Last month a client forwarded me a research note. Twelve pages. Nine tables. In six of those tables, every cell read "N/A — insufficient information."
The note had a disclaimer, a risk matrix, a glossary of technical terms, and two pages of regulatory caveats. It had no finding. No transaction hash. No contract address. No number that couldn't be lifted from a price widget. I ran it through the novelty-entropy script I originally built to strip wash-trading artifacts out of BAYC secondary sales. It scored 0.03 bits per sentence. A CoinMarketCap ticker scores 0.11. The note was less informative than the thing it was describing.
That shouldn't be possible. A human wrote it, a human approved it, someone paid for it. So I did what I always do when a number looks wrong: I assumed the anomaly was mine, and I went looking for the rule it had broken.
Then I pulled 1,200 more.
The market has been sideways for eleven weeks. Chop. No trend to report, no breakout to explain, no liquidation cascade to narrate. When price gives you nothing, the research industry has two options: publish less, or publish anyway. It chose the second one. It always does.
Between January and March of 2026, I scraped 1,200 published crypto research notes from 14 outlets — three tier-one financial publications, four crypto-native terminals, five paid newsletters, and two "intelligence platforms" charging institutional seat rates. I stripped headers, disclaimers, boilerplate methodology sections, and legal footers. Then I tokenized what remained.
The sector I'm auditing is structurally strange right now. Post-MiCA, stablecoin issuers publish reserve attestations on a schedule. Post-ETF, flow data lands at 4 p.m. ET whether or not it says anything. Post-agent, most on-chain volume isn't human. The supply of "things that happened" is roughly flat. The supply of "reports explaining what happened" has roughly tripled in eighteen months. That gap has to be filled with something. My question was: with what?
The templates are the tell. The nine-dimension scoring framework — technology, tokenomics, market, ecosystem position, regulatory, team, risk, narrative, supply-chain transmission — has become a de facto industry standard. I have now seen it in four different outlets with cosmetic renames. It's a good framework. It is also forty pages long, and a forty-page framework applied to a week in which nothing happened produces forty pages of "insufficient information." The template doesn't detect the absence of news. It absorbs it.
For each of the 1,200 notes I measured four things. Sentence-level novelty — how many sentences contained a claim absent from that publisher's prior 90-day corpus and absent from the other thirteen outlets. Hedge density — tokens like "could," "may," "potentially," "notably," "remains to be seen." Placeholder density — cells, bullets, or sentences containing "N/A," "TBD," "insufficient information," or bracket text. And on-chain groundedness — the presence of at least one verifiable artifact: a hash, a contract address, a block number, or a reproducible query.
The results were not subtle. 41.3% of the reports carried placeholder density above 25%. Nearly half of what shipped was scaffolding wearing a suit. 28% contained zero on-chain artifacts. No hash. No address. No callable state. Not one. Median sentence-level novelty across the full corpus was 6.2% — strip the boilerplate and the macro recap and 93.8% of every note is a restatement of something already published somewhere else.
One note deserves a name I won't give it. Tier-one publication, 22 pages, of which 14 were a regulatory section that concluded, in its own words, that "no material regulatory change occurred during the period." Fourteen pages. Zero bits.
To be fair to the corpus, the top decile was genuinely good. Notes carrying two or more on-chain artifacts and a reproducible query averaged 31% novelty — five times the median. They were also, predictably, the shortest. Median length 4.2 pages, against 11.6 for the bottom decile. In this corpus, length was inversely correlated with information. The best reports were the ones that stopped early.
The correlation that actually stopped me, though, was citation. Notes with at least one on-chain artifact scored 3.4x higher downstream reference — other analysts quoting them, dashboards ingesting them, terminals re-publishing them. Notes with zero artifacts were quoted almost never, yet published at the same rate. The market was paying for output volume and recycling only output substance. Those are not the same product, and nobody had noticed they'd been unbundled.
The second-ugliest correlation: zero-artifact notes spiked on days when on-chain activity was low. Not high. Low. When there was nothing to measure, output went up. Volume spikes don't correlate with information; they correlate with calendar pressure.
I've seen this shape before. In 2020 I scraped 5,000-plus Aave governance votes and found 12 entities controlling roughly 15% of voting power — a system with the appearance of distributed decision-making and the mechanics of a small committee. This corpus is the same architecture. A nine-dimension template, faithfully filled out, produces the appearance of rigor and the mechanics of a checklist. The template returns "N/A," and the analyst ships it, because shipping is the job.
The code doesn't lie. Templates do — by omission, structurally, at scale. When I traced the Parity wallet drain in 2017, I had four weekends, 14 wallet clusters, and one hard constraint: every claim had to survive a hash. That constraint is the entire discipline. Remove it and you get 1,200 documents that describe a market without ever touching it.
Here is where I have to argue against my own data.
The obvious read is that AI slop has colonized crypto research. That read is comfortable and mostly wrong. I checked publication timestamps against bylines and revision histories: 71% of the zero-artifact notes were human-authored and human-edited. The slop wasn't generated. It was requested.
The demand curve is the product. In a sideways market, readers don't buy information — they buy the feeling of being positioned. A report that says "nothing measurable changed this week" satisfies no one. A report with nine tables satisfies everyone and commits to nothing. We don't have a research problem. We have a demand problem wearing a research problem's clothes.
And there's a version where the N/A is honest. A framework that returns "insufficient information" is a framework refusing to fabricate. That's correct behavior. The failure is that it shipped anyway, with a cover page and a price tag attached.
Which brings me to the part that isn't philosophy. My 2026 work on agent-to-human interaction ratios showed roughly 40% of DeFi lending activity is already initiated by algorithmic agents. Those agents retrieve. They ingest corpora. A knowledge base seeded with 41% placeholder-density documents isn't neutral — it's a noise injector with a structured schema, which is the worst kind, because humans discount filler instinctively while retrieval systems rank it by formatting quality. Nine tables and a risk matrix look exactly like signal to a retriever.
We don't have time to fix the writing. The ingestion is already happening.
Watch citation half-life, not publication volume.
I'm tracking one number over the next four weeks: how long a note keeps getting referenced after it drops. Reports with on-chain artifacts were still being cited 19 days out. Reports without them died in 40 hours. If that gap widens while publication volume holds flat — and my model says it will — the market is quietly re-pricing substance for the first time in two years.
Between the hash and the human, there is a silence. Most of what's published right now lives inside it. The question isn't whether it gets called out. It's whether anyone is still reading closely enough to notice.