How Morrowfly works, in the open.

Morrowfly follows world events: a closed strait, a drought, an export ban. For each one it traces the path the shock travels, through chokepoints and commodities, to the markets pricing it. Then it publishes a view, and when the outcome lands, the view is graded in public, next to the ones that missed.

It keeps a smaller set of stories and follows them for months, reporting what changed and what it got wrong.

How we grade ourselves

A view only counts if it was written down before the outcome could be known. Every call is registered with our probability, the market's probability at the time, and the exact condition that settles it. A weekly Bookkeeper grades what has resolved. Probability calls get Brier scores, published as they land; the first resolve on Oct 31 and Dec 31. Band calls are graded hit, miss, or partial against the criterion written down when the call was made.

When a record is wrong, the original stays up, the change is logged as a revision, and challenges are answered on the record itself. If we do not have the data, the page says coverage is incomplete; a missing feed is never scored as zero risk. If we research a market and decide not to take a view, we publish that too, with the reason.

The record so far: 51 graded calls, every one public. Read them on Receipts and the track record.

How a story is built

Every story starts from records, not from a headline. Ship traffic comes from IMF PortWatch, trade shares from UN Comtrade, prices from Polymarket, and a daily monitor scores supply stress across 17 commodities. Underneath, a knowledge graph connects events, scenarios, commodities, chokepoints, and markets, so a claim on any page can be traced back to where it came from.

Every story follows the same six steps: the shock, how it transmits, who is exposed, our view, the market, and the graded outcome.

Who is behind it

Morrowfly is built by Harsha Vashisht, its founder. At Google he led ML integrity and directed anti-abuse engineering for infrastructure running at millions of queries per second during the Gemini era; he also led engineering for YouTube Podcasts. Before that, he helped build Uber Eats from launch to more than 140 cities. He was also cofounder and CTO of Dyssonance AI, an AI lab for continuous reasoning under uncertainty. The insistence on graded calls and public corrections comes from that background.

Harsha's focus is self-learning loops: systems that get better the longer they run. Morrowfly is built as one. Every event it tracks, every call it makes, and every outcome that follows is recorded, graded in public, and folded back into a knowledge graph that grows a little sharper with each pass. The knowledge accumulates, the lessons carry forward, and the system keeps learning continuously, not in releases but every day. The Morrowfly you use next year will know more than today's, because it will have learned from everything in between.

What Morrowfly is not

Morrowfly is not a trading platform: you cannot buy or sell anything here. It is not financial advice: the views are published research with a public score, nothing more, and all trading here is paper trading, with no real money at stake. And it is not a news site.

Contact

The founder reads every message. For partnerships, corrections to the record, or questions about the data behind a page, write to harsha@morrowfly.com.

Harsha is also on LinkedIn.