Loop Engineering

The discipline of designing systems that continuously improve through feedback.

Not prompt tricks. Not single agents. Closed loops — observe, act, evaluate, update, repeat — made measurable, comparable, and engineerable.

Read the manifesto · Install the stack · Learning paths

---

The shift

Era	What got optimized	The ceiling
2020–2023	Prompt engineering	Single turn — no closure
2023–2024	Context engineering	Static information — no iteration
2024–2025	Agent engineering	Autonomous actors — no system-level improvement
2025+	Loop engineering	Self-improving systems at scale

Prompt engineering optimizes a single interaction.
Agent engineering optimizes autonomous actors.
Loop engineering optimizes systems that get better through feedback.

---

What Loop Engineering offers

Pillar	What you get
Theory	13 fundamentals, 6-level taxonomy, 14 patterns
Method	D-D-M-I-S framework — Design, Diagnose, Measure, Improve, Scale
Standards	LSS 1.0 — declare loops in YAML · LES 1.0 — score them on 8 dimensions
Evidence	Case studies — AlphaGo, GitHub PRs, Toyota, coding agents
Runnable stack	Specs, dataset, runtime, and public benchmarks — all open source

This repo is the narrative home: manifesto, patterns, case studies, and learning paths.
Machine-readable specs live in Loop Core Engineering — the canonical source.

---

The published stack

Everything below is live on GitHub and PyPI (v0.1):

flowchart TB
  DOCS["<b>Loop Engineering</b><br/><i>you are here</i><br/>manifesto · patterns · case studies"]
  CORE["Loop Core Engineering<br/>LSS · LES · validators"]
  NET["LoopNet<br/>500 trajectories"]
  GYM["LoopGym<br/>pip install loopgym"]
  BENCH["LoopBench<br/>pip install loopbench"]

  DOCS -.-> CORE
  CORE --> NET
  CORE --> GYM
  NET --> GYM
  GYM --> BENCH
  CORE --> BENCH

Loading

Repository	One line	Link
Loop Core Engineering	Specs & governance — the constitution	GitHub →
LoopNet	Dataset — ground truth for loops	GitHub → · Hugging Face →
LoopGym	Runtime — run loops in sim, live, or replay	GitHub → · pip install loopgym
LoopBench	Benchmarks — public scoreboard	GitHub → · pip install loopbench

Full install map: ECOSYSTEM.md · Canonical source policy: CANONICAL-SOURCE.md

---

The loop, formally

Every loop is a closed dynamical system:

observe → decide → act → evaluate → update state → repeat

Formalized as L = (S, A, O, T, E, M, τ) — state, actions, observations, transitions, evaluators, memory, termination.

→ What is a loop?

Declare it in LSS:

loop_name: code-repair-loop
version: "1.0"
objective: "Fix failing tests with minimal diff"
workers:
  - role: implementer
evaluators:
  - type: test_suite
termination_conditions:
  - type: all_tests_pass
  - type: max_iterations
    value: 10

→ LSS 1.0 (canonical)

---

Where to start

You are…	Path	Time
Curious	Manifesto → Fundamentals	~2 hours
Building	Patterns → pip install loopgym loopbench → first benchmark run	~1 hour
Researching	Case studies → LoopNet dataset	~1 day
Leading a team	D-D-M-I-S framework → LES scoring	~2 hours

---

Inside this repo

Section	Contents
manifesto/	Founding principles
fundamentals/	13-topic theoretical foundation
taxonomy/	Six-level loop classification
patterns/	14 design patterns with LSS specs
framework/	D-D-M-I-S methodology
case-studies/	AlphaGo, GitHub PRs, Toyota, coding agents
loop-library/	Production-ready loop YAML
implementations/	Python, LangGraph, CrewAI examples
research/	Open problems and roadmap

---

Loop library preview

Loop	Level	Use case
Research Agent	2	Literature synthesis
Coding Agent	3	Feature implementation
Autonomous Debugger	3	Test-driven repair
Startup Validator	2	PMF experiments

→ Full library

---

Tools

Tool	Purpose
les_calculator.py	Structural LES estimate (local mirror)
loop_validator.py	LSS validation (prefer canonical validator)
loop_diagram_generator.py	Mermaid from LSS

---

Contributing

New patterns, case studies, implementations, and benchmark results welcome.

→ CONTRIBUTING.md · GOVERNANCE.md

---

Citation

@misc{loop-engineering-2026,
  title={Loop Engineering: The Discipline of Self-Improving Systems},
  author={Loop Engineering Community},
  year={2026},
  url={https://github.com/KanakMalpani/Loop-Engineering}
}

Feedback is the fundamental unit of intelligence.
Loop Engineering makes it engineerable.

_{MIT License}