Somewhere inside a small crypto charity called Manifund, there is a reconciliation tool. It was written by an employee the platform knew as "Carol" โ a presence in the working files, a name on internal documents, a person nobody outside the organization needed to think about. By the platform's own account, that tool did its job. It surfaced a six-figure transaction that had been recorded in error, flagged it, and let someone correct the books before the mistake compounded.
A six-figure reconciliation error, caught by a woman whose name, three years earlier, was attached to an $8.7 billion hole.
"Carol" is Caroline Ellison. The former chief executive of Alameda Research, the star witness of the FTX prosecution, the person who stood in a Manhattan courtroom and described, in careful sentences, how she signed documents that made numbers agree with what the firm needed them to say. She is now, according to a post from the crypto outlet TBPN and subsequent reporting, working at Manifund โ anonymously, until she wasn't.
That juxtaposition is the whole story, and it is a story about governance, not about mercy.
Start with what is actually established. Ellison completed her prison term. She joined Manifund on a trial basis on July 13, moved to full-time on August 10, and the arrangement became public on September 11 โ roughly a month after the promotion, and not because anyone at the charity announced it. Manifund co-founder Austin Chen issued an apology for the use of a pseudonym, framing the hire as a bet on redemption over rรฉsumรฉ. Manifund is an effective-altruism-adjacent donation platform focused on effective giving and AI safety, and it markets itself on what it calls radical transparency: open finances, open data, open source code. Chen has said elsewhere that the FTX Future Fund was an early seed funder of the project and shaped the way it approaches giving.
The legal picture is tidy, and that tidiness is the interesting part. Three layers closed in sequence. Criminal: a forfeiture order of $11.02 billion. Civil: the CFTC closed its case on August 19, extracting a five-year trading ban and a ten-year registration ban, with no additional fine โ explicitly credited to cooperation and to the fact that the criminal forfeiture already existed. Administrative: the SEC barred her from serving as an officer or director of a public company and โ this is the sentence worth reading twice โ did not restrict her from working at a nonprofit.
Meanwhile, the machinery of restitution grinds on independently. The FTX Recovery Trust has distributed more than $11 billion to creditors. It has also asked a Delaware court to block additional payout claims brought under an older fraud theory, with a re-hearing set for October 20.
Now the analysis.
Every one of the three sanctions against Ellison governs an exit. None of them governs an entrance. Forfeiture takes the proceeds away. The trading ban removes her from the market. The registration ban removes her from professional finance. The officer-and-director bar removes her from the boardroom. Each is a wall built at a door she is leaving through. And so, when she walked through a door nobody had walled โ the door of a small nonprofit that publishes its finances and does not publish its hiring decisions โ she walked through legally, quietly, and under a name that was not hers.
I have spent a great deal of my professional life on the failure modes of governance, and I have come to believe that the exit-entrance asymmetry is the single most reliable tell of a system that is performative rather than protective. It shows up everywhere. In DAO governance, where I work, we pour enormous energy into the exit: how do you revoke a role, slash a stake, rage-quit a treasury, unwind a position. Exits are legible. They have addresses. They emit events, they generate evidence, they settle into a block explorer where anyone can point at them. Entrances are diffuse. An entrance is a decision that happens inside a room โ sometimes a small one, occasionally a room with a single person in it. You cannot index a decision.
Which brings us back to the reconciliation tool.
There is a version of this story that treats the tool as proof of rehabilitation: a gifted numbers person, now using her gifts for good. That version is emotionally satisfying and analytically useless. The honest observation is stranger and more uncomfortable: the capability that hides an $8.7 billion shortfall and the capability that finds a six-figure misbooking are the same capability. Both require an intuitive grasp of a complex ledger, a feel for where entries can drift without being noticed, and the discipline to sit with a set of accounts long enough to understand its internal logic. The skill does not carry a moral signature. It carries a memory of the context in which it was learned โ and that memory does not transfer, and it does not stay behind either. It simply travels with the person.
I audited more than fifty whitepapers during the 2017 mania, and I remember the specific moment I understood that the dangerous part of a project was rarely the cryptography and almost always the bookkeeping. A team can implement a zero-knowledge proof correctly and still have no idea what its own treasury actually contains. I wrote a piece back then called "The Ethics of Empty Vests," aimed squarely at retail investors, arguing that a project's philosophical emptiness precedes its technical emptiness โ that the first thing to fail in a doomed token sale is not the circuit but the honesty of the ledger. I was not wrong, but I was incomplete. I had not yet seen that the same ledger discipline which fails honestly can also succeed dishonestly, and that the discipline itself is indifferent to which one you are doing.
So Manifund did not hire a redeemed accountant. It hired a very good accountant and got the redemption question for free.
That leads to the second structural problem, and the one I find genuinely instructive.
Manifund's brand is transparency, and its hiring process was the least transparent thing about it. I do not mean hypocrisy in the lazy sense. I mean a specific, recognizable system failure that anyone who has built decentralized governance will recognize instantly: the centralized sequencer problem, expressed in human resources. You can open-source every line of your treasury code, publish every grant, stream every financial record to a public dashboard โ and still make the one decision that matters most in a closed room, unilaterally, with no timelock and no notice period. Publishing your code is a promise. Publishing your decisions is a constraint. The first is cheap, scales beautifully, and generates excellent press. The second is expensive, because it costs you the option to do a compassionate thing quietly.
I have run weekly governance literacy workshops in Paris since the DeFi summer of 2020 โ over two hundred participants across the years, most of them ordinary users trying to understand why their vote mattered and how to cast one that counted. The lesson I keep returning to in those sessions is that transparency is not a property of a document. It is a property of a decision process. A fully public ledger tells you what happened. It never tells you what was considered and discarded.
There is a third layer here, and it explains why the hire happened at all. Manifund's seed funding came, by its co-founder's own account, from the FTX Future Fund. The organization that hired the woman who ran the trading firm that supplied the money that seeded the charity is, in a very literal sense, completing a circuit. That is not a coincidence and it is not corruption. It is what dense moral communities do. Effective altruism is not a large movement; it is a network, and networks recruit from their own edges. When an ecosystem defines itself by a shared project โ in this case, doing the most good with the most rigor โ it develops internal trust relationships that are genuinely strong and genuinely inward-facing. Those relationships read a second chance very differently from the outside world.
I built a platform once called SoulBound Stories, alongside three other women, funded by community grants rather than venture capital specifically so that we could not be pressured on mission. We raised โฌ150,000, and the essay I wrote for it argued that tokens should encode belonging and social consensus rather than pure transferable value. What I learned from that experience, and what I see operating at Manifund, is that the people inside a mission-driven community are not being cynical when they make an inward-looking decision. They are being loyal. The problem is that loyalty and governance are different functions, and an organization that lets one perform the other has not eliminated the risk. It has relocated it somewhere the public cannot see.
And then there is the part of this story that involves actual money, and which almost nobody is discussing.
The FTX Recovery Trust has already distributed more than $11 billion to creditors. A criminal forfeiture of $11.02 billion sits alongside it. Creditors are, functionally, implicit equity holders in a bankruptcy estate, and their unlock schedule is a Delaware courtroom calendar. The October 20 re-hearing on whether to block additional claims under an older fraud theory is a live dispute about distribution fairness โ between people who have been paid and people who want more. Notice the arithmetic nobody has reconciled in public: the initial customer shortfall was reported around $8.7 billion, and distributions plus forfeiture now comfortably exceed that. Either asset recovery went substantially better than the worst-case estimates, or interest and priority are producing over-recovery for certain classes, or the figures are simply not comparable in the way retail observers assume. Any of those possibilities matters more to more people than whether a former executive uses a pseudonym at a charity. But it does not trend.
Here is the counterintuitive read, the one I would defend in a room full of people who disagree with me.
The question everyone is asking โ should Caroline Ellison be allowed to work in crypto? โ is not a hard question. It is a question our rules already answered, and they answered it the way they were designed to. Yes, at a nonprofit. Yes, with a trading ban that keeps her away from markets for five years. Yes, presumably with no custodial access to donor funds. The disclosure was unflattering in timing and presentation, but the arrangement itself is not a legal scandal. It is a legal non-event.
The hard question is upstream of all of that, and almost nobody is asking it: why did Manifund's governance have no entrance protocol for the most reputationally loaded hire in its history? Not a ban โ a protocol. A process by which someone, somewhere, with a name attached, reasoned about the costs to donors, to staff, to the grantees whose funding continuity depends on the platform's credibility, and then either approved the hire with disclosure or approved it in a form the organization could defend out loud. None of that appears to have happened, and the pseudonym is the evidence that it did not. A pseudonym in this context is not a privacy measure. It is an avoidance measure. It exists to prevent the decision from having to be made publicly.
Decentralized governance is often accused โ sometimes by me โ of over-engineering entry. Quorum thresholds, proposal bonds, timelocks, execution delays, delegated voting, veto councils. It is genuinely exhausting, and a fair amount of it is theater. But every one of those mechanisms exists for one reason: a system that governs only exits can be gamed by anyone willing to walk in the front door with a plausible story. We have a saying in this industry that has hardened into reflex: do not govern the exit โ govern the entrance. Manifund ran the experiment in reverse and produced the expected result. An open ledger with an opaque hiring process is a treasure map with a secret room. You published the map. You never published the room.
Code is law, but people are the soul. That line gets quoted as though it were a plea for leniency. It is not. It is a warning about where the enforcement has to live. You cannot write a smart contract that governs a human decision you have not bothered to specify. You can only write the decision down.
October 20 will pass, and the recovery trust will either receive its order or it will not, and most people will not notice either way. The SEC's officer-and-director bar will keep not reaching nonprofits, and at some point โ probably after a worse case, involving someone less sympathetic โ a regulator will look at that seam and decide whether to sew it shut. And somewhere in a small office, a person with a debt that cannot be paid in dollars will keep reconciling a ledger she understands better than anyone else in the building, and the question of whether that is a mercy or a vulnerability will stay open.
The question is not whether a person can change. People can. The question is whether an institution that claims radical transparency can be trusted to say so out loud, before anyone asks.
