swe-pruner-rollout-gate

To evaluate telemetry data for a rollout flip, process JSONL records to compute metrics like pipeline count, days since merge, and regression count. The evaluation checks gating thresholds and outputs a pass, fail, or insufficient data status for operator review.

The minimum telemetry thresholds to pass a rollout gate evaluation require collecting data from at least 50 pipelines over 14 days. The evaluation also requires zero regressions and successful data-quality checks to authorize the enforcing mode flip.

A JSONL telemetry rollout gate handles insufficient data or regressions by outputting a fail or insufficient data status. It provides guidance for manual operator review, ensuring risky flips to enforcing mode are blocked until gating criteria are met.

Yes, you can assess deployment readiness across multiple pipeline runs. The gate evaluates JSONL telemetry from multiple runs, computing metrics like n_pipelines and eval_regression_count to determine if the rollout meets the enforcing mode criteria.

The purpose of an operator review path in a telemetry rollout gate is to provide clear exit conditions for risky flips. It outputs status and guidance for manual review when telemetry data is insufficient or fails gating criteria.

You should run a rollout gate evaluation on pipeline telemetry after collecting at least 50 pipelines over 14 days. This ensures enough JSONL data exists to accurately compute metrics and determine readiness to flip from advisory to enforcing mode.