THREADS — public test bundle

THREADS = Temporal Historical Relational Engine for Adversarial Deterministic Systems.

This folder contains the exact Python source snapshot plus public test/benchmark scripts needed to reproduce the THREADS checks.

Requirements

• Python 3.10+
• NumPy 1.24+

No model weights, GPU, network access, or external solver are required for the core tests.

Fastest test

From this folder:

python THREADS_test_runner.py

This runs the original regression suite plus a lightweight diagnostic subset of the 60-category benchmark.

Full 60-category benchmark

python THREADS_test_runner.py --micro60

Applied tests

python THREADS_test_runner.py --applied

Paper-10 scientific benchmark

python THREADS_test_runner.py --paper10

Warning: Paper-10 contains the million-event distractor experiment and can require roughly 1 GB of process memory on the reported implementation.

Everything

python THREADS_test_runner.py --full

Test fairness

THREADS receives the structured facts/events/constraints and the requested operation. Expected answers and independent oracles are used by the benchmark only to grade the result after THREADS has computed it. Ordinary relation walks receive the relation program as part of the query; the bounded program-induction test is the exception that searches a supplied finite operator library.

Internal package name

The Python package is still named srmh to preserve the tested source snapshot and benchmark history. The research project is published as THREADS.

Installation and COUNT semantics

Install the package in editable mode with:

python -m pip install -e ./source

Graph is an explicit multigraph, so COUNT counts parallel edge witnesses. Engine and Ledger store observations of logical transitions; Engine COUNT counts each distinct positive subject/relation/object transition once. Repeated or independently lineaged observations therefore do not manufacture additional engine edges. This preserves the evidence-copy policy used by the regression suite and is intentionally different from Graph COUNT.

Licensing

This repository uses a split noncommercial licence. See LICENSING.md, LICENSE-CODE.txt, and LICENSE-DOCUMENTATION.txt before reuse.