Run page backend: full-text search over a run's traces, token records, prompts and reports; a dev-only editor flag for local viewing.

Rebuild the dashboard as an app over two SQLite databases (live and mock), with the source in settings.

Organizer controls on the board and the hill climb

Fixes from the researcher fresh-eyes test

Fixes from the participant fresh-eyes test

Submissions: deep links, search, per-suite deltas, what the model learned per task, real gate data

Submissions are the arena's main content

Dashboard on expected data: a projected mid-competition dataset and a fuller overview

dev/check_pipeline_configs.py: each challenge's config must load under the pipeline commit it pins

Deploy check also refuses on reconnecting relays and running PostTrain jobs; read-only dev server needs a token

One budget for the guard and the dashboard; build stamp; log links; deploy check

Dashboard styled to the posttrain.com brand

Dashboard: minimal overview with one chart and clickable runs

Dashboard: MiMo-style skeleton (overview, metrics, about) at /, Agent Collabs board at /board

Dev tools for the dashboard: read-only launcher, offline mock, screenshots