🔴 Overview
From manual calibration to self-learning
"The marks of a mature control system are not its performance under known conditions — it is its grace under unknown ones." — Samir Baladi, April 2026
ENTRO-EVO introduces adaptive entropy weighting for self-calibrating intelligence systems. Instead of manual calibration, the system learns optimal control weights online via gradient descent. The AEW optimizer updates weights w = [w₁, w₂, w₃] at each timestep to minimize the loss function L(t) = (Ψ_norm(t) - Ψ*)².
0.075
Best Ψ (v11.0)
Closest to target = 0.1
0.33
Best Balance (v7.0)
[0.33, 0.33, 0.34]
71.5%
Transfer Speedup
Cross-domain adaptation
🎯 Core Innovation
Adaptive Entropy Weighting
w(t+1) = clip(w(t) - η·∇L, w_min, w_max)
w(t+1) ← w(t+1) / ‖w(t+1)‖₁
L(t) = (Ψ_norm(t) - Ψ*)²
η(t) = η₀ / (1 + κ·t)
📐 AEW Optimizer
Online Gradient Descent
∂L/∂w₁ ≈ 2·(Ψ_norm - Ψ*) · (1 - u(t))
∂L/∂w₂ ≈ 2·(Ψ_norm - Ψ*) · dΨ/dt
∂L/∂w₃ ≈ 2·(Ψ_norm - Ψ*) · d²Ψ/dt²
u(t) = w₁·σ(Ψ_norm - θ) + w₂·tanh(dΨ/dt) + w₃·tanh(d²Ψ/dt²)
| Parameter | Symbol | Value | Description |
|---|
| Initial Learning Rate | η₀ | 0.01-0.05 | Controls adaptation speed |
| Decay Coefficient | κ | 0.001 | Learning rate decay |
| Target Entropy | Ψ* | 0.339 | Target equilibrium state |
| Min Weight | w_min | 0.1 | Prevents weight collapse |
| Max Weight | w_max | 0.8 | Prevents dominance |
🌡️ Dynamic Thresholding
Stability-History Adaptation
H(t) = (1/T_h) · Σ 𝟙[|Ψ_norm(τ) - Ψ*| < δ_h]
θ(t+1) = θ_base + (1 - H(t))·Δθ_max
🔄 Cross-Domain Transfer
Environment Fingerprinting
F(e) = [μ_Ψ, σ_Ψ, μ_{dΨ}, σ_{dΨ}]
w_init = λ·w_reg(e*) + (1-λ)·w_default
λ = sim(F(e), F(e*))
📊 Validation Results
Three-Regime Performance
| Regime | Steady-State Error | Budget Violations | Final Weights |
|---|
| Scraper (High Volatility) | 0.365 | 0 | [0.20, 0.10, 0.70] |
| LLM (Slow Saturation) | 0.062 | 0 | [0.80, 0.10, 0.10] |
| Phase Shift (Abrupt Transition) | 0.367 | 0 | [0.17, 0.10, 0.73] |
94.1%
Performance Recovery
vs offline-optimal
71.5%
Transfer Speedup
Cold-start vs transfer
58.7%
Noise Reduction
Dynamic thresholding
📈 Version Evolution
ENTRO-EVO v11.0 Results
| Version | Key Innovation | Final Ψ | Weights |
|---|
| v5.1 | Dual-Objective | 0.246 | [0.38, 0.22, 0.40] |
| v7.0 | Softmax Normalization | 0.697 | [0.33, 0.33, 0.34] |
| v11.0 | Thermodynamic Controller | 0.075 | [0.02, 0.02, 0.97] |
"The marks of a mature control system are not its performance under known conditions — it is its grace under unknown ones. ENTRO-EVO gives entropy-regulated intelligence the grace to learn the physics of its own environment."
📦 Installation
Quick setup
pip install entro-evo
git clone https://github.com/gitdeeper10/ENTRO-EVO.git
cd ENTRO-EVO
pip install -r requirements.txt
pip install -e .
python -c "from entro_evo import __version__; print(__version__)"
🔧 API Reference
Python interface
AdaptiveEntropyWeighting
Main AEW optimizer class
from entro_evo import AdaptiveEntropyWeighting
aew = AdaptiveEntropyWeighting(
eta_0=0.01,
target_psi=0.339
)
weights = aew.step(
psi_norm=0.5,
d_psi=0.1,
d2_psi=0.05,
u_t=0.3
)
print(f"Weights: {weights}")
DynamicThreshold
Adaptive activation threshold
from entro_evo import DynamicThreshold
dt = DynamicThreshold(
base_threshold=1.2,
delta_max=0.6
)
theta = dt.step(psi_norm=0.5)
TransferProtocol
Cross-domain transfer learning
from entro_evo import TransferProtocol
transfer = TransferProtocol()
fp = transfer.compute_fingerprint(psi_values, dpsi_values)
w_init = transfer.initialize_weights(fp)
🧩 Core Modules
ENTRO-EVO architecture
aew_optimizer.py
AEW Optimizer
Online gradient descent
dynamic_threshold.py
Dynamic Threshold
Stability-based adaptation
transfer_protocol.py
Transfer Protocol
Cross-domain transfer
simulator.py
Simulator
Three-regime simulation
👤 Author
Principal investigator
🔴
Samir Baladi
Interdisciplinary AI Researcher — Theoretical Physics, Statistical Mechanics & Information Theory
Ronin Institute / Rite of Renaissance
Samir Baladi is an independent researcher affiliated with the Ronin Institute, developing the Rite of Renaissance interdisciplinary research program. ENTRO-EVO is the fifth project (E-LAB-05) in a ten-project research program, building directly on ENTROPIA (E-LAB-01), ENTRO-AI (E-LAB-02), ENTRO-CORE (E-LAB-03) and ENTRO-ENGINE (E-LAB-04).
No conflicts of interest declared. All code and data are open-source under MIT License.
📝 Citation
How to cite
@software{baladi2026entrovevo,
author = {Samir Baladi},
title = {ENTRO-EVO: Adaptive Entropy Weighting for Self-Calibrating Systems},
year = {2026},
version = {11.0.0},
publisher = {Zenodo},
doi = {10.5281/zenodo.19464489},
url = {https://doi.org/10.5281/zenodo.19464489},
note = {E-LAB-05. Builds on E-LAB-01 through E-LAB-04}
}
@article{baladi2026entropia,
title = {ENTROPIA: Statistical Dynamics of Information Dissipation},
author = {Samir Baladi},
year = {2026},
doi = {10.5281/zenodo.19416737},
note = {E-LAB-01}
}
@software{baladi2026entroai,
author = {Samir Baladi},
title = {ENTRO-AI: Entropy-Resistant Inference Architecture},
year = {2026},
version = {2.0.0},
doi = {10.5281/zenodo.19284086},
note = {E-LAB-02}
}
@software{baladi2026entrocore,
author = {Samir Baladi},
title = {ENTRO-CORE: Regime-Dependent Entropy-Augmented Control},
year = {2026},
version = {0.1.0},
doi = {10.5281/zenodo.19431029},
note = {E-LAB-03}
}
@software{baladi2026entroengine,
author = {Samir Baladi},
title = {ENTRO-ENGINE: Entropy Flow Regulator for Coupled Systems},
year = {2026},
version = {1.0.0},
doi = {10.5281/zenodo.19441032},
note = {E-LAB-04}
}
"A single engine regulates itself. A fleet of engines requires a coordinator. But an intelligent engine learns the physics of its own environment."