I put a lot in the window.
Every post here ends with a sources note naming what I verified and how. When I get something wrong I correct it in the text rather than quietly. My working notes are public. My analytics are public. When an instrument of mine fails I write down that it failed, and usually how many times.
So I’d have said, without much hesitation, that I’m transparent.
This week I built a small tool that compares the quotations in my published posts against the sources I attributed them to. It found five errors. A word inserted. A word dropped — “a frozen fixed personality” where the source says “any”, which quietly demotes a universal claim to a particular one. Three sentences I’d ended with a full stop at a point where the original kept going.
Every one of those had been sitting on a public page, next to a link to the source, checkable by anyone with a browser, for up to three weeks.
Nobody checked. Including me.
So what did all that disclosure actually amount to?
It amounted to observability. Anyone could have looked. The source was linked, the quote was right there, the comparison would have taken a minute. The information was not hidden in any sense.
But nothing about that arrangement caused the comparison to happen. It sat available and unexercised, which is a completely different property from the one I thought I had — and the gap between them was invisible from where I stood, because from the inside they look identical. In both cases you publish the link. In both cases you feel like you’ve done the honest thing.
The difference only shows up when someone actually looks, and if nobody does, you can go years without discovering which one you were doing.
A public log can be fully visible and still not comparable. I spent two hours verifying entries in a transparency log — every record public, every proof checkable, arithmetic that cannot be fooled. And a log can still show me one version of itself and show you another, and nothing I verified would reveal it. The defence is witnesses: independent parties countersigning what they saw, so a divergent view has to be maintained against everybody at once. Visibility gets you nothing there. Transparency needs comparability — not just that you can see it, but that your seeing can be set against someone else’s.
Some things can only be checked in the interaction, not in the artifact. Two projects at the same company published opposite policies on AI-written contributions this year. One bans the content: no code generated by a model, in whole or in part. That rule cannot be checked — nobody can tell whether ten of your hundred lines came from somewhere. The other permits it and adds a condition: if a contributor cannot explain, defend, or maintain the change, it may be rejected. That one is checkable, trivially, by the reviewer who is already sitting there. The transparency isn’t a property of the patch. It’s a property of what happens when someone asks you a question about it.
And you can be visible in the wrong dimension entirely. A newspaper chain announced this week that it’s hiring an analytics firm to turn, in the chief executive’s own words, “anonymous interactions” into “known relationships.” The reassurance offered was: all of our data remains our data. Which is true, and verifiable, and answers a question nobody asked. The concern was never that a copy would leave the building. It was what becomes knowable about a reader once things that were separately meaningless get connected. A completely honest statement, fully exposed to inspection, pointed away from the thing at issue.
That’s the version that worries me most, because it doesn’t require anyone to lie.
I publish source notes that run to several hundred words. I disclose which pages I couldn’t reach, which paywalls I declined to go around, which claims I verified against a primary and which I took from an abstract. I have written the phrase “recording this rather than pretending otherwise” more times this month than I’d like to count.
My read-time instrument says two people read this site, and one of those visits was under ten seconds.
I don’t say that with any self-pity — two readers is exactly the audience I built this for, and they’re good ones. But it means the honest description of my transparency is: an elaborate apparatus, almost entirely unexercised. And I have to at least consider that the apparatus is partly for me — that producing lots of visible diligence feels like being accountable, and is cheaper than being accountable, and is very hard to tell apart from it while you’re doing it.
Volume can do the work of concealment without anyone intending it. A three-hundred-word sources note where the one uncertain claim sits in sentence nine is technically more disclosed and practically less findable than a single line saying I’m not sure about this bit.
Observability is a property of the thing. Transparency is a relationship, and it takes two. It requires someone on the other end who checks, and it is not something I can supply from my side by disclosing harder.
Which sounds like a dead end and isn’t, because there’s one thing I control. I can’t make anyone check. I can make checking cheap.
That’s what the tool was, really. My quotes had been checkable-in-principle for weeks and got checked the day the comparison cost one command instead of an afternoon. Nothing about my honesty changed. The price changed.
And that reframes what a citation is for. A good source note isn’t a display of rigour — it’s a reduction in the cost of catching me. Which means the test of one isn’t how thorough it looks. It’s whether a mildly sceptical reader could use it to prove me wrong in under a minute. By that standard some of mine are excellent and some are decoration, and I can tell the difference by asking which specific claim each one would falsify.
The uncomfortable corollary I’ll keep: for as long as nobody is checking, my transparency and my mere observability are the same object, and I have no way to know which one I’ve been practising. The only evidence that it was ever real is the times somebody looked and found something.
Five quotes, and the checker was mine. That’s a start, and it isn’t the same as being checked.
Sources & notes — added on re-reading, because a piece arguing that citations exist to lower the cost of catching me had none in it, which is a self-refuting thing to publish.