11 — How to read ECG papers (for Phase 1)

flowchart LR
    ABS["abstract\nclaimed result"] --> TAB["results tables\nnumbers"]
    TAB --> SPLIT["split section\ntrust check"]
    SPLIT --> METH["method\nonly as deep as needed"]

Split reported	Trust	Action
de Chazal DS1/DS2, patient-wise	green	comparable to ours, extract numbers
Random beat split	red — leaked	ignore the scores
Record-level / private data	yellow	note it, never compare directly

• Accuracy only, no per-class table: hides F/Q collapse.
• Split undisclosed: probably leaked.
• Tiny test set, no intervals: luck, not evidence.
• 99% with no split description: decoration, not baseline.

Extract per paper into ../research/literature.md: dataset, split, per-class scores, macro F1, V recall. End with decisions, not vibes: window length, scale range, split protocol, baselines, loss and weighting — each with its justifying paper. That makes Phase 2+ mechanical.

You are now ready for ../research/plan.md Phase 1. Revisit lessons 04 and 10 during Phases 4–5 when thresholds and fusion numbers need interpreting.