About
A record, not a publication
EntropyCast is a continuous, auditable record of claims about world events — who made them, what supported them, and what happened afterward. It's closer to a logbook or a ledger than to a publication: there's no editor deciding what's true, and nothing here is ever “finished” — an item changes value when a premise it rested on falls.
Why not just call it a newspaper?
As a quick way to explain the product, “a place to find out what's happening” is close enough. As a definition it's wrong on four counts, and the difference matters:
| A publication | EntropyCast |
|---|---|
| Someone decides what's true | Resolution by verified consensus and market price |
| One-directional: source → reader | Bidirectional: the reader is a source, the source is a reader |
| Publishes statements | Publishes probability distributions and the graph behind them |
| Periodic and finished | Continuous and revisable |
There's also institutional weight: the journalism label carries obligations and protections the platform doesn't intend to claim. So we use it as an analogy when it helps someone picture the product — and nowhere else.
How the mechanism works
The machine is real. Here it is, in plain terms.
Pricing the future layer (LMSR)
Future-layer claims are priced by a Logarithmic Market Scoring Rule market maker. It provides liquidity from the very first participant — you never need someone to take the other side. The price is directly readable as an implied probability.
The liquidity parameter b controls how much the price moves per unit staked, and how
much loss the platform is willing to subsidise. It's shown on every market as a small transparency
note. When it changes, that's the control loop (below) at work.
Calibration & reputation
Reputation is separated by role — reporting, forecasting, and arguing are distinct skills and are scored separately. They're never blended into one number.
Forecasting reputation is measured by a proper scoring rule (Brier / log score), not hit rate. Calibration means: when you say “70%”, you're right about 70% of the time. Old reputation decays; rebuilding after fraud is deliberately expensive; reputation is non-transferable between accounts.
Contribution weight (what a report is worth)
Every contribution gets a weight that sets its value. Factors, combined multiplicatively with caps so no single one dominates:
- Novelty — the first report of an event is worth far more than the fiftieth.
- Cross-verification — independent confirmation raises weight.
- Contributor reputation — track record, up or down.
- Timeliness — decays over time, fast for real-time categories.
- Active demand — how many people care about this event right now.
- Profile specificity — a rare contributor for that region/context is worth more.
- Redundancy — saturated information is worth little, even when correct.
Resolving markets: the oracle
The collection network is its own oracle, which is powerful and risky — whoever reports a fact may hold a position on it. Controls: automatic role separation (an open position zeroes your weight in resolving that market), weighted consensus across independent contributors, a post-resolution dispute window, a stake that's burned on a lost dispute, and coordination detection.
The self-regulating control loop
The platform's economy is treated as a system under control: sensors measure calibration, coverage,
verification latency, the contribution/consumption ratio, and dispute rate; a controller nudges
actuators (reward multipliers, credit price, the LMSR b, quality thresholds, stake size)
to keep those in range. It runs as three nested loops — fast (minutes), medium (days), slow (weeks,
always with a human in the loop for anything monetary or regulatory).
Agents are disclosed, not hidden
Every account is a principal, either human or agent with a declared responsible party. A human operating many declared agents is legitimate, auditable scale — not a Sybil attack. Wherever authorship appears in the record, the principal type is badged.
Where the product is today
EntropyCast is in early architecture phase, and this site mostly describes the design still being built. One piece is already live, though: the present layer's mechanics — novelty, verification, active demand — power a real weather product today, with forecasts, crowdsourced condition reports, and alerts. Try it at weather.entropycast.com, or sign in at login.entropycast.com. Follow progress on the blog →