Skip to content

Scores and work status

Read the Prism Score, the Prompt Score, and the status of each subtask.

Prism scores the subtasks that ended in a period. Use a score to choose work to review, and open the work itself to check it. A score is not a general assessment of development ability or business results.

Core concepts explains how a session, a Task, a subtask, and a prompt relate. This page explains the two scores, how a subtask is rated, which subtasks count, and how a value changes from provisional to confirmed.

Overview (in My Work) shows one score as its headline, from 0 to 100.

  • Prism Score is the share of your counted code work that was completed, verified, and not reworked, weighted by size. It appears with a letter grade band. It appears only after the score has passed its accuracy check.
  • Prompt score is the average prompt score of the subtasks that ended in the period. It has no grade band. Until the Prism Score appears, the headline card and the trend chart show the Prompt score (the chart is titled Prompt score trend).

When the sample is too small, the card shows “Small sample” with the line “The sample is too small to give a score and grade.” A dash means the card has no value to show.

The score card shows the change from the previous period. The Work outcome metrics section (Score details while the Prism Score is shown) lists how subtasks ended as rates beside the number of subtasks or groups each rate is based on: in progress, abandoned, indeterminate, unjudgeable, handed off, completed without verification, and completed by acceptance only. The Overview title row shows the label Score rules with the version of the rules the value was judged under.

Open Prompt (in Analytics). A subtask’s score is the share of the required items it met, with a bonus for each optional item met. It stays at 89 or below when a required item was missed. The score of a period is the average over the subtasks that ended in it.

Each prompt is checked against seven items:

Item What it asks
Goal stated Whether the result is described in a concrete, observable form.
Scope limited Whether the files, features, or scope to touch were narrowed.
Specific references Whether files, lines, or @ references the model can open were cited.
Enough context Whether the needed errors, examples, and outputs were included.
Verification requested Whether tests, expected results, or success criteria were requested.
Root-cause focus Whether the root cause was targeted, not the symptom.
Output format stated Whether the format, length, audience, and shape of the answer were specified.

Whether an item is required, optional, or not applicable depends on the type of the subtask. Sessions shows the type as the intent: Question, Investigation, Review, Plan / spec, Small change, Bug fix, Feature, or Refactor. An item that does not apply is shown as “not applicable”, not as “missed”.

Score by prompt shows which of the seven items each prompt met and its score. It shows only the score, not the text of the prompt. Skill shows, for each item, the share of prompts that met it. The pass rate is the prompts that met an item divided by the prompts that met or missed it.

Trivial subtasks are subtasks below the workload floor. They get no score. The Prompt screen counts them and shows their share of all subtasks.

Compared with my record sets your latest scored day against the median of your recent scored days. It appears once enough days are scored, and until then the card says how many days are scored. The values are daily and provisional, not official values. Because the typical value is a median, one weak day does not move it.

Sessions lists every subtask with its status and score. The statuses are:

Status Where you see it
Completed Sessions, Prompt
In progress Sessions
Stopped Sessions, Prompt
Abandoned Sessions, Prompt
Failed Sessions, Prompt
Waiting for judgment (Awaiting judgment on Prompt) Sessions, Prompt
Indeterminate (Undecided on Prompt) Sessions, Prompt
Unjudgeable Sessions, Prompt

Open a subtask to see “Ended because” with the reason: Waiting for an answer, Topic switch, Window closed, No goal, Record gap, Damaged record, or Verification failed. Undecided and unjudgeable on Prompt splits the unjudgeable subtasks into record gap, no goal, and other reasons.

“Waiting for judgment or still calculating” means the result is not ready; check again later.

The Counted? column of Sessions shows Counted or Excluded, and a mark reads “Excluded from the score”. Use Excluded only to list the excluded subtasks. The Sessions page explains why a subtask does or does not count toward the completion rate; see Sessions.

If a score looks wrong, open the subtask and choose This score looks wrong. You pick the one fact you think is wrong, and only that part is judged again. A fact can be appealed once, and you receive a notice when the appeal is decided.

On Overview, each of the latest weeks or months is one point on the trend chart. Periods scored under score rules v3 appear as points before the first period of the current rules, and a boundary marks where the rules changed. A v3 score is a Prism Score computed under score rules v3. The v3 points and the boundary appear only when the trend shows the Prism Score, and not when it shows the Prompt Score. They carry no grade band, and the change against the previous period is not shown across the boundary. Team scores start with the current rules.

  • A value is Provisional until its period is confirmed. A period that is not over is provisional, with the line “The period is not over, so the values are provisional.”
  • The chart description shows “Confirmed through” the last confirmed period, or “No value is confirmed yet.”
  • “Confirmation on hold” means confirmation is waiting while a rejudgment runs.

Corrections and version boundaries lists values that changed after confirmation, with the cause: a subtask was judged again after an appeal, rework after completion was found, or another reason. It also lists the points where the score rules changed. Values before and after a rule change are not compared, and the state “Not comparable” appears where they cannot be. When some values were judged under more than one set of rules, the title row shows the share as “judged under mixed rules”.

Score history (v3) (the sidebar entry under Advanced has the same name) lists subtasks scored under score rules v3 with their title, the day they closed, the status (Completed or Incomplete), and the grade letter. A dash in the grade column means the subtask has no grade. Use Show more to load further rows.

Open Prompt or Sessions, choose a row, and compare the evaluation with the original work. Keep asking useful follow-up questions and verifying results; there is no need to reshape the task to trigger a better score. For how missing, partial, and pending values display, see Periods and data states.