Engineering Article 6 min read
Building ShipStory: Eliminating Hallucinations in Automated Release Marketing
Published on July 2026 by Solarc Product Engineering
How we engineered a deterministic verification layer on top of LLM outputs to guarantee zero hallucinated release features.
The Problem with Automated Release Notes
Generative AI models excel at prose generation, but in product release marketing, a single invented feature or wrong PR attribution damages customer trust and compliance audits. Traditional LLM wrappers hallucinate commit details.
Our Evidence Verification Layer
ShipStory enforces a two-pass architecture. First, claims are extracted from git commits. Second, an evidence verification parser cross-references every sentence against source pull requests, outputting a 0-100% Evidence Score.
Results in Production
Teams using ShipStory reduced announcement drafting time by 95% while maintaining a 100% compliance pass rate across release cycles.