Glossary

The operating vocabulary used across the proof ledger, defined once. Terms marked inline anywhere on this site link back to these definitions.

Core dependency

A technology whose removal would break the record. Distinguished from supporting (replaceable) and incidental (touched once), so the default view stays short.Technology: role, usage, and limits →

Evidence log

The per-run record of prompts, tool calls, diffs, and which gate approved. It is what makes a proof record checkable rather than asserted.Detailed post planned: How an evidence log makes a claim checkablenot written yet

Gate

A mandatory review a change must pass before it can proceed. A lane can never approve its own output.Detailed post planned: What it takes to trust an agent with a merge to mainnot written yet

Lane

One specialist agent with a scoped toolset, running as its own session. Lanes cover planning, implementation, review, security, QA, architecture, and release.Detailed post planned: A day in the life of running an agent workforcenot written yet

Proof record

One numbered entry in the ledger, stating what shipped, how it was verified, and where the evidence lives.The proof ledger →

Re-verification

Checking a record's evidence still holds. Records go stale on a clock and are flagged in the ledger until checked again.Detailed post planned: Re-verification on a fixed clocknot written yet

Task contract

The committed file that states what a lane was asked to do and what counts as done. Because it is committed, a run can be reproduced later.Detailed post planned: Task contracts: how I stop an agent from redefining its own jobnot written yet

Trust ratchet

Autonomy earned per task rather than granted per agent. A lane that has shipped cleanly still clears the same review on its next task.Detailed post planned: How the trust ratchet worksnot written yet