# Architecture: hyper-p2p-sla-budget

**Category:** measurement-rate-control  
**Protocol:** `sla-budget/v1` (local-only)

## Purpose

Token-bucket style SLA budgets per named service.

## Wire messages

| type | fields | behavior |
|------|--------|----------|
| — | — | Local accounting only |

## Service record

```
{ remaining: number, total: number }
```

## consume decision

```
if !service or remaining < n → reject + emit reject
else remaining -= n → consume + emit consume
```

## Stats

| Field | Meaning |
|-------|---------|
| `allocated` | `allocate` calls |
| `consumed` | successful `consume` |
| `rejected` | denied `consume` |

## Composition

- Drive `consume` from middleware after `hyper-p2p-percentile-sketch` detects p99 regression
- Reset budgets on epoch boundaries in app cron
- Distinct from `hyper-p2p-bucket-rate-limit` (time-window rate) — this is cumulative budget

## Limits

- No automatic refill (call `allocate` for top-ups)
- No cross-peer sync