All field guidesPortfolio evidence governance · Prove and scale

A model price changed before the rate table: contain the estimate gap

A rate-change timeline bounds the 150-minute estimate gap, updates one verified source, and reconciles affected usage to billing.

Updated 2026-09-01 · 4 min read
Written for
Finance, procurement, or privacy lead
Article format
Price-change incident timeline
Take-away
rate-change control log

Open a controlled change at the first verified notice

At 09:00 UTC, capture the provider's official notice or pricing page with retrieval time, affected model IDs, effective time, billing unit, and old and new input and output rates. A social post or third-party calculator can alert the team, but it is not the configuration source.

Identify every local rate table and report that uses the model. Record version, owner, currency, and whether the value drives monitoring, an approval threshold, or an enforced rule. The same stale number has different consequences in a dashboard and in an automated decision.

Bound the 150-minute gap

If configuration changes at 11:30 UTC, export usage from 09:00 through 11:30 by model and separate input, cached input where applicable, and output units. Multiply each quantity by the difference between old and new verified rates. This is the maximum known estimate correction before billing adjustments.

Do not back-edit raw historical estimates. Preserve what operators saw at the time and add a corrected view with rate version and reason. Auditability requires both the original decision signal and the later reconciliation.

Control-log eventRequired valueEvidence
Notice received09:00 UTCofficial provider URL and capture
Effective timeprovider-stated timepricing notice
Old/new input rateenter verified ratesversioned configuration
Old/new output rateenter verified ratesversioned configuration
Local update11:30 UTCdeploy/change record
Gap usageunits by model and typeprovider/local exports
Invoice variancefinal amountmonth-end bridge

Limit decisions only where the gap is material

Estimate the worst credible dollar difference over the time until update. If it cannot change an approval or breach a budget tolerance, keep monitoring and update normally. If it can, pause only estimate-dependent automation for the affected model while deterministic request or token ceilings remain in place.

Avoid disabling customer AI because a display rate is stale. The operational risk is incorrect cost interpretation, not necessarily higher traffic or a broken request path. State which controls remain reliable during the gap.

Update one source and test every consumer

Change the canonical verified rate source once, increment its version, and propagate it through supported consumers. Test input-only, output-heavy, cached-input, and zero-usage examples where applicable. Confirm rounding and billing units rather than comparing only one familiar request.

Have a second person verify the model identifier and rate direction. A correct rate assigned to the wrong model can create a larger silent error than the original lag. Record the review and rollback value.

Reconcile after the invoice arrives

Compare corrected gap estimates with the provider invoice and assign differences to actual usage, discounts, credits, tax, rounding, or unresolved items. Use the invoice as authority for payment while retaining the control log to evaluate how the monitoring performed.

Close when every affected model has a verified version, consumers pass tests, the 150-minute window is quantified, and invoice variance is explained. The product pricing page is a next step for plan costs; provider model rates must continue to come from the provider's official source.

Use the rate-change control log from “A model price changed before the rate table: contain the estimate gap” on a real first installation. Download AI Cost Circuit Breaker for free, begin in Monitoring, and move to enforcement only after the expected signals and rollback are verified.

Primary sources checked

Next field guideRenewal and cancellation evidence for subscriptions owned by changing staff