Confidence, Evidence and Verification

Understand provenance categories and attribution confidence.

Confidence indicates how strongly available data supports a conclusion, not a guarantee that a forecast or recommendation will materialize.

Evidence families

Observed billing: cost facts from a validated export and an authoritative published period. Imported source: records supplied from external legacy systems or spreadsheets, with actual/estimated qualification. AWS recommendation: opportunity and supporting context produced by a native service at an observation time. User action:an authorized actor's decision, implementation confirmation or verification record.

Value claims

An AWS recommendation's projected monthly impact is potential value. A user confirming a change is an implementation claim. Customer-attested value depends on human evidence. Independently system-measured value needs a comparable baseline and after window, consistent measure/currency, and attribution that avoids confounding changes.

Incomplete evidence

Missing sources, weak dependency data, absent cost coverage or inconsistent historical labels should produce unknown, review or unverified states — not optimistic default values.

Documentation accuracy rule

Public documentation may describe a control as part of the product only when the deployed release behaves that way. If an integration is code-complete but not live validated, say so on the relevant page rather than inventing external approval or functionality.