When Facts Freeze
Why many facts are just decisions with short time horizons, and what happens when we stop revisiting the conditions underneath.
I was thinking about this the other day while walking past a large boulder.
The kind that’s been sitting there longer than anyone remembers. If the person next to you says, “That’s never going to move,” they’re probably right from their point of view. It didn’t move yesterday. It won’t move tomorrow. As far as their lived experience goes, it’s fixed.
From a geological point of view, though, that statement is nonsense.
It will move. Gravity, erosion, freeze–thaw cycles, tectonics. The only thing missing is the time horizon. We’re just not scoping long enough to see it.
That’s how a lot of our “facts” work.
Facts are often just decisions with short time horizons
In most organisations, facts don’t emerge. They’re decided.
Someone looks at the data available at the time. They weigh it against constraints, incentives, and urgency. Then a conclusion is reached, written down, and quietly promoted from “current understanding” to “known fact”.
Once that happens, something subtle changes.
People stop revisiting it. They stop questioning the assumptions underneath. They stop asking whether the world it was true in still exists.
The fact freezes.
Not because it’s eternally true, but because nothing has forced it to thaw.
Software defaults are a great example
Think about a default setting in a piece of software.
At the moment it was chosen, it probably made sense. Performance constraints. User behaviour at the time. Security trade-offs that felt reasonable.
Then the system ships.
The default becomes “the way it works”.
Years later, people argue fiercely about it as if it were a law of nature. “That’s just how the system behaves.” Nobody remembers why it was chosen. Nobody re-runs the trade-offs. Changing it feels dangerous because so much now depends on it.
The fact didn’t survive because it was optimal.
It survived because nothing broke loudly enough to force a review.
Prices do this too
A price gets set. Maybe for a product, a service, or even labour.
At the time, it reflects costs, market conditions, and negotiation power. Then it gets reused. Quoted. Budgeted around.
Years later, everyone treats it as a fixed property of the world. Inflation changes. Supply chains shift. Value moves elsewhere.
But the price stays, defended with phrases like “that’s what it’s always been” or “the market expects it”.
The fact isn’t wrong.
It’s just operating on an outdated map.
Models freeze faster than rocks
AI systems accelerate this problem.
A model is trained on a snapshot of the world. It captures patterns that were real then. Those patterns get encoded into weights, scores, and confident outputs.
Once deployed, the model produces answers that sound timeless. Clean. Finished.
Unless you build in decay, review, and explicit time-scoping, those answers become frozen facts almost instantly. They get reused in dashboards, reports, and downstream automation long after the conditions that shaped them have shifted.
The boulder didn’t move.
We just stopped watching.
Why this matters more now
The faster systems move, the more dangerous frozen facts become.
When decisions took months and systems changed slowly, there was time for friction to surface problems. Now answers propagate in seconds. They get embedded before anyone has the chance to ask whether they’re still load-bearing.
We mistake stability for truth.
We mistake persistence for correctness.
We mistake “nothing broke” for “nothing changed”.
A better question than “is this true?”
I’ve found that the most useful question isn’t “is this fact true?”
It’s: over what time horizon is this true?
And closely behind it: what would have to change for this to stop being true?
If you can’t answer those, you don’t have a fact you can rely on. You have a convenient assumption that hasn’t been challenged yet.
The boulder will move
That boulder will move. Not on our schedule, not in our lifetime, and not in a way that matters for most day-to-day decisions.
But pretending it’s immovable because we’re not scoping long enough is a category error.
A lot of what we call “facts” work the same way.
They’re true at human scale, false at system scale, and dangerous when we forget which one we’re operating in.
As our tools get faster and our answers travel further, we don’t need fewer facts.
We need facts that remember time exists.
And systems that are willing to ask, occasionally, whether the rock has started to shift.