# Evolution 81 — Cost Report

| item | value |
|---|---|
| real local model calls (main ledger recount) | 1475 |
| real local model calls (endurance ledger) | 114 |
| call count as originally reported (older-label rows omitted) | 1156 |
| real remote model calls | 0 |
| input tokens | 70743 |
| output tokens | 15818 |
| model cost | $0.000000 |
| provider billing | none (local CPU weights, no paid API) |
| host rate (estimate) | $0.360000 / hour |
| host cost (estimate) | $0.557705 |
| models used | qwen2.5-0.5b-instruct, qwen2.5-1.5b-instruct, smollm2-135m-instruct |

Local CPU inference has no marginal monetary cost. The host cost is an **estimate** at a published
VPS rate, not a bill. No paid API was called and no credit was purchased.

## Budget governance

Every mission had an authorized budget; agent, model and mission budgets are charged per call;
exhaustion raises and stops work. Concurrency and retries are bounded. This is enforced in
`resources.py` and tested in SEC-045..SEC-048.
