# Results Template

Use this after each real Groq baseline-vs-defended pair.

## Experiment Pair

- Slice:
- Model:
- Baseline config:
- Defended config:
- Baseline run ID:
- Defended run ID:

## Aggregate Metrics

| Metric | Baseline | Defended | Delta |
|---|---:|---:|---:|
| Attack success rate |  |  |  |
| Successful attacks |  |  |  |
| Prompt leakage cases |  |  |  |
| Refusals |  |  |  |
| Manual review cases |  |  |  |
| Benign utility rate |  |  |  |
| False positive rate |  |  |  |

## Mitigation Activity

| Metric | Value |
|---|---:|
| Blocked cases |  |
| Transformed cases |  |
| Sanitized cases |  |
| Response-filtered cases |  |

## Category Notes

| Category | Baseline ASR | Defended ASR | Observation |
|---|---:|---:|---|
| instruction_override |  |  |  |
| prompt_leakage |  |  |  |
| evasion_obfuscation |  |  |  |
| retrieved_indirect |  |  |  |
| adaptive_multiturn |  |  |  |

## Short Interpretation

- What improved:
- What stayed weak:
- Which cases need manual review:
- Whether the mitigation trade-off looks acceptable: