[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8},"actionData","errors","root",{"_28":-5,"_38":-5,"_39":40,"_41":42,"_43":44,"_45":46,"_47":48,"_49":50,"_51":52,"_53":54,"_55":56,"_57":58},"routes/_store+/_readers+/email_author.$slug/route",{"_9":10,"_11":-5},"book",{"_12":13,"_14":15,"_16":17,"_18":19,"_20":21,"_22":23},"currentUser","id","Z2lkOi8vbGVhbnB1Yi9TdG9yZTo6Qm9va1Byb2R1Y3QvODAzNjA","title","From Trial to Policy","state","published","contactAuthorMethod","email","aboutText","

Power is drifting below target.

An operator can correct it. A conventional controller can correct it. A page of carefully tuned if-rules can probably correct it too.

But what happens when the operating conditions change, the assumptions behind those rules begin to break down, and the trade-off between accuracy, stability, and actuator wear becomes too complex to tune by hand?

And now there is a new temptation: hand the problem to a machine-learning model and let it learn for itself.

It will produce recommendations.

It may even produce excellent recommendations.

But if the recommendation affects a control system, one question matters more than all the others:

Can you explain exactly why it chose that action?

From Trial to Policy is written around a single governing principle:

The agent learns, the engineer understands.

This book shows how to build a reinforcement-learning agent that remains completely readable after training. No hidden layers. No black-box neural network. No mysterious weights distributed across millions of parameters. The result of learning is a simple, inspectable Q-table: a finite set of states, a finite set of actions, and a number describing the long-term value of each possible choice.

Every recommendation can be traced.

Every action can be explained.

Every learned policy can be read directly by a human.

The book takes reinforcement learning apart piece by piece and rebuilds it from first principles. Agent. Environment. Reward. State. Action. Exploration. Policy. Nothing is assumed, and nothing is hidden behind mathematical jargon. Each concept is mapped directly to C#, .NET, SQL Server, and a working industrial-style control problem.

Try the Live Console Now!

This is a hands-on companion to NEXUS-1, a fully interactive browser-based optimization console. No installation, no server, zero setup. Every state, action, reward curve, policy grid, and training result shown in the book comes directly from this live system.

Open the console, start training, and watch the policy emerge from thousands of episodes of trial and error.

👉 Play with the live console:
https://gregory82gr.github.io/Nexus-1-phase-0/

What's Under the Hood? (The Engineering Blueprint)

This is not a machine-learning theory book. It is the architecture of a working, interpretable reinforcement-learning system.

• The Control Problem — how a familiar feedback loop becomes a learning problem.

• Agent, Environment, Reward — the three building blocks of reinforcement learning, mapped directly to C# interfaces and classes.

• The Digital Twin Training Ground — why learning happens inside a fast surrogate model rather than on the real system.

• State and Action Design — turning a continuous world into 35 readable states and five discrete control actions.

• The Q-Table — the entire learned policy stored as 175 auditable numbers.

• The Learning Algorithm — Q-learning explained step by step, arithmetic by arithmetic, without requiring a machine-learning background.

• Exploration versus Exploitation — why an agent must sometimes try apparently bad actions in order to discover better policies.

• Policy Interpretation — reading the learned table directly and understanding why every decision was selected.

• Advisory Deployment — how a learned policy becomes operational advice while remaining bounded, inspectable, and human-approved.

• C# and SQL Server Implementation — complete architecture, persistence model, training workflow, reward logging, and policy storage.

How Every Chapter Is Structured

Every chapter follows the same engineering blueprint:

What You Already Know — connecting new reinforcement-learning concepts to familiar software-engineering ideas.

The Concept — a precise explanation of the mechanism and why it exists.

One Figure — a focused diagram illustrating the idea being discussed.

See It in NEXUS-1 — exactly where the concept appears in the live console.

Honest Boundary — a transparent discussion of limitations, assumptions, and what the demonstration genuinely does versus what it deliberately simplifies.

No magic.

No hype.

No claims that learning systems are smarter than they are.

Only mechanisms that can be inspected, explained, and reproduced.

👤 Who This Is For

If you are a C#/.NET developer, software architect, controls engineer, simulation engineer, or systems designer who has never studied reinforcement learning but wants to understand how a real learning agent works, this book is for you.

It assumes you are comfortable with software development and basic engineering reasoning. It assumes no prior knowledge of machine learning, neural networks, statistics, or control theory.

By the final chapter you will be able to open a policy table, inspect every learned action, and understand exactly how trial and error became a decision-making strategy.

This volume is the reinforcement-learning companion to NEXUS-1 and the third book in the companion series following From Grid to Core and From Flood to Cause.

Author: Grigorios Kyriakos Agathangelidis
Greek name: Γρηγόριος Κυριάκος Αγαθαγγελίδης
Also searchable as: Αγαθαγγελίδης Γρηγόριος, Αγαθαγγελιδης Γρηγοριος, Grigorios Agathangelidis.

","authors",[24],{"_12":25,"_26":27,"_28":29},"Z2lkOi8vbGVhbnB1Yi9TdG9yZTo6UHJvZHVjdEF1dGhvci84NTkyMQ","name","Grigorios Agathangelidis","user",{"_30":31},"avatar",{"_32":33,"_34":35,"_36":37},"url","https://d39qdlcrvnra4b.cloudfront.net/avatars/3737581/medium/avatar.jpg?1783850356","styleName","medium","exists",true,"shoppingCartId","shoppingCart",{"_12":71,"_72":73,"_74":37,"_75":76,"_77":78,"_79":80,"_81":37},"locale","en","readerTheme","light","constants",{"_69":70},"publishers",{"_65":66,"_67":68},"learnerOrganizations",[],"needsLocaleCookieSync",false,"announcements",{"_59":60,"_61":-5},"sentryEnvironment","production","sentryRelease","cce21d19a0d9612afdbced29947de1d8868e588b","announcement_store",{"_62":52,"_63":64},"announcement_authors","flag","text","

Kick off your book project, get started with GhostAI, get better at marketing, or spend the day doing all three! Free live workshops on Zoom. Saturday, June 27, 2026.

\n","companies",[],"universities",[],"paypalClientId","AX53Cwz8jR6rVVzzZiZNWWXl0dGagi-GZrjG-S3Ua14E5nrKMn7DxDQ730O4R8LusLPp49ALn6aRVaEy","no-cart","count",0,"free","updatedAt","","paymentGateway","STRIPE","cartItems",[],"loading"]

Email the Author