Witness

Keep evidence, inference, and AI output distinct.

Open Source

Developer Tools

Artificial Intelligence

Visit WebsiteSee on Product HuntTwitter ⧉Github ⧉

Hi, I’m Sean, the builder behind NextEleven. I built Witness because a recorded observation, an interpretation, and a generated summary can all sound equally authoritative. I want people to be able to inspect the steps between them. Witness keeps three categories distinct: Observed, Inferred, and Generated. You can inspect record payloads, declared premises, methods, and falsifiers through a local dashboard or the APIs. It includes content identifiers and optional signatures, with explicit limits on what those mechanisms establish. Start with the disposable local demo in INSTALL.md. It creates two clearly labeled synthetic readings, an inference with its premises and a falsifier, and a synthetic narrative. It also checks that an inference referencing a missing record is rejected. You need Rust/Cargo and Python 3; no model API key is required. This is a prototype, not a truth detector or a production evidence service. A hash checks bytes; a signature can establish control of a signing key. Neither establishes that a measurement or conclusion is true. The current dashboard displays stored signature data but does not verify signatures on read. Humanity Grid is a related pre-alpha project. Its frozen local Proof A demonstrates a computational-artifact-to-Witness record path; it does not establish a scientific discovery or an operating volunteer network. We’re seeking independent reviewers, research collaborators, and funding partners for an independently designed and reviewed Proof B. The goal is a reproducible experiment with a public result, including failures or inconclusive findings. Try Witness: https://github.com/seanebones-la... Review and collaboration: https://github.com/seanebones-la... What would you need to inspect before trusting an evidence trail in your own workflow?