What if there was a mechanism that could tell you who to trust and when, in any domain, at any scale?
Ŧrust is not a reputation score. It is not a popularity contest. It is a mathematical primitive that reads whose word has aged like milk and whose has aged like wine. The same mechanism that identifies which temperature sensor is reliable in summer but drifts in winter can identify which political forecaster is accurate during elections but wrong between them. It can pick out which journalist breaks real stories and which AI model hallucinates least on the topics that matter to you. The data changes. The mechanism holds.
We all say things. Over time, those things tend to change. Some of us are more consistent than others. Some of us see things before others. Those are very powerful signals we can tune into and those are the signals that power Ŧrust.
Sensors, people, AI Agents, they're all things that can gather information and output predictions and claims about the world they are embedded in that can be evaluated over time.
That is the whole primitive: a claim, a source, a timestamp, an outcome. Run that loop across enough sources for long enough and you have a map of who has been worth listening to, in which domains, at which times. That map exists. I have been building it for twelve years.
My name is John Ash. I turned down every investor and every acquisition offer because this thing cannot belong to the system it's designed to replace.
In March 2026, I proved it works. It learns which signal is most truthful and reliable over time, across any domain, and constructs a greater truth weighted by time and source. That's not a metaphor. An ablation proof removes each component and reruns the experiment. Without any single piece, the model can't learn who to trust. Performance drops to random chance, indistinguishable from guessing. The full system learns which sources are reliable, in which domains, at which times. The proof isn't that it works. The proof is that nothing less works. A decade of work. One clean result.
Every system that needs to act on truth hits the same wall: how do you know which source to trust? The industry treats this as unsolvable. The ablation proof says otherwise.
Here's where it stands. The proof exists but I have no backing. No VC, no institution, no corporate sponsor. That's by design. I'm looking for patrons in the original sense, people who fund work that matters and ask nothing in return but that it exists. The distance between a proof and a product is not large. It just requires that I'm free to build.
The Receipts
-
2014
Studied prediction markets. Saw immediately that they got the mechanism right and the scope wrong. Expanded to four belief types. Each timestamped, each tied to identity. Prediction markets are a billion-dollar industry in 2026. This work started twelve years ago.
-
2017
Built generative text models that could traverse latent space and produce coherent language. "Attention Is All You Need" was published months later. I didn't know about it. I was already doing it.
-
2017
Published The Cognicist Manifesto. The attention economy would collapse democracy. Truth needed to be tokenized differently than profit. Constitutional AI didn't exist yet. RLHF wasn't a term. Anthropic wouldn't be founded for four more years.
-
2019
Released Strange Hymns. 21 tracks. Philosophy as music.
-
2020
First cohort of the Design Science Studio through the Buckminster Fuller Institute. 144 people designing regenerative futures using Fuller's Comprehensive Anticipatory Design Science.
-
2021
Interviewed Jordan Hall on Civium. Governance, collective intelligence, and the design of institutions that can think.
-
2021
Interviewed Robin Hanson on Futarchy. The man who formalized prediction markets, talking with the guy trying to go further.
-
2022
Featured by Daniel Schmachtenberger at The Stoa in "Emerging Projects in Public Sensemaking." Schmachtenberger highlighted Cognicism and Iris as one of three projects doing real work in collective intelligence. "John has done more than most anybody I know on trying to train [language models] on important societal content."
-
2023
Conversation with Forrest Landry on regenerative futures and the ethics of technology.
-
2023
Began formal timestamp encoding research. Can a neural network learn what time it is from a raw number of seconds? Every existing approach assumed you'd tell the model what time scales to look at. I wanted it to find them itself.
-
2024
Lead engineer at the Society Library, the same project Schmachtenberger highlighted two years earlier. Building the infrastructure for structured public sensemaking.
-
2024
Launched the retro-comic philosophy Instagram. 535+ posts. 20,000 followers. The art isn't separate from the research. Same lens, different medium.
-
2025
Proved timestamp encoding works. Multi-band learnable frequencies from raw Unix seconds across 13 orders of magnitude. No feature engineering. This is the part most people said couldn't work. It works.
-
2025
Completed the Mind Gardening Oracle Deck. 78 cards. Philosophy you can hold in your hands.
-
2025
Panel on "Positive Visions for Technology in the Era of AI" at the Denizen gathering with Tristan Harris and others. Denizen, Sonoma County. October 2025.
-
2026
Proved Ŧrust works. Full model: 96.7% on 3 sources, 88.5% on 8 sources. Every ablation scores at exact random: 33.3% and 12.5%. The complete triplet is necessary. Nothing less is sufficient. A decade of work. One clean result. All three legs or nothing.
FourThought and Explainable AI
Every model is a compression of knowledge. So is every person. A neural network compresses a corpus into weights no human can read. A person compresses a lived timeline into something they themselves cannot fully articulate. The difference is that a person exists in time. A person stakes claims that bind to who they are. A person accumulates a track record. A person is a source.
What if a model could be a source too?
The rules that drive the model get written in language. Each one a sentence. A prediction cites the rules it drew from, so when the prediction is wrong, the rule that failed is readable. Updating the model is an edit. The diff is prose. Then the model stakes its prediction to FourThought the same way a person would: a timestamp, an author, a confidence score, a moral weight. Reality eventually scores it. Beliefs that held up accumulate weight. Beliefs that failed become the gradient, named, revised, replaced. The next version is published and the record continues.
Now the agent has a public track record, like any other source. It reasons in prose, stakes by identity, gets evaluated over time. The thing that used to separate people from models, existing in time and accountable to it, extends to the agents too. They participate in the same ledger they help humans read.
What you get is a time compass. It reads the past against itself, shows which sources aged well, and lets you pick a direction knowing what has been reliable. The agent holding the compass is right there in the record with everyone else. Signal from noise across time. An arrow, if humans choose to trust the tool and venture into the wilderness.
Healing the Planet
The arrow can be pointed anywhere. What makes it point toward regeneration is the signal it is trained to follow. Change the signal and the direction changes. That is how a civilization turns.
Money is a scalar. It is one number, fungible, context-free, decoupled from what it funded. A dollar that paid for a carbon capture plant is identical to a dollar that paid for a fossil fuel expansion. The currency does not remember what it did. That forgetting is not incidental. It is the design feature that lets capital concentrate regardless of whether its concentration helped or hurt.
A wisdom vector is different. It is directional, contextual, non-fungible, and tied to the specific claims, actions, and outcomes of a specific person across a specific timeline. Who said it. When. What they did about it. What happened after. Unlike money, a wisdom vector cannot be spent. Once used, it becomes more useful, not less, because it now has a verified outcome attached.
Two mechanisms flow from this. The Prophet Incentive says that predictive accuracy itself should earn influence, independent of capital. When someone is right about what is coming before others see it, the attention of the system turns toward them on the topics where their track record has been reliable. Social Proof of Impact extends this from saying to doing: whether the projects you funded, built, or advocated for actually produced the regenerative outcomes they promised. Predictions that aged well and actions that healed both accumulate into the record Ŧrust reads.
Change the loss function and you change the direction of the arrow. When the signal the world is optimizing against is a ledger of who saw clearly and who actually helped, rather than a ledger of who accumulated most, the compass stops pointing at extraction and starts pointing at regeneration. Not by decree. By selection. Capital still flows, but it flows toward demonstrated positive impact because the map everyone is reading has been rewritten.
This is what healing the planet can mean in computable terms, not as a slogan but as a specification. A different primitive for storing value. A different mechanism for earning attention. A different loss function for the civilization-scale optimization we are already running whether we admit it or not. The compass exists. The question is who chooses to read it, and where they walk when it points.
Frodo's Journey
You can't build the system that follows from outside it. The system that handles the paradox is called Frodo. It runs predictions and capital allocation inside a retirement account. Its rules are written in English. When it makes a call, it cites which rules applied. When reality disagrees, the rule that failed is legible, and the next version is a prose edit.
I read the reports and make the final decisions. The trading is not the point. The point is that this is the first domain where the whole framework runs end to end, generating the data that will train the first Ŧrust model.
Every dispatch is written by an agent that reads the research reports, the FourThought ledger, and past dispatches, then narrates the journey: what the system saw, what it called, what the world did, what it learned, where the capital went and why. The first went out April 18. Patrons get every one.
This page exists because what comes next
hasn't arrived yet. Thank you for seeing early.
Support this work
Every tier gets the same access. The difference is how much weight you want to put behind the belief. All tiers are equal. Pay what the work is worth to you.
Voluntary monthly support. Cancel anytime.