Every decision has a half-life. Your search bar doesn’t know that.

The wrong answer your team is still citing has a name. Every company is carrying it, and none of them are counting it.

Somewhere in your company, right now, someone is confidently citing a decision that was reversed eight months ago.

They are not careless. They did the responsible thing. They searched before they spoke. They found a document, and the document was real, and the document said what they said it said. Everything in that chain worked exactly as designed.

The decision had simply stopped being true, and nothing in the system was built to notice.

I want to be precise about what went wrong there, because the obvious diagnosis is wrong. It looks like a search problem. It isn’t. Search worked. It returned a genuine artifact, written by a real person, in the right project folder, with the right keywords in it. If anything, better search would have made this worse — it would have surfaced the dead decision faster and with a cleaner summary attached.

The failure is that a decision is not a document. A decision is a thing with a lifespan, and we store it in a format that has no concept of one.

Two clocks, one of which nobody keeps

Every decision you make carries two separate times inside it.

There is the time it was recorded — the meeting, the doc, the Slack thread, the timestamp on the file. This one we’re good at. It’s automatic. Every tool you own captures it for free.

And there is the time it was true — the window during which that decision actually governed behaviour. It opened when the team started acting on it. It closed when something replaced it.

We track the first clock obsessively and the second one not at all. So when someone searches, they get a perfect answer to the wrong question. They asked “what did we decide?” and the system answered “here is a document that was written.” Those are not the same question, and the gap between them is where every stale-answer disaster lives.

The March pricing decision and the June pricing decision both come back. Same confidence. Same clean formatting. Same authoritative tone. Nothing in either one knows the other exists. The reader has to reconstruct the relationship between them from timestamps and vibes — and they usually don’t, because both documents read as current, and the reader has eleven other things to do today.

Supersession debt

I’ve started calling the accumulated version of this supersession debt: the gap between what your organisation decided and what it can still prove it decided.

It behaves like technical debt in the ways that matter. It’s invisible on any dashboard. It compounds silently. It’s cheap to incur and expensive to service. And it gets paid off, eventually, by whoever happens to be standing there when it comes due — usually someone junior, usually in public, usually in a meeting where they said a confident sentence that turned out to be eighteen months out of date.

But it differs from technical debt in one uncomfortable way. Technical debt announces itself. The build slows down. The test suite flakes. Someone opens a ticket. You can feel a codebase rotting.

Supersession debt produces no signal at all until the moment it costs you something. The decisions don’t decay noisily. They just sit there, well-formatted and fully searchable, looking exactly as authoritative on the day they became wrong as they did on the day they were made.

Here’s the part I find genuinely interesting: the better your tooling gets, the less visible this becomes.

When institutional knowledge lived in one person’s head, everyone knew it was fragile. You could feel the risk. “Ask Priya, she’d know” is an unstable arrangement and everyone in the room understands it’s unstable. That anxiety was doing real work — it made teams write things down, over-communicate reversals, and double-check before acting on anything old.

Then we got good search. And the anxiety went away, because the interface stopped signalling uncertainty. A search bar that returns results feels like a system that knows things. It doesn’t distinguish between a document that is current and a document that merely exists. It has no way to. Absence of a “this was superseded” flag isn’t evidence that nothing superseded it — the flag was never a field in the first place.

That is the actual damage. Not that we forget more than we used to. That we now forget with the interface confidence of a system that remembers.

Four questions

Try these against your own team. Time yourself. The scoring is the point — it’s not how good your answers are, it’s how long they take.

1. Name the last significant decision your team reversed. Now find the moment of reversal. Not the new decision — the reversal. The artifact that says “we are no longer doing the thing we said we’d do, and here’s why.” Most teams have the new decision. Almost none have the reversal, because nobody’s job is to write the sentence that closes a door.

2. Pick a decision from a year ago. Does it know it was superseded? Go and look at the original doc. Is there anything on it pointing forward to what replaced it? Or does it still read as live? This is the one that tends to land — people go and look, and the doc is just sitting there, perfectly confident, wrong.

3. Who signed off? Not who was in the room — who signed. Attendance is not agreement. A recording proves someone was present while words were spoken. It doesn’t prove anyone accepted them. If your record of a decision is a transcript, you have evidence of a conversation, not evidence of a commitment.

4. If the person who made it left tomorrow, is the reasoning recoverable? Not the decision — the reasoning. The three options considered and the two rejected. Because the next time this comes up, someone will re-propose one of the rejected options, and without the reasoning there is no way to short-circuit that except by re-running the entire argument with less information than you had the first time.

Score is out of four. Under five minutes each.

Most teams I’ve put this to score one. Occasionally two. The ones who score three are almost always regulated — finance, pharma, aviation — where somebody external forced them to build this muscle and they’d never have grown it voluntarily.

I would very much like to hear if you score four. I’m not sure I believe it exists outside of compliance.

The thing nobody wants to hear

The reflexive fix is to buy something. Better search. Better summarisation. Point a model at the whole corpus and ask it what’s current.

It won’t work, and the reason is structural rather than technical. A model reading your documents can only tell you what your documents say. If no document records that March was replaced by June, no amount of intelligence applied at read time can recover that fact. It isn’t hiding in the corpus, waiting to be found by a sharper retrieval strategy. It was never written down. The information does not exist.

You can’t retrieve your way out of a recording problem. And every year we spend making retrieval better without touching the recording layer, the debt grows while the interface gets more reassuring — which is the worst possible combination, because it means the confidence curve and the accuracy curve are moving in opposite directions.

Which leaves the genuinely hard question, and it’s the one I’ll take up next: someone has to write the sentence.

Somebody, at some point, has to actually mark that March was superseded by June. That’s not a technical act. It’s a person, doing a small piece of unglamorous work, at a moment when the reversal feels obvious to everyone present and therefore not worth recording. The whole problem lives in that gap — between the moment a thing is obvious to the room and the moment it becomes invisible to everyone who wasn’t in it.

Whether that work survives contact with a real team, under real deadlines, with no one’s bonus attached to it, is the only question that actually matters here. I’ve been building against it for a while now, and I’ll write about what I’ve found — including the parts that aren’t working.

So: what’s the oldest wrong answer still circulating in your company?

Not a hypothetical. I mean the specific one — the thing people still say in meetings that stopped being true a while back, that everyone repeats because it’s written down somewhere and nobody’s got around to writing the sentence that kills it.

I’ll read every one of these.

Every decision has a half-life. Your search bar doesn’t know that. was originally published in Bootcamp on Medium, where people are continuing the conversation by highlighting and responding to this story.