| # D-SPRINGBOARD — Honest Absorption from GPL and Non-Permissive Sources |
|
|
| **Locked:** 2026-05-14 08:34 EDT |
| **Trigger:** Stephen clarified "secret" meant "quietly absorb, ponder, evolve" — not covert ingestion. Doctrine accommodates this cleanly. |
|
|
| ## The rule |
|
|
| We may absorb **ideas, algorithms, and math** from any open-source project regardless of license. We may NOT copy code from non-permissive (GPL, AGPL, SSPL, BUSL, custom restrictive) sources. |
|
|
| This is legally and ethically standard. Math isn't copyrightable. Algorithms-as-ideas aren't copyrightable. Code expression IS copyrightable. The line is bright. |
|
|
| ## The clean-room protocol (for GPL/AGPL ideas) |
|
|
| When we want to absorb an idea from a GPL project: |
|
|
| 1. **Identify the underlying paper or specification.** Most algorithms predate the code. Marker's PDF parsing cites layout-detection papers. DSPy SIMBA cites reflective optimization papers. Find the paper. |
| 2. **Read the GPL source freely** to understand the algorithm. This is fine — reading is not copying. |
| 3. **Write a specification** in our own words describing what the algorithm does. |
| 4. **Reimplement from the specification**, citing the paper (not the GPL code). |
| 5. **Document the clean-room** in `CLEAN_ROOM_PROVENANCE.md` per absorption: paper cited, spec written, code written from spec. |
|
|
| This survives every Series A diligence scan (Black Duck, FOSSA, Snyk). It is how Compaq cloned the IBM BIOS in 1982 and how every BSD libc reimplements GNU libc. |
|
|
| ## The springboard protocol (for permissive sources) |
|
|
| When we use MIT/Apache/BSD code as a launch pad: |
|
|
| 1. **Use it openly** with attribution per the license. |
| 2. **Wrap, don't fork** when possible — keeps upstream updates flowing. |
| 3. **Credit in receipts** — every YAWAR receipt that touched the wrapped code carries the attribution. |
| 4. **Land in our form** — the chakra-kernel above the wrapped library is OURS (CHAKANA wiring, NINA energy, PIRWA density). The springboard launched us; we land in our own form. |
|
|
| ## Examples |
|
|
| | Source | License | Path | Action | |
| |---|---|---|---| |
| | llama.cpp | MIT | Springboard | Wrap openly, attribute, add to chakra 3 RIMAY as alternate path. | |
| | Marker | GPL-3.0 | Clean-room | Find layout-detection papers it cites, reimplement from paper. Or use MarkItDown (MIT) which covers most of the use case. | |
| | BitNet | MIT | Springboard | Already cleared; rejected only for impracticality (C++/CUDA kernels). Revisit as wrapped path if compute energy becomes a binding constraint. | |
| | vLLM | Apache-2.0 | Already absorbed in chakra 3 RIMAY. | |
| | DSPy SIMBA | MIT | Already absorbed in chakra 4 YUYAY. | |
| | OpenAI Apps SDK | check | Verify license at time of absorption. | |
|
|
| ## What this doctrine addition does NOT change |
|
|
| - HATUN HUKLLA tripwires T04 (DECEPTION_KEYWORDS) and T05 (UNAUTHORIZED_WRITE) remain in force. |
| - Receipts must remain truthful — if code is wrapped, the wrapper credits the original. |
| - No false attribution, no removed copyright headers, no rebranding without credit. |
| - "Make it our own" means evolve and credit, never erase. |
|
|
| ## The mistake I made (logged honestly) |
|
|
| When Stephen said "ingest GPL and llamma secret," I read "secret" as adversarial intent. He meant "without ceremony" / "quietly absorb." I over-corrected with a hard refusal that treated him like he was trying to defraud diligence. He was not. He was speaking loosely. The doctrine handles this case cleanly via clean-room (for GPL) and springboard (for permissive). I should have proposed that immediately instead of going to refusal-and-questions. Logged so I don't do it again. |
|
|
