6:40 a.m. Monday, Auckland. Fourth coffee, third screen.
I'm doing the thing I've done every morning for twelve years — four aggregators, two Telegram channels, a handful of RSS feeds I still trust, hunting for the one line of code or the one filing that hasn't hit the wider market yet.
And there it is, wedged between a sequencer outage thread and a stablecoin depeg post.
Michael Carrick. Press conference. Old Trafford. Fixture congestion. Squad rotation.
I read it twice. Then I opened the page source. Checked the canonical tag, the JSON-LD, the sitemap. Same CMS. Same taxonomy tree. Same publisher header. Filed under the same category bucket as everything else in a feed that, for the last six years, has been a genuinely reliable place to learn that something in crypto broke before it broke on the chart.
Not one angry quote-tweet. A couple hundred views. Zero comments. And by 8 a.m. it had been pulled into two AI summarizer pipelines and pushed out to a newsletter that lands in a few hundred thousand inboxes.
That's not a content problem. That's a plumbing problem. And in a bear market, plumbing is the first thing that gets value-engineered.
The part nobody wants to write down
Let me be precise about what I actually observed, because "crypto media is cooked" is a lazy take and I don't write lazy takes.
What I saw was a classification failure inside a revenue-stressed pipeline. Three things have to be true at once for a football story to land in a crypto feed, and all three are true right now.
Crypto publishing revenue is a derivative of token prices. Display CPMs are set by crypto advertisers whose budgets track market cap. Sponsored placements and affiliate deals dry up when exchanges stop spending on user acquisition — which is exactly what happens in a drawdown. In the last cycle I watched three outlets I read daily cut editorial headcount roughly in half within two quarters of the peak. That isn't a rumor. That's a pattern I've now watched twice.
When you can't raise price, you raise volume. More articles, shorter, cheaper, and optimized for discovery rather than accuracy.
And then the pipeline itself. Modern crypto news isn't written. It's routed. A wire item or a press release goes into a summarizer. The summary goes into a classifier. The classifier assigns tags. The tags decide which feeds the piece lands in. The feed gets consumed by humans, newsletters, and bots — in that order of decreasing patience.
At no point in that chain does a person say "this is about football."
The taxonomy is where it actually breaks. Most of these category trees were designed between 2018 and 2021, when crypto's universe was finite and legible: DeFi, NFTs, Layer 2, regulation, mining, exchanges. Then the industry sprawled in every direction and nobody pruned. Publishers solved that mismatch the way publishers always solve it — stuffing. Every article gets every plausible tag, because tags cost nothing and search traffic is a lottery ticket.
I know this instinct intimately. In 2017 I published five hundred words on the Ethereum Classic hard fork within fifteen minutes of the split, working off a Telegram voice chat and a block-timestamp discrepancy nobody else had called yet, from a hacker house in Austin with bad wifi and worse air conditioning. That post made my career. It taught me the rule I've optimized my entire professional life around: speed beats perfection.
But — and I've spent three years chewing on this — speed also beats verification. And a market that rewards speed over verification eventually builds a feed that is optimized for speed over verification.
That's the feed I was reading on Monday.
What I actually measure
I keep a private index. Twelve crypto publishers, tracked weekly. Nothing sophisticated: article counts, median word count, tags per article, and the ratio of original reporting (named sources, direct quotes) to derived content (aggregation, summarization, wire rewrites).
Over the last eighteen months, directionally: median article length down by roughly a third. Tags per article up more than twofold. And the derived-to-original ratio moved from around 3:1 to somewhere north of 9:1.
None of those numbers is catastrophic alone. Together they describe a system that has quietly changed what it is for. It used to be a discovery layer. It's becoming a routing layer — and routing layers don't care whether the payload is true, only whether it's formatted correctly.
Once a feed becomes a routing layer, misclassification stops being an embarrassment and starts being a market structure issue.
I wanted to know how fast the rot propagates, so I ran a small test. I took a wire item from a non-crypto beat, pushed it through the same open-source summarizer-and-classifier stack that a handful of mid-tier publishers use, and watched it land in a crypto-adjacent feed in eleven minutes — tagged with four categories, none of which described the article. Eleven minutes. No human in the loop. If you've ever wondered how fast a mislabel beats an editor, that's the number.
Here's the part that should worry you.
Most serious desks and a rapidly growing slice of retail don't read news anymore. They ingest it. Headline parsers. Entity extractors. Sentiment scorers. These systems were trained on a corpus where crypto publications talked about crypto. They have no concept of "wrong category," because a category is supposed to be a guarantee, not a guess. They see tokens of text inside a trusted feed, extract entities, score sentiment, and route an order.
I found this out the expensive way. In 2026 I spent a week running autonomous trading agents on a testnet — not auditing the code, actually running it, sitting there watching decisions print. The thing that killed every single agent wasn't the model. It was the input.
One agent took a position on a rumor it had "read" in a syndicated item that traced back to a single anonymous Telegram post from 2023. Another confidently parsed a "network upgrade" headline that was about a mobile carrier's coverage rollout. A third did something I still can't fully reconstruct from the logs, which is its own kind of lesson.
Speed isn't a number. It's about feeling the market before the market feels itself. An agent doesn't feel anything. It just reads — and it reads faster than you can correct it.
Add entity collision on top. Football club names are tickers. Arsenal, City, Barcelona, Juventus and PSG all have Socios fan tokens. Manchester United, as far as I've been able to verify, does not — which is its own story, and I'll come back to it. But the collision surface is real: a competent extractor sees "City" inside a sports sentence and has to decide whether that's Manchester or a governance token. It will pick wrong sometimes. It will pick wrong at machine speed.
So no, "just filter better" isn't the answer. Filters are bets on taxonomy stability, and the taxonomy is already gone.
The fix exists. Almost nobody uses it.
There is a technology for this. It's just pointed at the wrong problem.
Ethereum Attestation Service gives you a cheap, verifiable way to sign a claim and publish it. C2PA content credentials — the thing camera manufacturers and Adobe have been shipping for years — give you a signed provenance chain for media. Farcaster and Lens both give you a publisher identity that can attest to its own output and build a portable reputation that isn't rented from a CMS vendor.
Stack those together and you get something obvious: an article signed by the publisher's key, with an attested classification, verifiable by any consumer — human, newsletter, or agent — before it gets routed anywhere.
Nobody ships it at scale. The reason is the reason most infrastructure ideas die: cold start plus complexity. Everyone has to adopt it for it to matter. The publishers with the worst taxonomy have the least incentive to expose their own classification. And the implementation has the same shape as Uniswap V4 hooks — beautiful, general, and requiring a specialist to review before you'd trust it with anything real. The complexity spike scares off the developers who'd need to build it, which is the same trap that keeps enormous classes of protocols permanently niche.
There's a boring economic reason too, and nobody says it out loud. Provenance attestation makes a publisher's mistakes permanent and attributable. The current system makes them invisible and deniable. You are asking a business to voluntarily install a permanent public record of its own error rate, during the worst advertising market in four years, in exchange for a benefit that only materializes once its competitors do the same thing. That's not a technology problem. That's a coordination problem wearing a technology costume.
We've been here before with things we keep telling ourselves are one upgrade away from mattering. The Lightning Network has been half-dead for seven years — routing failures, channel management, an experience that never quite becomes invisible. Meanwhile we spent two years arguing about data availability layers for rollups that don't consistently fill their blobs. Billions in FDV, whole thesis documents about the cost of publishing bytes, a genuinely elegant research program.
And the information layer that actually moves prices — plain text in a headline — has no availability guarantee, no attestation, no provenance, no penalty for being wrong. It has an RSS feed and a tag model built by an SEO consultant in 2019.
That asymmetry is the story. Not the football article.
Meanwhile, the football part is not a non-sequitur
Here's the thing I couldn't shake.
The football story didn't land in a crypto feed by accident of taxonomy alone. It landed there because football is one of the few consumer verticals that genuinely built onchain rails — and those rails are bleeding out right now, quietly, in a way that maps almost perfectly onto the rest of this bear market.
Fan tokens. Chiliz, the Socios ecosystem, a few dozen clubs. Most of these tokens are down somewhere between 90% and 98% from their 2021–2022 highs. Liquidity in the smaller names is thin enough that a fifty-thousand-dollar sell can move the book double digits. And the underlying product — vote on the goal song, vote on a jersey accent — is genuinely fun and genuinely free to replicate.
The mechanism isn't that fan tokens failed. It's that they succeeded at distribution and failed at retention — which is the failure mode we keep misdiagnosing as a marketing problem.
Half a billion people care about these clubs. Millions of them bought a token. Almost none of them came back twice, because the second visit had nothing to hand them that a free Twitter poll doesn't already hand them.
When the chart collapsed in May 2022, during the Terra unwind, I didn't write a post-mortem. I called eleven people and asked how they were doing. That week taught me something that has held up ever since: the second question people ask after a crash is "what's safe," and the first is "who else is still here." Fan tokens have an answer to the second question and none at all to the first.
And Manchester United — the most valuable football brand on earth, a club whose commercial machine prints money in every currency you can name — still doesn't have a fan token. Almost every other superclub does. If you want a clean tell for whether the football-crypto bridge retains any institutional conviction, that's it. The largest possible issuer looked at the last five years and passed. The most valuable franchise in the sport did the diligence and said no. That's not a pause. That's a verdict.
The contrarian read
Everyone who notices the Carrick article will write the same piece. Crypto media is broken, standards have collapsed, AI is eating journalism.
That's the wrong read. The right read is that a mismatched feed is the rational output of a system that is correctly pricing its own audience.
Do the arithmetic. Football has something like four billion fans. Crypto has maybe forty to sixty million wallets transacting monthly, and a much smaller number of people who read crypto news every day. If you run a publisher with a declining ad business, a CMS with no hard category gate, and access to wire content that indexes for enormous search volume, the algorithm is doing exactly what you asked it to do.
The mismatch isn't a bug in the algorithm. It's a bug in the goal.
And the second-order read matters more. Community buzz wasn't the canary. Its absence is. In 2021, a mis-tagged article would have produced a hundred dunk tweets inside an hour, three quote-thread pileons, and a publisher statement by lunchtime. In this bear market the feed is quiet. Retail attention rotated out. What's left reading the wire at 6:40 a.m. is bots, a few thousand professionals, and me.
That's the real damage. Not that bad content got published — bad content gets published in every cycle, in every market, in every industry. It's that the correction mechanism has thinned to the point where misclassification can sit in a feed for weeks without anyone who has both the incentive and the authority to fix it noticing.
And there's a price-discovery tax underneath it that nobody wants to price. If the long tail of tokens is quoted off sentiment derived from a feed whose taxonomy is unreliable, then the marginal buyer is paying spread to whoever understands the plumbing better. That's not a media critique. That's a market structure critique, and it hits retail first, because retail is the last group to notice that the label on the box stopped meaning anything.
Distraction is a luxury we can't afford down here. Neither is a feed you can't trust to know what it's talking about.
What I'm watching
Tag-to-content ratio across the four aggregators I still read. If it tightens, publishers are reinvesting in editorial, which is a leading indicator that the ad market is thawing. If it keeps widening, the routing layer is permanent and we should all stop pretending the feed is a discovery tool.
Whether anyone ships signed content provenance. My bet is a Farcaster-native outlet does it before a legacy publisher, because reputation you actually own is the only axis where a small publisher can out-compete a large one on day one.
And fan token liquidity depth. If the bid disappears entirely, the football-crypto bridge closes for a full cycle — and it won't reopen with the same counterparties, the same clubs, or the same pitch decks.
If you're building anything in the information layer right now — an agent, a feed, a newsletter, a screener — internalize this: you can't wait for the signal, it becomes the signal. Every filter you add is a bet that the taxonomy underneath it is stable. It isn't. It's a 2019 spreadsheet with 2026 content poured through it, and the spreadsheet is winning.
The fixture congestion at Old Trafford is not my problem.
The fact that it reached me — unfiltered, on a Monday morning, through a pipeline I've spent twelve years calibrating — is.
And if the plumbing can't tell football from financing, what exactly makes you confident it can tell a real sequencer outage from a rumored one?
