Quick Takes
The site reads the feeds of crypto outlets and of official sources such as the Fed and the SEC about every half hour. An AI model (Llama 3.3 70B on Cloudflare Workers AI) groups the stories by event, keeps those with an IMPA score of 6 or more, about eight a day at most, and drafts each Quick Take from the text of its story, using only the facts the story states.
- Every Quick Take credits the outlet whose reporting it summarizes and, when there is one, the official document behind it.
- A draft with a number that is not in its source, or with a sentence copied from it, waits for my review.
- I review the drafts in my dashboard. One I do not review goes out two hours later, never between 4:00 and 12:00 Buenos Aires time, and says on the page that it was drafted with AI and not reviewed yet. Search engines are asked not to index it until I review it.
- The Quick Takes I write or review carry my name, and only those go to the Telegram channel as alerts.
The IMPA score
IMPA, from 0 to 10, measures the impact of a story on the crypto market. It is the sum of four parts:
- Magnitude, 0 to 3: the size of the move or of the number against expectations or history.
- Breadth, 0 to 3: one token, protocol, or company (0 or 1), a sector such as DeFi, stablecoins, or miners (2), or Bitcoin and the whole market (3).
- Surprise, 0 to 2: expected (0), partly expected (1), or a shock (2).
- Persistence, 0 to 2: noise for a day (0), weeks (1), or a change in the outlook, the rules, or the flows (2).
A 9 or 10 is an exchange collapse or a record hack, a stablecoin losing its peg, an ETF approval or rejection, a Fed surprise, or a landmark law or ban. A 6 to 8 is a large ETF flow, a major upgrade or exploit, a data surprise that moves crypto, or a key regulatory step. Most crypto news scores 4 to 7, and a 9 or 10 happens only a few times a month. A Quick Take of 7 or more goes to the Telegram channel as an alert.
The Morning Brief
Its numbers are captured by the site at 01:00 Buenos Aires time and again at 07:40, from the public sources named under them: crypto over the previous 24 hours and US markets at their last close. The agenda comes from the calendar, and the topic of the day is mine.
Morning Brief calls
- Each brief can make one call: Bitcoin, ether, or Solana ends a UTC day above or below a level, with the probability I give it, from 50% to 99%.
- The call is written before the brief goes out and cannot be changed after that.
- It is scored with the last price the site saved that UTC day. Prices are saved every 30 minutes, so the last one is from 23:30 UTC. A day without a saved price leaves the call unscored.
- The track record shows the hit rate, the Brier score (the mean squared gap between the probability and what happened: 0 is perfect, and calling everything at 50% gives 0.25), and the calibration: of the calls made at about 70%, about 70% should come true.
Investment theses
Each thesis is published with its entry, target, deadline, and invalidation level, reviewed every Friday in its log, and scored in the track record when it closes: target hit, failed, or closed early. How the return is calculated is in the disclosures.
Data pages
The data pages update by themselves from the sources named on each one, and work out the change over the last week, month, and three months from the same data. The note at the top of each page is mine, rewritten every week on the day of its market, and shows the day I wrote it.
My positions
A brief or a thesis says at its end whether I hold the assets it discusses, as of the day it was published. The general disclosures, including conflicts of interest, are in the regulatory disclosures.