All pull requests

Pull requests

• v4 architecture realign: v4.1–v4.6 stack + trained SUT on GPU (merge gate FAILED — ship-blocking follow-up v4.7 identified)

  #30·FluffyAIcode opened Apr 22, 2026Owner·

  ·

• v3.46 · session-layer viability: prefix channel (A/C) is the ship track; B_ams_text is a diagnostic

  #29·FluffyAIcode opened Apr 22, 2026Owner·

  ·

• v3.46-trained: 60-step training + audit (18/26, −3 vs fresh 21/26)

  #28·FluffyAIcode opened Apr 21, 2026Owner·

  ·

• v3.46: revert [E] top1-exclusive content_bias (single Cfg flip) [draft pre-audit]

  #27·FluffyAIcode opened Apr 21, 2026Owner·

  ·

• v3.45 cond-buffer: fix 4.23 aliasing, audit 21/26 (+1, no regressions)

  #26·FluffyAIcode opened Apr 21, 2026Owner·

  ·

• v3.45 (staged): revert [B], align rare_keyword refresh timing on write [draft pre-audit]

  #25·FluffyAIcode opened Apr 21, 2026Owner·

  ·

• v3.44 rewrite (A/B/C/D/E/F) + audit: 18/26 under v3.49 runner, fresh init

  #24·FluffyAIcode opened Apr 21, 2026Owner·

  ·

• v3.49: revert 4.24 runner fallback, enforce context_descriptor substitution ban

  #23·FluffyAIcode opened Apr 21, 2026Owner·

  ·

• v3.48 stacked attention-sharing mechanisms 1+2+3+4: 120-step train + audit = 19/26 (prediction partially refuted)

  #22·FluffyAIcode opened Apr 21, 2026Owner·

  ·

• v3.47 mechanism-1 diagnostic: frozen Qwen attention pool outperforms learned MemoryContextEncoder by 30% on 4.24

  #21·FluffyAIcode opened Apr 20, 2026Owner·

  ·

• v3.46 de-overfit: fix probes 4.22/4.23/4.24 (held-out prompts / paraphrase queries / 4 domains); audit on v3.44-Trained ckpt = 19/26

  #20·FluffyAIcode opened Apr 20, 2026Owner·

  ·

• Runner v3.45 update: 4.23/4.24/4.25 corrected metrics + axis-coverage + determinism hook + audit on v3.44-Trained (19/26)

  #19·FluffyAIcode opened Apr 20, 2026Owner·

  ·

• Spec correction: compression-communication channel definition (Section 1.1) + rewrite of probes 4.22–4.25 + axis-coverage reporting

  #18·FluffyAIcode opened Apr 20, 2026Owner·

  ·

• v3.44-Trained black-box audit: 60-step CPU training breaks 17/26 plateau

  #17·FluffyAIcode opened Apr 20, 2026Owner·

  ·

• v3.43 black-box audit: I-1..I-7 convergence fixes for 9 v3.42 failures

  #16·FluffyAIcode opened Apr 20, 2026Owner·

  ·

• v3.42 black-box audit: H-1..H-5 convergence fixes for 9 v3.41 failures

  #15·FluffyAIcode opened Apr 20, 2026Owner·

  ·

• v3.41 black-box audit: G-1..G-7 structural fixes for 10 v3.40 failures

  #14·FluffyAIcode opened Apr 20, 2026Owner·

  ·

• v3.40 black-box audit

  #13·FluffyAIcode opened Apr 20, 2026Owner·

  ·

• v3.39 black-box audit

  #12·FluffyAIcode opened Apr 20, 2026Owner·

  ·

• v3.38 black-box audit — 15/19 original PASS (new best), 1/2/4 probe PASS/NI/FAIL

  #11·FluffyAIcode opened Apr 20, 2026Owner·

  ·

• Spec: add Section 7 (Reporting Discipline) + Cipher-System Structural Probes (4.20–4.26)

  #10·FluffyAIcode opened Apr 20, 2026Owner·

  ·

• v3.37 black-box audit — 14/19 PASS (new best), 4.16 retrieval fix confirmed

  #9·FluffyAIcode opened Apr 19, 2026Owner·

  ·

• v3.36 black-box audit: 12/19 PASS (4.10 flipped to PASS), 1169s CPU

  #8·FluffyAIcode opened Apr 19, 2026Owner·6 tasks done·

  ·

• v3.35 black-box audit: 13/19 PASS (best across v3.31-v3.35), 1176s CPU

  #7·FluffyAIcode opened Apr 19, 2026Owner·6 tasks done·

  ·

• v3.34 black-box audit: 12/19 PASS (best across v3.31-v3.34), 1262s CPU

  #6·FluffyAIcode opened Apr 19, 2026Owner·6 tasks done·

  ·