Yesterday, Decrypt reported that Defend American Jobs, the Fairshake affiliate that spends in Republican-friendly races, had disclosed roughly $7 million more in advertising aimed at the Ohio Senate contest. The filing it was reading was signed on 15 October. It lists a $5.2 million media buy opposing Sherrod Brown, a further $1.95 million buy five days later, and the production costs attached to both.
I pulled the filings rather than the coverage. All 297 itemised independent expenditures this committee has reported in the 2026 cycle, plus every Form 3X it has filed since January 2025, plus the same set for Fairshake itself. The raw electronic filings are free, they are plain delimited text, and they download in seconds.
What they show is a disclosure regime working exactly as designed, producing a record in which the single question a reader actually wants answered cannot be asked. Not hidden. Not redacted. Not late. Unaskable, because the record has no field for it and the money has no property that would fill one.
TL;DR
- Defend American Jobs has reported $122,252,474.23 of independent expenditures this cycle across 297 line items, 48 candidates and 24 states, every one of them itemised with a date, an amount, a vendor and a support-or-oppose flag.
- Its most recent disclosure of its own finances covers through 31 August 2026 and shows $5,659,903.73 in the bank. Since that date it has told the FEC about $54,598,590.75 of advertising, or 9.6 times its last published balance.
- Advertising is reported within 48 hours of dissemination under 11 CFR 104.4(b)(2). Money is reported monthly in arrears. The gap between the two clocks is currently 50 days.
- The statute that requires a filer to name the donors who gave for the purpose of furthering an independent expenditure, 52 U.S.C. 30104(c), applies to “every person (other than a political committee)”. The regulation that implemented it, 11 CFR 109.10(e)(1)(vi), now reads “[Reserved]”.
- Of the 297 expenditure rows, zero carry a back-reference to any receipt. The same filing format carries parent-and-child attribution rows on the money coming in, and does it well. The hole is on one side only.
What the filings actually say
Start with the spending, because that is the part the system reports beautifully. Defend American Jobs, FEC committee ID C00836221, has filed 297 Schedule E rows in this cycle totalling $122,252,474.23. Of that, $102,634,501.14 supports candidates and $19,617,973.09 opposes them. Two hundred and thirteen rows concern House races, 84 concern Senate races. Every row names a payee, and there are only five payees in the entire set. One of them, Main Street Media Group, received $110,874,833.61, or 90.69% of everything the committee spent.
Now the Ohio race, the one in the news. In calendar 2026 the committee has filed 17 rows touching that seat: $33,778,233.05 supporting Jon Husted and $12,894,063.13 opposing Sherrod Brown. Those two figures sum to $46,672,296.18.
That number will look familiar if you read yesterday’s coverage, which reported the filing as listing “Brown’s calendar-year-to-date total at about $46.66 million”. It does not. The Schedule E column it comes from is headed Calendar Year-To-Date Per Election for Office Sought, and the running total in the filings increments across both candidates in one sequence: $5,713,491.28 for Husted, then $11,426,982.55 after the Brown buy, then $11,444,032.55 after a production invoice for Husted. It is a figure for the race, not for the candidate. The actual total spent against Brown is $12,894,063.13, and more than half of it, $7,162,997.86, was filed on one day.
I am not interested in scoring a point off a reporter working a Saturday deadline. I am interested in the fact that the misreading is the most natural one available. The field is a per-office aggregate, the filing is a list of per-candidate rows, and a reader who wants to know how much was spent against one person has to know which of those two shapes the number has. In a story about who paid for what, the first number quoted was already an aggregate wearing a candidate’s name.
The two clocks
Here is the arithmetic that got my attention. The committee’s most recent report of its own financial position is the September monthly, filed on 20 September, covering through 31 August. It shows cash on hand at close of $5,659,903.73 and receipts for that month of $25,000.
Since 1 September it has filed 35 independent expenditure rows totalling $54,598,590.75. Every one of them is a 48-hour report. The median gap between the expenditure date and the filing date on those 35 rows is five days, and the fastest is one.
So the public record currently says two things at once, both official, both current, and mutually unreconcilable: this committee has $5.66 million, and this committee has spent $54.6 million since. More than nine times its own published balance, disclosed to the cent, with no accompanying statement of where the money came from, because that statement is not due yet.
It is due soon. The FEC’s 2026 calendar for monthly filers closes the October report’s books on 30 September with a deadline of 20 October, and the Pre-General report closes on 14 October with a deadline of 22 October. So the funding behind the 21 September buy becomes public 29 days after the buy, and the funding behind the 9 October buy becomes public 13 days after. The spending clock runs in hours. The money clock runs in weeks.
None of that is a scandal and I want to be precise about why. Fairshake, the parent committee, reported $108,293,831.76 in cash on 31 August and has transferred money to this affiliate six times in this cycle, in amounts from $1 million to $9 million. Anyone reading the record can infer with near certainty where the September and October money came from. But inference is what you do when the record does not say, and the gap between a well-founded inference and a filed fact is the entire subject of this series.
Ninety-six per cent of the money is a single line
Follow the funding one hop further back and the shape of the problem gets clearer.
Defend American Jobs has disclosed $38,394,280.07 of receipts in this cycle. Of that, $37,000,000, or 96.37%, arrives as transfers from Fairshake. The rest is refunds from media vendors and a handful of direct corporate cheques.
Fairshake itself has disclosed $169,815,102.93 of receipts in the cycle, and its donor list is short enough to read in one sitting: Ripple Labs at $48,000,000, Coinbase at $33,092,475, AH Capital Management at $23,800,000, Marc Andreessen and Ben Horowitz at $11,900,000 each, Digital Currency Group at $1,000,000, Uniswap Labs at $999,987.89 in UNI and Robert Leshner at $300,007.50 in COMP. Those itemised contributions come to $131,003,170.39. The remainder of the $169.8 million is $30,053,655.12 of proceeds from crypto and stock the committee sold, $8,539,879.03 of bank interest, because a nine-figure balance earns that, and $218,398.39 of refunds from media vendors.
Every one of those lines is public. Every expenditure line is public. And there is no row anywhere in the system that joins one to the other, because the thing that would have to be true for such a row to exist, that a particular dollar retains the identity of its giver, is not true of money in a bank account.
One detail makes the point sharper than any argument I could construct. Media vendors issue refunds, and refunds come back as receipts. Main Street Media Group, the firm that received 90.69% of this committee’s advertising spend, appears on the receipts side for $760,026.28. Targeted Platform Media refunded Fairshake $145,139.46 in August and $43,362.34 last year. So a dollar can leave as an expenditure against one candidate, return weeks later as a receipt from a media buyer, and go back out against a different candidate in a different state, acquiring a new date and a new counterparty at each step and shedding whatever lineage you imagined it had. Even first-in-first-out, the crudest accounting fiction available, cannot survive that.
A penny decided which candidate $11.4 million was about
On 21 September the committee bought media through Main Street Media Group and reported it as two rows: $5,713,491.28 supporting Husted and $5,713,491.27 opposing Brown. Same vendor, same date, same buy, split down the middle, with the odd cent going to the supporting side.
I have looked at that pair for longer than it deserves, because it is the whole post in miniature. Somebody had to decide how much of one media purchase was for one candidate and how much was against the other, there is no fact of the world that settles it, and the answer they filed was fifty-fifty plus a rounding convention. The $46.66 million figure that travelled through yesterday’s coverage sits on top of that choice and every other allocation decision in the same column.
This is not fraud and it is not even sloppiness. It is what happens when a legible number has to be produced out of something that does not come pre-divided. A reader downstream sees two authoritative figures to the cent and has no way to know that one penny of the split was an aesthetic decision.
The format can express lineage. It is simply not asked to here.
Here is the part that turns this from an observation about money into an observation about disclosure design.
FEC electronic filings have a native pointer. Any row can carry a back-reference to another row’s transaction ID, and the receipts side of these filings uses it properly. When Fairshake received 33 bitcoin from Phil Potter in 2023 and sold them through The Giving Block for $997,971.14, the filing contains a parent row for the sale, whose memo reads “33.0 Bitcoins (BTC) sold via The Giving Block – purchaser unknown”, and a child row pointing at it that names Potter. Same for 15,500,000 USDC from Coinbase sold through Coinbase Commerce. Same for 3.65607 bitcoin from ARK Invest. The committee has recorded $71,714,064.70 of crypto and stock liquidated through intermediaries since 2023, and on the 2023 rows the format carried the chain from the asset to the giver without complaint.
Andreessen Horowitz’s cheques carry the same machinery in a different form: each one is tagged “See attribution below” and lands on an odd cent, $5,000,000.01 and $1,500,000.01, because an LLC’s contribution has to be attributed to the people behind it and the arithmetic of splitting it leaves a penny somewhere.
So the schema can say this money is really from that party. It does say it, repeatedly, in these very files. Now count the back-references on the other side: of 297 independent expenditure rows, the number carrying a pointer to any receipt is zero. Not because the field is missing. Because nothing requires it to be filled.
And the law is explicit about who is exempt. 52 U.S.C. 30104(c) requires a filer to identify “each person who made a contribution in excess of $200 to the person filing such statement which was made for the purpose of furthering an independent expenditure”. That is precisely the donor-to-advert link, written into the statute. It applies to “every person (other than a political committee)”. Super PACs are political committees. The FEC’s own regulation implementing the provision, 11 CFR 109.10(e)(1)(vi), was vacated in CREW v. FEC in 2018 and affirmed on appeal in 2020, and today the Code of Federal Regulations prints the subparagraph as a single word: [Reserved].
A reserved paragraph is the most honest artefact in this entire story. It is the shape of a question that was asked, answered, litigated and then left as an empty slot with a number on it.
Why this is not a disclosure failure
I want to resist the easy version of this argument, which is that someone is hiding something. Nobody here is hiding anything. The FEC’s 48-hour independent expenditure regime is one of the fastest financial disclosure requirements in any democracy, the itemisation is to the cent, the files are free and machine-readable, and every figure in this post was reconstructed from them in an afternoon.
The problem is upstream of disclosure. Suppose tomorrow the law required real-time reporting of every receipt and every disbursement, published to a public ledger within a second. You still could not answer “which donor paid for the Brown advert”, because after Ripple’s $48 million and Coinbase’s $33 million enter the same account, there is no longer a fact about which of them bought anything. You would have to choose a convention. First-in-first-out gives you one answer, pro-rata by share of receipts gives you another, last-in-first-out gives you a third, and all three are arithmetic performed on an assumption rather than measurements of a thing.
Compare that with what a verifiable random function does. A VRF proof binds one specific output to one specific request and one specific published key. You do not pick a convention; you run a verification that terminates in yes or no, and it returns the same answer for me, for you and for a stranger who was not there. Attribution by proof terminates. Attribution by accounting terminates only at a choice somebody made and did not have to show you.
That distinction is also what separates this from the receipt nobody published, where the problem was a missing record of a spend that did occur, and from the flow nobody followed, where a rule was drafted and then never issued. Here the record is complete, timely, itemised and correct. The question still has nowhere to land.
What this has to do with who wins your raffle
Every provably fair page on every crypto gaming site in the world grades the same object: the mechanism that produced the number. Server seed, client seed, nonce, hash commitment, published proof, a verifier you can paste a string into. All of it answers one question, which is whether the operator could have changed the outcome.
Almost none of it answers the question underneath, which is where the money that pays you comes from.
Think about what a conventional platform’s bankroll is. It is a pool. Your winnings leave it alongside everybody else’s, and they entered it as other players’ losses, as treasury top-ups, as an investor’s bridge financing, as affiliate rebates coming back from a marketing spend, exactly like a media vendor’s refund landing back in a committee’s account. If you asked an operator “which deposit funded my payout”, the honest answer would not be a refusal. It would be that the question has no referent. The pool does not remember.
That matters more than it sounds, because the pooled answer and the proven answer fail differently. A fairness proof keeps verifying perfectly right up to the moment a platform cannot pay, since the proof was never about the money. The mechanism and the bankroll are two different objects, and the industry has spent a decade proving the first and asking you to take the second on trust. This is a different axis from whether your balance is segregated or merely equivalent: a segregation claim is about whether the asset is there, and this is about whether anything in the record connects it to you.
What Satoshie actually claims
The claim is narrow and structural, and it is about the prize path of a single round rather than about our company being virtuous.
When you buy into a Satoshie raffle, your stake goes into the contract for that round, not into a company account. The draw requests randomness from the Chainlink VRF coordinator named in the deployed code, which you can read before you stake. The coordinator verifies the proof on-chain before the callback is permitted to act. The winner is computed in that callback and paid in the same transaction, out of that round’s escrow. The result is that “where did my prize come from” is a transaction hash and a block number, permanently linked to a resolved game, rather than a withdrawal from a shared hot wallet about which nothing can be shown.
Ticket odds work the same way: they are arithmetic over tickets minted, readable before you buy, not a figure produced by an internal convention. If the round is thin, you can see that it is thin, which is a different condition from a number whose method you cannot inspect. As we have argued before about prize pools, the point is not that a pool is always wrong, it is that a pool whose contents are a function call should be read as a function call.
Honest limits
Three, and the first one is the one I would lead with if I were arguing against this post.
We run pooled things too. Our VRF subscription is a prepaid balance topped up from a treasury: 13 top-ups bought 12,946 draws, and no proof on earth says which top-up paid for your randomness. Gas comes from an operational wallet. The treasury is a pot like any other pot. The narrow claim above is about the path from your stake to your payout, and it does not extend one metre beyond that.
The gap is designed to close. The 50 days I measured is not a permanent darkness. On 20 October the September receipts land, and on 22 October the Pre-General report covers through 14 October. By the end of next week most of the money in this post will have a date and a name attached to it in aggregate. What will still not exist, then or ever, is the link between a given gift and a given advert.
This is not an argument that American campaign finance should be on a blockchain. It should not. Public ledgers are permanent and indiscriminate, and a system that made every political donation permanently traceable to an individual would produce harms that dwarf the ones described here. The useful transfer is narrower: a mechanism that binds an output to its input is a different class of object from a mechanism that reports inputs and outputs separately, and knowing which class you are holding is the whole skill.
Three questions worth asking
When somebody tells you a payout is funded, ask whether they mean a balance or a lineage. “The prize pool is fully funded” is a balance claim. “This payout came from this escrow for this round” is a lineage claim. The first survives any amount of commingling. The second cannot be made at all unless the system was built to make it.
Ask what convention produced the number. Any figure that splits a shared thing into named parts, a media buy across two candidates, a bankroll across players, a revenue share across partners, was produced by a rule somebody chose. Ask what the rule was. If the answer is a shrug, the figure is an opinion with decimal places.
Ask what the format could have recorded and did not. The most informative thing in these filings was not any number. It was that the back-reference field exists, is used correctly on the receipts side, and is empty 297 times out of 297 on the expenditure side. Absence inside a format that supports presence is a choice, and choices have authors.
A committee told the public, within 48 hours each time, that it had spent $54.6 million. The same public record says it has $5.66 million. Both statements are true, both are filed, both are signed under penalty of perjury, and the thing that would reconcile them is not late. It was never going to be written down.
Satoshie builds the other kind of system, where the number that decides who wins carries a proof of where it came from and the money that pays the winner carries a transaction hash that says the same. Have a look at how the draws work, and check one.
Photo by Diego Marín on Unsplash
