# Third-Party Licenses and Provenance

This document describes the origin of code and data shapes used by GridStamp
and certifies the absence of non-commercially-licensed research code.

GridStamp is released under the **Apache License 2.0** (see `LICENSE`).

---

## Clean-room declaration: 3D Gaussian Splatting

GridStamp contains **no code derived from, copied from, or translated from**
the Inria / Max Planck Institute reference implementation of 3D Gaussian
Splatting at
[`graphdeco-inria/gaussian-splatting`](https://github.com/graphdeco-inria/gaussian-splatting),
which is licensed for non-commercial use only.

What GridStamp *does* contain:

- **A TypeScript interface `GaussianSplat`** describing the 59-parameter splat
  data shape (3 position + 3 scale + 4 rotation quaternion + 1 opacity +
  48 spherical-harmonics coefficients). This is used purely as a data
  interchange format — a contract between producers and consumers of spatial
  data. Data formats are not copyrightable, and the 59-parameter shape is
  public scientific knowledge described across multiple independent papers
  and implementations released after the original 2023 paper.

- **A generic `gaussian(x, center, sigma)` math function** implementing the
  classical Gaussian bell curve `f(x) = peak * exp(-((x - center)^2) /
  (2 * sigma^2))`. This function pre-dates computer graphics by roughly two
  centuries (Gauss, 1809) and is common mathematical knowledge.

- **Placeholder integration points** in `src/memory/spatial-memory.ts` and
  `src/index.ts` where a future Gaussian splatting rasterizer could be
  plugged in. At the time of this writing, these integration points store
  empty arrays (`splats: []`) and perform no splatting computation.

GridStamp does **not** contain:

- Any CUDA kernels, PyTorch modules, or differentiable rasterizers from the
  Inria reference implementation.
- Any training or optimization code from `graphdeco-inria/gaussian-splatting`
  or its forks.
- Any model weights, scene files, or datasets derived from Inria publications.

If GridStamp adds a Gaussian splatting rasterizer in the future, it will be
sourced from one of the following permissively-licensed alternatives, or
developed clean-room from the public scientific literature:

- [`gsplat` by Nerfstudio](https://github.com/nerfstudio-project/gsplat) — Apache 2.0
- Other Apache / MIT / BSD-licensed implementations available at the time.

Any commercial use of the Inria reference implementation would require an
explicit commercial license from Inria (contact: `stip-sophia.transfert@inria.fr`).
GridStamp does not rely on such a license and does not require its users to
obtain one.

---

## Runtime dependencies

GridStamp has three runtime dependencies, all under permissive licenses:

- `merkletreejs` — MIT (for Merkle tree integrity verification)
- `sharp` — Apache 2.0 (for image processing in frame capture)
- `ssim.js` — MIT (for structural similarity comparison in spatial verification)

Development-only dependencies (used during build and test, not shipped to
users) are listed in `package.json` under `devDependencies`. Each is under a
permissive license (Apache 2.0, MIT, BSD, or ISC):

- `typescript` — Apache 2.0 (Microsoft)
- `vitest` — MIT
- `@vitest/coverage-v8` — MIT
- `eslint` — MIT
- `@types/node` — MIT

---

## Contact

If you believe GridStamp contains code in violation of a third-party license,
please contact: **omiagbogold@icloud.com**. We will investigate and remediate
any substantiated claim promptly.

---

_Last reviewed: 2026-04-09._