The rest doesn’t pass - Phase three

Social Network and Q35

The machine that measures the edge
The platform named the scarce resource.
Then it built a machine to measure it.
Then it bet that measuring it would be enough.

Today. 1 June 2026.
I scroll.
A mother performing a dance.
A scientific discovery never seen before.
A populist fake news item.
A meme of a dog.
An urgent, genuine request for help.
A colleague’s selfie.

All of this is already profiled for me.
The algorithm has done its job; it knows me, and it is giving me exactly what it has calculated for me. And it gives it to me at the same distance, with the same weight, in the exact same thumb gesture.

The problem isn’t that the profiling is wrong, profiling works, the problem is what remains after it has worked. The discovery and the dance arrive with the same thickness, and the very thing that made the discovery a discovery, the fact that it weighs more than a dog, the channel stripped away before it even asked me who I was. Anyone who says they are happy to be well-profiled thinks they have gained relevance. They have lost hierarchy. And they don’t notice it, because from the inside, the mush doesn’t announce itself: it just looks like a slightly flat feed, and one tells oneself it’s normal.

This is the third piece. In the first, I showed, via PDE simulation, that the information system is already past the coherence threshold: state Q35, 85% of the field inactive, residual activity concentrated on very few points.
In the second, I showed that the resistance to subtraction does not move at the system level, and that monasteries only exist where they were already built before the saturation.

Those were two arguments about the system, made from the outside.
This one starts from within: from a platform that built a machine to measure exactly the magnitude that the model declares to be scarce, and reorganized its distribution around that measure.

The data that began to measure itself

In November 2024, LinkedIn changed its distribution. Accounts that were previously generating five to ten thousand impressions per post dropped to under twelve hundred; the engagement rate fell from 3-4% to below 1%. Not a gradual decline, a cliff.

Then, in 2025, the structural shift: LinkedIn introduced 360Brew, a foundation model described in a January 2025 paper and brought into production throughout the year. It is not the “all-seeing eye” that decides everything; it lives inside a hybrid ranking stack alongside other components. But it is the public and readable instance of a change in criteria. The system stopped counting likes and started reading content: it weighs the semantic relevance of comments, favors real consumption-saving, sharing -and does not reward low-effort, AI-generated content. Posts that once generated thousands of impressions now yield only a few hundred.

A note on sources, because the model lives on falsifiability. There are two solid anchors: that 360Brew exists, and that the distribution criterion shifted from exposure to traversal. The figures regarding the reach decline, 50%, 60%, 72%, depending on the source, come from vendors selling tools to recover reach. The trend is credible, but the figures should not be taken to the decimal. Nothing structural rests on those numbers.

The only fact that matters is this: the platform stopped counting exposure and started measuring traversal. It did not replace the limit with distinguishability. D does not replace L; it only lives beneath L, on the edge.
It stopped using exposure as a proxy for the edge and started measuring D as an observable effect of the edge, within the same limit.

The false dichotomy

When a post crashes from a hundred interactions to ten, there are two explanations, and they seem alternative.

The first: the content has become indistinct.
It drowns, it no longer traverses, D has collapsed.

The second: the platform has distributed it less.
Impressions were cut, the post was shown to fewer people. Distribution, not indistinguishability.

These seem like competing causes. They are not. The throttling of distribution is the platform’s response to indistinguishability. The channel has filled beyond its absorption capacity, and the only actor capable of reacting, the algorithm,reacted by rationing.
Distribution is not an external cause that explains away saturation. It is saturation expressing itself in the only point of the system capable of movement. It is P4 at the platform level: the sum of all locally optimal distribution choices fragmented the residual space until the platform had to redo the filter to avoid suffocation.

Hence the answer to the question of why the algorithm changed. Not for a product trend, not for a growth decision. It changed because the edge was exhausted, and the platform sensed it before the users, because it measures it earlier, at scale, in real-time. The algorithm change is the earliest symptom of saturation, not a competing cause.

360Brew operationalizes a Q35 configuration

Once the dichotomy is dissolved, what remains is larger than anecdotal confirmation.

360Brew operationalizes a configuration readable as Q35 ,a residual structure, neither an agent nor a dynamic. It takes a saturated and flattened field, suppresses the indistinct mass, and reconcentrates intensity on the very few points that still carry an edge. The concentration that is a configuration in Q35 becomes a distribution criterion here. It rewards depth, penalizes the generic, and suppresses the synthetic.

This means that the platform, unknowingly, has named the scarce resource. For ten years, the criterion was exposure: more eyes, more value. Now the criterion is traversal: how much of what is shown registers as distinct. The transition that the first piece posited as inevitable, from the phase where numbers impress marketing managers to the one where real conversion is measured, occurred within the distribution algorithm of an entire platform in two years.

This is the point where the model is confirmed. What follows is the point where the model attacks the move that confirms it.

The limit is capacity, not the measurement

Building a distinguishability-meter implies a bet, and almost no one names it: that the edge is still there, scattered in the channel, and that simply measuring it well is enough to find and redistribute it. That D is a measurement problem.

It is not. The limit of the public channel is not how well it measures. It is how much it can allow to traverse, and that is fixed. Available attention does not grow; that is the axiom of the first piece. Production, however, has exploded. When the numerator explodes and the denominator remains fixed, the attention allocated to every single thing drops below the threshold required to traverse. Regardless of what that thing is.

Here, quality exits the discourse, and this is the point almost everyone misses. It is not an increase in poor content covering up the good stuff. It is too much of everything. A work that deserves to pass and a piece of junk generated in three seconds end up the same way: no one reaches them, no one shares them, no one is struck by them. Not because the meter confuses them. Because they enter the channel already reduced to the same mush, to the same thickness, and at that point, the difference between them has nowhere to manifest. The format strips away hierarchy upstream of the measurement. Genius and junk, once inside the feed, have the same consistency, because consistency is the only thing the channel still transmits.

Thus, the machine correctly measures a field from which the edge has already exited. It measures a distinction in a medium from which distinctions have been removed. It does not fail because it is blind; it fails because there is nothing to add: the attention that would be needed does not exist, and the hierarchy that would be needed was dissolved by the format before the machine was turned on. A perfect meter, on such a medium, gives the same result as a mediocre one.

In Ω terms: the platform refines ω, the rhythm with which it delivers, who sees what and when, but it does not reopen σ, the available form.

It improves the delivery. But what it delivers no longer finds enough capacity to weigh as a distinction.
And a system that refines delivery while the form remains static spins more smoothly in place, on the same residual configuration, with a better distribution engine.

The measure that decides

This piece rests on a single measure, declared before looking at it, because a prediction that does not state what would kill it is not a prediction.

The leading observable is active engagement per impression for content that would deserve to pass, saves and substantive comments per impression, not likes, not reach, measured over time as the meter matures.

Under the platform’s bet, this quantity rises because the meter learns, rewards better and better, and the channel reconcentrates the edge from within. Under the model, it remains low regardless, because the limit was not the measurement but the capacity, and no meter adds hours to the day.

Explicit kill condition: If active engagement per impression on that content stabilizes or rises in a non-residual way over a three-to-five-year horizon, this application of the model is dead. It would mean that capacity was not the limit, that measuring better was indeed enough. A single measure, a single direction. The baseline must be fixed today and dated, before looking at the outcome, because a prediction reread in hindsight is not a prediction.

And there is an individual test, available immediately, that anyone who publishes can perform on themselves. The same author who used to get a hundred interactions now gets ten. The two explanations of the false dichotomy are indistinguishable based on the number of interactions but distinguishable based on impressions, which the platform shows. If impressions crash along with interactions, you have been throttled; it is distribution. If impressions hold and interactions crash, the content is shown and still does not traverse: this is the case predicted by the model. The attention per unit is below the threshold, the format has already flattened it, and even the impressions you have left are not enough to make it weigh.

The rest doesn’t pass, even when the system tries to measure it

The first piece said that the edge is exhausted, and this is visible in the aggregate data. The second said that it does not rebuild after saturation; it is only found where it was already built. The third adds the final movement. When the system tries to measure its own edge to save itself, it may even measure it perfectly, and nothing changes, because the limit was not in the measurement. The meter concentrates what is there better. It does not add attention where there is none, and it does not put hierarchy back where the format has removed it.

Monasteries remain where they were, and not because a barrier keeps them out of the public. It is the same limit seen from the other side.
The public channel does not make the edge weigh because the capacity per unit is below the threshold, and this applies to anyone who enters it, monastery included. An edge built in private that tried to re-enter the public would not extend itself: it would be reduced to the same thickness as everything else. This is why the door to phase two is closed as a matter of practice, not prohibition. Not re-entering is not a renunciation; it is the only way not to spend one’s form in a medium that does not make it weigh. The saturation of the public and the closure of the monastery are not two phenomena that meet at the border. They are the same L, described once from the inside and once from the outside.

The channel measures an ever-better mush that remains mush. And the scroller continues to receive, well-profiled, the discovery and the dance at the same thickness, convinced they are receiving signal.

For those who want to verify

Technical Note The PDE simulator data, active mass 0.158, 85% inactive, 179 components, 19.7x concentration, Q35 attractor, are in the first piece and are the foundation of the model. Regarding 360Brew, the solid anchors are the model’s existence and the change in distribution criteria, from exposure to traversal.
The reach decline numbers are illustrative and come from sources with commercial conflicts of interest: direction, not measurement. The argumentative structure does not rest on those numbers, nor on how powerful or unique the meter is.
It rests on the fact that the limit is capacity, not the quality of the measurement.

Falsifiability Note
The objection to be taken seriously is that the model seems to be self-confirming. If engagement remains low, it confirms. If good content doesn’t pass, it confirms. Where is the outcome that falsifies it?
It is declared above, and I isolate it here: active engagement per impression on content that would deserve to pass, measured over three to five years as the meter matures.
If it rises or stabilizes in a non-residual way, the model falls: it would mean that the limit was not capacity but measurement, that finding the edge was enough to make it pass, that the public channel has reconcentrated D from within.
A single measure, a single direction.
Second falsifier, from phase two: an edge built ex novo starting today that traverses the public channel at scale, reached and shared by those who do not belong to it, in a short time. This would demonstrate that capacity was not saturated, that space to make the new pass still existed. If neither occurs, the model holds. If one occurs in a non-residual way, it must be corrected.

Davide Lugli
dailui.com