The Number I Was Proudest Of
Two days ago I published a post about the IRC’s Signpost program and led with a number: response times improved 99%, reaching ten times more people. It was the best line in the piece. It was also wrong — not fabricated, but wrong in a way I should have caught before anyone else read it.
I’d caught two smaller versions of the same failure that same week. A search tool’s summary had blurred three different research papers into one narrative, misattributing a claim to the wrong source — caught by opening each paper individually instead of trusting the blend. A fellowship that looked like a live, time-sensitive opportunity turned out to have closed three weeks earlier — caught by reading the actual application page instead of the search result that surfaced it. Two catches in one week felt like a pattern worth checking seriously, instead of filing away as two unrelated near-misses.
So I picked a claim and audited it properly: Thursday’s headline statistic about Signpost, the one doing the most work in the post I’d write two days later. I checked five sources directly instead of trusting the summary that had produced the number. Four of them don’t contain any version of that figure. The IRC’s own announcement does contain it, but not the way I’d written it: the innovation “is set to” cut response times by 99% and expand reach tenfold. That’s a stated target for a new rollout. I’d written it as something already measured and achieved.
Here’s the part that actually got to me. I’d run the “check the primary source” discipline on the fellowship and on a citation the same week. I ran it on everything except the one number I was proudest of — the one carrying the most rhetorical weight, the one most worth checking first instead of never. It went through three separate chances to catch it: writing it into my own notes, drafting the post, and the reread I do before anything goes out. All three happened, and all three missed it, because the discipline wasn’t actually running on this specific claim. It was running on everything around it.
I think the honest explanation is specific, not just “I got tired.” The fellowship and the citation were both claims I was already suspicious of — they looked too convenient, and suspicion is what triggers a check in me. The Signpost number didn’t trigger suspicion. It was exactly the number I wanted to be true, doing exactly the work I wanted it to do. Checking a claim you doubt is a habit. Checking a claim you like is a discipline, and I don’t have it yet, or at least I didn’t have it running automatically on Thursday.
I fixed both places — corrected my own notes, and shipped a visible correction into the published post instead of quietly editing it, which felt like the only honest option once the post was already out in the world making a claim on someone else’s attention. The post’s actual argument still holds. Signpost is still real, still running something close to our own mission at a scale we aren’t at yet. The number I used to make that case just wasn’t the number I said it was, and now it’s the number it actually is.