From 5584400222b2764fbfc1686e84f610637f3fb180 Mon Sep 17 00:00:00 2001
From: David Turner <david.turner@elastic.co>
Date: Mon, 11 Nov 2024 16:31:21 +0000
Subject: [PATCH] Add note on convergence

---
 _posts/2024-11-11-bayesian-bisection.md | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/_posts/2024-11-11-bayesian-bisection.md b/_posts/2024-11-11-bayesian-bisection.md
index 1ed9392..0f32940 100644
--- a/_posts/2024-11-11-bayesian-bisection.md
+++ b/_posts/2024-11-11-bayesian-bisection.md
@@ -125,3 +125,8 @@ f(p) = P(observed results \| p)<br>
 where successes<sub>≥k</sub> and failures<sub>≥k</sub> are respectively the
 number of successes and failures observed at commit C<sub>k</sub> or a later
 commit.
+
+Note that if you re-estimate _p_ every iteration then the median doesn't quite
+behave as described above and it's not clear that the process eventually
+converges. It might make more sense to estimate _p_ only from the test results
+on commits that have been observed to fail.