Prompt Score

Prompt Score is Prism’s headline number. It’s the average grade of your prompts, on a 0–100 scale with a letter grade, and it answers one question: how well, on average, are you directing the AI? A prompt earns a higher grade when it gives the model a clear goal, the context it needs, and a way to check its own work — so your Prompt Score rises as your prompting gets sharper, not as you send more prompts.

In the sidebar this page is Prompt, under Analytics. It’s where you watch that number over time and see what’s moving it.

The four cards

Across the top of the page:

• Prompt Score — the headline grade and 0–100 number for the window you’re viewing. Read it as a running average: one weak prompt barely moves it, while a run of sharp ones pulls it up.
• Trivial sub-sessions — how many stretches of work were too small to score (a quick question, a one-line tweak). Prism sets these aside so they neither pad nor drag your score; a high count just means much of your day was small stuff. See Trivia for the details.
• Merges — how many sub-sessions Prism joined because they were really one goal: closed and reopened within about ten minutes, touching mostly the same files (a /clear mid-task is one common cause). This keeps a single task from being counted as several.
• Crush Weight — the median number of tokens it took to finish one sub-session. Watch it fall over time: it means you’re reaching the same outcomes for less.

Below the cards sit three things:

• Prompt Score over time — a trend graph of how the headline score moved from period to period.
• Where you stand — a comparison card placing you among your teammates. Along with a by-habit view, it carries a Completion rate comparison (crushed substantive sub-sessions ÷ substantive sub-sessions), showing the teammate count, team average, middle-half range, and your percentile. It never shows names or individual score rows, and it falls back to your own recent history when too few teammates have scored work.
• The table of sub-sessions behind the score — click any row to see whether it counted toward your Completion rate (Counted or Excluded), and why.

Prompt Score and Completion rate

Prism keeps two numbers side by side because they measure different things:

• Prompt Score grades how you prompt — clarity, context, verification — turn by turn.
• Completion rate measures how you finish — the share of your sub-sessions that were crushed, meaning they reached their goal and weren’t redone afterward.

You can prompt beautifully and still not finish (Prompt Score high, Completion rate low), or grind out results with messy prompts (the reverse). Read together, they tell you which half to work on. Completion rate has no card of its own here; each row of the sub-session table is marked Counted or Excluded, and the rate itself appears in your Report and the Team Console.

Raising your Completion rate

A sub-session is crushed when it reaches its goal and nothing later undoes it. The judge is deliberately strict about what counts as “finished,” so a few habits help it recognize your wins:

• Say when you’re done, and how you know. A bug fix counts when the fix is applied and verified; a feature counts when it’s built and accepted (or you note that tests come later); a refactor counts when you show behavior didn’t change.
• Don’t leave a goal half-stated. If the model can’t tell what “done” looks like, the sub-session reads as incomplete.

For the full picture of how a sub-session is judged, see how scoring works.

Going deeper

Page	What it adds
Speed	How much finished work you crush per day, week, or month
Token Usage	Whether your finished work is getting leaner
Skill	How consistently you apply the habits behind a good score