Revisions & vintage

Published numbers are versioned. A source agency releases a preliminary figure and later revises it to a final one, so any honest index built on them must revise with them — logging what changed and when, never silently overwriting history. Vintage is which version of a number you are looking at: the release it came from and the date it carried.

Why it matters

If a number can change after you read it, you need to know which version you acted on. Treating every figure as a value plus a vintage is just good data practice — it keeps a past decision judgeable against what was actually known. The Cost Index follows a stated revision-and-correction policy: when an upstream series is revised, it re-vendors the new value with a fresh vintage and records the change rather than pretending the first number was always right.

Frequently asked

What does it mean that a published number gets revised?

Most economic numbers are released in stages. A source agency publishes a preliminary figure quickly, then revises it as fuller data arrives, until a final value settles. The number didn't change because anyone was wrong — it changed because the agency learned more. So a figure is never just a number; it is a number plus a date that says which release you are reading. Any honest index built on those sources has to revise alongside them rather than freeze the first print.

What is a vintage?

Vintage is which version of a number you are looking at — the release it came from and the date it carried. Last month's preliminary CPI and this month's revised CPI for the same period are two vintages of one figure. Recording the vintage is what lets you reconstruct what was known at the time instead of quietly replacing the old value with the new one, so a past decision can still be judged against the data that was actually on the table.

How does the Muntin Cost Index handle revisions?

The Cost Index follows a stated revision-and-correction policy. When an upstream series is revised, the next refresh re-vendors the new value with a fresh vintage date and the prior value stays recoverable in version control — the change is logged in an append-only revisions audit trail, never silently overwritten. A discovered error is corrected at the source and noted in the change log, and a discontinued series turns the ingredient absent with the reason rather than being frozen at a stale last value.

What is a data vintage?

A vintage is which version of a number you are looking at — the release it came from and the date it carried. Last month's preliminary figure and this month's revised figure for the same period are two vintages of one number. Recording it lets you reconstruct what was known at the time instead of quietly overwriting the old value with the new one.