About · Meadow Registers

Hi, I'm Victor.

I'm an entrepreneur in data and design, and the founder of Meadow Registers. Here is why and how I build it.

Victor Seryakov outdoors at sunrise

01Why this data

I first started collecting data for my own projects. Along the way I saw how powerful a system AI agents can build — I'll show you how it works below. So I asked myself: where could I be most useful? That's how Meadow Registers began.

I chose areas that matter to me: health, energy, computing, the materials we reuse. Their registers show problems we already live with — a medicine that runs short, a price that jumps, a power line that can't take more load. Better decisions here make the world a little better.

Funds, companies and governments make their decisions on this kind of data. I hope the history I keep helps someone make a few of those decisions better.

02What we collect

Four areas, read every day:

Medicine — drug shortages from national regulators.
Compute — GPU cloud prices and the rules around them.
Recycling — prices of recovered materials and producer registers.
Grid — queues to connect data centers and new load to the grid.

So far we have found 5,780 public sources in 196 countries. 105 of them can be read by a machine every day — those are in the collection. Every week we find more, 1,118 this week, and the readable ones join.

03One person, but a team

As I write this, it's been 73 days since I started. In that time, AI agents have written about 50,000 lines of code that now read 105 public sources every day. These days, about 20 changes ship daily — the pace of a full team, with one person behind it.

04Data quality — five levels

I believe AI changes data quality too. Self-driving cars already crash less than people: Waymo reports 82% fewer injury crashes. I see the same in code: an agent doesn't get tired, and it checks its own work before every run. And when fixing a reader takes minutes instead of weeks, a broken source gets repaired the same day.

Level 1 · Primary sources only. The publisher's own register or price list — never a repost, never an aggregator.

Level 2 · Every run checks itself. The agents don't touch the data: they write the code, and ordinary code reads the registers. Each register is read up to four times a day, with 2,090 checks before each collection, and every record is compared with the day before. If a day is missing, it stays marked as missing — I never fill it with a guess.

Level 3 · Agents watch for anything unusual. Eight times a day, agents review every run: a sudden jump or drop, a changed format, an empty answer. If a read fails, the system retries during the day. If it still can't recover, the agents fix the reader, or I get an alert and fix it myself.

Level 4 · Sealed every day. Each day's data gets a cryptographic fingerprint anchored in a public blockchain. Anyone can check that a record existed on that day and hasn't changed since.

Level 5 · Trust my agents, but verify. Every week I pick a few registers from different countries and compare our records with the source, line by line. So far I haven't found a mismatch.

Still, mistakes happen. Recently our homepage said a nasal spray had left the US shortage list. In fact, its maker had just changed its name. We fixed it the same day.

05How we can work together

Have a question, or need a custom dataset or long-term collection?

Let's talk: hello@meadowregisters.com