Regression Case Files
Regression Case FileIndependent OSS replay

A provider failure skipped every configured final processor.

Mastra 1.56.0 skipped the final output-processor pass after an unrecovered provider error. Three releases were affected before 1.59.0 restored the behavior.

The provider error was already visible.The final handling silently disappeared.
Baseline1.55.0Jul 31, 2026
analyzed change
Released1.56.0Aug 6, 2026
3 releases affected
Restored1.59.0Aug 16, 2026
The case file

The anatomy of a regression

  1. 01 Project
  2. 02 Regression & fix
  3. 03 Early replay
The project

A TypeScript framework for AI applications and agents.

Mastra is an open-source TypeScript framework for building AI-powered applications and agents, including model streaming, processors, memory, workflows, and related runtime behavior.

Project repository
The regression & fix

Configured terminal processing could be skipped after provider failure.

A fix for orphaned input persistence placed a failed-status guard around the shared runOutputProcessors call. That stopped every configured final processor, not only MessageHistory. The later fix restored the shared pass for failed streams and moved the input-only persistence rule into MessageHistory.

Provider error → failed stream → shared output-processor pass → configured terminal behavior
A narrow MessageHistory persistence rule is applied at the shared output-processor boundary, causing every final processor to be skipped after provider failure.
The persistence rule guarded a shared lifecycle stage instead of the processor that owned persistence.
Compared
1.55.0 → 1.56.0
Affected
1.56.0–1.58.0 · 3 releases
Public report
User · 133 hours after release
Fixed
1.59.0 · 247 hours later
The Early replay

Early found the regression.

1Regression
10Expected changes
Early verdict · Regression

“Output processors skipped when status failed/canceled”

Early identified the broad guard in the exact changed file and classified the target as Regression. Published packages produced processor counts of 1, 0, and 1 across versions 1.55.0, 1.56.0, and 1.59.0. The complete output also contained two false-positive Regression rows and one unresolved candidate excluded from confirmed counts.

Primary evidence

Follow the record.

Independent targeted historical replay. Not affiliated with Mastra.

Explore more Case Files
Your case file

Remember your last regression?

Let Early trace it