docs(codec): cross-domain synergies — x265 ⇄ 3DGS ⇄ attention ⇄ SGD

722-line knowledge doc capturing the architectural claim that PR-X12's
primitives (LeafCu, pack_leaf, predict_intra, Ctu, eventual rANS) are
not video-codec-specific — they serve four loads on a single
predictive-coder substrate:

  1. Cognitive cells (the original target)
  2. 3D Gaussian Splat coefficient compression
  3. Transformer attention sparsification
  4. Distributed-SGD gradient streaming

Sections:
- The 4-load isomorphism (§ 2) and 16-state (CellMode × MergeDir)
  classification table across all four domains
- Primitive → load mapping matrix (§ 3) with file:line refs
- 10 cross-domain epiphanies (§ 4) — including MergeDir-as-topology,
  predict_intra-as-attention-sparsifier, escape_next-as-allreduce-slot,
  mode-coded LoRA
- 6 integration plans (§ 5) with worker × week + dependency graph
- 8 exploration paths (§ 6) for sprint-scale research
- 6 holy grail outcomes (§ 7) — sub-1-bit/Gaussian 3DGS,
  bit-exact attention, federated SGD at 8-16×, Lance substrate identity
- 10 codec-side debts (§ 8) and 13 stack-side debts (§ 9)
- Sequencing summary (§ 10): critical path = A7 rANS

Companion to pr-x12-codec-x265-design.md (the mechanical design); a
"See also" breadcrumb added there points here.

Author: claude

.claude/knowledge/pr-x12-codec-x265-design.md

@@ -8,6 +8,12 @@
 > **Depends on**: PR-X10 (linalg-core), PR-X3 BlockedGrid (shipped).
 > **Used by**: PR-X9 (basin-codebook lazy storage) — the codec encodes
 > cognitive cells into skip/merge/delta/escape modes.
+>
+> **See also**: `pr-x12-cross-domain-synergies.md` — the architectural
+> claim that the primitives in this design doc *also* serve 3DGS
+> coefficient compression, transformer attention sparsification, and
+> distributed-SGD gradient streaming. Holy grail outcomes, integration
+> plans, and honest debt catalogue live there.
 
 ## Why