yhengaa/scpPower: scpPower 0.3.1 — TMT endpoint consistency and validation

Abstract

M2b and M4b TMT validation — amendment I Diagnostic J triggered amendment I under the fixed 5-pp rule before changes. Both TMT presets now use the endpoint aggregated mask for p_g/q_g; generating intensity, variance and endpoint definitions remain unchanged. Registered M4b: specht2019v3 72/72 (PASS); leduc2022_pSCoPE 69/69 (PASS) All predictions precede observations. Higher-power-window agreement and signed gaps are descriptive, not gated. The accepted v0.3.0 verdict and non-TMT results stand. Includes Diagnostic J, M2b and M4b tables, plots, seeds, hashes, versions, timing and cluster accounting. No M5 work. In carrier-TMT presets, unconditional power is validated only up to observed power of about 0.3. At twofold, v0.3.0 predictions ran above observation by up to 20 percentage points, associated with over-predicted testability. Q Exactive and timsTOF could not be validated at useful power for lack of cells. No TMT design numbers or new design examples are presented in the README or a vignette during this round. The 1-pp conditional-power description holds for mean signed TMT gaps: -0.82 pp for Specht and +0.94 pp for Leduc across counted points. Brunner averages +1.30 pp and Q Exactive -0.46 pp. It is not a bound on every point: among points with maximum power at least 0.2, maximum conditional gaps are 1.53 pp for Specht, 4.56 pp for Leduc and 6.05 pp for Brunner. Across all counted points the maxima are 7.77, 11.13, 7.57 and 2.32 pp respectively. The high-power agreement counts do reproduce: 3/8 Specht, 11/14 Leduc and 12/12 Brunner; Q Exactive has none. Testability is the main TMT discrepancy, while the statement that the entire gap is testability with universal agreement within 1 pp is stronger than the recorded columns support. This clarification does not change any data, criterion or accepted verdict. Software: R=4.6.1; scp=1.22.0; limma=3.68.5; edgeR=4.10.5; scplainer=workflow in scp 1.22.0; no separate package