The September 16 Markup: A Crypto Tax Bill's Real Payload Is Buried in the Word 'Broker'
Hook
A committee calendar entry is not a catalyst. It is a placeholder for a catalyst.
On September 16, a House committee is scheduled to mark up rules governing how the United States taxes crypto assets. That is the total factual payload of the item I am working from. There is no bill text. No committee name. No year attached to the date. No section numbering, no revenue estimate from the Joint Committee on Taxation, no vote margin, no sponsor list.
I have spent enough hours reading filings to know what a thin primary source means. When the document is empty, the headline becomes the product. And the headline circulating — "US moves toward crypto tax clarity, confidence improves" — is carrying weight the underlying facts have not yet earned.
So treat this the way I treat a whitepaper that ships without a repository: the missing parts are the signal. Not the announcement.
Here is what I actually know. A markup is scheduled. A markup is a procedural step, not a legal outcome. Everything else — whether the rules are friendly, punitive, or inert — sits inside text nobody outside the committee has published.
The tradeable insight is not the date. It is the definition of one word that will appear in that text: broker.
Context
Start with the boring part, because the boring part is where the money is.
The United States has taxed crypto as property since 2014. Notice 2014-21 said so in plain language: virtual currency is property, general tax principles apply, you owe capital gains on disposition. That notice ran about six pages. It created a decade of ambiguity in six pages, and every crypto accountant in the world has been pricing that ambiguity ever since.
Then came the rulings. Rev. Rul. 2019-24 addressed hard forks — receive new coins from a fork you did not control, and you have ordinary income at the moment you gain dominion and control. Reasonable in theory. Unworkable in practice, because "dominion and control" across a node operator's wallet set is a lawyer's abstraction, not an engineering state.
Then came 2021, and the infrastructure bill. This is the moment that should have taught the entire industry how this fight actually resolves.
The bill contained a broker reporting provision. The sector read it, realized the statutory definition of "broker" could swallow node operators, miners, validators, and wallet developers, and mounted the most intense lobbying campaign in its history to narrow it. What did they get? A Senate amendment that failed on a procedural vote. A parallel House effort. Then nothing. The definition went into law roughly as written, softened later by the IRS's own guidance rather than by Congress.
The lesson from 2021 is not "lobbying works." The lesson is that operative language usually survives the legislative process, and the practical boundaries get drawn years later by regulators, accountants, and the first taxpayer unlucky enough to be audited. The statute sets the shape. The guidance sets the pain.
Then 2023. The IRS proposed broker regulations covering digital assets. Then 2024 brought the final regulations and Rev. Proc. 2024-28, which quietly changed something enormous: it ended universal cost-basis pooling. From 2025 onward, taxpayers must track cost basis wallet by wallet — or account by account, as the service defines it.
That last sentence is the one most people skipped. I want to come back to it, because it is the most technically consequential piece of crypto tax policy in a decade, and it has almost nothing to do with a markup date.
One more piece of history, because it explains why institutions are not waiting on this vote. In early 2024, after spot Bitcoin ETFs cleared, I spent about three months reading prospectuses from the major asset managers. What I found interesting was not the fee war. It was the custody architecture and the creation/redemption mechanics, and specifically how each structure handled the investor's tax position.
Here is the thing nobody framed clearly at the time: an ETF is a tax wrapper. The fund does the basis tracking. The investor gets a 1099 from a broker they already have, in a format their accountant already accepts, for an asset sitting in an account that already exists. The entire wallet-by-wallet nightmare is dissolved at the wrapper layer.
That is not a small part of why ETF vehicles won the institutional flow war. It is a large part of it. Compliance is not what institutions buy. Packaging is what institutions buy, and packaging is compliance made invisible.
Hold that thought. It becomes the contrarian argument later.
Core
Here is the mechanical structure of the event, and why the surface reads as noise while the substrate carries signal.
First: what a markup actually is.
A markup is where a committee takes a chairman's draft, offers amendments, debates them, votes them up or down, and reports a bill out. It is the first point at which text stops being a press release and starts being a document with a number.
Historically, most bills that reach markup in a chamber controlled by the sponsoring party pass that chamber. Most of them also die later — in the other chamber, in conference, or on a calendar nobody schedules. So the base rate for "markup implies law" is low. Not zero. Low. And the base rate for "markup implies immediate price impact" is lower still, because the market has already been told a bill exists by the time the date is set.
A scheduled markup is a step on a staircase. It is not a door.
Second: the four clauses that actually matter.
Every crypto tax bill comes down to four variables. If you want to evaluate the September 16 text in thirty seconds, read for these and ignore everything else.
One — the definition of broker. A narrow definition captures centralized custodians that already file information returns. A broad definition captures any person who "regularly provides any service effectuating transfers of digital assets on behalf of another person." That second phrasing is the one that nearly caught miners and validators in 2021. It is also the phrasing that determines whether a non-custodial protocol has a reporting obligation it is structurally incapable of satisfying.
Two — cost basis method. Universal pooling versus wallet-by-wallet versus a hybrid. This is not an accounting footnote. It changes realized gains by double-digit percentages for active users, and it changes how much software every exchange, every tax tool, and every self-custody user has to build.
Three — the character and timing of staking, mining, and validation income. Ordinary income at receipt followed by capital gains treatment on later disposal is the current framework. The open questions are when income is recognized and how it is measured. Measuring one day of mining rewards is trivial. Measuring a validator's rewards across thousands of epochs, on a network that rebalances continuously, against a price that moves every twelve seconds, is a genuinely hard instrumentation problem.
Four — decentralized finance and non-custodial activity. Does the rule reach it? If yes, how? A protocol cannot file a return. It has no legal personality, no KYC pipeline, and no mechanism to collect a taxpayer identification number. Either the obligation lands on a front-end operator, or on a wallet provider, or it is simply unenforceable. And unenforceable rules do not disappear. They migrate.
Third: the wallet-by-wallet problem, which is already live.
I want to dwell here, because this is where my own experience is relevant and where the real analysis sits.
In 2020, during DeFi Summer, I wrote a Python script that watched Uniswap and SushiSwap pools for price divergence and executed small arbitrage trades. Over the season it ran more than five hundred trades and netted about forty-five thousand dollars. Good result for a script. Terrible result for my accounting.
Every one of those five hundred trades was a taxable disposal of one asset into another, and therefore a realized gain or loss in dollar terms at the moment of execution. Five hundred events. Each with a basis, a timestamp, a price, and a fee. And at the time, I was permitted to run all of it against a single pooled cost-basis figure across my holdings — a crude method, but a survivable one.
Now remove the pooling. Under wallet-by-wallet basis, that same activity has to be attributed to the specific wallet and the specific lot that funded each leg. If you have ever tried to reconstruct which USDC minted which liquidity position on which chain, you know the shape of the problem. It is not an accounting problem. It is a data reconstruction problem, and the data was never designed to be reconstructed this way.
Five hundred trades, one summer, one human writing a script. Now scale that to a market maker running continuous inventory rebalancing across six chains. Now scale it to an autonomous agent, which I will get to.
The industry responded by building indexers — tools that crawl on-chain history and reconstruct lot-level basis after the fact. That market exists because the rule exists. Notice what happened there. A tax rule created an entire software category, and the customers of that category are the participants who can afford to buy it.
That is not a side effect. That is the mechanism.
Fourth: what a broker obligation actually requires in engineering terms.
Strip away the legal language and a reporting requirement is a data pipeline with four stages.
Stage one: identity. You need a legal entity and a taxpayer ID attached to every account. This is precisely why custodial exchanges can comply and non-custodial tools structurally cannot. It is not a matter of will. It is a matter of whether the counterparty exists.
Stage two: event capture. Every transfer, trade, reward, airdrop, fork, and fee has to be recorded with a timestamp, an asset identifier, a quantity, and a fair-market value in dollars. On-chain this is messy. A single user intent can produce four or five state changes, and standard token interfaces do not tag which state change constitutes the taxable event. Anyone who has tried to label a Uniswap v3 position event stream knows the parser is doing interpretive work, not bookkeeping.
Stage three: basis propagation. When an asset moves between wallets, the basis has to move with it. When it enters a pool, basis has to be split across the position. When the position is burned, basis has to be recombined. When you pay gas in a native token you also hold as an investment, that fee is itself a disposal, which means gas consumption is simultaneously a cost and a capital event. Cross-chain bridges compound all of it, because now basis has to survive a change in ledger.
Stage four: form generation and filing. Form 1099-DA — the digital-asset information return the IRS has been building toward — looks like a two-page form. Behind it sits a reconciliation system that has to survive an audit.
Here is the part that gets underrated. Stage three is where the engineering difficulty concentrates. And it is nearly impossible to do correctly for self-custody activity without either the taxpayer maintaining perfect records in software, or a third party reconstructing the whole history from chain data.
Both of those are businesses now. Neither of those is free. And every exchange with a US user base has already had to rebuild its tax-center infrastructure once to accommodate the shift away from pooling. That rebuild is the real cost of this regulatory arc, and it landed before any of the current headlines.
I spent weeks in 2017 auditing ERC-20 contracts for an ICO raising twelve million dollars, and I found an integer overflow in the distribution logic that would have let miners mint unlimited tokens. The lesson I took from that was simple: the whitepaper is one document and the code is another, and only one of them executes. Tax law works the same way. The press release is the whitepaper. The section text is the code. Nobody has published the code.
Fifth: the DeFi question, stated precisely.
I want to be careful here, because this is where commentary gets lazy in both directions.
The bearish framing: if the rule defines broker broadly enough to include non-custodial front-ends, US-facing access to those front-ends becomes a compliance burden, and some operators will geo-block. That has precedent. It is not science fiction, and it has already happened to other categories of product.
The bullish framing: the rule cannot reach the protocol layer, only the interface layer, and interfaces can relocate, restructure, or decentralize their governance until the legal question becomes unresolvable.
Both framings are partly right, which is why the clause matters more than the headline. And note carefully: a markup date tells you nothing about which framing is correct. The information content of "a markup is scheduled" with respect to the broker definition is precisely zero.
Sixth: there is no price signal here, and pretending otherwise is a tell.
Legislative process events are the cheapest news to manufacture. A staffer schedules a hearing. A press release goes out. A wire service writes it up. A secondary outlet rewrites the wire service. A newsletter rewrites the secondary outlet. By the fourth hop, "a committee scheduled a markup" has become "US moves toward crypto tax clarity."
I have watched this pattern long enough to name it: narrative inflation through relay. Each hop adds confidence and subtracts detail. The final product sounds more actionable than the original document while containing strictly less information.
For calibration, compare classes of event. Regulatory enforcement — a lawsuit, a settlement, a denial of a registration — moves markets five to fifteen percent within hours, because it changes the operating environment immediately, on a specific entity, with a defined cost. A procedural legislative event moves markets under two percent, because it changes nothing today and may change nothing ever.
Scheduled markup dates belong to the second category. Treat any five percent move attributed to a markup date as something else wearing a costume.
Seventh: the externality nobody models — machine-speed taxable events.
This is where I stop reporting and start forecasting, and where my recent work is relevant.
In 2026 I built a prototype in which an autonomous agent negotiated data-access fees over a smart contract, holding a testnet wallet funded with ten thousand dollars. The agent priced access, signed transactions, and settled. It worked. And the moment it worked, I understood what I had actually built: a taxable entity that generates disposal events at machine speed with no human in the loop.
Every micropayment was a disposal. Every service fee was a disposal. Every swap executed to acquire compute was a disposal. At a human trading cadence, that is an accounting chore. At an agent cadence — hundreds of events per hour, across an unbounded set of counterparties, in assets the agent acquires and spends without ever intending to hold — it is a category the tax code has no vocabulary for.
Now place that next to wallet-by-wallet basis. An agent rotating inventory across wallets and chains generates a basis-tracking problem that cannot be solved by human reconstruction. It is only solvable by instrumentation: the agent accounting for itself, in real time, at the point of execution, in a format that a filing system can consume without translation.
I think that is the real long-run consequence of this entire regulatory arc. Not the rate. The instrumentation requirement. Whoever defines the standard ledger format that a machine agent writes to at execution time captures a category, because compliance will have to be embedded in the runtime rather than reconciled after the fact.
No one on a committee calendar is thinking about that. It is still where the value accrues.
Eighth: the lag, which is the most boring and most important fact.
The distance from markup to signed law to effective regulation to first enforcement action is typically twelve to twenty-four months. Not weeks. Two years.
Which means the correct holding period for any thesis built on this event is measured in quarters, not hours. Which means the correct instrument for expressing a view is not a leveraged position taken on the headline. It is a position taken in the businesses that will service the mandate once the mandate is real — the accounting layer, the indexing layer, the reporting layer.
That is not exciting. It is just true.
Contrarian
Now the part the consensus gets backwards.
The prevailing read on any crypto tax legislation is: clarity is bullish. Certainty lowers the risk premium. Institutions require it. Rules bring capital.
That is half true, and the half that is false inverts the causality.
Clarity is not a gift to the market. Clarity is a cost assigned to specific participants. The question that matters is never "is the rule clear?" The question is "who pays for the clarity?"
Read a reporting mandate and ask who can satisfy it. A custodian with a legal entity, a KYC pipeline, and a data warehouse can. A non-custodial protocol cannot, structurally, ever, at any budget. A small exchange can, expensively. A large exchange can, cheaply, amortized across more volume and a bigger compliance team.
Which means a reporting mandate functions as a regressive tax on small operators and non-custodial architectures, and as a competitive moat for large ones. The clarity is real. So is the redistribution. The industry narrative only ever mentions the first half, because the first half is the part that sounds like adoption.
Here is the second inversion, and it is the one that makes me skeptical of this specific news cycle. A markup scheduled is not clarity. It is the announcement of an intent to discuss clarity. The gap between the announcement and the document is where the entire narrative currently lives, and the narrative is being priced as though the document already exists.
The spread: market expectation, elevated. Actual delivery, one calendar entry. That is a wide spread, and wide spreads resolve toward the smaller number.
I have seen this shape before. In May 2022 I sat with a block explorer open and watched stablecoin minting interact with a token's supply mechanics hours before mainstream coverage started. The story that eventually got told was "algorithmic stablecoin failed." The mechanically accurate story was "the redemption path was structurally incapable of absorbing the exit." The consensus narrative arrived late and slightly wrong. The people who traded the narrative instead of the mechanism got the timing wrong in both directions.
Tax markups are the same species of event. The consensus narrative — clarity is coming — will arrive on schedule, because it always does. The mechanism — which clause reallocates cost to whom — will be decided in text nobody in the audience has read.
There is a third inversion worth naming. The "clarity is bullish for institutions" argument has already been partially satisfied by a different vehicle. If an institution wants crypto exposure with clean tax treatment today, the ETF wrapper already provides it, without waiting for a statute. Which means the marginal institutional flow that a tax bill unlocks is smaller than the narrative implies. Institutions did not wait. They routed around the problem through packaging, and packaging does not need a markup.
Arbitrage is just geometry disguised as finance. Right now, the geometry here is a single point: a date. A point has no shape. You cannot trade a shape that does not exist.
Takeaway
So what do you actually watch?
Not the date. The broker definition, first and hardest. Whether the text says "custodian" or "any person effectuating transfers" determines which architectures survive in US-facing form — and that clause will be written by staff counsel whose interests have nothing to do with your portfolio.
Second, whether wallet-by-wallet basis survives the amendment process or gets softened back toward pooling. That single line item decides the size of the on-chain accounting software market and the audit exposure of every active trader.
Third, the small-transaction exemption. A De Minimis threshold for payments has been requested for a decade and granted never. If it appears, the downstream effect lands on payments and machine-to-machine settlement rather than speculation — and that is the clause that would matter most in 2027, when agents start transacting at scale.
Fourth, whether the text addresses staking and validation income timing at all. Ambiguity there is not neutral. It is a liability that gets resolved by whoever is audited first, and the resolution will be expensive.
And the meta-question, the one I would actually ask if I had the chairman's ear: does the drafting assume a human is in the loop? Annual filing, cost basis, wash sales, custody — every framework we have assumes a person with a calendar, a preparer, and the patience to reconcile twelve months of history in April.
That assumption has maybe two years left. Whatever is marked up on September 16 will either embed it or ignore it. Those are the only two futures worth pricing, and one of them is already late.