Session 1: hardware research, content measurement, codec decision, MAME harness

Verified GVRAM is one word-access per pixel in ALL color modes; chose 256-color
256x192 with movem.l bursts (page 1 sacrificed as double-buffer).

Measured 8 scenes from the Blu-ray source: blit costs under 8% of the 12fps
cycle budget, so I/O is the bottleneck, not CPU. Naive delta+RLE reaches only
3.2:1 (365 KB/s, 470MB) -> decision to use 4x4 vector quantization (~30 KB/s).

"Shot on twos" assumption failed: the transfer has zero duplicate frames, so
12fps requires explicit decimation.

Documents three false measurement results and their root causes (per-frame
Floyd-Steinberg dithering, temporal denoise, exact-match dedupe on noisy source).

MAME Lua injection harness works and is reusable for cycle-cost measurement;
the IOCS _B_READ disk benchmark is blocked returning -1.

Claude-Session: https://claude.ai/code/session_01194oWYW8DQXK1SZ2DnChW6

tools/analysis/03_twos_detect.py

@@ -0,0 +1,11 @@
+import numpy as np, glob, sys
+from PIL import Image
+d=sys.argv[1]
+f=sorted(glob.glob(f"{d}/f*.png"))
+a=[np.asarray(Image.open(x).convert("RGB"),dtype=np.int16) for x in f]
+# noise-tolerant per-pair change: % of pixels differing by more than 8 levels
+p=np.array([100.0*np.count_nonzero(np.abs(a[i]-a[i-1]).max(axis=2)>8)/(a[0].shape[0]*a[0].shape[1]) for i in range(1,len(a))])
+print(f"  pairs={len(p)}  mean={p.mean():.2f}%  median={np.median(p):.2f}%  p90={np.percentile(p,90):.2f}%  max={p.max():.2f}%")
+ev,od=p[0::2],p[1::2]
+print(f"  even-idx pairs mean={ev.mean():.2f}%   odd-idx pairs mean={od.mean():.2f}%   ratio={max(ev.mean(),od.mean())/max(min(ev.mean(),od.mean()),1e-9):.1f}x")
+print(f"  first 16 pairs: {np.round(p[:16],2)}")