hillclimb chart — hillclimb

Live hillclimb chart: best score so far vs tested candidates.

With no argument and several charts to show (more than one problem, or the same problem in more than one run), first a table of them — one row per problem worked in a run, newest activity first; enter opens that chart, esc comes back to the table. hillclimb watch reaches the same chart with c from its runs, searches and candidates tables.

One staircase across every search of the problem, every scored candidate a dot (bright where it set a new best, dim where it missed), plus optional problem-config baselines; an experiment gets one line per arm instead. Refreshes as candidates land. Keys: r=refresh, t=toggle improvement text, d=detail (every scored candidate as a mark, parent edges, accepted lineage bold), h=toggle holdout/validation, c=cost overlay, p=switch problem, esc=back to the table, q=quit.

hillclimb chart [OPTIONS] [SEARCH]

Arguments

argument	what it is
`SEARCH`	latest (default), <run-id>, or <run-id>/<search-id> optional

Options

option	what it does
`--detail`, `-d`	One search only, with its exploration tree drawn on the curve.
`--holdout`	Force the holdout view. Default: a holdout-scored problem opens on holdout (the split its reference baselines live on), others on validation; h toggles either way.
`--cost`	Overlay cumulative agent tokens and verifier CPU-minutes on right-hand axes — what the climb cost as it climbed.