Update README to v4.0.0 — GodelReplay + Two-Layer Architecture

- DOI badge updated to zenodo.19886315 (v4.0.0)
- Latest Results: GodelReplay mem sweep table (sweet spot mem=200 +4.1%)
- Validation table: GodelReplay + mem sweep rows added
- Roadmap: Q2 2026 GodelReplay sprint marked complete
- Repo structure: strategies/, experiments/, results/ documented

Co-Authored-By: Rk/RNA (Claude Code) <noreply@anthropic.com>

Author: creator35lwb-web

README.md

@@ -7,7 +7,7 @@
 **🎉 EXTERNALLY VALIDATED** — Our C-S-P philosophy independently confirmed by [SimpleMem (UNC/Berkeley, Jan 2026)](https://arxiv.org/abs/2601.02553)
 
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
-[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.18048374.svg)](https://doi.org/10.5281/zenodo.18048374)
+[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.19886315.svg)](https://doi.org/10.5281/zenodo.19886315)
 [![Whitepaper](https://zenodo.org/badge/DOI/10.5281/zenodo.18053612.svg)](https://doi.org/10.5281/zenodo.18053612)
 [![MACP & LEP](https://zenodo.org/badge/DOI/10.5281/zenodo.18504478.svg)](https://doi.org/10.5281/zenodo.18504478)
 [![GitHub Discussions](https://img.shields.io/github/discussions/creator35lwb-web/godelai)](https://github.com/creator35lwb-web/godelai/discussions)
@@ -22,7 +22,26 @@
 
 ---
 
-## 🔥 Latest Result (April 2026)
+## 🔥 Latest Results (April 2026)
+
+### GodelReplay — Two-Layer Architecture Validated
+
+**GodelReplay = GodelPlugin (Fisher-scaled EWC-DR) + Avalanche Replay**
+
+Validated on PermutedMNIST (10 tasks, seed=42). Memory buffer sweep across [50, 200, 500]:
+
+| mem_size | Replay-only Forgetting | GodelReplay Forgetting | Delta |
+|----------|:----------------------:|:----------------------:|:-----:|
+| 50 | 0.3902 | 0.4038 | −3.5% *(below replay floor)* |
+| **200** | 0.2549 | 0.2443 | **+4.1%** ← sweet spot |
+| 500 | 0.1459 | 0.1419 | +2.8% |
+
+GodelPlugin's complementarity peaks at **mem=200** — where replay provides partial coverage and EWC-DR fills the weight-identity gap. Below mem=50, Fisher estimates become unreliable (< 5 samples/task).
+
+**Kaggle kernels:** [godelai-replay-permutedmnist-v1](https://www.kaggle.com/code/creator35lwb/godelai-replay-permutedmnist-v1) · [godelai-mem-sweep-v1](https://www.kaggle.com/code/creator35lwb/godelai-mem-sweep-v1)  
+**Results:** [results/GODELREPLAY_PermutedMNIST_v1.md](results/GODELREPLAY_PermutedMNIST_v1.md) · [results/GODELREPLAY_MemSweep_v1.md](results/GODELREPLAY_MemSweep_v1.md)
+
+---
 
 ### Conflict Data Proof — VERDICT: GO
 
@@ -206,14 +225,20 @@ GodelAI is unique in AI history — it was **co-created across five AI models**:
 godelai/
 ├── godelai/              # Core framework
 │   ├── agent.py          # GodelAgent with T-Score & Sleep Protocol
+│   ├── avalanche_plugin.py  # GodelPlugin (Avalanche SupervisedPlugin)
+│   ├── strategies/       # GodelReplay factory
+│   │   └── godel_replay.py  # create_godel_replay_strategy()
 │   ├── core/             # GodelaiAgent implementation
 │   ├── models/           # Model architectures
 │   └── reg/              # EWC and regularization
+├── experiments/          # Benchmark experiment scripts
+│   ├── permutedmnist_godelreplay.py   # 4-strategy comparison
+│   └── permutedmnist_mem_sweep.py     # Buffer size sweep [50,200,500]
+├── results/              # Validated benchmark results
+│   ├── GODELREPLAY_PermutedMNIST_v1.md
+│   └── GODELREPLAY_MemSweep_v1.md
 ├── datasets/             # Training & test datasets
-│   ├── conflict/         # Conflict data for C-S-P activation
-│   └── wisdom/           # YSenseAI integration (future)
 ├── notebooks/            # Interactive demos
-│   └── GodelAI_EWC_Demo.ipynb  # Mnemosyne Colab
 ├── tests/                # Test suite
 ├── docs/                 # Documentation
 ├── whitepaper/           # Technical whitepaper
@@ -228,9 +253,11 @@ godelai/
 | T-Score Formula | Correctly measures gradient diversity | ✅ Verified |
 | Sleep Protocol | Triggers at T < 0.3 | ✅ Verified |
 | EWC Integration | 21.6% forgetting reduction | ✅ Verified |
-| **Fisher Scaling + EWC** | **82.8% forgetting reduction on conflict data** | **✅ NEW RECORD** |
+| **Fisher Scaling + EWC** | **82.8% forgetting reduction on conflict data** | **✅ Verified** |
 | Cross-Platform | 0.0000 variance (Manus + Claude + Colab) | ✅ Verified |
 | **External Validation** | **C-S-P confirmed by SimpleMem paper** | **✅ Verified** |
+| **GodelReplay (PermutedMNIST)** | **+0.87% forgetting reduction vs Replay-only (mem=500)** | **✅ Verified** |
+| **GodelReplay Mem Sweep** | **Sweet spot: +4.1% at mem=200; boundary at mem=50** | **✅ Verified** |
 | Training Improvement | No improvement over baseline | ❌ Not proven |
 | Transformer Support | Not yet tested | ⏳ Pending |
 
@@ -256,9 +283,17 @@ godelai/
 - 🔄 YSenseAI integration research
 - 🔄 Community engagement
 
-### Q2-Q4 2026
+### Q2 2026: GodelReplay Sprint ✅
+- ✅ **GodelPlugin** — Avalanche SupervisedPlugin (Fisher-scaled EWC-DR + T-Score)
+- ✅ **GodelReplay** — `godel_replay.py` factory: Replay + GodelPlugin combined strategy
+- ✅ **PermutedMNIST benchmark** — 4-strategy comparison; HYPOTHESIS CONFIRMED
+- ✅ **Memory buffer sweep** — [50, 200, 500]; sweet spot at mem=200 (+4.1%)
+- ✅ **Two-Layer Architecture** — GodelReplay (training) + GodelAI-Lite (inference) validated
+- ✅ **Zenodo v4.0.0** — DOI [10.5281/zenodo.19886315](https://doi.org/10.5281/zenodo.19886315)
+
+### Q3-Q4 2026
 - 📋 Conflict data benchmarks
-- 📋 Research paper (focus: data requirements for C-S-P)
+- 📋 Research paper (focus: data requirements for C-S-P + Two-Layer Architecture)
 - 📋 Multi-modal data experiments
 - 📋 YSenseAI production integration