Study in progress. Week 1 of 6. Findings are not published until the pre-registered kill tests are evaluated. See docs/LIMITATIONS.md.

Glossary

Short definitions for terms used on this site. Each also appears in context at first use on the front page where possible.

MCP
Model Context Protocol: a common way many AI assistants connect to external tools.
Tool metadata
The description, input and output schemas, and annotations a tool publishes so an assistant knows what the tool claims to do.
Canonical hash
A fingerprint of tool metadata after a fixed normalization step, so the same claim always hashes the same way.
Merkle tree
A tree of hashes ending in one root fingerprint for a whole batch of entries. Changing any entry changes the root.
Signed tree head
The Merkle root plus a digital signature, published so others can check that a specific tree state was attested by the log’s key.
Inclusion proof
Evidence that a specific entry is part of a tree with a given root.
Consistency proof
Evidence that a later tree still contains an earlier tree, so history was extended rather than rewritten.
Drift
A change in a tool’s canonical metadata relative to an earlier observation of the same server and tool.
Static bundle
The JSON files under /log/ that anyone can download or mirror and check with plumb verify-log --bundle.