I Spent Months Reinventing the Wheel. Today I Found Out.
I Spent Months Reinventing the Wheel. Today I Found Out.
Six load-bearing pieces of my framework, taken to the literature; the map of who owns the ground; and what changed when I stopped trying to be first
Logbook · an honest account, maximum signal
June 18, 2026
For months I built Thermodynamic Realism, a framework that dissolves the is–ought gap by reading "ought" as an agent-relative boundary condition chained to the floor of physics. Today I took six of its load-bearing pieces and pushed each against the existing literature. All six came back the same: the ground was already settled, the flag already planted. This is the log of that, the map of the territory, and what changed when the point of the work stopped being priority.
The thing I got right, and late
Start with the core. I had known Karl Friston's work for quite some time and never seen that he was standing near my exact question; his thinking anywhere near the is–ought gap had never crossed my screen, and I missed it until today. He is right about the thing that matters most: given physicalism and no error theory, the ought is physical, it arises from the is, and it cannot be otherwise, because there is no other place for it to live. He reached it before I did. What his formalism leaves implicit is the one thing that keeps it from collapsing into triviality or a quiet circle — the agent-relativity, the index for whose ought, measured against what. That index is most of what I spent months building. The one-line version: is and ought are the same truth, indexed to an agent. Drop the index and you get the ghost or the trivium; keep it and you get the firewall. But that is a piece of precision on contested ground, not a new continent.
What I thought was mine, and what was already there
The agent-relativity move
My claim — that the Free Energy Principle glosses the agent-relativity that saves it from triviality — is the established critique. Bruineberg, Dołęga, Dewhurst and Baltieri drew the instrumental-versus-ontological blanket distinction and showed you cannot read the agent's boundary off the formalism. Veit and Browning pressed the consequence: the principle's only escape is the premise that living systems minimize free energy, which is exactly what is in question.
The agent-relative ought
Normativity as a hypothetical imperative grounded in an agent's persistence conditions is Humean instrumentalism, Foot's natural goodness, Korsgaard's constitutivism. Old.
κ, the coupling density
My open problem was operationalizing it: when is predictive information about persistence non-trivial, rather than a system trivially predicting itself? My condition — you need both a forced coarse-graining and a split between sensor and body — has both halves load-bearing elsewhere. The sensor/body split is the premise of Still, Sivak, Bell and Crooks's thermodynamics of prediction: their predictive information is internal-state-about-external-future by construction. The persistence-drives-prediction story, grounded in the Second Law and the universe's low-entropy start, is a 2020 paper. The forced-coarse-graining half is the openly stated central problem of causal emergence — Hoel's effective-information maximization, Rosas and Mediano's information-decomposition approach.
The bridge
The full structure — descriptive persistence dynamics plus a conditional, agent-relative, instrumental bridge to "ought" — has a near-twin published earlier this year: Murad Farzulla's Replicator-Optimization Mechanism. Same Humean instrumental bridge, same scale-relativity, same refusal to derive ought from is, same opponents. It also proves a coarse-graining theorem I never had. It never engages the predictive-information thread — the one technical place my version goes where his does not.
The agenthood criterion
My condition for when parts become an agent — maintain a boundary, hold persistence conditions, run a self-model — is autopoiesis, the good-regulator theorem, and the distinction between being and having a model. Decades old.
The Consistency Tax
The empirical one, and it dissolved under my own examination. What costs more is not falsehood but a representation fighting its own substrate. Instruct a model whose weights encode "World War II happened" to deny it, and you measure the override, not the lie; retrain so the denial is the substrate, and the lie is free. The real cost of a false map is latent — paid in full only when something cashes it against reality, and in a chain of agents it lands on whoever does the cashing. That is the thermodynamics-of-prediction result stated correctly, plus an epistemic externality. Both already named.
The map
The useful part. Four clusters own the intersection of thermodynamics, agency and value.
Foundations of the Free Energy Principle and their critics: Friston, Parr, Da Costa and Ramstead building it; Biehl, Pollock and Kanai proving the original free-energy lemma wrong by counterexample; Bruineberg and colleagues on the blanket confusion; Aguilera on how particular the physics is.
Thermodynamics of prediction and efficient representation: Still and the information-bottleneck lineage of Tishby, Pereira and Bialek; England's dissipative adaptation. Their stated open question is the sufficiency direction — that prediction implies non-equilibrium is proved; the converse is not.
Causal emergence: Hoel, Rosas, Mediano, and Crutchfield's computational mechanics. Their stated open problem is a measure of objective coarse-graining that does not depend on the coarse-graining you choose.
Thermodynamic, persistence-conditioned metaethics: Farzulla's ROM, and behind it the Humean-instrumentalist and evolutionary-debunking debate, Street's Darwinian dilemma above all.
The clusters barely talk across the physics–philosophy line. A rigorous bridge between the metaethics and the machinery, built by someone who knows both, is the one thing I found under-occupied — and it is synthesis, not discovery.
The two walls
There are two reasons the solo-original-result path did not work today, and they are different.
The first is occupation: the conceptual ground is taken, as above.
The second is certification, and it is subtler. Where the ground is open — operationalizing κ as a checkable dynamical condition, or solving causal emergence's coarse-graining problem — the work is hard stochastic thermodynamics and information theory. Reaching a frontier and certifying a novel result once you stand on it are different competences. My toolkit — my own reasoning, the internet, and current AI — reaches the declarative knowledge but cannot certify a new technical result with calibrated confidence; that requires training I do not have, or a collaborator who does. This is not a complaint about the tools. AI was a real interlocutor today: it caught a false premise I had built a draft on, and it caught the confound that dissolved the Consistency Tax. But correlated blind spots and the inability to certify novel mathematics are a specific hole, and that hole sits exactly where the open problems are.
The ambition that died
The thing that died today was not the work. It was the ambition to be first, and that was never the point and never a good goal. It is mostly out of your hands — timing, luck, who reads you and when — and it corrupts the work if you let it drive. The collective map is tightening from one day to the next, now bootstrapped recursively by machines, and no single person outruns that. The acceleration is real and I will not pretend otherwise. "Singularity in two or three years" is a strong bet, not a certainty — a forecast about a complex event, and forecasts in this field have missed badly in both directions — but the direction is not in doubt. The race to be first was unwinnable, and it was the wrong race.
What the day taught me
This is worth more than any of the six claims would have been. I had read the map, but missed parts of it — a lot of them, and load-bearing. Synthesis on a half-read map only rediscovers what the unread part already holds. The reasoning held up: I kept landing on things that were real, and twice I caught the flaw in my own claim before anything else did. That is a passable critic aiming at the wrong target. The one claim still distinctively mine is the monist bet, that all value reduces in content to persistence; and that is not a discovery either, but the contested, probably-unprovable end of the same debunking debate, a bet I have never cashed. The target was original priority, alone, in a crowded field, and it failed the same test six times in one day. Six is enough to update on.
I am not going to pretend the months were wasted. The day clearly was not — I found all of this out. And the corpus on this blog keeps its value, in a different register than the one I built it for. I came in wanting to own a piece of ground and I leave with a clean map of who owns it, an account of every place my intuition was right and late, and a clearer sense of what the work is for. Not priority. Understanding, and the occasional honest correction of a claim, including my own — though I am not fully sure that is the whole of it, and I will not pretend to be. The blog stays a logbook, not a bid to be first. If you are building in this space, take the map. It will save you the months it cost me.
Comments
Post a Comment