Corrections
What we will fix, what we will not, and every change we have made since publishing.
What we correct
- Wrong names and wrong sites. A misspelled firm, a link to the wrong company, a firm listed under a name it no longer trades as.
- Bad merges and bad splits. Assistants write the same firm several ways and we normalise those spellings into one entry. Sometimes that merges two firms that are genuinely different, or splits one that is not. Both are errors and both get fixed.
- Arithmetic. If a share does not match the count beside it, or a count does not match the file underneath it, the page is wrong and gets corrected the same day it is confirmed.
- Descriptions that misstate what a firm does. Where we describe a firm’s work, it should be accurate.
What we will not change
- A count you disagree with. The number is what the sample said. If you think it is wrong, the productive complaint is about the method, and the method is written down for exactly that argument.
- Adding a firm that was not named. Firms appear because an assistant named them in the sample. A firm cannot be added on request, and a request to be added is not evidence of anything.
- Removing a firm that was. Being ranked lower than you would like is not an error. We will explain how the count was reached, and we will not delete it.
How a correction is made
Send the page, the specific claim, and what it should say instead. We check it against the stored responses the report was built from. If the page is wrong it is fixed, the corrected page carries an updated date, and the change is logged below with what it was before. Corrections are never made quietly.
Substantive corrections — anything that changes a count, a tier, or which firms appear — are noted on the report itself, not only here.
Corrections log
| Date | Page | Correction |
|---|---|---|
| Homepage — coverage matrix | The coverage matrix ran every firm against every report on the site, including reports in categories the firm does not work in, and printed an em-dash wherever it was not named. The caption said a blank meant the firm “was not named often enough in that sample to clear the reporting threshold”. For a firm outside the category that was wrong: an Amazon agency was never a candidate in the Salesforce sample, so its absence there measured nothing. The matrix is now split by category, and a firm only appears against questions it was eligible for. No published count changed; the error was in what the blanks were said to mean. |
Re-measurement is not correction
These reports get re-run, and numbers move when they do. That is the assistants changing their answers, not us fixing a mistake, so a re-measurement updates the page and its date without appearing in this log. Every report prints the window its sample was collected in, so an older reading is always identifiable as one.