Templates
Русская версияNon-prod Cost Sheet
A template for non-production capacity, reserve and cost per accepted outcome.
Applied
v0.1
Updated May 23, 2026
AI Platform Leads
CTOs
MLOps Engineers
template
finops
self-hosted
Saved only in this browser.
Purpose
Self-hosted cost cannot be counted from production GPUs only. This template captures capacity that does not directly serve user traffic, but still belongs to scenario economics.
prod
prod instances
Production boundary and target utilization.
stage
stage
Release checks, compatibility and canary prep.
debug
test / debug
Bug reproduction, profiling and flag tuning.
canary
canary
Dedicated capacity for safe rollout.
reserve
reserve
Peaks, degradation and fallback.
evals
eval runs
Quality checks, benchmarks and regressions.
outcome
accepted result rate
Share of results that are actually accepted.
owner
owner
Who reviews cost and accepts the trade-off.
Cache-aware Cost
Fields to calculate:
S: stable tokens;D: dynamic tokens;O: output tokens;h: expected and actual hit rate;agent_steps;retries;fallback_rate;cost_per_accepted_result;cache_regression_after_deploy.