Omicau 0.5.2 evaluation results and reproduction package

Post-execution aggregate results and reproduction materials for the Omicau 0.5.2 software article. The archive contains a 4,800-dataset complementary-signal experiment performed with Omicau 0.5.1 and a preselected 90-case categorical bridge on 0.5.2; two complete 2,400-dataset diagnostics on 0.5.2; five released-CLI molecular cohorts with three seeds each; and task-matched DIABLO or multiblock-PLS comparisons across the same outer assessment groups. It includes aggregate data, case banks, scripts, and Figure 3–7 inputs and exports. Raw molecular matrices and row-level predictions are excluded.

Differences from the earlier frozen suite

1. Difference 1 — Evaluated release and execution scope. The molecular application and full diagnostic runs use the released 0.5.2 path. The 4,800-case Experiment A core used 0.5.1; its preselected 90-case bridge agreed categorically with 0.5.2. Version identity is attached to each result.
2. Difference 2 — Direct user workflow. Five public-data cohorts were run through the released omicau run command for three fixed model seeds each; their modality ledgers and reports are interpreted separately from the synthetic wrapper.
3. Difference 3 — Task-matched published comparator. A focused fixed-split DIABLO or multiblock-PLS wrapper uses the same outer assessment groups and training-only tuning. Its analysis is descriptive and does not replace the registered historical 35-contrast family.
4. Difference 4 — Diagnostic operating-characteristic grid. Both modeled batch and outcome-associated missingness families were executed as complete 2,400-dataset 0.5.2 grids. Their alarm proportions describe those generators, not universal leakage-detection sensitivity.
5. Difference 5 — Exploratory external evidence. An earlier TCGA-to-CPTAC assessment remains exploratory supporting context. No new external-holdout claim is assigned to the present released CLI evaluation.

This is a separate post-execution record. The earlier frozen suite remains available as historical protocol and public-data source-registry context. The comparison is descriptive and no universal software ranking, clinical utility, or molecular mechanism is claimed.