Problem
When hermes telemetry pricing drift --apply updates pricing.yaml with corrected prices, historical costs in the telemetry DB remain frozen with old (higher) prices. The dashboard mixes old-price and new-price costs, making budget tracking unreliable.
Real-world example
| Model |
Old price |
New price |
moonshotai/kimi-k2.6 |
$0.66/$3.41 |
$0.5472/$2.736 |
deepseek-v4-pro |
$0.435/$0.87 |
$0.348/$0.696 |
- Cron session with 4.9M tokens: dashboard shows $3.78, real cost ~$3.25 (−14%)
- Telegram session with 460K tokens: dashboard shows $0.30, real cost ~$0.19 (−37%)
Suggested behavior
After pricing drift --apply, recalculate cost_usd in llm_calls for rows where estimated=0 and provider_assumed=0 (they had explicit pricing at the old rate). Could be a --recalculate flag or a separate pricing recalculate subcommand.
Problem
When
hermes telemetry pricing drift --applyupdatespricing.yamlwith corrected prices, historical costs in the telemetry DB remain frozen with old (higher) prices. The dashboard mixes old-price and new-price costs, making budget tracking unreliable.Real-world example
moonshotai/kimi-k2.6deepseek-v4-proSuggested behavior
After
pricing drift --apply, recalculatecost_usdinllm_callsfor rows whereestimated=0andprovider_assumed=0(they had explicit pricing at the old rate). Could be a--recalculateflag or a separatepricing recalculatesubcommand.