Compare runs
Up to four runs side by side, down to every question: where each one finds the answer, where they disagree, and what each spends to get there.
Pick two to four runs
Choose runs above, or tick them in the leaderboard's Compare column. Runs on the same board answered the same questions with the same judge, so they compare question by question.
Side by side
The best value in each row is in bold.
Head to head
For each pair, the questions only one of the two got right. A lopsided split means a real difference; p is the chance of a split at least this uneven if both runs were equally good (exact McNemar test). Pick a count to see those questions.
Accuracy against context tokens
The vertical line is each run's 95% confidence interval.
Where it fails
Accuracy by question category, with 95% confidence intervals. A low category is where the memory system misses the context the question needs.
Every question
One square per question, grouped by category. Select a square to read the question, the gold answer, the judge's reason and the context it was given.
Loading questions…
Retrieval latency
The share of questions whose context came back within a given time. Further left is faster.