Belief should change
when evidence changes.
EIDOS treats price as a belief problem rather than a declaration of truth. For each selected Solana market, the engine defines 256 candidate price states, begins from a uniform prior, and updates the weight of those states as indexed market observations arrive.
Why 256 states?
A market price is continuous, but an inspectable model benefits from a finite hypothesis space. EIDOS divides the observed price domain into 256 states. At maximum uncertainty, each state receives probability 1/256. A uniform distribution across 256 outcomes has exactly 8 bits of Shannon entropy.
UNIFORM 256-STATE PRIOR = 8 BITS
MAP is not truth.
The maximum a posteriori state is simply the most supported hypothesis under the current evidence, likelihood model and price domain. EIDOS displays MAP beside entropy and the 95% credible range because a point estimate without uncertainty creates a false impression of knowledge.
Surprise and correction
Surprise measures how improbable a new observation was under the belief that came immediately before it. Contradictory market activity may increase entropy. That is not a failure of the engine; it is the system recognizing that its previous confidence is no longer justified. A material surprise becomes a correction record.
The implemented predictive prior mixes 95% of the previous posterior with 5% uniform mass. The Gaussian likelihood is normalized over the 256 possible observed states. Surprise is −log₂ P(E | predictive prior). The configured correction threshold is — bits; this is a disclosed policy, not an empirically calibrated performance claim. The first 100 verified observations calibrate the fixed price domain. Subsequent out-of-domain evidence stops publication of the posterior with MODEL DOMAIN EXCEEDED instead of silently clamping the price. Corrections exclude the initial calibration period.
Where the rule came from
INVERSE PROBABILITY
GENERALIZED INFERENCE
CONSISTENT PLAUSIBILITY
PROBABILITY AS LOGIC
Thomas Bayes wrote a short manuscript on revising belief from evidence. Richard Price published it after Bayes died. Later, Cox developed consistency arguments for representing plausible reasoning with probability, and E. T. Jaynes expanded probability as a disciplined language for incomplete information. EIDOS publicly chooses this framework; it does not claim that every possible intelligence system must be Bayesian.
What the implementation does
EVIDENCE
MODEL
EIDOS is a market diagnostics instrument, not a price oracle, guaranteed predictor, trading adviser, or promise of return. Its outputs are conditional on the selected evidence source and model assumptions. Public market data can be delayed, incomplete or unavailable.