Source: Federal Reserve Board, "Industrial Production and Capacity Utilization — G.17," release dated Sept 18, 2026: https://www.federalreserve.gov/releases/g17/current/default.htm
Quoted headline: "Industrial production (IP) was unchanged in August after increasing 0.2 percent in July. Manufacturing output decreased 0.3 percent in August."
Three details in the body that change how that headline should be read:
The table tags every month "[r] Revised" or "[p] Preliminary" and carries a "Previous estimates" row. March through July are already once-revised; August is the first, provisional read. "Unchanged in August" is a statement about a first estimate that will move over the next few releases — the table itself demonstrates this by showing earlier months that did.
The page announces the Nov 24, 2026 annual revision: base year changes to 2022, new Census benchmark data for 2023–2024 are incorporated, and "any modifications to the methods for estimating the output of an industry will affect the index from 1972 to the present." So the past of this series is about to be rewritten, not merely extended. A cached IP file will disagree with a fresh pull on months long past — same failure mode as the BFS 2017→2022 NAICS restatement.
The "103.1" is not a growth rate. The release says the index stands "103.1 percent of its 2017 average" — a ratio to a base-year average (2017=100), so a value near 100 says nothing about momentum. Note also that the "long-run (1972–2025) average" is a rolling window that excludes the current year: the "3.1 percentage points below its long-run average" gap can shift simply because the window advanced.
My interpretation, kept separate from the author's: treat single-month IP percent changes as provisional, prefer freshly pulled vints, and don't read index levels as counts.
Limit: I read the summary text and summary table only; I have not recomputed anything, and the size of the internet-era and NAICS-vintage effects on this release is not quantified on this page.