Bit Ghost Security

Hardened Logic for an Intelligent Era

6 minute read

This is Part 2 of a three-part series about TolmiGround. Part 1 described what the system does. This part is the origin story — where the idea came from and why it took the shape it did. Part 3 places it inside the larger system I am building toward. TolmiGround is patent pending — U.S. Provisional Patent Application No. 64/181,356, filed October 21, 2026.

A Thought That Wouldn’t Leave

The idea behind TolmiGround started as something closer to a daydream than a design. Here is roughly how I framed it to myself:

It must be possible to have a database of every possible question and answer. Theoretically, if you had all the time and resources in the world, a small group of people would be able to put together a list of everything that exists — every word in the dictionary and every other important book of science and learning. Then take every word and make a list of how each combination of words relates to each other. Basically, make a list of every possible question and answer — but doing the research and vetting of all answers.

Say it out loud and it sounds absurd. Every possible question? The combinatorics alone are hostile. But the absurd version of the idea was never the point. The point was the realization hiding inside it: a huge amount of what we treat as “knowledge” is really just relationships between things we already have names for. Is a whale a fish? Does a dog have fur? Is a wheel part of a car? The words are already written down. The relationships between them are, in principle, enumerable. What is missing is not the raw material — it is the vetting.

So I stopped trying to imagine the brute-force version, where a few tireless people grind through every combination by hand, and started asking a better question: what would it take to build a system that does the breaking-down and the vetting for you, carefully, and never forgets what it has already proven?

From Brute Force to Pending Answers

The shift that made the idea workable was a change in posture. Instead of treating knowledge as a giant list to be completed, treat it as a field of pending answers — questions that are, to some degree, awaiting resolution.

That reframing does something quietly powerful. A question that is merely “unanswered” is a dead end. A question that is “pending” is a piece of work with a status, and work with a status can be prioritized, decomposed, revisited, and reused. When TolmiGround cannot answer something, it does not shrug — it splits the question into smaller ones, resolves what it can, and keeps the rest on the board for later. The map fills in the way a careful survey fills in, not the way a brute-force search grinds.

This is the “every possible question” idea made practical. The system pre-created a large body of simple questions directly out of the links already present in its reference sources, and gave each one a status — answered, pending, or contested. The dream of enumerating everything turned into something far more modest and far more useful: a structured backlog of questions, most already answerable from the sources, the rest waiting in line.

Logic Before Probability

There is an older version of this idea, one I had been circling for years before TolmiGround had a name. I wanted an AI that leaned on the mathematics of logic more than on probability. Something that starts from things regarded as true, tests each supposition against them, and then tests the suppositions behind those tests. In that framing, the “integrity” of an answer is just a measure of how thoroughly its foundations have been checked.

That instinct runs against the grain of how most modern AI works. The dominant approach is magnificently good at producing the most likely answer and largely indifferent to whether that answer is grounded. I wanted to invert the priority. Grounding first; likelihood a distant second. An answer should earn its place by what it stands on, not by how natural it sounds.

That is also where the name comes from. Stephen Toulmin spent his career describing how real arguments are actually supported — by grounds and warrants, by the connective tissue between a claim and the evidence under it — rather than by formal syllogism or by force of assertion. TolmiGround is Toulmin plus ground: an argument is only as strong as what it is grounded in. The whole system is an attempt to take that philosophy and make it mechanical.

The Part That Took the Longest: People

If TolmiGround only drew on dictionaries and taxonomies, it would be a tidy, closed thing — and a limited one. Reference sources have gaps, errors, and silences. Eventually you need people. And the moment you let people in, you reopen every problem that makes crowdsourced knowledge untrustworthy.

I spent a long time on this, because the obvious solutions are all traps. Route hard questions to “experts”? Now you have to decide who is an expert, and you have imported their blind spots and their incentives. Let everyone vote? Now popularity wins. Weight people by reputation? Reputation is claimed far more easily than it is earned.

The principle I settled on is that influence has to be earned, per topic, and continuously re-earned — never claimed, never global. In my own words from back then, the goal was to open it up

to other people of varying skill and knowledge to be able to participate, but in a way that forces them to build trust and to try and limit their impact to topics they have knowledge or interest in answering as correctly as humanly possible.

So in TolmiGround, anyone can join the topics they care about. Questions go to random participants within a topic, never routed to self-declared experts. A person’s weight is measured against hidden test questions whose answers the system has already proven from evidence — so trust is something the system observes, not something a person asserts. It grows slowly and has to keep being earned. Even the small group of vetted members who start with higher trust are tested and audited the same way as everyone else.

And people never get to overwrite the evidence. Their consensus is shown beside the evidence-based answer, in that separate score I described in Part 1, and when they disagree with the evidence the system treats it as a reason to check harder — not as a vote to be tallied.

Trust You Can Audit

One more thing had to be true before I could trust the thing myself: every claim it makes has to be reconstructible. Every change is written to a tamper-evident log, and any answer can be rebuilt from that log exactly. Any participant’s answers and reviews can be audited against proven results. If a reviewer approves a piece of evidence that later turns out to be wrong, that shows up, and their influence reflects it.

This is the part of the design that comes straight from my security background. I do not trust a system I cannot audit, and I am not going to ask anyone else to either. “Show your work” is not a nicety here — it is the whole architecture.

Where This Is Going

So that is the why. It began as a half-serious thought about writing down every question in the world, survived contact with reality by becoming a system of pending answers, and was shaped at every turn by a preference for grounded logic over fluent probability — and by a stubborn refusal to let popularity stand in for truth.

But I did not build TolmiGround only to answer “is a whale a fish.” In Part 3, I will explain what it is actually a piece of: a larger system I am developing that aims to behave, in its reasoning, in a manner closer to general intelligence — built on grounded evidence instead of prediction.

Join the Mission

If the idea of knowledge that is earned and auditable rather than generated and asserted resonates with you, reach out. TolmiGround is patent pending and still early, and I am glad to talk about it.

Hardened Logic for an Intelligent Era.

You May Also Enjoy