Solvent License Boundary Architecture

The Problem

MIT-licensed software (PageIndex) integrated with proprietary IP (CFF, eigenmass, NUVMAP) must be architected so the MIT license does NOT propagate to proprietary code.

The Solution: Clean Module Boundary

/home/allaun/                                    ← Workspace root
├── third_party/                                 ← ALL MIT/LICENSE-FOREIGN CODE
│   └── pageindex/                               ← VectifyAI/PageIndex (MIT)
│       ├── LICENSE                              ← MIT license (UNTOUCHED)
│       ├── ...                                  ← MIT code (NEVER MODIFIED)
│       └── NOTICE                               ← Attribution notice
│
├── cff/                                         ← PROPRIETARY CODE
│   ├── core/
│   │   ├── __init__.py                          ← Proprietary header
│   │   ├── cff.py                               ← Proprietary header
│   │   ├── fingerprint.py                       ← Proprietary header
│   │   └── schema.py                            ← Proprietary header
│   ├── gpu/
│   │   └── eigenmass_engine.py                  ← Proprietary header
│   ├── adapter/                                 ← ADAPTER LAYER (Proprietary)
│   │   └── pageindex_adapter.py                 ← Interfaces with MIT code, proprietary itself
│   ├── nuvmap/
│   ├── ingestion/
│   └── ...
│
├── LICENSE                                      ← Proprietary global license
├── THIRD_PARTY_NOTICES.txt                      ← Lists ALL third-party licenses
└── IP_BOUNDARY.md                               ← This document

Key Rules (Legally Defensible)

1. NEVER modify files inside third_party/

PageIndex source lives exactly as cloned
No patches, no inline changes, no forked copies
Configuration is done via env vars / config files OUTSIDE third_party/

2. Adapter layer NEVER includes MIT code

pageindex_adapter.py imports PageIndex as a library
Does NOT copy, paste, or inline MIT-licensed code
Contains ONLY proprietary glue logic

3. Proprietary modules NEVER import from MIT code directly

Only the adapter layer touches PageIndex
CFF core, eigenmass engine, NUVMAP engine remain clean
This prevents "derivative work" contamination

4. Third-party notices file is COMPLETE

Lists ALL foreign licenses, their terms, and attributions
Updated whenever a new third-party dep is added

5. Copyright headers on EVERY proprietary file

Date, entity, and explicit prohibition of copying

Why This Works

MIT license requires:

Copyright notice included in copies of the MIT software → SATISFIED (third_party/ untouched)
Permission notice included → SATISFIED (LICENSE file preserved)

MIT does NOT require:

Derivative works to be MIT-licensed
Combined works to be MIT-licensed
Proprietary code that uses MIT code to be disclosed

The adapter pattern ensures our code "uses" PageIndex (imports its API) without "modifying" or "distributing" it in a way that triggers copyleft.

Verification Checklist

All third_party/ files retain original copyright headers
All proprietary files have proprietary headers
No MIT-licensed code copied into proprietary directories
Adapter layer imports only, no inlining
THIRD_PARTY_NOTICES.txt is current
Root LICENSE file reflects proprietary status

3.6 KiB Raw Blame History