The AI didn’t invent my client update. It promoted it.

Description and use

Receipts automatically generates weekly client updates, verifies each sentence against the cited commit, and flags unsupported statements, reducing manual effort and improving accuracy.

Problem

Reconstructing a weekly client update from dozens of commit messages and diffs is time‑consuming and error‑prone.

Outcome

Receipts produced 36 bullets across two prompt variants; 13 were flagged for unsupported commits, 4 for other errors, and 17 were correct, demonstrating effective detection of misattribution and hallucination.

Workflow

• Run the CLI command to generate a report for the target repository
• Review the six‑bullet output
• Inspect flagged sentences and their receipts
• Adjust the prompt or commit history if necessary

Evidence

Receipts reads a week of commits with a local model and writes the update. Then it does the part I actually care about: it checks every sentence against the commit it cites, and tells me which sentences the commit does not support.

Source

The demo is a saved run baked into a static page, because the model is 3.4GB and GitHub Pages does not have a CPU.

Source

Five real outputs from the saved runs. Left is what the model cited; right is what it said.

Source