housecast
docs
Bring your cases. A person grades what comes back.
housecast is the grading half of an evaluation. You bring the cases, the profile, and the runner, and it puts each answer in front of a person with its target beside it, then records the grade. Grading stays apart from running, so a board can be regraded without being run again.
These pages are early. Several carry a settled structure and an explicit list of what has not been written yet, and they are mounted verbatim from the repository rather than rewritten for the web.
If you are still deciding whether any of that is your problem, the case for housecast is the better place to start. These pages assume you already decided.
Synced from housecast 0be7810, 2026-10-02. A copy, so the repository is the source of truth if the two disagree.
Getting started
Grading
- Grading The human half, and why it ships in an extra.
- Grading surfaces One set of rules across a terminal and a browser.
- Grading evidence The public and private halves of a single run.
- The deck housecast grade deck ROUNDS --run RUNDIR --out DECK builds
- The grading page housecast/grade/page/index.html renders a committed run and
- The grading page, payloads and delivery How grading-page.md gets data, and the test proving it needs
- The grading page palette Where the colour and type come from, and where they deviate
- The grading page, colour and type How grading-page.md treats evidence, verdicts, and scale
- Presenting housecast grade present serves a built deck to a room, taking
- Publishing to PyPI How a housecast-v tag becomes a release
- Non-scores A cell that is decided and is not a verdict, in three reasons
- CI How .forgejo/workflows/ci.yml gates main and pull requests
- MCP loop tasks What a task file under housecast/mcpeval/tasks/ is, and what
- The live room housecast room serve --subjects subjects.json --log
- The room's pages The three pages room.md serves from housecast/room/page/
- The room as a static site How the room pages ship without the room server serving them
- Commitments and prepared cases Every prompt carries a commitment (empty string when absent)
- Room load test just room-load-test holds a dev or rehearsal room under
- A model per subject, and the fallback Both are off until set, and with neither set every subject