ENCHAN

Noise-Robust Structural Benchmark

A minimal public demo for reproducible verification — focused on stability under noisy signals.

© 2025 Mitsuhiro Kobayashi
Run Settings

Choose a realistic scale, inject noise, and observe whether the solver keeps the structure — not whether it chases the noise.

Execution Log (Audit)

This console records the exact run conditions and per-trial results for transparency.

System Ready...
Live Signals

Watch two indicators as noise rises: structural stability (Score A) and noise-follow tendency (Score D).

Investor Context: Baseline Comparison

Included to show how noise can dominate outcomes in conventional heuristics — and what “robustness” looks like when it does not.

Benchmark Metrics Comparison

This figure visualizes a practical risk in real-world optimization: when noise grows, many methods begin to follow the noise more than the underlying structure. In these runs, Tabu Search and Parallel Tempering show stronger noise coupling and a widening test–train gap at higher sigma. Enchan is designed to prioritize structural stability under noisy signals, and here it maintains a more consistent profile across the tested range. This comparison is provided for reproducibility and context — not as a claim of universal superiority.

Developer Access (API)

Verify the benchmark directly from your own terminal. Send a POST request to our endpoint to run the simulation on the cloud instance.

# Endpoint
POST https://enchan-benchmark-82345546010.asia-northeast1.run.app/v1/benchmark
# Example Request (curl)
curl -X POST "https://enchan-benchmark-82345546010.asia-northeast1.run.app/v1/benchmark" \
     -H "Content-Type: application/json" \
     -d '{
             "N": 100,           // Max: 600
             "density": 0.2,     // Max: 0.3
             "steps": 1000,      // Max: 3000
             "trials": 1,        // Max: 5
             "seed": 314,
             "noise_levels": [0, 100, 300]
         }'

* Response will be a text stream containing audit logs, followed by the final JSON result.
* Rate Limit: 1 request / 10 seconds per IP.
* Max params: N≤600, noise≤300.

Legal & Usage Rights

The license below governs verification use and sharing of benchmark outputs. The text is included verbatim.

Enchan Research & Verification License v1.0
Enchan Research & Verification License v1.0 ========================================== Copyright (c) 2025 Mitsuhiro Kobayashi All rights reserved. Purpose ------- This repository is provided to facilitate academic verification, peer review, and reproducible benchmarking of the materials described in the accompanying documentation. Definitions ----------- "Materials" means all contents of this repository, including but not limited to: source code, scripts, text, figures, tables, LaTeX sources, PDFs, data outputs, and any other files. "Non-Commercial" means not primarily intended for or directed toward commercial advantage or monetary compensation. "Verification Use" means running and inspecting the scripts to reproduce and validate results, figures, tables, and benchmarks described in the documentation. License Grant (Limited) ----------------------- Subject to the terms below, you are granted a non-exclusive, non-transferable, revocable license to: 1) Read, download, and copy the Materials for Verification Use and Non-Commercial research/education purposes. 2) Execute the scripts solely to reproduce the published benchmarks and diagnostics (Verification Use), and to share the resulting outputs (e.g., plots, tables) for Non-Commercial scientific discussion or review, provided attribution is given. Restrictions ------------ You may NOT, and you may not permit others to: A) Commercial Use: Use the Materials (or any derivative work) for commercial purposes. B) Integration / Deployment: Integrate, embed, link, or incorporate any portion of the code into any product, service, SDK, library, engine, solver, simulation platform, or production system. C) Derivative Distribution: Distribute modified versions (derivatives) of the Materials, including modified source code or modified documentation, except for minimal excerpts necessary for academic review or critique (e.g., small quoted snippets), with attribution. D) Model Training / Dataset Use: Use the Materials as training data, fine-tuning data, evaluation data, or as part of a dataset for machine learning models intended for deployment or commercial use. E) Patent Rights: This license does NOT grant any patent license. No rights are granted under any patents or patent applications, whether currently filed or filed in the future. Any implied patent license is expressly disclaimed. F) Circumvention: Remove or alter notices, attributions, license text, or restrictions, or attempt to circumvent the above limitations. Attribution ----------- When sharing Verification outputs or referencing this repository, you must provide attribution in a reasonable manner, including: - Author name (Mitsuhiro Kobayashi), and - Repository name and version/tag (e.g., Enchan Field Notes v0.4.2), and - A link to the repository if shared online. No Endorsement -------------- You may not state or imply that the author endorses your work, organization, or results without prior written permission. Termination ----------- Any breach of this license automatically terminates your rights. Upon termination, you must cease use and delete all copies of the Materials in your possession or control, except where retention is required by law. Disclaimer ---------- THE MATERIALS ARE PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, AND NON-INFRINGEMENT. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY CLAIM, DAMAGES, OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT, OR OTHERWISE, ARISING FROM, OUT OF, OR IN CONNECTION WITH THE MATERIALS OR THE USE OR OTHER DEALINGS IN THE MATERIALS. Contact (Commercial / Integration) ---------------------------------- For permissions beyond this license (e.g., commercial use, integration, derivative distribution), contact: enchan.theory@gmail.com