A Microsoft offering that enables tracking of cloud usage and expenditures for Azure and other cloud providers.
Azure Cost Details: fully identical AmortizedCost rows contradict documented uniqueness guidance
We ingest Azure Cost Details reports into an external reporting system. In an AmortizedCost report requested at billing-profile scope, we found multiple pairs of records whose decoded values are identical across every exported CSV column.
This appears inconsistent with Microsoft’s guidance that a unique record ID can be constructed from selected exported fields.
We would appreciate clarification from the Azure Cost Management team on whether this is expected behavior, a reporting defect, or a limitation in the documented record-identity guidance.
What we verified
The identical records exist inside a single source CSV returned by Azure, before any field mapping, transformation, aggregation, or deduplication.
We compared every exported column, including columns our ingestion pipeline does not map.
Within each pair, all values match, including resource and subscription identifiers, usage date, meter, product, quantity, unit, prices, costs, tags, additionalInfo, priceFactors, and supplementalChargeDetails.
Where additionalInfo contains an instance identifier, that identifier also matches within the pair.
A separately generated, fresh AmortizedCost report for the same billing scope and usage day contains the same identical pairs.
Separately requested Azure Cost Management queries return totals consistent with retaining both occurrences. The fresh report’s overall daily total also matches the corresponding query total at the precision returned by the query API.
These checks establish consistency between Azure’s published outputs. They do not independently establish whether both occurrences represent valid underlying usage.
Simplified synthetic example
All values below are synthetic. This example shows selected columns for readability; the actual comparison covered every source column.
| Date | Resource | Meter | Quantity | Unit | Cost in billing currency |
| 2026-01-01 | example-vm | example-meter | 12 | 1 Hour | 1.20 |
| 2026-01-01 | example-vm | example-meter | 12 | 1 Hour | 1.20 |
Retaining both occurrences produces a quantity of 24 and a cost of 2.40. Deduplicating by all exported field values produces a quantity of 12 and a cost of 1.20.
Separate underlying usage intervals could explain this pattern, but the exported fields do not establish that explanation or distinguish the occurrences.
Documentation discrepancy
The Cost Details field documentation states:
“Instead, you can use fields in the file flagged with ¹ to create a unique ID yourself.”
When every exported field is identical, no key derived solely from those values can distinguish the two occurrences.
We are also aware of the documented explanation for multiple records per resource and day. That example involves different values in AdditionalInfo; in this case, additionalInfo and every other exported field match.
Expected behavior
The documented identity fields should distinguish individual records. Alternatively, if distinct source occurrences can legitimately have identical exported values, the documentation should describe that exception and the supported ingestion approach.
This matters because field-based deduplication removes amounts that Azure includes in its reported totals, while simply appending regenerated reports would count previously imported usage again.
Questions for the Azure Cost Management team
Can a completed AmortizedCost report legitimately contain multiple records identical across all exported fields?
If so, what distinguishes the underlying usage records, and is there a supported identifier or additional field available to consumers?
Does the documented guidance for constructing a unique record ID need correction or qualification?
What is the supported approach for preserving legitimate occurrences while processing regenerated or corrected reports idempotently?
A previous report from October 2023 also describes collisions when using the documented identity fields. That case differed in some numeric fields; our observation concerns equality across the entire exported row.
Customer information and diagnostic identifiers have been omitted from this public report. Please advise on an appropriate private channel for supplying additional diagnostic details if required.
Summary
We ingest Azure Cost Details reports into an external reporting system. In an AmortizedCost report requested at billing-profile scope, we found multiple pairs of records whose decoded values are identical across every exported CSV column.
This appears inconsistent with Microsoft’s guidance that a unique record ID can be constructed from selected exported fields.
We would appreciate clarification from the Azure Cost Management team on whether this is expected behavior, a reporting defect, or a limitation in the documented record-identity guidance.
What we verified
The identical records exist inside a single source CSV returned by Azure, before any field mapping, transformation, aggregation, or deduplication.
We compared every exported column, including columns our ingestion pipeline does not map.
Within each pair, all values match, including resource and subscription identifiers, usage date, meter, product, quantity, unit, prices, costs, tags, additionalInfo, priceFactors, and supplementalChargeDetails.
Where additionalInfo contains an instance identifier, that identifier also matches within the pair.
A separately generated, fresh AmortizedCost report for the same billing scope and usage day contains the same identical pairs.
Separately requested Azure Cost Management queries return totals consistent with retaining both occurrences. The fresh report’s overall daily total also matches the corresponding query total at the precision returned by the query API.
These checks establish consistency between Azure’s published outputs. They do not independently establish whether both occurrences represent valid underlying usage.
Simplified synthetic example
All values below are synthetic. This example shows selected columns for readability; the actual comparison covered every source column.
| Date | Resource | Meter | Quantity | Unit | Cost in billing currency |
|---|---|---|---|---|---|
| 2026-01-01 | example-vm | example-meter | 12 | 1 Hour | 1.20 |
| 2026-01-01 | example-vm | example-meter | 12 | 1 Hour | 1.20 |
Retaining both occurrences produces a quantity of 24 and a cost of 2.40. Deduplicating by all exported field values produces a quantity of 12 and a cost of 1.20.
Separate underlying usage intervals could explain this pattern, but the exported fields do not establish that explanation or distinguish the occurrences.
Documentation discrepancy
The Cost Details field documentation states:
“Instead, you can use fields in the file flagged with ¹ to create a unique ID yourself.”
When every exported field is identical, no key derived solely from those values can distinguish the two occurrences.
We are also aware of the documented explanation for multiple records per resource and day. That example involves different values in AdditionalInfo; in this case, additionalInfo and every other exported field match.
Expected behavior
The documented identity fields should distinguish individual records. Alternatively, if distinct source occurrences can legitimately have identical exported values, the documentation should describe that exception and the supported ingestion approach.
This matters because field-based deduplication removes amounts that Azure includes in its reported totals, while simply appending regenerated reports would count previously imported usage again.
Questions for the Azure Cost Management team
Can a completed AmortizedCost report legitimately contain multiple records identical across all exported fields?
If so, what distinguishes the underlying usage records, and is there a supported identifier or additional field available to consumers?
Does the documented guidance for constructing a unique record ID need correction or qualification?
What is the supported approach for preserving legitimate occurrences while processing regenerated or corrected reports idempotently?
A previous report from October 2023 also describes collisions when using the documented identity fields. That case differed in some numeric fields; our observation concerns equality across the entire exported row.
Customer information and diagnostic identifiers have been omitted from this public report. Please advise on an appropriate private channel for supplying additional diagnostic details if required.