How the Karma Score works
Your Karma Score (0–100) is objective. It’s built only from validation actions that were observed or logged — never from feeling-based self-report. It answers one thing: did you verify the AI’s work before trusting it?
Design principles
Section titled “Design principles”- Action-based. Points come from real validations (tests, build, lint, type check) you actually ran — not from how much AI you used.
- Validation-dominated. ~90 of 100 points are validation; a soft prompt-clarity hint contributes at most 10.
- Traceable. Every point maps to a named rule, so the score is never a black box.
- Fair (no vacuous truth). A check only earns points if it actually ran — you can’t score “clean” by simply never testing.
The rules
Section titled “The rules”The table below is rendered directly from the extension’s engine — it is the exact rule set the extension uses, so these docs can’t fall out of sync with the code.
| Rule | Max points | What it rewards |
|---|---|---|
| Tests run | 25 | You ran a test command during the session — the core act of verifying AI output. |
| Tests passed — observed | 10 | A real, observed passing test exit code — never self-reported. Observed beats logged. |
| Build / type-check ran clean | 20 | A build or type check ran and did not fail — evidence the AI's change compiles / type-checks. |
| Lint ran clean | 15 | A linter ran and did not fail — catches issues the AI quietly introduced. |
| Test added/updated alongside code | 15 | You saved both a test file and non-test code — the change arrived with its test. |
| Change measured | 5 | A git diff summary was captured, so the size of the change is known. |
| Prompt hygiene hint | 10 | A soft, low-weight hint on prompt clarity — the only non-validation input, capped at 10%. |
| Total | 100 / 100 | The remaining points are reserved; validation dominates by design. |
Self-comparative trend
Section titled “Self-comparative trend”The headline number you see over time is a self-comparative EMA — you against your own rolling average, not a global leaderboard. With only a few sessions it stays a calm “forming” note rather than grading a sparse history.
Want to feel it? Try the Validation Dojo — it scores your choices with this exact engine.