feat(E3): Muon A/B verdict — 4.8287 (−2.727pp from the optimizer alone); factorial cell 4 queued

run-004-pkm-muon vs run-003-pkm differs only in optimizer: 4.8287 vs
7.5553 full-set windowed DER. Adopt gate (≥0.3pp) exceeded 9x. With E2:
gap decomposition ≈0.70pp capacity + 2.73pp optimization + 2.25pp
residual. Training CE ~0.007 vs ~0.02; ~1.2s/step vs ~3.4s.

run-005-muon (vanilla + Muon) queued to close the 2x2 factorial —
orchestrator ARMS extended; completed arms skip via markers (idempotent).

Author: Ronald Tse

docs/EXPERIMENTS.md

@@ -110,10 +110,25 @@ All rows passed the CER parity gate at release. Readings:
 
 ## E3 — Muon optimizer A/B (run-004-pkm-muon)
 
-- **Status:** queued behind E2 (same A10G slot, serialized; the
-  server-side `qwen_next_chain` orchestrator sequences both arms from
-  durable volume markers — best/config.json, final_eval.json,
-  chain_log.jsonl — with 20-min stall detection and respawn).
+- **Status:** COMPLETE (2026-08-28). **Outcome: ADOPTED — the gate is
+  exceeded 9x.**
+- **Measured (full 1,200, windowed zero-skip, greedy):** Muon arm
+  **4.8287** DER-CE vs the single-variable AdamW arm (run-003-pkm)
+  7.555 — **−2.727pp from the optimizer alone** (adopt gate ≥0.3pp).
+  Against the shipped vanilla student (8.259): 3.430pp of the 5.677pp
+  canonical gap closed (60.4%). Teacher reproduced at 2.5997 in this
+  container (range across eval containers 2.5793–2.5997, bf16
+  autocast; protocol consistent).
+- Training-side corroboration: CE ~0.007 vs the AdamW arm's ~0.02 at
+  equal steps; step time ~1.2s vs ~3.4s (the <15% overhead gate met
+  with margin — Newton–Schulz is cheap next to 1450-byte windows); no
+  stability events.
+- **Combined with E2, the gap decomposition at the ByT5-small rung:**
+  capacity ~0.70pp (PKM) + optimization ~2.73pp (Muon, on the PKM
+  architecture) + residual ~2.25pp (domain coverage — the subset
+  lesson). Caveat carried: the A/B isolates the optimizer ON the
+  memory architecture; the vanilla+Muon cell (run-005-muon) is queued
+  to complete the 2x2 factorial.
 - **Hypothesis:** orthogonalized-momentum updates (Newton–Schulz; the
   optimizer Qwen3.8-Flash-Next / LongCat report) help even in
   knowledge-limited distillation fine-tunes — unmeasured territory for

docs/RESULTS.md

@@ -238,8 +238,27 @@ zero-init gates) on the ByT5-small decoder — single-variable.
 near-zero added compute — below the pre-registered ≥1.0pp win bar, so
 the memory axis is real but not the dominant term of the gap. Gates
 verified engaged (0.034-0.053 at completion). Not shipped; the vanilla
-client rung stands. The optimizer A/B on the same architecture
-(run-004-pkm-muon) is recorded in EXPERIMENTS.md E3.
+client rung stands.
+
+### run-004-pkm-muon — optimizer A/B on the memory student (2026-08-28)
+
+Identical to run-003-pkm except the optimizer (Muon on 2D hidden
+matrices, AdamW group for embedding-like params; EXPERIMENTS.md E3):
+
+| Model | DER-CE (full 1,200) |
+|---|---|
+| Teacher (r6, this container) | 2.5997% |
+| **ByT5-small + PKM + Muon (run-004)** | **4.8287%** |
+| ByT5-small + PKM + AdamW (run-003) | 7.5553% |
+| ByT5-small vanilla + AdamW (run-002) | 8.2590% |
+
+**−2.727pp from the optimizer alone** — the adopt gate (≥0.3pp)
+exceeded 9x; 3.430pp of the 5.677pp canonical gap closed (60.4%)
+combining memory + optimizer. Training CE ~0.007 vs ~0.02 at equal
+steps; ~1.2s/step vs ~3.4s; no stability events. The teacher-student
+gap at this rung decomposes: ~0.70pp capacity + ~2.73pp optimization
++ ~2.25pp residual (domain coverage). The vanilla+Muon factorial cell
+(run-005-muon) completes the decomposition.
 
 Leaderboard context (SadeedDiac-25, Misraj evaluator, zero-skip,
 harakat-projected DER-CE): the teacher tier (r6, 580M) at 2.5793

docs/paper.adoc

@@ -262,7 +262,7 @@ A fourth question completes the frontier analysis: at the shipped rung, is the r
 |r6 teacher |580M |2.582
 |===
 
-Memory closes 0.70pp of the 5.68pp teacher–student gap — a real, controlled improvement at near-zero added compute, but below the pre-registered 1.0pp bar under which we would have called the capacity axis *the* fix. The honest conclusion: the distillation gap at this rung is dominated not by parameter capacity but by optimization and domain coverage (the student's training labels span one corpus's domain; the full-benchmark subset lesson of <<section-leaderboard>> measures exactly this exposure). Capacity is a lever on this frontier — a modest one, now measured.
+Memory closes 0.70pp of the 5.68pp teacher–student gap — a real, controlled improvement at near-zero added compute, but below the pre-registered 1.0pp bar under which we would have called the capacity axis *the* fix. The same pre-registration discipline then isolated the *optimization* term: an identical arm differing only in optimizer — Muon (orthogonalized momentum via Newton–Schulz) on the hidden matrices, AdamW for embedding-like parameters — scored **4.829**, a further −2.73pp. The gap at this rung decomposes into ≈0.7pp capacity + ≈2.7pp optimization + ≈2.2pp residual (domain coverage; the full-benchmark subset lesson of <<section-leaderboard>> measures exactly this exposure). Two practical notes travel with the number: Muon's training CE was ~3× lower at equal steps *and* ~2.8× faster per step on this workload (1450-byte windows dominate; Newton–Schulz is cheap next to them), and the adopt gate (≥0.3pp, set before training) was exceeded nine-fold. Optimization — not parameter count — is the dominant recoverable term of the distillation gap at this rung; capacity is a modest second lever, now measured. (This speaks to the *distillation* gap on a pretrained backbone; it does not reopen the from-scratch collapse of finding 1, which is an initialization effect.)
 
 == The decode protocol is part of the measurement
 [[section-decode]]

src/gpu/modal_distill.py

@@ -185,6 +185,26 @@ def _ensure_src_path() -> None:
         "mode": "sequence",
         "note": "Muon A/B arm; same gate and verdict rule as run-003-pkm",
     },
+    "ara-diac-small-muon": {
+        # factorial completion (EXPERIMENTS.md E3 caveat): the A/B was
+        # PKM+Muon vs PKM+AdamW; this cell measures Muon WITHOUT memory
+        # so the 2x2 {vanilla, pkm} x {adamw, muon} closes cleanly
+        "teacher": "rababa_arabic_byt5/run-006-morph/best",
+        "teacher_volume": "rababa",
+        "student_init": "google/byt5-small",
+        "optimizer": "muon",
+        "muon_lr": "0.01",
+        "train": "r5-units/domain.txt",
+        "train_extra": ["r5-units/replay.txt"],
+        "unit_limits": [24000, 6000],
+        "max_len": 1450,
+        "label_beams": "1",
+        "out": "rababa_arabic_distill_small/run-005-muon",
+        "labels_file": "teacher_labels_v2.jsonl",
+        "labels_complete": "true",
+        "mode": "sequence",
+        "note": "vanilla ByT5-small + Muon (factorial cell 4)",
+    },
     "ara-diac-tiny": {
         "teacher": "rababa_arabic_byt5/run-006-morph/best",
         "teacher_volume": "rababa",
@@ -1651,6 +1671,7 @@ def qwen_next_chain() -> dict:
     ARMS = [
         ("ara-diac-small-pkm", "rababa_arabic_distill_small/run-003-pkm"),
         ("ara-diac-small-pkm-muon", "rababa_arabic_distill_small/run-004-pkm-muon"),
+        ("ara-diac-small-muon", "rababa_arabic_distill_small/run-005-muon"),
     ]
     ROOT = Path("/checkpoints")