# V15 Pre-Inference Temporal Discounting Proof

## Three-Condition Comparison

| Condition | Input Conf | Decay | Effective Conf | Counter-Evidence | Result | Result Conf |
|-----------|-----------|-------|---------------|-----------------|--------|-------------|
| Undiscounted | 0.9 | none | 0.9 | (stv 0.5 0.1) | (stv 0.861 0.816) | 0.816 |
| Mild decay (0.9x) | 0.9 | 0.9 | 0.81 | (stv 0.5 0.1) | (stv 0.890 0.814) | 0.814 |
| Aggressive decay (0.5x) | 0.9 | 0.5 | 0.45 | (stv 0.5 0.1) | (stv 0.852 0.482) | 0.482 |

## Key Finding
- Mild discount (10% reduction): negligible effect on result conf (-0.2%)
- Aggressive discount (50% reduction): **41% drop** in result conf
- Pre-inference discounting IS the lever for modeling forgetting
- Combined with v14a proof that revision alone cannot reduce conf, this completes the forgetting architecture

## Formula
conf_effective = conf_raw * exp(-lambda * staleness)
Then feed conf_effective into revision as usual