Skip to content
Home
Services
Work
Resources
AboutContactBook a Strategy Call
Tips & Guides

Receipts beat hype: how we log every AI output

If the site says an agent did something, the visitor can see what, how, and when. Here's the schema and the policy.

  • Tips & Guides
  • practitioner
  • Mar 28, 2026
  • 6 min read
  • By Alex Colon

The receipt schema

Every agent call writes a row: model ID, prompt hash, tool calls, input and output token counts, cache hit ratio, cost estimate, latency, timestamp, and the human reviewer if one signed off. The row lives in a Notion database we built for this purpose.

What we surface publicly

On the site, visitors see a compact receipt: model, tokens, and when. On client deliverables, they see the full record. No claim about AI on this site exists without a matching receipt.

What stays internal

Raw prompts, tool-call payloads, and any client data. Hashes go public so the receipt is verifiable; the underlying prompt stays private. The goal is auditable trust without turning every run into a spectacle.

Want this applied to your business?

Thirty-minute working call. We map the idea to your situation and leave you with a prioritized next step.