shadcn.io is not affiliated with official shadcn/ui
Prompt Performance Score Monitoring Block for Solid
Judge scores for three prompt versions over twelve weeks as lines told apart by dash, hollow diamonds for weeks under 30 samples, a Quality, Adherence or Tone toggle, a week cursor, and a table of each version's latest trusted score.
Installation
Overview
See whether a new prompt version is really better, and since when, with this score panel for Solid with Ark UI. It is Promptgrove, a prompt evaluation tracker (made up), showing three versions of one support prompt scored by a judge each week for twelve weeks. A hairline frame with a plus mark on each corner holds a line per version on a 70 to 100 scale, told apart by dash and weight: v2.1 solid, v2.0 dashed and v1.4 dotted. A week with fewer than 30 samples is a hollow diamond instead of a filled dot, and the reading says "too few samples to tell" rather than quoting a score. A Toggle Group switches the metric between quality, adherence and tone, and the header sentence compares the two newest versions, each on its latest week with enough samples. A hidden range over the plot is the cursor that reads one week for every version. Under it a table lists each version's latest trusted score, its week and its samples. Guides draw the frame and its bands; the entrance is CSS only. For teams that ship prompt changes and want evidence, not a feeling.
FAQ
How do I use my own scores?
Edit VERSIONS: each version has the index of its first week (start), the samples the judge scored in each week it ran, and one array of scores for each metric (quality, adherence, tone) of the same length. WEEKS holds the 12 labels and MIN is the sample floor. The lines, the markers, the header sentence, the cursor reading and the table all derive from them.
Why are some points hollow diamonds?
Because the week had fewer than 30 scored samples, too few to tell versions apart. Those weeks stay on the line so you can see the rollout, but they are never quoted as a score: the reading says "too few samples to tell" with the count, the figures show a dash, and the table uses the latest week that had enough.
How does the header sentence decide who leads?
It takes the two newest versions, finds each one's latest week with at least 30 samples, and subtracts the scores for the chosen metric. It says "leads" or "trails" and the gap in points. Switch to Tone to see v2.1 trail v2.0 while it leads on Quality.
How does the cursor work?
It is a native range named Week with a value text of the week's date. Arrow keys move it one week, Home and End jump to the ends. The line under the figures reads every version for that week, including the ones that did not run yet, and a polite status line speaks it after a short pause.