Full-dataset slicing run (Phase 3)

uv run driftwatch data slice all --nT 4 --nE 12 --stride 1 --ratio 0.10 --min-malware 50, code version 4ce5cba (clean tree), data manifest sha256:50bf7ae3... (cache v2), 11 Oct 2026, 8 GB Windows laptop, memory cap 1.5 GB. Stride 1 is the heaviest setting (most window overlap, most reloads), chosen as a stress test. This run exercises the slicer only: no model, no AUT. Per-slot manifests are in results/slicing/ (not committed).

	LAMDA (month)	BODMAS (2 weeks)
Grid slots / usable / skipped	140 / 97 / 43	29 / 26 / 3
Windows (nT 4, nE 12, stride 1)	47	2
Evaluation slots yielded	564	24
Evaluation rows after C3	2,466,046	67,065
Training rows loaded	1,422,358	34,229
Wall-clock	28.6 s	2.9 s
Peak memory	0.40 GB	0.69 GB
Feature matrices densified	0	0 (dense by format)
C1, C2, C3 violations	0	0
Rows excluded by the placeholder rule	0	5,877 benign (2020-01-01)

Skipped BODMAS slots: 2w:2019-12-30 (323 benign after exclusion, 35 malware kept at 10%), 2w:2020-01-13 (27 benign), 2w:2020-01-27 (14 benign). This reproduces ADR-0004 decision 5. The first BODMAS run (9 slots) is shorter than nT + nE = 16, so it hosts no window at this setting.