Let's kick this off.
Noisy is a bet. I was on a ski trip a few weeks ago, with my son pretending to be asleep on the pull out while an old friend from high school and I massaged cannabis cream into our tired thighs (I'm a snowboarder; my thighs felt worse) while nursing zero cal beer from Quebec. One or the other of us expressed our annoyance, our disdain, our frustration that there is so much noise out there in the media designed to attract eyeballs or "drive engagement" and that there is, probably, a lower layer of signal underneath all of that noise, and if you could tap into that signal, you could maybe, if you were lucky, get a sense of what is really happening.
To mix a couple pop phrases, the truth IS out there, but it's not evenly distributed.
And even when it is distributed, it tends to land in a 400-page regulatory filing with a title like "Regulatory Capital Rules: Regulatory Capital and Standardized Approach for Risk-Weighted Assets." Nobody asked if that was a good title. Nobody seems to have asked if any of this needed to be readable. People put on their "I'm writing a policy paper" hat, and even when that's not their intention, they start writing this way. The hat goes on and the person disappears.
There used to be people whose whole job was to read this stuff. They worked at think tanks, or they charged $2,000 a year, or they burned out on a single beat. Noisy is an attempt at something else: one operation, a range of domains, general audience. An experiment, honestly. We don't know yet what it becomes.
If you scan this first issue it won't escape your notice that we're finding a lot of AI-related signals. That's partly real — AI is probably the largest structural shift happening across multiple domains simultaneously right now — and partly a pipeline artifact we're actively working on. The more interesting observation is where these signals are coming from: not just research preprints, but government procurement notices, World Bank reports, education policy filings, central bank papers. The topic has spread into every kind of document we read. We read a deliberately wide range of sources for exactly this reason — when the same structural shift shows up in a regulatory filing and a research paper in the same week, that's a different kind of signal than either one alone.
Signals are the items we surface from the feed — individual documents from sources most people never open: government data releases, regulatory filings, research preprints, central bank papers. Each one went through our pipeline, got scored, and landed in the queue because something in it looked structurally significant. Not dramatic. Not breaking news. Just a document that quietly changes what's possible, or what's true, or what the baseline is.
The US Energy Information Administration — a government agency that publishes energy data releases essentially nobody reads — reported that utility-scale battery storage capacity jumped from 6,446 megawatts in August to 43,232 megawatts in January. Whether that reflects new installations, reclassification of existing assets, or some combination, the number that utilities are planning their grids around has permanently changed. The forecasts built on last year's assumptions are now working from the wrong baseline. Read more →
We also run what we're calling pattern detection — a separate pass that looks across everything we've read in a given week for moments when multiple documents from completely unrelated places converge on the same structural finding. When that convergence is real, it tends to be more significant than any single document.
Three separate research groups found, from different angles, that AI tools make people faster in the short run and measurably worse at the underlying skill over time. Ten minutes with an AI assistant makes people give up sooner when it's taken away. AI-assisted peer review may be eroding the expertise of reviewers themselves. A formal economic model suggests the skill decay accelerates as adoption spreads. The US Department of Education, apparently working from a different reading list, announced the same week that it would now prioritize AI tools in school grants over evidence-based interventions.
Nobody appears to have read the papers. See the full pattern →
That's four domains — cognitive science, scientific publishing, economics, education policy — all pointing at the same structural tension in the same week. That's what we're looking for.
Noisy is a public ledger of signal detection. The homepage is today's newspaper. The archive is the track record. If you discover this in 2028 and scroll back, you should see: Noisy was reading the obscure reactor filings and procurement notices before that, eventually, told the tale.
Weekly. Free. For now.