Guide — the product documentation
Written for someone using Everdict, not maintaining it. This is the tree the public docs site
publishes as /docs; the design records under ../architecture/ are a separate,
maintainer-facing section. See ../architecture/docs-site.md for how
the two map onto the site.
Get started
- Section index
- What is Everdict — what it does, what it refuses to do, and who it is for
- Quickstart — the whole stack on your machine in one command
- Your first scorecard — dataset × harness → a verdict you can defend
- Bring your own agent — the on-ramp per agent kind, ending at a registered harness
- Bundles — a whole evaluation setup as one document you apply
- Connect an agent — MCP, API keys, and the Claude Code plugin
Core concepts
- Overview — the seven nouns, and how one run flows through them
- Run · Harness · Dataset · Grader & Judge · Scorecard · Verdict · Workspace · Runtime
Your workspace
- Section index
- Workspace agents — the agent that runs inside Everdict, and how it wakes itself
- What the agent knows — the context a turn starts with, and inherited knowledge
- The workspace filesystem — one file tree per workspace, with attributed revisions
- Environments —
repo·prompt·browser·os-use, and how to choose - Secrets — credentials by name, and the two kinds treated differently
- Image registry
- Browser profiles — a captured session so cases do not start at a login wall — publishing images and keeping their provenance
Integrations
- Section index
- MCP — drive Everdict from any agent
- Claude Code plugin — two commands, tools plus domain context
- Running Codex
- Desktop app — a runner that lives on your machine — Codex as the agent under test, via a declarative harness
Operate
- Section index
- Track the work — issues that close citing the scorecard that proved it
- Schedules & monitoring — a scorecard on a cron, and durable resume
- Analysis & views — save the question, not the answer
- Durability & Temporal — resume instead of restart, and when you need it
- Products & releases — scores on the version axis, and a gated ship
- Budgets & cost — what evaluation actually spends
- Notifications — being told instead of checking
Self-hosting
- Overview — deployment shapes, what each one needs, and what to decide first