimate and ranges assume past score errors transfer to this exam. October accuracy is not independently validated. See Evidence and limits.
Separate assumption-based sensitivity calculation
This older probability calculation does not provide a calibrated interval around the rule estimate.
- Conditional passing estimate
- Historical 80% band
- Historical 95% band
Possible whole-number scores
Bell curve is a normal approximation. Actual scores are whole numbers; reported bands still use observed score errors.
Modeled probability for each answer
New: five-exam historical tests and consensus mistakes. Each answer’s details include a separate historical equal-vote probability comparison.
How these probabilities were tested
April’s 318 usable survey responses produce one set of weighted votes for each question. The calibration uses 78 single-answer questions, including the shared-consensus error on Q19. It learns vote-margin sensitivity and an allowance for consensus mistakes; there is no fixed target of five errors.
Each test withholds one question’s official answer, fits on the other 77, and predicts all four choices for the withheld question. The weighting method was already developed using April, so this is an internal check—not validation on a new exam.
| Method | Log loss | Brier score |
|---|---|---|
| April-fitted margins + error allowance | 0.112 | 0.031 |
| Raw weighted vote shares | 0.274 | 0.127 |
| Same historical probability for every leader | 0.095 | 0.026 |
The margin model improves on raw vote shares but does not beat the simpler leader-only baseline. With Q19 withheld, it assigns 98.9% to the wrong leading answer. This model allows consensus to be wrong; it has not demonstrated an ability to identify those failures in advance.
April contains only one wrong leading answer in these 78 questions. Priors stabilize this limited evidence. Each question shows how its leading probability changes under alternative prior settings. Multiple accepted answers and universal credit are not forecast.
Evidence and limits
How the estimate works
For each question, the model ranks voters by agreement on the other 79 questions. It uses the top 10%. Weights move halfway toward equal votes. A small adjustment reduces repeated influence from similar answers. Cutoff ties share the remaining places. All 80 October questions count.
Confidence and limits
Score bands use April 2026 results: 231 calibration sheets and 58 check sheets, including all 80 questions.
The passing percentage counts historical score errors that leave you with at least 60 correct answers. One prior pass and one prior fail prevent 0% or 100% estimates from limited history.
These estimates assume historical errors transfer to October. October accuracy is not independently validated. Agreement can include shared mistakes. Your own survey submission may remain in the weighted pool. Vote support is not the probability that an answer is correct.
Read the methodology and accuracy details.
Privacy
Your PDF and answers stay in this browser. Nothing is submitted or saved. Clear removes entered answers. Scanned PDFs require manual answer entry.