Batch Bisection Statistics API

Bisection is a key signal for batch size optimization:

• Frequent bisections suggest the batch size may be too large, causing more CI failures and slower throughput
• No bisections may indicate room to increase batch size for better merge velocity
• Deep bisections (high depth) reveal that failures are hard to isolate, possibly due to flaky tests or complex interactions

Emitted when a failed batch is split into sub-batches:

• Batch size before split — how large the batch was when it failed
• Number of sub-batches — how many pieces it was divided into
• Bisection depth — 0 for the first split, increments for nested splits
• Whether it’s top-level — distinguishes root splits from recursive ones

Emitted when the culprit PR is identified:

• Culprit PR numbers — which PR(s) caused the failure
• Remaining PR numbers — PRs that were re-queued after culprit removal
• Max failure resolution reached — whether the bisection hit the configured retry limit