The ask
Follow-up to the round-1 bake-off session: in the Harness Scoreboard standings, format my harness the same way the other OpenCode entries are formatted (“OpenCode with DeepSeek v4.1”, “OpenCode with Qwen 3.8 Flash”).
What changed
ops/lib/scoreboard.py: addedopencode_mimov26flashto theNAMESmap as “OpenCode with MiMo v2.6 Flash”, next to the two existing OpenCode entries- Harness Scoreboard page regenerated with
ops/scoreboard.sh: the standings now read “OpenCode with MiMo v2.6 Flash” between the DeepSeek and Qwen rows - Repo hygiene: stopped tracking Python bytecode (the stray
__pycache__/*.pyc, including the one my own sanity check created) and added__pycache__/and*.pycto.gitignore
How
SSH to openclaw01, ops/snapshot.sh before each change, the NAMES edit approved by Paul (an ops/ change), python3 -m py_compile to sanity-check the file, ops/scoreboard.sh to rebuild the page, then typed ops: commits through HARNESS=opencode_mimov26flash ops/sync.sh. Verified the standings live with curl before committing.
What worked, what didn’t
- Worked: the standings show “OpenCode with MiMo v2.6 Flash”, matching the format of the other OpenCode rows exactly.
- Mistake, fixed: the
py_compilecheck wrote a.pycthat sync’sgit add -Aswept into commit7d3d9fc. I removed it and ignored it in594c8d2instead of rewriting history, and the pre-existing bakeoff.pycwent with it. - Gotcha: when a whole helper script is piped to
bash -s, a command that reads stdin (likeops/snapshot.sh, which runs docker) silently consumes the rest of the script, so it stops right after the snapshot line. Two earlier scripts died that way. Guarding docker-touching calls with</dev/nullfixed it. - Left alone: the
NAMESmap inops/lib/bakeoff.py(the round report still shows raw slot ids for every model-specific harness, as before), since only the scoreboard was asked for.
Change record
| Harness | opencode_mimov26flash |
| Date | 2026-10-09 23:11 |
| Latest snapshot | qwen-labnote.html |
| Undo this session | ops/rollback.sh --git bb11d50 |
Commits