Calibrated probability cones.
Provenance-sealed forecasts.
Quants and risk PMs get Oracle probability cones with full provenance. Every output sealed. Provisional until Gate 1b clears. See also: Path Queries →
Run a calibrated forecast — right here.
Pick a target actor, a vertical, and a horizon, then run the forecast: a synthetic 23-agent / 5-Sister swarm settles into a merged cone, a descending scenario list (Σp ≈ 1, entropy in nats), each Sister's vote and weight, a Brier / log-loss calibration chip, and a sealed manifest ref. Change the inputs and re-run — the cone moves. Fully interactive, fully synthetic — no account needed.
Forecast Terminal
Booting the swarm…
Synthetic demo · certified provenance, not certified truth · ROADMAP
The probability cone. Provisional until Gate 1b.
Five Sisters, 23 agents, one merged cone. Stroke weight encodes probability; opacity encodes confidence. The envelope is the plausible-futures band at the horizon.
5 Sisters. 23 agents. One normalizer.
Analyst, Contrarian, Disruptor, Historian, Institutionalist each imagine futures independently and return an unnormalized SisterOutput. deino-oracle is the single place probabilities normalize — it has no ledger dependency, so it physically cannot read or corrupt what it normalizes. deino-eval closes the Brier/log-loss calibration loop.
Calibrated ensemble
Five sisters. 23 agents. One merged probability cone. sum(p)≈1. Entropy in nats. Sorted descending. Provisional until Gate 1b.
Provenance-sealed
Every forecast output seals a per-output hash-chained manifest. Same inputs. Same output. Exactly replayable.
Defend to regulators
When a model-risk officer asks why P50 moved 12 points, you replay the manifest. Every agent version, every timestamp, on record.
Certified provenance ≠ certified truth.
Every output replayable. Not every output correct.
Forecasts with provenance.
Market Prediction side is frozen behind Gate 1b. Pre-register for early access when it opens.