Over a forty-one-hour window in the second week of this quarter, a single unlinked claim about Algorand's leadership traveled further through crypto's information layer than every statement the Algorand Foundation has published in the last nine months combined. The claim was simple, company-shaped, and entirely false: that Algorand had appointed William Herkelrath, a former Chainlink business development executive, as chief executive officer. No filing. No Foundation post. No quote from Herkelrath. No quote from anyone. The artifact that started it was a screenshot of a sentence, and the sentence had no parents.
I watched it move, because sitting in Telegram rooms and X lists with a scraper running and a stopwatch is what my job has degenerated into. In the sample I captured โ roughly 9,400 posts across X, Telegram, and four Discord servers I monitor โ the median time from original post to first repost was under eleven minutes. The median time to primary-source verification was never. Three separate 'news' articles appeared inside the window, with bylines that resolve to nothing and prose that reads like a model summarizing a model summarizing a screenshot. Crypto Briefing eventually did the unglamorous thing and published the correction. Algorand has not appointed William Herkelrath as CEO. That is the fact. It is also, and I want to be precise about this, the least interesting part of the story.
The rumor was wrong. The reason it worked is not.
What follows is not a media-ethics sermon. There are enough of those. What follows is an attempt to take a fabricated personnel announcement seriously as a piece of market structure, because that is what it was. Somebody bid ALGO on it. Somebody sized a position on it. Somebody's model, or somebody's group chat, or somebody's aggregator feed, treated a screenshot as an input. And the deeper problem โ the one that outlives this particular lie โ is that the industry built a settlement layer with perfect finality and then ran its identity layer on vibes.
History rhymes, but the code doesn't. The code, in this case, had no opinion at all about who runs Algorand. That is the whole problem in one sentence.
The Ledger of a Lie
The mechanics of this class of rumor are now stable enough to be modeled. Stage one is the orphan claim: a sentence with no source link, posted where deletion is cheap and screenshots are permanent. Stage two is aggregation, where the sentence is stripped of what little context it had and rendered as a headline. Stage three is synthesis, where the headline becomes the source material for a longer piece that reads as though it were researched. By stage three, the original absence of a source is invisible, because the new article cites the aggregator, and the aggregator cites the screenshot, and the screenshot cites nothing.
I have started calling this citation collapse. It is distinct from ordinary misinformation in one important respect: nobody in the chain is lying. Each node is repeating something it believes it received from a more authoritative node upstream. The lie is a property of the network topology, not of any individual actor. Which is precisely why it is so hard to prosecute, and why debunking it feels like shouting into a cave.
What makes crypto uniquely permeable to this is the collapse of the discovery layer. Ten years ago, a leadership change at a protocol would surface through a handful of publications with human editors who had, at minimum, an email relationship with the entity in question. Those publications still exist, in reduced form, and they still do this work โ Crypto Briefing's correction is evidence of that. But they no longer control distribution. Distribution now belongs to feeds, and feeds optimize for engagement per unit of latency. A claim published in minute eleven outperforms a correction published in hour forty by roughly an order of magnitude in reach, and roughly two orders of magnitude in cost-adjusted attention.
In the sample I pulled, the correction reached an estimated eight to twelve percent of the audience that saw the original claim. I want to be careful about that number โ my methodology is a scrape, not a census, and audience overlap across platforms is genuinely hard to estimate. But the order of magnitude is not in dispute, and it is not new. What is new is the speed of the asymmetry. The gap between claim and correction used to be measured in days. Now the claim has fully saturated its addressable audience before the correction has been drafted.
The market response was instructive in a way the rumor itself was not. ALGO ticked, thinly, and gave it back. The tick was not a vote on William Herkelrath. It was a measurement of the order book. A rumor can only move a price as far as the book is shallow, and Algorand's book in the current regime is shallow enough that a screenshot can produce a visible print. In a market with real depth, this rumor would have died in the mentions. It moved because there was nothing standing behind it.
That is the first genuine signal in this entire episode, and it has nothing to do with Chainlink.
What Algorand Actually Is
The reason the rumor was illegible rather than merely false is structural, and to explain it I have to do a thing I usually avoid, which is recap. Bear with me, because the recap is the argument.
Algorand launched its mainnet in June 2019, built on work by Silvio Micali, who shares a Turing Award with Shafi Goldwasser for foundational contributions to cryptography. The design is genuinely elegant in a way that few Layer 1s are. Committee selection uses verifiable random functions to sortition a small, unpredictable subset of stake-weighted participants for each round, which gives the chain probabilistic security without the energy profile of proof-of-work and without the long fork-choice ambiguity that has defined Ethereum's rollup-centric era. Blocks finalize in a single round. There are no forks to resolve, no reorgs to reason about, no uncle blocks. As a piece of protocol engineering, it is closer to a well-specified distributed database than to a political settlement layer, and I mean that as a compliment.
ALGO has a maximum supply of ten billion tokens. The distribution began with a Dutch-style auction in mid-2019 at a clearing price around twenty-four cents, followed by a multi-year release schedule covering early backers, the Foundation, and a participation reward program. In 2021 the Foundation accelerated the vesting of early backer rewards, an episode that did real and lasting damage to community trust and is worth remembering whenever someone tells you that Algorand's problem is purely narrative. It is not purely narrative. Some of it is a specific, dated decision that people have not forgiven.
The price history is the standard arc of the 2019 vintage. An all-time high near three dollars and thirty cents in November 2021, a drawdown that took it below ten cents during the 2023 trough, a partial recovery through 2024 that never approached the prior peak, and then the grind of the current regime. I am not going to pretend to know where it goes next, because I do not, and because anyone who tells you they do is selling something.
Now the part that matters. 'Algorand' is not one organization. It never was, and the ambiguity is not incidental โ it is baked into how the project was legally and commercially structured from the beginning.
There is Algorand Technologies, formerly Algorand Inc., headquartered in the Boston area, which holds core protocol intellectual property and employs a substantial share of the core engineering talent. And there is the Algorand Foundation, a Singapore-based non-profit, which controls a large allocation of ALGO, funds ecosystem grants, runs governance programs, and functions as the public face of the project in most institutional contexts. The two entities are related by contract and by shared history, not by a command hierarchy. The Foundation does not appoint the Technologies CEO and vice versa.
On top of that, 'Algorand' colloquially includes the node network โ a permissionless set of validators whose operators have no institutional affiliation whatsoever โ and the ecosystem of independent teams building on top of the protocol, none of whom report to either entity. Depending on how you count, you can get to four or five distinct organizations all reasonably described as 'Algorand' in a headline.
This is not a criticism of Algorand specifically. Most large networks have some version of this structure, and for good reasons involving liability, jurisdiction, and the fact that non-profits and venture-backed companies have incompatible fiduciary obligations. But it has a specific consequence for information: the sentence 'Algorand has appointed a new CEO' is underdetermined by construction. It does not specify which Algorand, and therefore it cannot be verified or falsified without a follow-up question that the headline format does not permit.
Which is exactly what a fabricated rumor needs. Ambiguity is the rumor's habitat.
The leadership history reinforces the pattern rather than clarifying it. Steve Kokinos, who ran the commercial entity, departed in 2022 and resurfaced weeks later as chief executive of Celsius Network, in the window immediately after Celsius froze customer withdrawals. That story was true, and it was so strange that it did more to shape public perception of Algorand's leadership than any actual governance decision before or since. When people say Algorand has a leadership question, that is often the episode they are circling without naming. Meanwhile the Foundation side has been run by Staci Warden, and the organization has been through at least one significant restructuring in the intervening years, with the usual consequences for institutional memory and public communication cadence.
I want to flag something about my own posture here. I have been skeptical of Algorand's commercial execution for years, and I have said so in writing, and I have been wrong about the timing of that skepticism, which in this industry is the same as being wrong. The protocol is good. The market has never rewarded it in proportion to its quality, and the reasons for that are a tangle of distribution, community, token mechanics, and the fact that 'technically superior' has never once been a durable narrative in this sector.
Which brings us to the second real signal in this episode. The rumor found a buyer because Algorand's leadership question is, in a meaningful sense, genuinely open โ not because a CEO is missing, but because the entity structure makes it impossible to answer 'who decides' with a single name. A vacuum, in information terms, is a place where anything can be true. And nature, or at least the content industry, abhors a vacuum.
The Plausibility Engine
Why William Herkelrath? This is the question that separates a useful analysis from a shrug, because the choice of name was not random. Fabricated rumors that name real people are not cheap to produce โ they carry defamation exposure, they invite a correction from the named individual, and they can be checked. The people generating this content are optimizing for something. Understanding what they were optimizing for tells you what the market is hungry for.
Herkelrath's public profile is a business development profile at Chainlink. That matters enormously. In the taxonomy of crypto careers, the oracle business development executive occupies a specific and unusually legible niche: they are the people who spent the last several years walking into banks, asset managers, and enterprise consortiums to explain why a decentralized price feed is a prerequisite for anything interesting happening on-chain. They are, by profession, the translators between the institutional world and the protocol world.
The archetype the rumor was borrowing is real. There is a well-populated career path from infrastructure business development into senior leadership at a Layer 1. It happens often enough that the pattern is recognizable. And the specific adjacency โ oracle person takes over a chain โ carries a semantic payload that a hedge fund person or a payments person would not.
Run the inference a market participant would run, at speed, on a phone. If a Chainlink business development executive is running Algorand, then Algorand is probably deepening its oracle integration. If oracle integration is deepening, then price feeds and cross-chain messaging are becoming first-class primitives. If those are first-class, then real-world asset issuers and institutional DeFi teams have a reason to look at Algorand again. If institutions are looking, then stablecoin float and TVL might inflect. If TVL inflects, the token re-rates.
Not one link in that chain follows from the previous one. Oracle integration is a config change, not a strategy. Institutional interest has never in the history of this industry been gated on oracle availability โ it has been gated on legal clarity, custody, and counterparty demand, none of which a business development hire addresses. And a business development executive, by training and by temperament, is a person who sells a product that already works, not a person who fixes a product that does not.
The rumor was not just false. It was structurally illiterate โ it proposed a solution from the wrong category to a problem it never specified.
That is the part nobody will say out loud, because saying it requires asserting that the market misread a story it never actually read. But look at what happened. The claim circulated in a form โ a headline, a screenshot โ that carried only the narrative-adjacency payload and none of the organizational specificity that would have allowed anyone to evaluate it. The name was real. The prior employer was real. The title was real. The composite was fiction, but every component was borrowed from reality, which is how you manufacture credibility without producing evidence.
I have a name for this pattern from my NFT provenance work, where it showed up constantly. I called it credential laundering: assembling an entity out of individually verifiable fragments whose combination is entirely fabricated. In 2021 it was pseudonymous artists with real gallery representation that turned out to be a group show. In 2026 it is executives. The mechanics have not changed. The collateral has gotten more expensive.
And there is something else, subtler. The rumor was attractive in part because it was flattering to the holder. If you hold ALGO โ and in a bear market, the people still holding are the people who have been holding for a while โ a story about a serious infrastructure executive taking over is a story about your thesis being validated. It does not ask you to change your mind. It asks you to feel confirmed. Rumors that flatter the position they move are the ones that move the position.
What the Chain Can and Cannot Tell You
Here is where my actual professional practice becomes relevant, because I spent part of the last decade doing on-chain verification work and I have a routine for evaluating claims like this. It takes about ninety minutes and it is not glamorous.
First, primary source or nothing. I go to the entity's own channels โ the Foundation blog, the official X account, the corporate registry filings in Singapore and Massachusetts, the DNS records on the domains the Foundation actually uses. If a leadership change has occurred, one of those surfaces has a record. A leadership change that appears nowhere in any primary surface did not occur. This check costs roughly eight minutes and eliminates the overwhelming majority of claims.
Second, key custody. Organizations are defined less by their org chart than by their signing authority, and signing authority is often observable. Multisig configurations on treasury addresses, governance key rotations, changes to the addresses authorized to post on-chain governance proposals โ these are the mechanical traces of an actual transfer of authority. I pulled the relevant addresses. No rotation. No change in the signer set. No new proposal permissions.
Third, the repository. GitHub organization membership, commit rights, release-signing keys. Core protocol releases are signed, and the signing keys belong to specific people. A genuine change at the top of the engineering organization produces visible churn in this layer eventually, even if it is not immediate. Nothing.
Fourth, regulatory and corporate filings, which for a Singapore non-profit and a US corporation are public but lag reality by weeks to months. This check is inconclusive by design; I record it as null rather than negative.
Fifth, and this is the one that people forget, I check whether anyone is acting like it happened. Counterparties behave differently the day a CEO changes. Partnership announcements pause. Conference appearances get scrubbed from calendars. The entity goes quiet for a news cycle. Algorand did not go quiet. Which is itself weak evidence, but evidence.
Five checks. Roughly ninety minutes. And the result is the same as the correction published by Crypto Briefing, obtained by a completely different route. That redundancy is the point. Verification is cheap. It has been cheap for years. The reason it does not happen is not that it is hard โ it is that it does not pay.
Now the harder structural observation, and the one I keep coming back to.
The chain can tell you, with perfect fidelity, that a transaction occurred. It can tell you the exact state of every account at every round, with finality, without a reorg, forever. Algorand is unusually good at this. It is arguably the best in the industry at this.
What it cannot tell you is who has the authority to speak for it.
That information lives off-chain, in corporate registries, employment contracts, board resolutions, and email. It is notarized by nothing. It is verifiable only by the slow, human process of asking the entity and waiting for a response, and that process is exactly what the current information economy has devalued to zero.
We built a settlement layer with single-round finality and then ran the identity layer on top of it with no finality at all.
History rhymes, but the code doesn't. The code has no opinion about who is CEO. The code has no opinion about whether a press release is real. The code has no opinion about whether the person whose name is in the headline agreed to be in the headline. And every time we conflate the trustworthiness of the ledger with the trustworthiness of the story about the ledger, we hand the adversary the same opening.
There is a specific irony here that I cannot let pass. Algorand's entire commercial proposition is correctness. Deterministic finality. No forks, no ambiguity, no probabilistic settlement. It is a chain sold on the premise that you should not have to wait and see. And yet the question of who runs it โ the most basic governance fact an institution could ask โ is resolved by off-chain assertion and community memory. The protocol finalizes in one round. The org chart finalizes never.
That gap is not Algorand's fault alone. It is the industry's. But it is the gap the rumor walked into.
The Zero-Marginal-Cost Article
I want to spend real time on the production side, because I think the production side is where this gets solved or where it gets much worse, and most of the discourse skips it entirely in favor of scolding the consumers.
Before 2023, generating a plausible crypto news article had a floor cost โ a writer, an editor, a fact-check, a CMS. That floor was maybe a few hundred dollars for a short piece, and it imposed a hard constraint: the number of articles that could exist was bounded by the number of people willing to write them. Misinformation existed. It was expensive and therefore relatively rare.
That floor is gone. The marginal cost of a well-formed, grammatically clean, superficially researched crypto article is now effectively zero. Not low. Zero, plus the amortized cost of the API call. And the routing layer โ the aggregators, the feeds, the app widgets that surface 'news' next to a price ticker โ has no mechanism for distinguishing between content that cost four hundred dollars to produce and content that cost four hundredths of a cent.
So what gets produced? Whatever template has the best engagement-per-cost ratio. And the fabricated-executive-appointment template is, from a purely economic standpoint, a near-perfect artifact.
Consider its properties. It names a real company, borrowing the company's accumulated credibility. It names a real person, borrowing the person's verifiable employment history. It asserts a positive catalyst, which biases holders toward belief. It is short, which means it can be produced and syndicated cheaply. And it is unfalsifiable without a phone call, which means the window between publication and correction is measured in hours โ hours during which the article accrues impressions, and impressions are the revenue event.
The economics do not require the rumor to be believed for long. They require it to be believed for as long as it takes to load an ad.
I ran a rough version of this model myself last year, and I want to be honest about how it startled me. If you assume a realistic CPM on programmatic display inventory and a modest syndication footprint, the expected revenue from a fabricated executive-announcement article is positive even if it is corrected within two hours and even if the domain is eventually deindexed. The negative outcome is a domain that stops working, and domains are cheap. The correction is not a deterrent. The correction is a rounding error in the cost structure.
Now add the synthesis loop. Model A produces the article from a headline. Model B summarizes Model A. Model C generates a listicle from Model B. Each generation is cheaper than the last, and each one launders the provenance a little further, until the claim is circulating in a form whose original source cannot be located by anyone, including the people repeating it. I sampled three of the articles that appeared during the window. All three had the same sentence structure in the lede. None of them linked to a source. One of them had a fabricated quote attributed to a real person.
That last detail is the one that should worry you most, because attributing fabricated statements to a real named individual is not a content problem. It is a legal problem that the industry is not structured to litigate, because the defendants are disposable domains with no assets and no jurisdiction.
And then there is the 2026 problem, which is the one I actually spend my working hours on.
For the past year I have been modeling autonomous economic agents โ systems that hold assets, evaluate information, and execute without a human in the loop. The part of that model that keeps breaking is not the execution. Execution is solved; smart contracts have been perfectly good at execution for a decade. The part that keeps breaking is ground truth. An agent that reads a headline and sizes a position has no mechanism for distinguishing a verified claim from a synthesized one, and the verification step โ the ninety minutes I described earlier โ is exactly the kind of task that a human does slowly and that a model does confidently and wrongly.
I argued in a piece last year that human oversight would become the bottleneck in high-frequency agent-to-agent transactions. I stand by that, but I got the direction slightly wrong. The bottleneck is not oversight. It is attestation. We do not have a machine-readable, adversarially robust way for an agent to ask 'is this claim about a company's leadership true' and receive an answer with a signature attached.
When that gap closes, this class of rumor becomes expensive, because the arbitrage โ publish before verification catches up โ gets eaten by the verification layer. Until it closes, the arbitrage gets wider every year as the cost of production falls and the cost of belief stays constant.
Algorand's Actual Condition
Stepping back from the media forensics, because the underlying question is the one readers in a bear market actually care about, and it is not 'who is the CEO.' It is 'is this thing going to survive.'
I will give you the honest read, which is less exciting than either the bulls or the rumor-mongers want.
Algorand's technical fundamentals are intact and its commercial fundamentals are thin. That has been true for four years and the rumor did not change it. Total value locked is a metric I distrust in general โ it double-counts, it is gameable, and it says more about incentive programs than about organic demand โ but the trend matters, and the trend in Algorand's DeFi ecosystem through the current regime has been flat-to-down, consistent with an ecosystem that has real builders and not much mercenary capital. Stablecoin float on the chain is a better indicator because it is harder to fake and reflects actual payment and treasury activity; it is present, it is not negligible, and it is not growing at a rate that would justify a re-rating on its own.
Node distribution is genuinely one of the healthier profiles in the industry. That is a real asset in a bear market where chain liveness is being stress-tested and where the credible responses to regulatory pressure involve having no single point of failure to subpoena. Developer activity, as measured by commits to core repositories and by the number of independent teams shipping, shows a smaller, more durable cohort than the 2021 peak โ which is characteristic of the whole industry right now, and not a specific indictment.
The token mechanics matter for survival in a way that gets less attention than it should. The Foundation holds a substantial allocation, and the question that should be on every holder's list is runway: how many years of operating expense does the remaining treasury cover at current prices, and does the release schedule require selling into a market with this little depth. I have not seen a clear public answer to that question recently, and in a bear market, the absence of a clear answer is itself a data point. The protocols that survive the next eighteen months will be the ones whose treasuries are denominated in something they can actually spend, and very few chains have that.
The honest summary is that Algorand is not dying and it is not re-rating. It is a technically excellent protocol with a structurally difficult distribution problem, sitting on a narrative vacuum that it has not filled and that other people are now filling for it, badly.
Which is the thing I keep circling back to. In a bull market, a narrative vacuum is an opportunity โ someone shows up and fills it with a story about what the chain will become. In a bear market, a narrative vacuum is a liability, because the cheapest thing to fill it with is a rumor. The rumor about William Herkelrath was not an attack on Algorand. It was a scavenger arriving at an unoccupied property.
The Wrong Frame
Here is where I part company with most of the commentary you will read on this, including the commentary from people I respect.
The consensus framing is that this is a misinformation problem: bad actors produce false content, platforms fail to stop it, readers fail to verify it, and the fix is better fact-checking and stronger platform policy. That framing is correct as far as it goes, and it is almost useless as an explanation, because it treats the phenomenon as a supply-side failure when the actual driver is on the demand side.
The rumor found buyers because it answered a question that is genuinely open. Who runs Algorand, and toward what end? That question is not idle. It determines treasury strategy, it determines ecosystem priorities, it determines whether the institutional-partnership thesis gets another two years of runway or gets wound down. Institutional readers ask it constantly. And the publicly available answer has, over the past several years, become harder to reconstruct rather than easier.
When the authoritative answer to a question narrows, the market does not stop asking. It substitutes. It builds the answer out of whatever fragments are lying around โ a familiar name, a plausible adjacency, a prior that feels right โ and it writes that answer into a headline. The headline is not a lie about the world. It is a lie about the state of public information, which is a different and more interesting thing.
A rumor is a market's way of telling you it is running out of authoritative source material.
Now the counter-contrarian move, because I can feel certain readers preparing to convert the previous paragraph into a license. The fact that a rumor is directionally interesting does not make it true, and the move from 'this reflects a real anxiety' to 'therefore there is something to it' is exactly how misinformation gets laundered into consensus. It is the most common epistemic failure in this industry and it is committed constantly by people who consider themselves rigorous.
Let me be unambiguous. There is no CEO change at Algorand. William Herkelrath has not been appointed to anything. He is a real person whose name was used as a credibility prop by someone who has never met him, and the appropriate response to that is not to generate a take about what it all signifies. It is to stop repeating it.
The laundering pattern goes like this: a fabricated claim circulates. Someone observes that the claim is fabricated but adds that it reflects a genuine institutional weakness. The observation gets quoted without the first clause. Within a week the claim is 'controversial' rather than false. Within a month it is a thing that 'people said.' This is how a lie becomes a data point. I have watched it happen at least four times in the last two years with protocols whose names I will not repeat, because repeating the setup is half the mechanism.
The victims here are not abstract. There are the retail holders who bought a thin book on a screenshot and are now down on a position they never would have taken otherwise. There is Herkelrath himself, whose professional identity was appropriated without consent, and who now has a search engine problem that will trail him for years. And there is Algorand, which now has to spend a news cycle denying something nobody officially said, which is the least productive use of an institution's communication bandwidth in a bear market where communication bandwidth is already the scarcest resource.
And there is a fourth victim that is harder to name. Every episode like this erodes the only asset a non-yielding Layer 1 actually has, which is credible neutrality. If you cannot trust the chain's official statements because you cannot distinguish them from synthesized ones, you do not fall back on the chain. You fall back on nothing, and you size down. The cost of the rumor is not its direct market impact. It is the aggregate probability that the next real announcement gets discounted by people who have learned, correctly, that this industry's headline layer is untrustworthy.
What to Watch
I do not want to end on a lament, because that is cheap and I have no interest in it. Concrete things to watch, in rough order of how much they would change the picture.
Whether any of the large data aggregators implement a primary-source requirement for organizational announcements. This is a policy decision, not a technology problem, and it is the highest-leverage intervention available. A feed that will not render a leadership change without a link to the entity's own channel would eliminate most of this class of rumor overnight, at the cost of some latency. The question is whether any of them will accept that trade, because latency is exactly what they compete on.
Whether attestation registries for organizational identity get traction. The primitives exist โ Ethereum Attestation Service and its equivalents โ and what is missing is not infrastructure but adoption discipline: entities willing to publish signed, revocable claims about their own leadership, keyed to addresses, and third parties willing to treat the absence of an attestation as meaningful. This is the kind of thing that looks trivial until it becomes table stakes, and institutions are usually the ones who force it, because their compliance teams cannot operate on scraped headlines.
Whether agent-mediated news consumption arrives before or after the verification layer. I have modeled both orderings. In the bad ordering, autonomous systems trade on synthesized headlines for a full cycle before anyone builds the filter, and the market impact is no longer a thin tick on a small book โ it is a liquidity event, because agents do not hesitate and they do not read comment sections. In the good ordering, the verification layer gets built first because it becomes an arbitrage opportunity, and the people who build it make money by being right faster than everyone else is wrong.
And whether Algorand itself closes the gap it is standing in. Not by hiring anyone โ the rumor does not need to be answered with a hire. By publishing, in one place, clearly and durably, who has authority to speak for which entity, what the treasury runway looks like at current prices, and what the next two years are actually for. That is not a communications exercise. It is the governance primitive the entire industry is missing, and the protocol with the best claim to correctness is the one that should have built it first.
History rhymes, but the code doesn't. The code finalized every block this week, exactly as specified, with no forks and no ambiguity, and it did that while the human layer around it argued about who was in charge and got the answer wrong.
The ledger is fine. It is the story about the ledger that is broken.
Which raises the question I cannot answer and will keep asking: if the verification layer is not built by the people selling correctness, who builds it, and what exactly are they selling?