Who Owns the Algorithm?

Earlier this year, a method for routing vehicles beat the best published results on a problem class the logistics industry has been attacking for more than half a century. The margin over the previous state of the art was, by the account of the people who run the benchmark, very significant. The method did not come out of DeepMind, or Amazon's routing research, or a university optimization group with a corporate sponsor. It arrived through the submission queue of a crypto protocol, where it was published for anyone on earth to read, and the people responsible were paid for it in a token that currently changes hands for about eighty cents.

I want to ask a rude question about that algorithm. Who owns it?

The question is rude because the obvious answers are all wrong. The inventors do not own it, exactly; they assigned the invention to a foundation as the condition of a reward. The foundation owns a patent application, perhaps, in some jurisdictions, if the examiners cooperate. The token holders own nothing in the legal sense, and the protocol's own materials have always been unusually candid about that. And yet the entire design exists to make the answer something other than "no one." The Innovation Game, the protocol in question, has spent two years building what its foundation believes is a machine for owning open algorithms, and it has now published the document that explains how the machine is supposed to work. The document is called, without much poetry, Capturing the Value created by The Innovation Game's Algorithms. I read it in August. It is the most serious answer crypto has produced to the question this blog keeps asking, and it deserves a serious reading.

The exception

In July I published a long reckoning with my own portfolio, and the question behind it was accounting all the way down. What do I actually own? Governance influence over assets controlled elsewhere did not count. A good story did not count. If a project's success can happen somewhere else while the token holder receives only advisory influence, then the token's price is a bet on future buyers rather than a claim on the thing being built. By that test, most of what I had held failed. DEUS gave holders votes over assets legally controlled elsewhere. Bittensor's subnet tokens were emissions wrapped in a story about AI. The essay swept the altcoin market into a few cold categories and walked away toward bearer assets.

One experiment was deliberately left out of the sweep.

TIG was the exception, and the reason was the shape of its ambition. The protocol runs a market for algorithmic invention: innovators submit methods, benchmarkers run them against hard computational problems, adoption is measured, rewards follow adoption. That mechanism has now produced at least one result, the vehicle-routing method, that the outside world would have to call real. But the interesting claim was never the mechanism. The interesting claim was that the protocol could accumulate intellectual property around the inventions its market induced, keep them open for research, charge commercial users for the privilege of staying closed, and route the proceeds back through the token economy. If that worked, the token would be backed by something like rent on a growing portfolio of useful algorithms, which is a different animal from the fee revenue of a commoditized rail.

At the time, that claim existed as fragments: a whitepaper section, a licensing page, an interview, a guide in the monorepo. The new document assembles the fragments into a full argument, with the lawyers' fingerprints all over it. What follows is the argument, then the two admissions inside it that surprised me, then the part where I explain why I still cannot tell you whether any of it will work.

The machine

The licensing architecture has three layers, and the first thing to say about it is that it is genuinely well designed, by people who have clearly read their open-source history.

Everything enters through the game licenses. Submit code to the protocol and you have accepted the inbound license, which is the rights hook: it secures the Foundation's claim to the intellectual property in what you submitted, and it guarantees that your submission will always remain available under the project's open license. Code already inside the game can be used as a starting point for new submissions under a separate outbound license, which is how the commons compounds without leaking.

Everything leaves through one of two doors. The first door is the TIG Open Data License, and it is the cleverest artifact in the whole design. Modeled on the CERN open hardware license, it carries a share-alike obligation on the code, and then it goes somewhere copyleft has never quite gone: if you distribute the output of running the algorithm, or a product derived from that output, you must also publish the input data and whatever else is needed to reproduce your result. The license is openly engineered to be comfortable for academics, for whom disclosing data and methods is already the norm, and intolerable for most commercial users, whose input data is often the crown jewels. It fails the Open Source Definition's non-discrimination clause, and the Foundation says so without apology. The failure is the point.

The second door is the Commercial License: closed use, no disclosure obligations, a fee payable in TIG, set by a non-discriminatory rate card. For algorithms that are nearly finished products, the licensee is the end user. For the majority, which need adaptation to a specific customer's circumstances, the document introduces an integrator layer, intermediaries who license the algorithms, close what the document calls the customisation gap, and may distribute only to customers who hold a Commercial License of their own.

Step back and the shape of it is elegant. Most hybrid licensing schemes try to balance openness against adoption inside a single license, and the balance always leaks. TIG refuses the balance. The open channel is uncompromised, even militant, and the closed channel is simply priced. The open license's disclosure burden is the sales force for the paid one. You can have the algorithm free, or you can have your data private. For businesses that want both, the paid door is the only one open.

That is the machine, and the same document then makes two admissions that most projects would have buried.

The first admission: copyright is dead

The document states, in flat lawyerly prose, that copyright will not protect the portfolio. Copyright guards the expression of an idea; the idea itself, which is the valuable part, sits outside its reach. Because the protocol must publish its source code, and because coding agents are getting good at reading published code and reimplementing the underlying method in fresh expression, the Foundation expects copyright protection for its algorithms to fail. The concession is correct, and it is a strange thing to read in a document whose purpose is to argue that the portfolio will be defensible. The cheap layer of intellectual property, the one every software company relies on without thinking, is conceded to be worthless for the thing TIG produces most.

So the weight shifts to patents. Algorithms as such are not patentable in most jurisdictions, since the law declines to monopolize abstract ideas, but patents on the technical effects of algorithmic methods are, as the document notes, commonplace. The procedural design follows: a provisional patent application filed before an advance is published, exploiting the twelve-month grace period in the United States, with full applications pursued later where the market signal, meaning actual adoption inside the game, justifies the cost. Assignment of the invention to the Foundation is not a condition of playing; it is the price of claiming the five-hundred-token advance reward, which is how the portfolio gets built without conscription.

Notice what the admission does to the thesis. An open project has conceded that openness destroys one of its two legal legs by design, which means the entire rent hypothesis now rides on the most expensive and slowest instrument in intellectual property law, and the most jurisdictionally patchy. American examiners apply Alice with real hostility toward anything that smells like an abstract idea; the European Patent Office's technical-effect doctrine is friendlier; other jurisdictions are a patchwork. Prosecuting a single patent family across the jurisdictions that matter runs into the tens of thousands of dollars, and the entity expected to fund that work is a foundation whose token currently values the whole network at roughly twenty-five million dollars. The document does not address that arithmetic. I suspect it cannot, because the arithmetic depends on a rate card that does not yet exist.

The second admission: the courts are not the plan

Intellectual property you cannot enforce is a rumor, and enforcement through courts is ruinously expensive for an organization of this size. Here the document makes its most original move, and it is so unexpected from a foundation with lawyers on retainer that it is worth slowing down for: the primary enforcement mechanism is social, and the argument for why social enforcement might work occupies the document's appendix.

The argument runs like this. Because the protocol must publish its algorithms, infringement is visible. Because infringement is visible, the community itself can punish it, through four channels the document names without embarrassment: the infringer loses technical cooperation from researchers and experts whose tacit knowledge they need; the commercial actors who want TIG to succeed as a counterweight to closed labs withdraw support and turn hostile; the crypto community's considerable talent for reputational amplification gets aimed at consumer-facing free-riders; and governments, which the Foundation hopes will come to see TIG as supplementary science funding and anti-monopoly infrastructure, learn to disfavor companies seen stealing from it. The courts remain in the design, but demoted to an asymmetric backstop: the threat of an injunction, or of specific performance of the open license's disclosure obligations.

The document calls this community enforcement, and it supplies a theory of how such a community comes to exist. Three ingredients. A cause: keeping frontier algorithmic methods open. A means to act: the protocol's own market, the fair reward-allocation mechanism that Eric Raymond identified in The Magic Cauldron as the missing piece of open-source economics. And perceived legitimacy, which comes in two kinds. Legal legitimacy means IP rights that are plausibly valid and plausibly infringed. Social legitimacy means the community believes its cause is just and its allocations impersonal in the Hayekian sense, determined by the market, since a committee making intentional allocations will eventually be caught playing favorites.

The historical anchor is Linux. Corporate patronage of Linux, the document argues, became rational the moment companies understood it as a defensive move against a Microsoft monopoly; the same reflex should operate in favor of TIG against the prospective monopoly of a single firm over AI-assisted algorithm development. And that monopoly is not a fantasy risk, the document insists, because AI-assisted discovery creates a data flywheel: the assistant that captures the tacit knowledge of the experts it works with compounds its advantage the way early search compounded Google's, and a compounding advantage in the production of algorithms is a compounding advantage in everything algorithms touch.

I have turned this analogy over more than any other part of the document, and I keep finding the same crack in it, though the crack is not quite where I first looked. My initial objection was that Linux had a present and visible monopolist to organize against, while TIG asks its community to cohere against a monopoly that has not yet formed, and defensive patronage is a known reflex only when the threat is already priced into everyone's head; but the more I sit with it, the more I think the AI case may be the exception to that objection, because the threat at the algorithm layer is unusually legible, every gated model and every closed lab is a public demonstration of what enclosure would look like, and the flywheel argument, while exactly the sort of thing every crypto project says about itself, happens to be the same argument serious people outside crypto are making about recursive AI improvement. The crack, when I finally found it, is somewhere else: the Linux patronage reflex produced money and code, but TIG's theory requires the reflex to produce enforcement, which is a much stranger ask. Funding a commons is a positive act with a clear price. Punishing a free-rider is a negative act with diffuse costs, and the history of open source says the negative act happened rarely, through a handful of patient enforcers, gpl-violations.org and the BusyBox suits and the Software Freedom Conservancy, organizations with lawyers and donors and decades of accumulated credibility. TIG is betting that a token community can grow that norm faster than the norm has ever grown. Maybe. The bet has never been tested, because no free-rider of consequence has yet appeared to be enforced against.

What the token owns

Here is the accounting.

License fees are recycled into the protocol's reward pools. They are not distributed to token holders. There is no dividend or buyback, and no mechanism by which commercial success reaches a passive holder's wallet. What the holder owns, if the machine works, is twofold: structural demand, because commercial licensees must acquire TIG to pay their fees, and governance, which today covers challenge ratification and which the document promises will eventually extend to the rate card itself, putting the pricing of the entire portfolio to a token-holder vote.

That second item deserves a fair hearing. DEUS sold governance-theater: votes that floated free of anything owned. A rate-setting right over a real portfolio is attached to the asset. It is an economic right with teeth, and if the portfolio ever becomes valuable, the right to price it is a right worth having. But it is still not equity, and the distance between a rate-setting vote and a cash-flow claim is exactly where the last cycle's token theses died. Between the token holder and the rent stands the Foundation, which holds the IP, files the patents, sets the first rate card, and decides what enforcement means in practice. John Fletcher has described the contributors' IP assignment as conditional, granted on the condition that the algorithms remain available under the open and commercial licenses forever, which is a meaningful guardrail against the Foundation pulling an OpenAI and closing the commons. It is also a legal promise, enforced by courts the design hopes never to need. The chain records attribution with real elegance; every submission carries a unique identifier, and provenance is perfect. The ownership layer sits elsewhere, in filings and contracts.

My own framework survives this design, softened but standing. The blockchain fallacy says a tokenized claim is not the property, and the custodian does not disappear because the claim is well engineered. TIG's answer is not to deny the custodian but to surround it with incentives and conditions, and to post a watchful crowd at the door. That is a better answer than most projects have managed, but it still asks you to trust the custodian's good behavior.

In the interest of hygiene: I hold a small amount of TIG, and I submitted code to one of its challenges earlier this year. Neither fact is a reason to trust my reading. Both are reasons I read the documents closely.

The four events

The document's real virtue is that it makes the thesis falsifiable, which almost no crypto document does. Four events would tell us whether the machine works.

The first is patents. Watch whether provisional applications are actually filed on the strongest advances, and whether any survive examination. The second is a licensee. The first commercial entity to pay a fee in TIG for the right to keep its data closed is the moment the demand side stops being theoretical, and as of this writing there is no public evidence it has happened. The third is an enforcement incident: a free-rider of consequence, and a community that either does or does not make the free-riding expensive. Everything in the enforcement theory is conjecture until that day. The fourth is the rate card. If pricing power is genuinely handed to token holders, the governance right becomes real; if the handoff is always one more roadmap item away, it was marketing.

The market, for what it tells us, is pricing none of this. The token trades around eighty cents, a market capitalization near twenty-five million dollars, down from where it sat in the spring. That is not how the market would price a protocol it believed owned a growing portfolio of state-of-the-art algorithms. It is how the market prices an interesting experiment, and for now the experiment is the honest description.

The unresolved question

The bearer-asset essay ended on the observation that the test of an asset is who can suspend the rules it sits on. This essay has been the mirror image of that one. Its question is whether a rule no state has been asked to enforce can hold at all, and whether a community's belief in its own legitimacy can do the work that courts do for ordinary property.

I do not know the answer, and I have stopped expecting to learn it from documents. The answer arrives the day someone with something to lose takes a TIG algorithm, declines the license, and discovers what the community is made of. Until then, the most honest thing to say about the machine for owning open algorithms is that it is beautifully designed and candid about its own weaknesses, and that it is waiting for the one test its architects cannot run on themselves.

Can you own something you have given away? The law's answer has always been: only if someone with power agrees to punish the taker. TIG's answer is that the punishment can come from the crowd, if the crowd believes. I hope we get to find out what the crowd believes.

---

Primary-source note: the licensing architecture, community-enforcement theory, and pricing intentions described above are drawn from the TIG Foundation's value-capture document ("Capturing the Value created by The Innovation Game's Algorithms") and the "Anatomy of TIG Licensing" guide in the tig-foundation monorepo. The vehicle-routing result is the Foundation's own characterization. Token price and market-capitalization figures are CoinGecko's, as of August 11, 2026. The legal characterizations (Alice, EPO technical-effect doctrine, the Open Source Definition) are matters of public record, simplified for readability.