Model Prediction Cost Ledger

How to use Model Prediction Cost Ledger

Different models and failed runs can follow different billing rules. Record the units and rates from the actual relevant billing source rather than assuming every request has one fixed cost.

1. Read the exact model’s billing rules and your observed usage records.
2. Enter a unique prediction ID, billed unit, checked rate and billed/not-billed/unknown state.
3. Resolve unknown rows before comparing the exact same record scope with an invoice.

Good to know

• Rates are manual inputs, not live vendor pricing. Rows can use different unit types only when each rate matches its own units and every charge uses the one declared currency. Nonlinear charges, minimums and unentered fees are excluded.
• Each field is limited to8,000 characters and each CSV to100 data records. Use the exact displayed header; quoted commas, doubled quotes and quoted newlines are supported. IDs and short task descriptions are limited to 200 characters. Numbers use nonnegative ordinary decimal notation with at most 9 fractional places, magnitude at most 999,999,999; no scientific notation. Whole-count limits can be stricter. Results above 1 quadrillion or nonzero below 0.000000000001 are refused.
• Independent deterministic ToolJovia workflow. It does not call an AI model, contact a vendor, fetch sources, send messages, save input or execute code. Editable outputs and downloads contain your data; review before sharing. A prepared plan does not verify customers, revenue or a vendor’s internal workflow.