Skip to content
hipoturesPublic

About

Structural Graph Conjecture Lab: reproducible AI-directed searches for counterexamples to the Erdős–Gyárfás conjecture.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

161 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Structural Graph Conjecture Lab

Structural Graph Conjecture Lab (sglab) is a Linux research system for reproducible searches for finite counterexamples to structural graph conjectures. Its implemented research target is the Erdős–Gyárfás conjecture:

Every finite simple graph with minimum degree at least 3 contains a simple cycle whose length is a power of two.

The project is an engineering and experimentation system. It does not claim that the conjecture has been resolved.

What the system does

  • runs bounded graph-search lanes with deterministic checkpoints;
  • uses an AI Research Director to choose reviewed search actions;
  • keeps the Director stateless and supplies a bounded scientific-memory snapshot on every turn;
  • retains promising candidates and protects candidate references with immutable snapshots and pins;
  • submits finalists to independent Python and C++ exact verifiers;
  • certifies success only through the M4 two-path verification boundary;
  • stores durable state in a workspace-local SQLite database;
  • exposes a local dashboard for campaigns, attempts, lanes, candidates, verification, comparisons, and live scientific visualizations;
  • resumes the same scientific campaign through new execution attempts, with optional resource changes.

Documentation entry points

You are… Start here
Running experiments User guide
Operating a server Operator guide
Understanding the design Architecture
Looking up a command or state Reference
Changing the code with Codex AGENTS.md and Codex guide
Reviewing implementation evidence Reports

Installation

Python 3.12 or newer and a C++17 compiler are required.

uv venv --python 3.12
source .venv/bin/activate
uv pip install -e .
make cyclecheck

Optional SAT and reference packages:

uv pip install -e '.[sat,reference]'

Verify the installation:

make doctor
make test
make check
make benchmark-smoke
make dashboard-smoke

First campaign

Create a workspace and prepare a bounded campaign:

sglab init --workspace ./workspace
sglab research-campaign prepare   --workspace ./workspace   --time-limit 1h

Preparation creates a campaign ID, an immutable plan, and a plan fingerprint without reading credentials or calling a model. Review the plan before authorizing it.

Start the local dashboard in another terminal:

sglab serve   --workspace ./workspace   --host 127.0.0.1   --port 8788

Continue with the campaign quickstart.

Core safety boundary

A heuristic score is not a mathematical result. A candidate becomes COUNTEREXAMPLE_VERIFIED only when the exact M4 verification boundary records complete agreement from the independent verifier paths. A timeout, memory limit, malformed artifact, or verifier disagreement is never treated as proof of cycle absence.

Repository status

See Implementation Status for the current schema, completed gates, known limitations, and links to evidence reports.

About

Structural Graph Conjecture Lab: reproducible AI-directed searches for counterexamples to the Erdős–Gyárfás conjecture.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Contributors

Languages