Conditional hypothesis
Execution records → changes → Reported success proportion
The original article's section on abort reasons and source attribution proposes a conditional denominator effect: including previously omitted stopped attempts could lower reported success without reducing task capability. This is an alternative explanation, not an observed CrewAI regression. Source 6 supplies the logging-update background, not proof of the metric effect.
Read the evidence
If a test counts only runs that received a verdict, tasks that stopped early may disappear from the results. Even if the new version does not change task capability, simply bringing these attempts back into the count may lower the reported success proportion.
Background sources for the article's reasoning: [S6] CrewAI
Read the full research