# V19 W2C Grounding Prototype

## Core: Confidence FROM Observation Counts
c = w / (w + k)
where w = positive observation count, k = evidential horizon parameter

## Example: Maxworld NPC
| Encounters (w) | k=10 | Confidence |
|---|---|---|
| 1 | 10 | 0.091 |
| 5 | 10 | 0.333 |
| 10 | 10 | 0.500 |
| 25 | 10 | 0.714 |
| 50 | 10 | 0.833 |
| 100 | 10 | 0.909 |

## Frequency FROM Positive/Total Ratio
f = w_positive / w_total
NPC sees player 50 times, player attacks 15 times:
f(player_hostile) = 15/50 = 0.30
c(player_hostile) = 50/(50+10) = 0.833
Result: (stv 0.30 0.833) — confident the player is mostly NOT hostile

## Why This Answers Kevin M Challenge
- No hand-picked stv values
- Confidence EMERGES from encounter count
- Frequency EMERGES from positive/total ratio
- k is the only tunable parameter (evidential horizon)
- All other values are measured, not estimated

## Integration with v16-v18
- v16: g_eff uses confidence (now grounded)
- v17: staleness resets on re-observation (w increments)
- v18: reinforcement = new observation (w++)
- v19: WHERE the confidence number comes from

## Architecture Position
v14a(revision math) -> v15(discounting lever) -> v16(transport decay) -> v17(staleness metadata) -> v18(reinforcement signals) -> v19(grounded confidence from data)