mrs-core 0.1.1

MRS-Core: A deterministic modular reasoning engine.

Modular Reasoning System (MRS-Core)

A deterministic, operator-based reasoning engine for LLMs and autonomous agents.

MRS-Core provides a transparent, modular, reproducible reasoning substrate built from a small set of reusable operators:

• Transform
• Reflect
• Evaluate
• Rewrite
• Summarize
• Inspect
• Filter

MRS makes reasoning traceable, auditable, and deterministic all without requiring chain-of-thought exposure or hidden model internals.

---

Why MRS-Core Exists

Every agent framework today suffers from the same structural failures:

• No consistent reasoning sequence
• No deterministic backbone
• No visibility into intermediate states
• No enforceable phases or operator logic

MRS-Core solves this by introducing:

• Explicit operator-level reasoning
• Strict phase transition model
• Complete execution trace
• Deterministic, reproducible outputs
• Plug-and-play integration for ANY agent system

MRS is not an alignment system.
MRS is not a sandbox.

MRS is a reasoning substrate.

---

Features

Deterministic Reasoning Chains

• Operators execute in strict order.
• Output is repeatable.

Transparent Logs & History

MRS records:

• final text
• operator log
• phase trace
• structured history of every step

Simple, Extensible Operators

• Each operator is a small Python class registered via the Operator Registry.

Drop-In Presets

simple, reasoning, full_chain ready for production use.

---

Install

pip install mrs-core

---

Quick Start

from mrs.engine import MRSCoreEngine
from mrs.presets import get_preset

engine = MRSCoreEngine()
ops = get_preset("reasoning")

result = engine.run_chain(ops, "MRS-Core is a modular deterministic reasoning pipeline.")

print(result["text"])
print(result["log"])
print(result["phase"])

---

Example Output

[REFLECT] [TRANSFORM]
MRS-CORE IS A MODULAR DETERMINISTIC REASONING PIPELINE.

[EVAL CHARS=95 WORDS=14]

PHASE: rewrite

---

Running a Manual Operator Chain

from mrs.engine import MRSCoreEngine

engine = MRSCoreEngine()

ops = [
    ("transform", {}),
    ("reflect", {}),
    ("evaluate", {}),
    ("rewrite", {}),
]

result = engine.run_chain(ops, "Explain symbolic reasoning.")

print(result["text"])
print(result["history"])

---

Using Presets

MRS-Core includes preset chains:

• simple
• reasoning
• full_chain

from mrs.engine import MRSCoreEngine
from mrs.presets import get_preset

engine = MRSCoreEngine()

for name in ["simple", "reasoning", "full_chain"]:
    ops = get_preset(name)
    result = engine.run_chain(ops, "MRS-Core preset test")
    print(f"=== {name.upper()} ===")
    print(result["text"])

---

Operator Anatomy

Every operator:

• receives the current ReasoningState
• modifies state.text
• appends to state.log
• updates state.phase
• records an entry in state.history

Example:

@register_operator("reflect")
class ReflectOperator(Operator):
    phase = ("transform", "reflect")

    def run(self, state):
        state.text = f"[REFLECT] {state.text}"
        state.log.append("Reflect applied")
        return state

---

Project Structure

mrs-core/
    engine.py
    registry.py
    state.py
    presets.py
    exceptions.py
    operators/
        base.py
        transform.py
        reflect.py
        evaluate.py
        rewrite.py
        summarize.py
        inspect.py
        filter.py
tests/
examples/

---

What MRS-Core Is Not

MRS-Core is NOT:

• an alignment system
• a safety guarantee
• a sandbox
• a replacement for secure execution layers

MRS-Core IS:

• a deterministic reasoning layer
• an operator execution engine
• an audit-friendly cognition scaffold
• a missing substrate for agent stability

---

Contributing

PRs welcome, especially new operators, presets, or diagnostics.

---

License

Apache 2.0 (see LICENSE file).

Copyright 2026
Ryan Sabouhi

---

Notice

This software contains original work developed as part of the
Modular Reasoning System (MRS-Core) by Ryan Sabouhi.
See the NOTICE file for attribution details.