Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/deskiziarecords/QUIMERIA-HYPERION/llms.txt

Use this file to discover all available pages before exploring further.

The SMK imposes two concentric rings of authority around all trade execution. Ring -1 sits above the analytical layers and decides whether the market regime is fit to trade at all. Ring 0 sits below the layers and fuses every sensor signal into a final go/no-go decision. Both rings have unconditional veto power: if either ring objects, r['veto']['trade_allowed'] is set to false and the AEGIS bridge takes no action.

​
Ring -1: Bayesian orchestration

Ring -1 is implemented in core/kernel/gmos_hmm_engine.py and surfaces in the pipeline as the gmos key of the step result. It runs three concurrent sub-systems:

​
HMM regime tracker (7ZERO)

The Hidden Markov Model tracks market regime persistence. It classifies each bar into one of three states:
StateValueMeaning
SIDEWAYS0Choppy, range-bound — no trending structure detected
STABLE1Stable regime — trending conditions confirmed
STABLE2Stable-trending — high-persistence momentum confirmed
When HMM Regime = 0 (SIDEWAYS), the CLAUDE.md breakage-prevention rule requires forcing Accumulation logic and monitoring for λ₁ exhaustion before attempting any directional trade. Never increase Kelly position sizing unless regime is STABLE (1 or 2).
The HMM persistence prior ranges from 0.6 to 0.95, controlling how “sticky” regime transitions are. A tighter prior prevents the model from flip-flopping on volatile bars.

​
GMOS Governor (Hamiltonian energy)

The GMOS (General Market Orchestration System) Governor measures market exhaustion using a physics-inspired Hamiltonian energy metric H_t. It accepts six sensor probabilities per bar and updates a Bayesian posterior over execution confidence. The telemetry fields exposed in r['gmos'] are: 
r['gmos'] == {
    "execute":       True,      # False if GMOS vetoes execution
    "action":        "EXECUTE", # "EXECUTE" | "HOLD" | "VETO"
    "p_fused":       0.73,      # GMOS-internal fused probability
    "coherence":     0.88,      # Sensor coherence score (0–1)
    "hamiltonian":   0.04,      # Kinetic exhaustion measure
    "regime_stable": True,      # Whether HMM+topology agree
    "drawdown":      0.01,      # Normalized drawdown pressure
    "phase":         1,         # AMD phase integer (0–3)
    "available":     True,      # False if JAX is not installed
}
available: false indicates the GMOS governor is running in numpy fallback mode because JAX is not installed. The pipeline continues operating — it passes p_fused from the Lambda Fusion Engine through instead of computing the Hamiltonian energy independently.

​
OBNFE kill switch

The OBNFE (Optimal Bayesian Non-parametric Fusion Engine) Kill Switch monitors whether the combined posterior probability of a market reversal exceeds a hard threshold. If sensor reliability drops or the posterior for regime stability falls below the configured prior, OBNFE halts execution before any lambda sensor runs. This prevents the pipeline from acting on structurally invalid data.

​
Ring 0: fusion and veto authority

Ring 0 is implemented across core/kernel/lambda_fusion_engine.py, risk/mandra_kernels.py, detectors/kl_divergence_detector.py, and detectors/topological_fracture_detector.py. It runs after all lambda sensors have fired and combines their outputs into the final execution decision.

​
Lambda Fusion Engine

The Lambda Fusion Engine collects all eight λ signals — each with a score, confidence, and veto flag — and computes a single fused probability p_fused: 
signals = {
    "λ1_vol_decay":    {"score": 0.9, "confidence": 0.85, "veto": False},
    "λ2_session":      {"score": 0.7, "confidence": 0.80, "veto": False},
    "λ3_harmonic":     {"score": -1.0,"confidence": 0.75, "veto": True },  # Liar State
    "λ4_manipulation": {"score": 0.8, "confidence": 0.70, "veto": False},
    "λ5_displacement": {"score": 1.0, "confidence": 0.85, "veto": False},
    "λ6_bias":         {"score": 1.0, "confidence": 0.88, "veto": False},
    "λ7_macro":        {"score": 0.6, "confidence": 0.70, "veto": False},
    "λ8_light_cone":   {"score": 0.5, "confidence": 0.80, "veto": False},
}
The result is stored in r['fusion']: 
r['fusion'] == {
    "p_fused":       0.68,        # Combined probability (–1 to +1)
    "confidence":    0.82,        # Weighted fusion confidence
    "veto_active":   False,       # True if any λ issued a hard veto
    "active_lambdas": ["λ1", "λ5", "λ6"],
    "regime":        "STABLE",
    "status":        "FUSION_OK",
}

​
Mandra Gate (information energy constraint)

The Mandra Gate applies an information-theoretic check: trade execution is only allowed if the information energy gain ΔE is positive and exceeds the minimum threshold of 0.02. Formally, ΔE = E_current − E_previous, where energy is derived from p_fused² × confidence. 
r['mandra'] == {
    "open":          True,    # True = gate is open, trade allowed
    "delta_e":       0.034,   # Information energy gain this bar
    "size":          0.02,    # Clamped position size
    "regime_stable": True,
    "status":        "GATE_OPEN",
}
ERR_MANDRA_ENTROPY (delta_e < 0) means the market is too chaotic to trade. Stop execution immediately and wait for the energy signal to recover. Never override a negative ΔE — the CLAUDE.md breakage-prevention rules prohibit execution without Mandra Gate confirmation.

​
KL Divergence detector

The KL Divergence detector monitors for regime fractures — moments when the current price distribution has drifted so far from the calibrated manifold that the model’s assumptions no longer hold. It is calibrated on the first 60 bars of a dataset via kl.calibrate_manifold(). A KL:REGIME_FRACTURE veto fires when kl['score'] > 1.3 × threshold and kl['stable'] == False. Error code ERR_REGIME_FRACTURE requires manual manifold recalibration.

​
Topological Fracture detector

The Topological Fracture detector uses persistent homology (via the ripser library) to detect H₁ loop formation in the price manifold — a sign that the market structure has broken down into a non-linear, looping topology. A TOPO:H1_FRACTURE veto fires when h1_score > threshold.
HALT_H1_LOOP errors indicate topological fracture. Wait for manifold compaction (H1 reset) — do not attempt to trade through a fracture event.

​
The six Ring 0 veto conditions

All six conditions are evaluated in _veto() in smk_pipeline.py. If any single condition is met, trade_allowed is set to false.
The system will halt execution immediately if any of these conditions is triggered. They are checked every bar, in order, with no overrides.
ConditionTriggerSource module
MANDRA:ΔE<0Negative information energy gain (mandra['open'] == False)risk/mandra_kernels.py
TOPO:H1_FRACTUREPersistent homology loop sum exceeds thresholddetectors/topological_fracture_detector.py
FUSION:LAMBDA_VETOAny λ sensor issued a hard veto flagcore/kernel/lambda_fusion_engine.py
L3:LIAR_STATEFFT phase inversion exceeds π/2 (harmonic['inverted'] == True)lambda_sensors/harmonic_trap_detector.py
KL:REGIME_FRACTUREKL divergence score exceeds 1.3× thresholddetectors/kl_divergence_detector.py
CONF:INSUFFICIENTFusion confidence below 0.2 (fusion['confidence'] < 0.2)core/kernel/lambda_fusion_engine.py

​
Reading the veto decision

The final veto output lives in r['veto'] and is checked by the AEGIS bridge before any order action: 
r['veto'] == {
    "decision":      "Halt",                          # "Proceed" | "Halt" | "Reset"
    "trade_allowed": False,                            # AEGIS checks this field
    "reasons": [
        "TOPO:H1_FRACTURE",
        "λ₇: DXY correlation veto"
    ]
}
decision can take three values:
  • Proceed — all Ring 0 conditions passed, AEGIS may execute
  • Halt — one or more veto conditions triggered, no execution
  • Reset — the AMD R_MASTER flag is set (retracement reset in progress), AEGIS holds
The AEGIS bridge uses this conditional check: 
if r['veto']['decision'] == 'Proceed':
    r['execution'] = bridge.evaluate(r, bars)
else:
    r['execution'] = self._null_execution(r['veto']['decision'])
All veto decisions are written to logs/veto.log (5 rotating backups, 10 MB each) for post-session audit.

Build docs developers (and LLMs) love

Get started for freeTalk to us