# The engine

The quantum language describes classical statistics. This system does not use quantum hardware, quantum advantage, or predictions with guaranteed outcomes. Signals are estimates, not knowledge of future prices.

## Data and observability

PumpPortal events provide prices, signed sizes and wallet identities. DexScreener discovers pairs. A token needs 128 valid filtered observations before scoring. Unknown safety metrics prohibit entry. Historical prices are not simulated trades. A missing readout is displayed as unavailable, never replaced with invented activity.

| Module | Formula / inputs | Meaning |
| --- | --- | --- |
| Kalman | x=[log price, velocity]; prediction Fx, covariance FPFᵀ+Q; innovation y−Hx; K=PHᵀ/(HPHᵀ+R) | A filtered observable expectation value. Rolling return variance tunes Q/R. |
| Wave evolution | ∂p/∂t=−v∂p/∂x+D∂²p/∂x², D=variance/2; Bayesian likelihood update and normalization | A classical probability density over 201 log-return bins, downsampled to 32 display points. Stable CFL substeps preserve mass. |
| Superposition | magnitudes softplus(w·f+b), phase atan2(short velocity,long velocity) | Three classical amplitudes represent pump, flat and dump hypotheses. Cross-sectional z-scores are clipped at ±3. |
| Interference | I=abs(sum complex amplitudes)²−sum magnitude²; features whitened by C^−1/2 | Correlated features should not count repeatedly. A shrunk correlation matrix and Jacobi eigensolver provide the whitening transform. |
| Born measurement | Pk=abs(ak)²/sum abs(a)² | Entry requires pump threshold, dump probability ≤0.22, entropy check, available slots and edge above fees. These are model scores, not calibrated guarantees. |
| Hurst / fractal | R/S and detrended fluctuation analysis; D=2−H | H>0.55 permits momentum, H<0.45 requires a retracement entry; random regimes require θ+0.03. Short-window estimates are noisy. |
| Entropy | −sum pi log pi, normalized | Signed size bins and wallet concentration distinguish diverse flow from concentration. Concentration is not proof of a bundle. |
| Tunneling | T=exp(−2 sqrt(max(V−E,0)) width); V=log(R/price)/sigma | A momentum breakout metaphor. Current resistance uses fractal pivots; volume-profile integration remains a deployment blocker for the complete brief. |
| Fibonacci | retracements high−leg·r; extensions low+leg·r | Dip zone 0.382–0.618; partial exits at 1.382 and 1.618, then a trailing exit. |
| Entanglement | return correlation and optional shared deployer/funder/narrative links | Related tokens share an exposure group. Runtime currently uses return correlation; provenance and narrative feeds still require integration. |
| Decoherence | exp(−age/tau); C=Ppump(now)/Ppump(entry) | Flow ages out with tau=45s, slow data with tau=300s; coherence below 0.6 requests an exit. |
| Kelly / phi | quarter Kelly: 0.25·(bp−(1−p))/b | Risk-capped tiers 0.01, 0.0162, 0.0262 SOL. Two wins advance a tier; a loss reduces it. Sparse outcomes use priors. |
| Annealing | energy=−EV+lambda risk+mu pair correlation; 300 iterations, cooling 0.97 | A classical stochastic selector subject to slots, group cap, exposure and remaining loss budget. This is not a quantum optimizer. |
| Learning | online logistic gradient updates; isotonic calibration | At least 50 new closed labels are required; updates are versioned every two hours. The label records whether TP1 was hit before the stop. |

## Risk and execution

Position sizes stay within 0.01–0.03 SOL. Ten default slots, maximum twelve. Reserve is at least 0.04 SOL, exposure at most 60%, and realized plus unrealized daily loss is capped at 0.1 SOL. A daily loss lock lasts until the next UTC day. Entry threshold adapts within 0.52–0.75; it never relaxes integrity filters.

Stops are checked each second. Maximum stop is 8%, Fibonacci can tighten it; trailing activates at +6%, with a 4% minimum distance. Maximum hold is six minutes, or three minutes when the return stays within ±2%. Exits may fail or slip in real markets; a configured stop is not a guaranteed execution price.

Postgres stores the order intent and exact signed bytes before broadcast. One pending order per mint and an advisory worker lease prevent duplicate submission. Recovery reconciles recorded signatures; it does not create a second buy. Positions remain visible if exits cannot complete.

## Current integration limits

The pure engine and risk unit tests run. This is not yet an accepted production trading system. The six-hour quote-based paper run has not happened. A trusted integrity adapter is still needed for holder concentration, developer sales, first-slot bundles and pre-buy sell simulation. Unknown values fail closed. Pump curve transactions are explicitly blocked until their decoder is audited. Helius migrated-token stream normalization and deployer/funder attribution are not yet complete. Holder, curve and social features must not be advertised as active until their actual feeds are wired.

No profitability, cadence, sub-800ms execution, 60-second live flattening or Lighthouse score is claimed without a measured report. The website shows real wallet reads and market discovery when available, with missing trade fields left blank.