Merge pull request #72 from AdaWorldAPI/claude/qwen-claude-reverse-eng-vHuHv

feat: index Reader LM 1.5B + BGE-M3 for OSINT pipeline

jinaai/reader-lm-1.5b (safetensors, 1 shard, 3.1 GB):
  HTML→Markdown local model. No Jina API needed.
  bgz7 → palette → O(1) HTML structure recognition.

CompendiumLabs/bge-m3-gguf (GGUF F16, ~1.2 GB):
  Multilingual embedding model. Replaces DeepNSM for non-English.
  bgz7 → palette → O(1) semantic similarity.

Together: Reader LM reads the web, BGE-M3 embeds it, AriGraph
stores it as SPO triplets, AutocompleteCache routes it at 17K tok/sec.

https://claude.ai/code/session_01M3at4EuHVvQ8S95mSnKgtK

Author: AdaWorldAPI

CLAUDE.md

@@ -72,3 +72,26 @@ src/
 - **880 lib tests passing**, 2 doctest failures out of 302
 - **Build currently fails (exit 101)** — needs investigation
 - See blackboard for detailed per-module test counts
+
+## Session: Qwen3.5 × Opus 4.5/4.6 Reverse Engineering (2026-03-31)
+
+### New Modules
+- `src/hpc/styles/` — 34 cognitive primitives (rte, htd, smad, tcp, irs, mcp, tca, cdt, mct, lsi, pso, cdi, cws, are, tcf, ssr, etd, amp, zcf, hpm, cur, mpc, ssam, idr, spp, icr, sdd, dtmf, hkf). Each is `fn(Base17, NarsTruth) → result`. 49 tests.
+- `src/hpc/causal_diff.rs` — CausalEdge64 (u64 packed), scaffold_to_palette3d_layers(), quality scoring (GOOD/BAD/UNCERTAIN), NARS self-reinforcement LoRA, PAL8 serialization (4101 bytes).
+- `.cargo/config.toml` — `target-cpu=x86-64-v4` (AVX-512 mandatory).
+- `src/simd.rs` — compile-time AVX-512 dispatch via `cfg(target_feature = "avx512f")`.
+
+### Key Data
+- 5 Qwen3.5 models indexed: 685 MB bgz7 from 201 GB BF16 safetensors
+- GitHub Release `v0.1.0-bgz-data` on AdaWorldAPI/lance-graph: 41 bgz7 files
+- 4 diffs: FfnGate dominant (0.6%), v2 reverts v1, K stable at 27B, K shifted at 9B
+
+### Benchmark
+- SPO Palette Distance: 611M lookups/sec, 1.8 ns/lookup, 388 KB RAM
+- 17K tokens/sec (triple model, 4096 heads, Pearl 2³)
+
+### Architecture Rule
+- ndarray = hardware (SIMD, Palette, Base17, SpoDistanceMatrices, read_bgz7_file)
+- lance-graph = thinking (NarsTruth, NarsEngine, TripleModel, AutocompleteCache)
+- causal-edge = protocol (CausalEdge64, NarsTables, forward/learn)
+- p64 = convergence highway (both repos meet here)