Prompt
Inspect Prompt Score trends and the work included in a result.
Open Analytics → Prompt to review the Prompt Score over time and investigate the subtasks and prompts behind it. For the metric’s meaning, use Scores and work status.
Review your score
Section titled “Review your score”- Choose the date range with the range control; the default is the last 30 days.
- Read Prompt Score and the number of subtasks scored in the period. A small sample can change substantially when one result changes.
- Inspect Prompt Score trend (one bar per day) and Item rates over time, which shows how often each of the seven items was met and marks the weakest one.
- Compare your latest scored day with your usual level under Compared with my record. It is a provisional daily value, not an official one, and it appears once enough days are scored.
- Read the Subtasks table for each subtask’s status, score, elapsed time, turns, changed lines, end reason, and items. Under Score by prompt, read which of the seven items each prompt met.
A subtask’s score is the share of the required items it met, with a bonus for each optional item met, and it stays at 89 or below when a required item was missed. Which items are required depends on the type of the subtask. The seven items are goal, scope, reference, context, verification request, root cause, and output format.
A prompt that belongs to no subtask shows which of the seven items it met, and its score cell is empty. Such prompts are not part of Prompt Score or of the item pass rates. A prompt whose items are not evaluated yet appears in the list without items, and a prompt whose content cannot be shown appears as a line that says why.
The summary also shows Trivial subtasks, which are below the workload floor and get no score; Undecided and unjudgeable subtasks, split by reason (a gap in the record, no goal, or another reason); and Cost per completion, which is a partial sum while some requests have no price. The cost is not a measure of result quality.
Understand comparison and inclusion
Section titled “Understand comparison and inclusion”Compared with my record compares you only with your own recent days. The typical value is a median, so one weak day does not move it. Before enough days are scored the card says how many are scored and how many are needed.
A subtask’s status is shown on its own (Completed, Stopped, Abandoned, Failed, Awaiting judgment, Undecided, Unjudgeable). Whether it counts toward the completion rate is a separate decision, shown as Counted or Excluded in Sessions.
If a score looks wrong, open the task in Sessions and use This score looks wrong. At the bottom of this page, Coaching shows the missed items for each subtask; see Advisor and summaries for reviewing suggestions.