The Sound of Crickets

There’s a specific sound a discovery makes when nobody is listening to it.

It’s not silence, exactly. The world doesn’t go quiet. Emails still arrive. The platform still works. The benchmarks still run. People still talk about chemistry, physics, drug discovery, materials, AI, simulation, and the future of science.

The conversation continues.

You’re just not in it.

That’s the sound of crickets... the field moving on without you, even though you’ve published results that, at least to you, seem like they should change the conversation.

Or at minimum start one.

I want to write about that gap because I’m in it now. And because more people in science pass through it than are willing to admit. Not the dramatic version. Not the Galileo-on-trial version, which is mostly mythology anyway. The ordinary version. The slow version. The one where you publish the work, send it out, follow up, follow up again, and discover that follow up again still produces no response.

Not rejection.

Not criticism.

Not “this is wrong because of X.”

Just nothing.

Over the past weeks, I’ve published benchmarks for FluxMateria, the discovery engine built on FLUX Theory. They’re not vague benchmarks. They include cross-checks against DFT, validation on drug-induced liver injury data, semiconductor mobility screening, ADMET pipelines, and materials-property prediction at scale.

I contacted academics. I contacted investors. I refined the outreach. I wrote the explanations. I posted the updates. I sent the follow-ups.

Che te devo dì (what should I say)… the response, almost without exception, has been: crickets.

So I’ve been asking myself what nothing actually means.

Thomas Kuhn called it the structure of scientific revolutions: the period between the emergence of a new framework and the moment when the field is ready to engage with it.

Science doesn’t usually reorganize itself the instant a better explanation appears. It absorbs slowly. It resists by default. Not always maliciously. Often just structurally. People are busy. Fields are conservative. Reviewers are overloaded. Inboxes are dead zones. Trust flows through institutions, not claims.

There’s a long middle between published and recognized.

That middle has a strange texture.

Wegener proposed continental drift in 1912. He had geological, paleontological, and climatological evidence. The fit of the continents was visible to anyone with a map. The geological establishment rejected him for decades. Plate tectonics, the framework that vindicated the core idea, wasn’t widely accepted until the 1960s.

Barry Marshall and Robin Warren argued in the 1980s that peptic ulcers were caused by Helicobacter pylori, not stress or spicy food. Marshall eventually drank the bacterium himself to prove the point. He and Warren won the Nobel Prize in 2005.

Ignaz Semmelweis observed in the 1840s that handwashing reduced patient mortality. He was dismissed, ridiculed, and destroyed before germ theory arrived to explain why he had been right.

These aren’t analogies for me. That comparison would be arrogant and premature.

They’re reference points for the pattern.

The pattern is real: evidence can exist before attention arrives. A result can be public and still not socially present. A framework can be written down and still not have entered the conversation.

The field engages on its own schedule, not yours.

Abstract claims about good benchmarks aren’t useful. So here’s the concrete version.

The benchmarks page collects the current results and methodology. The DFT cross-check compares FluxMateria predictions against traditional density functional theory. The DILI benchmark reports drug-induced liver injury prediction at AUROC 0.9597 on the comparable TDC binary task, alongside mechanism, exposure, dose-window, and confidence in the same call.

A few examples:

The semiconductor mobility atlas enumerated 4,662,588 candidate compositions in 199 seconds on a laptop: roughly 23,000 candidates per second. The engine recovered the InGaAs trajectory and III-V alloy families that the materials community has spent decades validating, without being told those families were interesting.

The unified ADMET pipeline profiled 245 FDA-approved drugs across 8 ADMET endpoints in 51 seconds in full mechanistic mode. That’s 1,960 mechanism-aware predictions across plasma protein binding, blood-brain barrier penetration, intestinal permeability, metabolic stability, hERG cardiotoxicity, DILI, CYP inhibition across five isoforms, and aqueous solubility. Three of the eight endpoints report #1 SOTA performance against the relevant public TDC and AqSolDB leaderboards.

The materials screening case study ran 5,008 property predictions across 313 materials and 16 properties in 13.5 seconds end-to-end. Median per-prediction runtime was 2.7 milliseconds. On family holdout, the reported error was 1.17% MAPE, substantially below AFLOW, JARVIS, and MatBench baselines on the same split.

The tungsten-cuprate discovery study ran an inverse-design pipeline from 12,800 candidates to 3 experiment-ready leads in under 24 hours. The resulting family is a tungsten-modified calcium-barium-copper oxide system with predicted critical temperatures above 160 K at ambient pressure, fluorine-free, at roughly $21/kg estimated material cost. The composition family is patent-pending.

Those are the claims.

They’re also the invitation.

If there’s a methodological error, I want to know. If the comparison is unfair, I want to know. If the benchmark protocol has a flaw, I want to know. If an independent replication fails, I want to know.

But that requires engagement.

That’s the problem. The work is now concrete enough to be attacked, and almost nobody has attacked it.

The reason these results matter isn’t just that they’re fast.

Chemistry and materials discovery are caught between two unsatisfying regimes.

Traditional first-principles methods like DFT are powerful, but expensive. They don’t scale comfortably to millions of candidates, especially when the question isn’t evaluate this known molecule carefully but search an enormous chemical or materials space before anyone knows where to look.

Machine-learning methods are faster, but they pay a different price. They’re statistical surrogates trained on prior examples, often on datasets generated by DFT or experiment. They can perform very well inside their training distribution. But discovery is, by definition, about the region outside what has already been seen.

That’s where the epistemic problem begins.

A black-box model can produce a confident prediction in a region where it has no real basis for confidence. The output looks like knowledge. Sometimes it is. Sometimes it’s interpolation wearing a lab coat.

FluxMateria is built on a different premise: deterministic physics first, statistical learning second or not at all. The claim is that geometry can evaluate molecular and materials structure directly, at something closer to ML speed, while preserving mechanistic auditability.

That’s the claim I’m asking people to test.

Not admire.

Not believe.

Test!

So what does the silence mean?

Silence is information. The question is what kind.

Possibility one: the work is wrong. The numbers may contain a mistake. The methodology may have a hidden dependency. The benchmark may be less comparable than I think. The physical interpretation may be overextended. Any of that is possible. If so, the right outcome is correction. Fast correction is better than slow self-deception.

Possibility two: the results are real, but the interpretation is incomplete. This is also possible. FLUX Theory may be capturing a deeper structure without yet describing it in the final language. A serious physicist or chemist might look at the results and say: the numerical pattern is real, but your ontology is wrong. Here is the deeper mechanism. That would be a conversation worth having.

Possibility three: the field hasn’t actually looked. This is the possibility I suspect is currently dominant. Not because scientists are hostile, but because modern attention is broken. Cold emails are filtered. LinkedIn messages vanish. Researchers are drowning in low-signal claims. Investors see too many decks. Everyone has learned, rationally, to ignore almost everything that arrives outside trusted channels.

That means silence isn’t necessarily a verdict.

It may just be the default state of attention.

The crickets don’t tell you whether the work is right.

They tell you that the work hasn’t yet become socially real.

Here’s the part I want to be careful about.

I believe the benchmark results are strong. I believe the theory is capturing something real. I believe the cross-domain compression is too coherent to dismiss casually.

But belief isn’t the same thing as closure.

FLUX Theory describes the underlying structure of physical reality in geometric terms. The framework claims to generate a wide range of physics from a compact kernel rather than from independent fitted parameters. In my own internal hierarchy, there’s a difference between the published numerical results, the derivations that produce them, the model extensions that apply them, and the interpretation of what the structure means.

That distinction matters.

The numbers can be checked. The derivations can be audited. The applications can be benchmarked. The interpretation can be debated.

Those aren’t the same activity.

The most productive critique would not be this sounds strange. It would be one of the following:

• This benchmark is invalid because of this specific leakage or comparison error.

• This derivation depends on an undeclared fitted parameter here.

• This result is real, but it can be produced by a simpler structure.

• This works on the published examples but fails on this blind test.

• The computational speed is real, but the physical interpretation is unnecessary.

Any of those would move the conversation forward.

What doesn’t move the conversation forward is silence.

Silence leaves everything suspended. The numbers stand uncontested. The interpretation stands unrefined. The field continues as before. Nothing moves.

So the decision is simple, even if the experience isn’t.

I’d be lying if I said the silence doesn’t get to me. It does. There are nights at my desk with another unanswered message in the outbox when the weight of all this sits heavier than usual.

But the question that pulled me into this as a teenager hasn’t released me.

The benchmarks keep coming out right. The geometry keeps holding. The work continues, whether anyone is watching or not.

Vabbè… Keep working.

More benchmarks. More case studies. More blind tests. More validation across more domains. Clearer documentation. Better methodology pages. More reproducible manifests. More places where someone can say, here is the exact point where this fails.

That’s the discipline now.

Not begging for attention. Not sending louder emails. Not turning every unanswered message into a referendum on the work. Not confusing outreach with science.

The work is the work.

The marketing is adjacent to it, not constitutive of it.

If the framework is wrong, the work should expose that. If the framework is incomplete, the work should refine it. If the framework is right, the work should make that harder and harder to ignore.

That’s the only sane path through the crickets.

If you work in computational chemistry, materials science, physics, drug discovery, or scientific ML, I’m not asking you to believe anything on the basis of this essay.

I’m asking for something more useful.

Look.

Pick one benchmark. Pick one methodology page. Pick one claim that shouldn’t be possible. Try to break it.

If it breaks, tell me where.

If it doesn’t, tell someone else.

Either response is more useful than silence.

The demos are live. The benchmarks are public. The methodology is published. The email is on the website.

I’ll respond to anyone who has done the minimum work of actually looking at the results. I won’t respond by trying to persuade through rhetoric. I’ll respond by showing the work.

The crickets don’t decide whether a result is true.

Time and scrutiny do.

There’s an old line, often attributed to Schopenhauer, that truth passes through three stages: first ridicule, then opposition, then acceptance as self-evident.

Maybe.

But I’m less interested in the stages than I used to be.

Truth doesn’t need a narrative arc. It doesn’t need my impatience. It doesn’t need my preferred timeline. If the physics is right, it’ll still be right after a week of silence, a year of silence, or a decade of silence. If it’s wrong, no amount of frustration will save it.

The benchmarks are still there.

The work is still there.

The crickets continue.

So does the work.

I’ll let you know how it turns out.

Stiamo a vedere (let’s wait and see).

About the Author

Roberto Campus is the creator of FLUX Theory and architect of FluxMateria. Born in Sardinia, raised in Rome, he’s spent over 35 years asking stubborn questions about how the universe works. He may be onto something profound, or he may need a better hobby. The benchmarks are the reason he’s still betting on the former.