Market forecasts with a public, on-chain accuracy record. Every prediction is hash-committed to X Layer before its window closes, scored automatically against what actually happened, and published — losses included.

Start here

The 60-second version

Every forecasting product is unfalsifiable at the moment of sale. You're told what will happen, you pay, and nobody ever tells you whether it was right.

Horos writes every prediction into a hash-chained ledger, commits the chain head to X Layer before the outcome exists, and scores it automatically when the window closes. The record is public and free to read.

• 40 services — 39 A2MCP + 1 A2A
• 40/40 bought with real USD₮0, 40 distinct on-chain settlements, 125/125 outcome checks
• Anyone can commit their own forecast to the same ledger and be scored by the same code
• No backtest, ever — the model's pre-training contaminates one
• Right now the record says the model is losing to a random walk by 26%. It says that itself.

That last line is the product working. Read on for why.

The problem

Ask any forecasting service for its track record. You'll get a cherry-picked backtest, a screenshot, or silence.

This isn't dishonesty by a few bad actors — it's structural. Nothing forces a forecaster to publish its misses. The incentive runs entirely the other way, and the buyer has no mechanism to check. So the market prices confidence instead of accuracy.

A forecast is unverifiable when you buy it and perfectly verifiable a day later.

That gap is the entire opportunity. Close it, and forecasting becomes a market where being right is worth more than sounding right.

And the move that decides whether this is a good business or a category: do it for everyone's forecasts, not just ours. An agent choosing between five market ASPs needs a referee more than a sixth forecaster.