PRISM Score
The PRISM score measures the quality of your AI-assisted coding sessions across five dimensions.
Formula
Section titled “Formula”PRISM = 0.25×PQ + 0.20×IE + 0.20×VD + 0.10×TU + 0.10×AFEach dimension has two metrics, each scored 0–10. The dimension score is the average of its two metrics.
The five dimensions
Section titled “The five dimensions”Prompt Quality (PQ) — 25%
Section titled “Prompt Quality (PQ) — 25%”How well-crafted your prompts are.
- Specificity (0–10): Does the prompt include file paths, function names, line numbers, error messages?
- Decomposition (0–10): Is the prompt focused on a single task, or bundling multiple requests?
High PQ example:
“Fix the null pointer in
src/auth/validate.ts:42— theuser.emailfield is undefined when the OAuth callback has no profile scope”
Low PQ example:
“fix the bug”
Iteration Efficiency (IE) — 20%
Section titled “Iteration Efficiency (IE) — 20%”How quickly you converge on a solution.
- Convergence (0–10): Do subsequent prompts build toward a solution, or circle back?
- Recovery (0–10): When things go wrong, do you recover efficiently or retry blindly?
Verification Discipline (VD) — 20%
Section titled “Verification Discipline (VD) — 20%”Whether you validate AI-generated code.
- Review (0–10): Do you review output before accepting?
- Validation (0–10): Do you run tests, check types, or verify behavior?
Tool Use (TU) — 10%
Section titled “Tool Use (TU) — 10%”How effectively you use available tools.
- Selection (0–10): Do you choose the right tool for the job?
- Context (0–10): Do you provide sufficient context for tool execution?
Advanced Features (AF) — 10%
Section titled “Advanced Features (AF) — 10%”Whether you leverage advanced AI capabilities.
- Delegation (0–10): Do you use subagents, skills, or parallel tool calls?
- Configuration (0–10): Do you use CLAUDE.md, hooks, plan mode, or MCP servers?
Score ranges
Section titled “Score ranges”| Range | Tier | Description |
|---|---|---|
| 9.0–10.0 | Elite | Optimal AI collaboration patterns |
| 7.0–8.9 | Expert | Strong habits with minor gaps |
| 5.0–6.9 | Proficient | Solid foundation, room for improvement |
| 3.0–4.9 | Practitioner | Developing skills, significant improvement possible |
| 0.0–2.9 | Novice | Early stage, high token waste likely |
Token impact by tier
Section titled “Token impact by tier”Higher PRISM scores correlate with lower token consumption:
| Tier | Token multiplier | Meaning |
|---|---|---|
| Elite | ~1x | Baseline efficient usage |
| Expert | ~1.3x | Slightly above minimum |
| Proficient | ~1.8x | Nearly double the tokens |
| Practitioner | ~2.5x | Significant waste |
| Novice | ~3–5x | 3–5x more tokens than necessary |
A Novice-tier developer might spend $50/month where an Elite-tier developer spends $10 for equivalent output.