How The Discovery Stack Emerged
Most frameworks arrive fully formed.
This one didn’t.
It emerged.
Slowly at first.
Then all at once.
It began with a simple observation:
Something about AI-mediated discovery felt different.
The language people were using didn’t quite fit what the systems appeared to be doing.
The dominant conversation focused on visibility.
How do we rank?
How do we optimise?
How do we get mentioned?
How do we become retrievable?
It felt familiar.
A continuation of search.
But the behaviour of the systems hinted at something deeper.
They weren’t simply finding information.
They were resolving toward answers.
GEO Doesn’t Go All The Way
One of the earliest intuitions was that retrieval wasn’t the prize.
Being present in the candidate set wasn’t the same as being chosen.
Two companies could be equally visible.
Equally retrievable.
Equally cited.
Yet one consistently emerged as the answer.
The question became:
What happens after retrieval?
That observation gave rise to the first important distinction:
Retrieval is not resolution.
Visibility mattered.
But it wasn’t sufficient.
Something else was happening.
The Shift From Lists To Answers
Traditional search presented options.
Users explored.
Compared.
Decided.
The journey remained visible.
AI systems increasingly compressed that process.
Intent.
↓
Interpretation.
↓
Resolution.
The list didn’t necessarily disappear.
But increasingly, it disappeared from the user’s experience.
The system performed the work of discovery on their behalf.
This shifted the strategic question.
No longer:
How do we appear?
But:
Why are some entities repeatedly chosen?
Defaults
Repeated observations suggested another pattern.
Certain answers became easier to reuse.
They required less explanation.
Less uncertainty.
Less deliberation.
Over time, systems appeared to stabilise around trusted pathways.
Defaults emerged.
Not because alternatives ceased to exist.
But because the cost of reconsidering them became unnecessary.
The insight was simple:
Reuse compounds.
Priors
The next question naturally followed.
What happens when defaults become sufficiently stable?
What if repeated resolution shapes expectation itself?
The answer appeared in a concept borrowed from statistics:
Priors.
The assumptions that exist before new evidence arrives.
The starting point through which fresh information is interpreted.
Suddenly, the picture became clearer.
Retrieval wasn’t operating inside an empty system.
The system already understood something about the world before the query arrived.
Pressure Improves The Theory
The framework didn’t emerge in isolation.
It emerged through critique.
Through testing.
Through disagreement.
Ideas that couldn’t survive scrutiny were discarded.
Others became sharper.
Questions evolved from:
Is this true?
To:
What mechanism would make it true?
Some of that pressure came from people.
Some of it came through dialogue with AI systems themselves.
The very systems the framework was attempting to describe became part of the process used to refine it.
Arguments were challenged.
Definitions tightened.
Mechanisms clarified.
Weak ideas fell away.
Stronger ones survived.
In that sense, the experiment began long before publication.
The loop had already started.
Observation.
↓
Hypothesis.
↓
Critique.
↓
Refinement.
↓
Compression.
↓
Release.
The framework wasn’t simply written about AI-mediated discovery.
It emerged through interaction with it.
The Discovery Stack
Eventually, the pieces aligned.
Discovery wasn’t one process.
It was multiple layers interacting simultaneously.
The Index.
The Context Window.
The Weights.
Resolution.
Each operating on different clocks.
Each contributing to the answer the user ultimately experiences.
The Discovery Stack wasn’t a prediction.
It was an attempt to describe the machinery already in motion.
The Experiment
Perhaps the strangest part of this journey is that the framework now applies to itself.
It exists as:
→ essays
→ conversations
→ critiques
→ responses
→ examples
→ summaries
Some people will encounter fragments.
Others will encounter the whole.
Models may eventually inherit some version of it.
The question is no longer:
Was it written?
The question is:
What survives?
If the framework is wrong, the corpus will forget it.
If it is useful, others will sharpen it.
Test it.
Challenge it.
Extend it.
Either way, the mechanism remains the same.
The world decides.
A Different Kind Of Ending
This isn’t the end of the story.
Frameworks evolve.
Models change.
New evidence emerges.
The map gets redrawn.
Perhaps the most satisfying part of all of this is that the framework makes no special exception for itself.
It predicts the conditions under which ideas survive.
Then subjects itself to those same conditions.
The internet becomes memory.
The models inherit it later.
You write the first draft of your prior.
The world ratifies or vetoes it.
And what survives synthesis becomes the thing future systems remember.
Truth isn’t what avoids scrutiny.
Truth is what survives it.
What happens next isn’t up to the writing anymore.
The experiment is running.