| Age | Commit message (Expand) | Author |
| 2026-04-08 | Add BP+penalty control (round 19's #4 critical experiment) | YurenHao0426 |
| 2026-04-08 | PAPER_OUTLINE: §4 rewrite under 'two distinct failure modes' framing | YurenHao0426 |
| 2026-04-08 | 3-seed multi-seed verification of penalized DFA deep cos = +0.17 | YurenHao0426 |
| 2026-04-08 | DISAMBIGUATION: vanilla DFA early-epoch checkpoints + cos measurement | YurenHao0426 |
| 2026-04-08 | Add vanilla DFA early-epoch checkpoint training (round 19 disambiguation) | YurenHao0426 |
| 2026-04-08 | Add null calibration script: training-Bs vs fresh-Bs cos on penalized DFA | YurenHao0426 |
| 2026-04-08 | MAJOR: penalized DFA deep-layer cosine is +0.17, NOT zero | YurenHao0426 |
| 2026-04-08 | Add PAPER_OUTLINE.md: §1-§6 draft reflecting round 17 + 18 | YurenHao0426 |
| 2026-04-08 | Add penalty lambda 3-seed summary script + checkpoint save in penalty test | YurenHao0426 |
| 2026-04-08 | Add penalty λ sweep figure: shows λ-dependence of (d) verdict | YurenHao0426 |
| 2026-04-08 | EVIDENCE_SUMMARY: add (d) threshold sensitivity finding (round 18) | YurenHao0426 |
| 2026-04-07 | Add (d) frozen-baseline threshold sensitivity — IMPORTANT new finding | YurenHao0426 |
| 2026-04-07 | CHECKLIST pitfall #6: layer-0 dominance is ResMLP-specific, not universal | YurenHao0426 |
| 2026-04-07 | EVIDENCE_SUMMARY: round 18 language softening on CNN + penalty audit | YurenHao0426 |
| 2026-04-07 | CHECKLIST: add pitfall #6.5 — silent self-cosine fallback when aux nets not... | YurenHao0426 |
| 2026-04-07 | EVIDENCE_SUMMARY: add §3.7 CNN cross-architecture audit results | YurenHao0426 |
| 2026-04-07 | Add CNN third-architecture audit: BN, no terminal LN | YurenHao0426 |
| 2026-04-07 | Add minimal worked example: end-to-end protocol usage tutorial | YurenHao0426 |
| 2026-04-07 | Add §4 penalty rescue figure: visual two-failure-modes story | YurenHao0426 |
| 2026-04-07 | EVIDENCE_SUMMARY: add §3.5 sensitivity, §3.6 cross-width, §4 separability,... | YurenHao0426 |
| 2026-04-07 | Add §2/§3 hero figure: 5-method audit horizontal bar chart | YurenHao0426 |
| 2026-04-07 | Add d=512 ResMLP audit table (3 seeds): cross-width validation | YurenHao0426 |
| 2026-04-07 | Add fast direction-quality measurement on existing DFA checkpoints | YurenHao0426 |
| 2026-04-07 | Partial protocol audit on penalized DFA: (a)+(b) pass, (d) still fires | YurenHao0426 |
| 2026-04-07 | Add EVIDENCE_SUMMARY.md: consolidated snapshot of all protocol evidence | YurenHao0426 |
| 2026-04-07 | Add §3 cross-architecture temporal evolution figure | YurenHao0426 |
| 2026-04-07 | Add threshold sensitivity analysis: (a) 63x gap, (b) 24338x gap | YurenHao0426 |
| 2026-04-07 | Add ViT-Mini DFA training script that saves checkpoint + Bs | YurenHao0426 |
| 2026-04-07 | Add reproducers for pitfalls 4-6 (Bs reproducibility, aggregation, layer-0) | YurenHao0426 |
| 2026-04-07 | Add training-monitor early-stop demo: 96% compute savings on DFA | YurenHao0426 |
| 2026-04-07 | Cross-architecture temporal validation: 3 archs x 3 seeds x 2 methods | YurenHao0426 |
| 2026-04-07 | Protocol diagnostic (a): use max per-block growth, not max/min ratio | YurenHao0426 |
| 2026-04-07 | Add reproducers for pitfalls 1-3 in CHECKLIST.md | YurenHao0426 |
| 2026-04-07 | Temporal evolution 3-seed: protocol fires at DFA epoch 3-4 on all seeds | YurenHao0426 |
| 2026-04-07 | Add temporal diagnostic evolution: protocol fires at epoch 4 of DFA | YurenHao0426 |
| 2026-04-07 | Add random-init sanity check: protocol does not flag untrained networks | YurenHao0426 |
| 2026-04-07 | Audit table extension to 3 seeds (s42/s123/s456) | YurenHao0426 |
| 2026-04-07 | Add protocol decision-utility ablation table | YurenHao0426 |
| 2026-04-07 | Add DFA direction-quality direct test (codex round 13 option c) | YurenHao0426 |
| 2026-04-07 | Add audit table example: protocol applied to BP/DFA/SB/CB/EP | YurenHao0426 |
| 2026-04-07 | Add FA diagnostic protocol reference implementation | YurenHao0426 |
| 2026-04-03 | Add 5 extra seeds to synthetic cross-state distance (now 10 seeds for all met... | YurenHao0426 |
| 2026-04-03 | Recompute all CNN diagnostics with fixed BP gradient flow | YurenHao0426 |
| 2026-04-03 | Fix CNN compute_bp_grads: remove inter-layer detach so gradients flow to all ... | YurenHao0426 |
| 2026-04-03 | Complete EP data: 10-seed synthetic + 6-seed CIFAR persample + cross-state | YurenHao0426 |
| 2026-04-03 | EP synthetic 10 seeds complete: 30 JSONs + 30 checkpoints + cross-state distance | YurenHao0426 |
| 2026-04-03 | Add checkpoint saving to ep_synthetic.py | YurenHao0426 |
| 2026-04-03 | Add EP cross-state distance for CIFAR + verify CNN summary | YurenHao0426 |
| 2026-04-03 | Add EP synthetic per-seed CSV + synthetic cross-state distance | YurenHao0426 |
| 2026-04-03 | Add EP to BP gradient sparsity analysis | YurenHao0426 |