S143: 5 novel composition experiments (Exp 097–101) + doc sweep + V96 handoff

Axis 2 "Novel Compositions (No New Math)" complete:
- Exp 097: Digester×Anderson coupling (ESN + Anderson localization)
- Exp 098: Isomorphic reservoir ensemble (ESN + LSTM glucose + LSTM weather)
- Exp 099: WDM ensemble QS (surrogates + game theory + Anderson)
- Exp 100: HMM introgression on NN layers (PhyloNet-HMM + weight statistics)
- Exp 101: Attention Anderson spectral (attention + eigh + Anderson IPR)

5 Python controls (66 checks), 5 Rust modules, 10 validation binaries
(220 CPU + 54 GPU checks). 1085 lib tests, 245 binaries, 46 modules, 0 clippy.

Doc updates: README, CONTROL_EXPERIMENT_STATUS, EVOLUTION_READINESS,
EXTENSION_PLAN, BARRACUDA_USAGE, PAPER_REVIEW_QUEUE, experiments journal.
V96 ToadStool/BarraCUDA handoff. V94/V95 handoffs archived.
4 orphan binaries restored to Cargo.toml (nucleus_pcie, toadstool_s94b,
live_dashboard, ecosystem_dashboard).

Made-with: Cursor

Author: eastgate

CONTROL_EXPERIMENT_STATUS.md

@@ -1,11 +1,11 @@
 # neuralSpring — Control Experiment Status
 
-**Last updated**: March 10, 2026 (Session 142 — Upstream Rewire + `enable f64;` PTXAS Fix + Cross-Spring Evolution. Sprint 2 absorption. `enable f64;` regression diagnosed and fixed: 5 fused f64 ops restored, dispatch parity 48/55 → 55/55. HMM fused path workaround. DF64 tolerance characterization. V95 toadStool/barraCuda + coralReef handoffs. barraCuda `83aa08a`, toadStool S142 (`a86bc546`), coralReef Iter 29 (`2779c88`). 1048 lib + 71 forge + 9 integration tests. 0 clippy. 233 binaries)
+**Last updated**: March 10, 2026 (Session 143 — Extension Phase: 5 Novel Composition Experiments. Exp 096 Digester×Anderson coupling, Exp 097 Isomorphic reservoir ensemble, Exp 098 WDM ensemble QS, Exp 099 HMM introgression on NN layers, Exp 100 Attention Anderson spectral. 5 new Rust modules, 10 new validation binaries, 5 Python controls. 1085 lib + 71 forge + 9 integration tests. 0 clippy. 245 binaries)
 **Gate**: Eastgate (i9-12900K, 32 GB DDR5, RTX 4070 12 GB + TITAN V 12 GB NVK, Pop!_OS 22.04)
 **Python**: 3.10.12, PyTorch 2.9.0+cu128, NumPy 2.2.6, SciPy 1.15.3
 **Rust**: Edition 2021, clippy pedantic + nursery, unsafe_code=forbid
-**Grand Total**: 331/331 Python PASS + 3400+ Rust+GPU validation PASS = **4100+ total validation checks**
-**Library**: 1048 lib tests + 9 integration tests + 71 forge tests | 41 modules + config + streaming/ + search/ + visualization/ + gpu_ops/ + gpu_dispatch | 233 validation/bench binaries
+**Grand Total**: 397/397 Python PASS + 3600+ Rust+GPU validation PASS = **4500+ total validation checks**
+**Library**: 1085 lib tests + 9 integration tests + 71 forge tests | 46 modules + config + streaming/ + search/ + visualization/ + gpu_ops/ + gpu_dispatch | 245 validation/bench binaries
 **CPU↔Python Parity**: 41/41 PASS — `validate_cpu_math_parity` (9 primitives + 10 paper kernels + 6 Dispatcher cpu_only checks, all within 1e-10)
 **Dispatch Overhead**: `bench_dispatch_tiers` — 9/10 ops ≤1.04× overhead (CPU dispatch is transparent), per-call GPU driver-bound for small workloads (motivates pipeline batching)
 **baseCamp**: 6 biophysical AI modules + 11 validators (162/162 CPU + 14/14 GPU + 19/19 dispatch + GPU pure 6/6 sub-theses PASS) — Sessions 50, 54, 56, 77, 104b, 107
@@ -14,7 +14,7 @@
 **GPU Promotion**: 47 CPU→GPU ops via `gpu_dispatch::Dispatcher` (~97% of production math, +3: hill_gate, multi_obj_fitness, swarm_nn_forward)
 **Pure GPU All-Domains**: 11/11 PASS — `validate_gpu_pure_workload_all` (13 GPU domains across all Phase 0++ papers + Paper 026 LSTM + determinism check, scalar-only readback). **WDM+coralForge Pure GPU**: 24/24 PASS — `validate_gpu_pure_wdm_coral` (nW-01 MLP, nW-02 EOS, nW-03 LSTM, nW-05 ESN, coralForge attention, TriMul, AF3 pLDDT, AF3 PAE, AF3 diffusion forward, PF FFN, PF TriMul + determinism)
 **WDM Surrogates**: 5 Python baselines (33/33 PASS) + 7 Rust validators (160/160 PASS incl. GPU) + 4 GPU Tensor validators (nW-01 transport, nW-02 EOS, nW-03 S(q,ω), nW-04 transfer, nW-05 ESN) — `wdm_surrogate.rs`, `wdm_transport.rs`, `wdm_sqw.rs`, `wdm_esn.rs` modules
-**Publication Experiments (S88+)**: Exp-050 (Py 11/11, Rs 12/12, GPU 9/9), Exp-052 (Py 8/8, Rs 14/14, GPU 10/10), Exp-053 (Py 11/11, Rs 18/18, GPU 11/11). Pure GPU pipeline + metalForge cross-system: 13/13. Mixed-hardware NUCLEUS: 43/43. Phase 4 shader validation: 22/22. Streaming spectral pipeline: 28/28. **Dispatch parity: 53/53 (S115). ComputeDispatch bridge: 14/14. NUCLEUS PCIe bypass: 38/38. Mixed-hardware dispatch: 47/47. WDM+AlphaFold3 dispatch: 57/57. coralForge Evoformer dispatch: 47/47.** **232 binaries, 220/220 validate\_all**
+**Publication Experiments (S88+)**: Exp-050 (Py 11/11, Rs 12/12, GPU 9/9), Exp-052 (Py 8/8, Rs 14/14, GPU 10/10), Exp-053 (Py 11/11, Rs 18/18, GPU 11/11). Pure GPU pipeline + metalForge cross-system: 13/13. Mixed-hardware NUCLEUS: 43/43. Phase 4 shader validation: 22/22. Streaming spectral pipeline: 28/28. **Dispatch parity: 53/53 (S115). ComputeDispatch bridge: 14/14. NUCLEUS PCIe bypass: 38/38. Mixed-hardware dispatch: 47/47. WDM+AlphaFold3 dispatch: 57/57. coralForge Evoformer dispatch: 47/47.** **245 binaries**
 **NUCLEUS Compute Dispatch**: Tower discovery + Node eigensolve/Anderson/Hessian + Nest provenance + mixed atomic coordination + PCIe bypass: **43/43 PASS**. `validate_nucleus_compute_dispatch`. **NUCLEUS PCIe mixed pipeline**: Tower→Node→Nest chain + GPU↔NPU bypass + biomeOS graph coordination: **38/38 PASS** (S115). `validate_nucleus_pcie_mixed_pipeline`. **NUCLEUS Tower Mode**: JSON-RPC folding + GPU dispatch + cross-primal forwarding + concurrent: **22/22 PASS** (S133). `validate_nucleus_tower`
 **ToadStool Absorption Readiness**: CPU correctness (eigh/Anderson/Hamiltonian) + GPU parity (3 matrix sizes) + batch scaling + mixed substrate: **294/294 PASS**. `validate_toadstool_spectral_absorption`
 **biomeOS Integration**: neuralSpring registered as science primal. 7 capabilities (spectral\_analysis, anderson\_localization, hessian\_eigen, agent\_coordination, ipr, disorder\_sweep, training\_trajectory). `neuralspring_primal` JSON-RPC server. `validate_biomeos_spectral`: **29/29 PASS**. **biomeOS graph coordination**: DAG pipeline engine (`graph.rs`) with topological execution, 3 canonical pipelines (spectral, popgen, folding). `validate_biomeos_graph`: **32/32 PASS** (S133). NUCLEUS ready (all plasmidBin primals built)
@@ -338,7 +338,7 @@ Target progression (following hotSpring): **Python < CPU < GPU**
 | 0 | Synthetic baselines (48 checks) | **COMPLETE** |
 | 0+ | Scholarly reproductions (31 checks) | **COMPLETE** |
 | 0++ | Paper reproductions (127 checks) | **COMPLETE** |
-| 1a | neuralSpring Rust validation (1048 lib + 9 integration + 71 forge tests, 233 binaries, 41+ modules + gpu_ops/ + gpu_dispatch/) | **COMPLETE** |
+| 1a | neuralSpring Rust validation (1085 lib + 9 integration + 71 forge tests, 245 binaries, 46 modules + gpu_ops/ + gpu_dispatch/) | **COMPLETE** |
 | 1b | BarraCUDA validation (272 checks) | **COMPLETE** |
 | 1c | Fused ToadStool pipeline (46–78×) | **COMPLETE** |
 | 1d | 3-way benchmark + double-buffered shaders | **COMPLETE** |
@@ -1187,3 +1187,53 @@ Closed the GPU Tensor validation tier for AlphaFold3 diffusion and Pairformer pr
 - nS-01 weight spectral CPU benchmarks: eigh\_f64 on 64×64, 128×128, 256×256 Hamiltonians
 
 **Re-validated**: `validate_weight_spectral_real` **12/12 PASS**. `validate_all` **200/200** (198 PASS + 2 pre-existing wright\_fisher WGSL parse). 216 binaries, 3500+ total checks.
+
+---
+
+## Session 143 — Extension Phase: 5 Novel Composition Experiments (Experiments 097–101)
+
+**Scope**: Axis 2 "Novel Compositions (No New Math)" — compose existing validated modules into new scientific applications. Each experiment has Python control baseline, Rust module with unit tests, CPU-only validator, and BarraCUDA CPU+GPU validator.
+
+**New modules**:
+
+| Module | Experiment | Composes |
+|--------|-----------|----------|
+| `digester_anderson.rs` | 097 — Digester×Anderson coupling | ESN (Paper 027) + Anderson (Paper 023) |
+| `isomorphic_reservoir.rs` | 098 — Isomorphic reservoir ensemble | ESN + LSTM glucose + LSTM weather |
+| `wdm_ensemble_qs.rs` | 099 — WDM ensemble QS | game theory + WDM surrogates + Anderson |
+| `introgression_nn.rs` | 100 — HMM introgression on NN | HMM (Paper 018) + weight statistics |
+| `attention_anderson.rs` | 101 — Attention Anderson spectral | attention + eigh + Anderson IPR |
+
+**New binaries** (10):
+
+| Binary | Type | Checks |
+|--------|------|--------|
+| `validate_digester_anderson` | CPU | 55/55 PASS |
+| `validate_barracuda_digester_anderson` | GPU | 22/22 PASS |
+| `validate_isomorphic_reservoir` | CPU | 35/35 PASS |
+| `validate_barracuda_isomorphic_reservoir` | GPU | 13/13 PASS |
+| `validate_wdm_ensemble_qs` | CPU | 81/81 PASS |
+| `validate_barracuda_wdm_ensemble_qs` | GPU | 7/7 PASS |
+| `validate_introgression_nn` | CPU | 15/15 PASS |
+| `validate_barracuda_introgression_nn` | GPU | 6/6 PASS |
+| `validate_attention_anderson` | CPU | 34/34 PASS |
+| `validate_barracuda_attention_anderson` | GPU | 6/6 PASS |
+
+**Python baselines** (5):
+
+| Control | Checks |
+|---------|--------|
+| `control/digester_anderson/digester_anderson.py` | 17/17 PASS |
+| `control/isomorphic_reservoir/isomorphic_reservoir.py` | 17/17 PASS |
+| `control/wdm_ensemble_qs/wdm_ensemble_qs.py` | 11/11 PASS |
+| `control/introgression_nn/introgression_nn.py` | 11/11 PASS |
+| `control/attention_anderson/attention_anderson.py` | 10/10 PASS |
+
+**Key findings**:
+
+- **Spectral universality**: ESN/LSTM ensembles from 3 domains produce identical spectral profiles (eff\_dim CV = 0.003)
+- **Disagreement = disorder**: Surrogate ensemble variance maps to Anderson localization parameter
+- **Algorithm isomorphism**: PhyloNet-HMM (genomics) detects anomalous NN layers (TPR = 0.97, FPR = 0)
+- **GPU↔CPU parity**: All 5 experiments ≤ 2.5e-4 max diff (f32 vs f64)
+
+**Totals**: 1085 lib tests + 9 integration + 71 forge tests. 245 binaries. 46 modules. 0 clippy. 0 unsafe. 4500+ total checks.

Cargo.toml

@@ -752,6 +752,61 @@ path = "src/bin/validate_glucose_prediction.rs"
 name = "validate_wdm_esn"
 path = "src/bin/validate_wdm_esn.rs"
 
+# ── Paper 027: Digestion prediction (Wang/Liao 2020) ─────────────
+
+[[bin]]
+name = "validate_digestion_prediction"
+path = "src/bin/validate_digestion_prediction.rs"
+
+[[bin]]
+name = "validate_barracuda_digestion"
+path = "src/bin/validate_barracuda_digestion.rs"
+
+# ── Experiment 096: Digester-Anderson coupling ────────────────────
+[[bin]]
+name = "validate_digester_anderson"
+path = "src/bin/validate_digester_anderson.rs"
+
+[[bin]]
+name = "validate_barracuda_digester_anderson"
+path = "src/bin/validate_barracuda_digester_anderson.rs"
+
+# ── Experiment 097: Isomorphic reservoir ensemble ─────────────────
+[[bin]]
+name = "validate_isomorphic_reservoir"
+path = "src/bin/validate_isomorphic_reservoir.rs"
+
+[[bin]]
+name = "validate_barracuda_isomorphic_reservoir"
+path = "src/bin/validate_barracuda_isomorphic_reservoir.rs"
+
+# ── Experiment 098: WDM surrogate ensemble QS ─────────────────────
+[[bin]]
+name = "validate_wdm_ensemble_qs"
+path = "src/bin/validate_wdm_ensemble_qs.rs"
+
+[[bin]]
+name = "validate_barracuda_wdm_ensemble_qs"
+path = "src/bin/validate_barracuda_wdm_ensemble_qs.rs"
+
+# ── Experiment 099: HMM introgression on NN layers ────────────────
+[[bin]]
+name = "validate_introgression_nn"
+path = "src/bin/validate_introgression_nn.rs"
+
+[[bin]]
+name = "validate_barracuda_introgression_nn"
+path = "src/bin/validate_barracuda_introgression_nn.rs"
+
+# ── Experiment 100: Attention Anderson spectral ───────────────────
+[[bin]]
+name = "validate_attention_anderson"
+path = "src/bin/validate_attention_anderson.rs"
+
+[[bin]]
+name = "validate_barracuda_attention_anderson"
+path = "src/bin/validate_barracuda_attention_anderson.rs"
+
 # ── baseCamp Sub-thesis validators ────────────────────────────────
 [[bin]]
 name = "validate_weight_spectral"
@@ -1066,6 +1121,24 @@ path = "src/bin/validate_barracuda_basecamp.rs"
 name = "diagnose_f64_regression"
 path = "src/bin/diagnose_f64_regression.rs"
 
+# ── Orphan binaries restored (cleanup S143)
+
+[[bin]]
+name = "validate_nucleus_pcie_mixed_pipeline"
+path = "src/bin/validate_nucleus_pcie_mixed_pipeline.rs"
+
+[[bin]]
+name = "validate_toadstool_s94b_wgpu28"
+path = "src/bin/validate_toadstool_s94b_wgpu28.rs"
+
+[[bin]]
+name = "neuralspring_live_dashboard"
+path = "src/bin/neuralspring_live_dashboard.rs"
+
+[[bin]]
+name = "neuralspring_ecosystem_dashboard"
+path = "src/bin/neuralspring_ecosystem_dashboard.rs"
+
 [lints.rust]
 unsafe_code = "forbid"
 

EVOLUTION_READINESS.md

@@ -1,11 +1,11 @@
 # neuralSpring — Evolution Readiness
 
-**Date**: March 10, 2026 (Session 142 — Upstream Rewire + `enable f64;` Fix + Cross-Spring Evolution)
+**Date**: March 10, 2026 (Session 143 — Axis 2 Novel Compositions + Extension Phase)
 **barraCuda**: v0.3.3 at `83aa08a` (`../barraCuda/crates/barracuda`). 719 WGSL shaders, wgpu 28, Sprint 2 APIs (activations, rng, tridiagonal_ql), healthSpring domain, batched logsumexp, CoralReefDevice. `PrecisionRoutingAdvice` with `F64NativeNoSharedMem` Ada Lovelace reclassification, `WORKGROUP_SIZE_1D` constant, cross-spring provenance registry, `BatchedOdeRK45F64`, `mean_variance_to_buffer`, coralReef Phase 10 IPC. Three-tier precision: F32/F64/Df64 (lean 3-tier model, F16+templates removed). Deep debt: typed errors, named constants, `Arc<str>` hot-path, `RwLock` compiler, ring buffer back-off, streaming pipeline completion. **Known issue**: `enable f64;` in WGSL triggers PTXAS silent-zero regression on Ada Lovelace — fix implemented locally in `pipeline_cache.rs`, pending upstream absorption.
 **ToadStool**: S142 at `a86bc546`. Hardware testing, PCIe transport, ResourceOrchestrator, 19,900+ tests. Absorbed neuralSpring `pipeline_graph` DAG + hotSpring `streaming_dispatch`. Dual-write discovery (canonical + coralReef-compatible). `GpuDevice` enrichment (render_node, driver, arch). `gpu.dispatch` + `shader.compile` + `orchestration` capabilities. Compute triangle unblocked. Deep debt, zero-copy.
 **coralReef**: Iteration 29 at `2779c88`. NVIDIA last mile pipeline, SSA repair, multi-GPU sovereignty, 1200+ tests. Sovereign shader compiler (WGSL → native GPU binary). NVIDIA SM70-SM89, AMD RDNA2+ (E2E GPU dispatch verified on RDNA2). Three-tier precision architecture: f32 native, f64 DFMA+polynomial lowering, df64 preamble auto-prepend. `Fp64Strategy` in `CompileOptions`. Built-in `df64_preamble.wgsl`. 8 neuralSpring shaders in corpus.
-**neuralSpring**: 1048 lib + 71 forge + 9 integration tests, 233 binaries, 0 clippy warnings (pedantic+nursery), 0 doc warnings. All files ≤1000 LOC. AGPL-3.0-or-later. Pure Rust deps (ecoBin compliant).
-**S142**: Upstream rewire + `enable f64;` fix + cross-spring evolution. Sprint 2 API absorption (activations, fused\_ops\_healthy). `Precision::F16` removed. coralReef bridge aligned. 54 binaries with provenance blocks. `enable f64;` PTXAS regression diagnosed and fixed locally (5 fused ops restored). HMM fused path workaround (0.0 detection → per-step fallback). Glucose pearson tolerance adjusted for DF64. Dispatch parity: 48/55 → **55/55 PASS**. V95 toadStool/barraCuda + coralReef handoffs written.
+**neuralSpring**: 1085 lib + 71 forge + 9 integration tests, 245 binaries, 46 modules, 0 clippy warnings (pedantic+nursery), 0 doc warnings. All files ≤1000 LOC. AGPL-3.0-or-later. Pure Rust deps (ecoBin compliant).
+**S143**: 5 novel composition experiments (Exp 097–101): digester×Anderson, isomorphic reservoir ensemble, WDM ensemble QS, HMM introgression NN, attention Anderson spectral. Axis 2 complete. 209 files with barracuda imports. V96 handoff. Ready for Axis 1 (real data) and ToadStool pipeline graph absorption.
 **S138**: Industry gap closure — streaming FASTA parser (16 tests), CPU-reference BLAST pipeline (`search/kmer_index`, `search/seed_extend`, 19 tests), `bench_kokkos_parity` GPU benchmark harness (9 ops × production scale). `INDUSTRY_TOOL_GAP_ANALYSIS.md`, `BLAST_LIKE_SEARCH_SCOPE.md`, `MSA_PIPELINE_SCOPE.md`. V92 handoff written.
 **S137**: Upstream rewire + deep debt execution. Reviewed 27+ Mar 8–9 wateringHole handoffs (ToadStool S139, barraCuda Sprint 2 + deep debt + concurrency, coralReef precision architecture, groundSpring V98/V100 sovereign rewire). **Rewires**: hardcoded `256` → `WORKGROUP_SIZE_1D` (library+forge, 15 sites). **Absorption docs**: 7 WGSL shaders updated to "absorbed upstream" status. toadStool S139 `pipeline_graph` absorption acknowledged. **Deep debt**: `gpu_or_exit()` async helper eliminates 5-line GPU init boilerplate (~75 binaries); duplicate `max_abs_diff` in `validate_gpu_promotion` eliminated; 2 largest GPU binaries refactored to use `gpu_or_exit()`. **Full audit**: zero unsafe, zero TODOs, zero mocks in production, zero hardcoded paths, zero non-Rust deps (wgpu is the GPU bridge), all library files < 800 LOC. V92 handoff written. 968 lib + 71 forge tests PASS.
 **S136**: Deep audit + evolution — `PetalTonguePushClient::headless()` (socket hardcoding eliminated), `Gpu::read_buffer_u32` (upstream parity), `validate_gpu_pure_workload_all` refactored (976→940 LOC, raw wgpu eliminated), industry GPU parity gap documented (`specs/BENCHMARK_ANALYSIS.md`), Kokkos/Polybench/cuBLAS gap formally requested in V92 handoff. 968 lib tests (+2 headless client). All casts audited (f64→f32 for GPU, usize→f64 no From impl — legitimate).
@@ -26,8 +26,8 @@ Mixed-hardware (mH), and Multi-GPU (mG).
 
 | Category | Count | Status |
 |----------|-------|--------|
-| Python baselines | 331/331 | **COMPLETE** |
-| Rust native validation | 1048 lib + 9 integration + 71 forge tests, 41+ modules, 233 binaries | **COMPLETE** |
+| Python baselines | 397/397 | **COMPLETE** |
+| Rust native validation | 1085 lib + 9 integration + 71 forge tests, 46 modules, 245 binaries | **COMPLETE** |
 | BarraCUDA primitives | 272/272 | **COMPLETE** |
 | BarraCUDA CPU (bC) | **24/25** papers (96%) | **ALL GREEN** |
 | BarraCUDA GPU Tensor (gT) | **23/25** papers (92%) | **ALL GREEN** |