# results_online_moment_sw2

**Claim:** `online-moment-sw2` — past-only causal EWMA μ/σ **affine-to-train** or **whiten** on Acc tip streams (CPU):
1. published SW2 `fuse_general` a=0.5
2. Acc leader `hard_abs_swmix|t1` from `sw2_swmix_blend`

**τ∈{5,15,30,60}s** + scale-only @τ=30. **DISTINCT** from `online-moment-hp30` (4-seed linear_hp30) and Acc-stream soft/hard blend re-grids (no SW2↔swmix w re-grid).

HMM preset **0.994 / 0.8**. Code: `online_moment_sw2.py`. JSON: `results/research/online_moment_sw2.json`.

## Selection (mean LODO Acc then H; never AASD/Carta)

**SELECTED `tip_raw`** — mean LODO H **0.7916** Acc **72.19**

| Held-out | H | Acc % |
|---|---:|---:|
| AVGC | 0.7159 | 67.94 |
| DTU | 0.8390 | 74.67 |
| KUL | 0.8064 | 73.41 |
| UHD | 0.7821 | 72.55 |
| USTC | 0.8147 | 72.38 |
| **mean** | **0.7916** | **72.19** |

### Variant table (LODO Acc desc; selected first)

| Variant | LODO H / Acc | AASD H / Acc | ΔAcc vs 53.53 | ΔAcc vs 53.12 | Carta H / Acc |
|---------|--------------|--------------|---------------|---------------|---------------|
| `tip_raw` **←** | 0.7916 / 72.19 | 0.6424 / 53.53 | +0.00 | +0.41 | 0.6460 / 59.71 |
| `sw2_raw` | 0.7914 / 72.09 | 0.6439 / 53.46 | -0.07 | +0.34 | 0.6470 / 59.61 |
| `tip_affine_t30_scale` | 0.7921 / 71.92 | 0.6435 / 53.50 | -0.03 | +0.38 | 0.6448 / 59.78 |
| `sw2_affine_t30_scale` | 0.7915 / 71.56 | 0.6440 / 53.35 | -0.18 | +0.23 | 0.6423 / 59.47 |
| `sw2_affine_t60` | 0.7775 / 68.72 | 0.6462 / 53.36 | -0.17 | +0.24 | 0.6462 / 60.29 |
| `sw2_whiten_t60` | 0.7836 / 68.40 | 0.6439 / 53.22 | -0.31 | +0.10 | 0.6558 / 60.23 |
| `tip_affine_t60` | 0.7728 / 68.13 | 0.6467 / 53.47 | -0.06 | +0.35 | 0.6577 / 60.45 |
| `tip_whiten_t60` | 0.7803 / 67.83 | 0.6449 / 53.45 | -0.08 | +0.33 | 0.6600 / 60.44 |
| `sw2_affine_t30` | 0.7511 / 66.62 | 0.6456 / 53.39 | -0.14 | +0.27 | 0.6589 / 60.69 |
| `sw2_whiten_t30` | 0.7641 / 66.39 | 0.6435 / 53.29 | -0.24 | +0.17 | 0.6578 / 60.72 |
| `tip_affine_t30` | 0.7452 / 66.00 | 0.6477 / 53.49 | -0.04 | +0.37 | 0.6637 / 60.85 |
| `tip_whiten_t30` | 0.7598 / 65.69 | 0.6451 / 53.36 | -0.17 | +0.24 | 0.6607 / 60.91 |
| `sw2_affine_t15` | 0.6849 / 63.15 | 0.6465 / 53.40 | -0.13 | +0.28 | 0.6606 / 61.30 |
| `sw2_whiten_t15` | 0.7015 / 62.96 | 0.6421 / 53.21 | -0.32 | +0.09 | 0.6543 / 61.33 |
| `tip_affine_t15` | 0.6765 / 62.44 | 0.6466 / 53.46 | -0.07 | +0.34 | 0.6579 / 61.41 |
| `tip_whiten_t15` | 0.6951 / 62.28 | 0.6439 / 53.29 | -0.24 | +0.17 | 0.6558 / 61.44 |
| `sw2_affine_t5` | 0.5897 / 59.35 | 0.6474 / 53.44 | -0.09 | +0.32 | 0.6516 / 60.83 |
| `tip_affine_t5` | 0.5724 / 58.16 | 0.6471 / 53.43 | -0.10 | +0.31 | 0.6451 / 60.47 |

## Targets

| Target | Result |
|---|---|
| AASD Acc vs tip 53.53 / hp30 53.12 | Acc **53.53** (Δ +0.00 / +0.41); H 0.6424 |
| Carta vs linear_hp30 60.54 | Acc **59.71** (Δ -0.83); H 0.6460 |
| Acc>60? | **NO** (AASD Acc 53.53) |
| LODO mean | H 0.7916 Acc 72.19 |

## Verdict

Honest LODO Acc/H selector on causal EWMA moment match of Acc tip SW2 / `hard_abs_swmix|t1` does not create AASD Acc>60. Same qualitative flatness as online-moment-hp30, now on the Acc tip family (~53.46–53.53).

## Do not repeat

Do not rerun past-only EWMA affine/whiten on frozen SW2 fuse a=0.5 or `hard_abs_swmix|t1` tip streams at this HMM. Do not re-grid SW2↔swmix soft/hard.

## NEXT IDEA

Bottleneck is not 1st/2nd-order drift of Acc tip decision streams (online-moment flat on hp30 and SW2/tip). Prefer a **qualitatively new** Acc front-end that clears chance and lifts AASD acc1, or unlabeled emission-aware choice among *already scored* Acc leaders without soft-w re-grids. Keep tip AASD Acc 53.53 and `linear_hp30` Carta 60.54 as portfolio leaders.
