The Spread Collision: When LLM Output Silently Overwrites Your Workflow State

Description and use

A spread operator in LLM output silently overwrites metadata and real data, causing data integrity loss in the workflow.

Problem

A spread operator used to merge LLM‑generated JSON into an existing API envelope silently overwrites metadata and real data, causing silent corruption of workflow state.

Outcome

The spread collision caused silent data corruption in the workflow, leading to loss of provenance metadata and product information without any runtime errors.

Workflow

• Read product spreadsheet
• Build multilingual listing prompt per row
• Call chat‑completions endpoint
• Parse model JSON reply
• Merge parsed JSON into item using spread operator
• Write results to spreadsheet

Evidence

This one spread operator can corrupt your data without ever throwing an error.

Source

The signature of a spread collision is that everything looks fine. JSON.parse succeeded. The workflow ran green.

Source