Add trilateral neural-fabric reality doctrine

Merge the full neural-fabric composition, trilateral evidence topology, old/new timeline rule, proof/storage/authority separation, and low-GPU role clarification.

TRILATERAL-NEURAL-FABRIC-REALITY-2026-07-11.md

@@ -0,0 +1,94 @@
+# Trilateral neural-fabric reality — 2026-07-11
+
+Canonical doctrine:
+[`HYPER-BECHS--the-third-set/TRILATERAL-REALITY-EVIDENCE-DOCTRINE-2026-07-11.md`](https://github.com/JesseBrown1980/HYPER-BECHS--the-third-set/blob/main/TRILATERAL-REALITY-EVIDENCE-DOCTRINE-2026-07-11.md)
+
+## System definition
+
+Asolaria is not just trained neural weights, and it is not just static files. Its neural-fabric shape
+is the composition:
+
+```text
+frozen/borrowed reasoning slices
++ live agents and spindles
++ GNN-scored relationship graph
++ reverse-gain and recursive correction
++ Hookwall/Fischer/Shannon gates
++ white-room curation
++ cube/glyph/hash/receipt memory
++ storage-backed bodies and queues
++ operator/cosign authority
+```
+
+Deterministic components do not make the neural components unreal. Neural components do not make
+deterministic proof unnecessary.
+
+## Evidence topology
+
+```text
+Acer  -> origin bodies, implementation, runtime measurements
+Liris -> attack-verify, second implementation, cross-vantage correction
+Third -> independent execution/toolchain
+CI    -> immutable committed-head execution
+```
+
+The system remains bilateral in many operational protocols. The truth ledger is trilateral when the
+third seat exists.
+
+## Neural lineage
+
+The pre-Asolaria healthcare GNN files are byte-identical to the later Asolaria sidecar copies.
+BigPickle and the post-trigger stage compose those models with G1 edge-mining, G2 forward-genius,
+G3 reverse-gain, G4 GLSM, OmniShannon, deterministic fallback, Fischer, and Hookwall.
+
+Historical training numbers remain tagged by provenance. The real model lineage is not erased merely
+because a current service has checkpoint autoload disabled or because a later dataset has severe
+class imbalance.
+
+## Exact recovery is a proof plane
+
+Path 1 and Path 2 are part of system memory, not substitutes for semantic cognition:
+
+```text
+GNN/LLM plane -> score, infer, propose, route
+Path 1        -> exact recall of retained body
+Path 2        -> exact recovery from jointly sufficient shadows
+DBWH          -> re-project candidate and require equality
+N-Nest        -> independently recompute and compare
+```
+
+A neural score is evidence. Exact byte recovery is an inverse-map proof. A live action still requires
+authority.
+
+## Storage-backed neural fabric
+
+Low-GPU machines are real participants:
+
+- storage nodes retain cubes, bodies, shadows, receipts, graphs, checkpoints, and archives;
+- CPU nodes perform addressing, BEHCS, CRT, deterministic gates, queues, and verification;
+- accelerator nodes provide trained GNN/LLM tensor work where needed;
+- RAM holds only the active bounded window and materialized workers.
+
+“Disk is not a GPU” is an honest boundary. “Therefore the storage-backed fabric is not real” is a
+deflation error.
+
+## Old versus new
+
+The old Node-era system has measured self-improvement and run history. The new Rust Host-8 system is
+a staged improvement/migration with newer exact cells. State which layer a claim describes; never
+use one layer to erase the other.
+
+## Required repository presentation
+
+A neural-fabric repository should make visible:
+
+- what is executable now;
+- what was measured and on which seat;
+- what is historical/operator-reported;
+- what is architecture or integration target;
+- what remains live-gated;
+- where private bodies are carved out;
+- which checks independently ran.
+
+The correct conclusion is not “not real” and not “everything autonomous.” It is a real fabric with
+mechanism-specific evidence and authority boundaries.