ICC Decision Suite VS9 — StatLab pilot
30 October 2026

INSTALL
Extract to a NEW permanent folder and run BUILD_SUITE.cmd.
Disable older suite/ribbon XLAM add-ins, restart Excel, then load
  dist\ICC_Decision_Suite_VS9.xlam
Keep dist\suite_assets beside it. Keep VS2 XLL loaded for RiskSim LIVE formulas;
StatLab itself needs no XLL. Existing RiskSim and Sensitivity commands are retained.
VS6 form properties and VS7 font fixes remain in the builder.
This is a source/build package; build/compile and test in your Windows Excel.

NEW WORKED EXAMPLE
Open examples\ICC_StatLab_VS9_Worked_Examples.xlsx and read Start.
- Descriptions/correlations: select Data!B7:D13; choose Yes for headings.
- Regression: select Data!B7:C13; choose Yes. FIRST column is Y, SECOND is X.
The suite menu's StatLab tile offers the three analyses. There are also ribbon
buttons Describe Data, Correlations and Regression.
Known results: n=6, Y mean 17, X mean 3.5, slope 2, intercept 10,
residual sum of squares 4, residual standard error 1, R squared 0.945945946,
Pearson r 0.972597525, slope standard error 0.239045722,
slope two-sided p 0.001116055242. Expected tab supplies these references.

DESCRIPTIVE STATISTICS
Each selected column is summarized independently: numeric count, missing count,
mean, median, sample standard deviation/variance, min, inclusive Q1/Q3, max,
standard error, 95% Student-t confidence interval for the mean, and sum.
Blank cells and formulas returning empty strings are missing. Other text, numeric
strings, booleans and Excel errors are rejected with cell context.
One observation permits mean/median/min/max, but sample variance/SE/CI are n.a.
Mean CIs assume independent observations; a normal-population model is required
for exact small-sample t coverage. Descriptive summaries are NOT simulation stats.

CORRELATION
Pearson matrix based on rows complete across ALL selected columns. This listwise
rule uses the same observations for every coefficient. Report includes complete
and omitted row counts. Constant columns yield n.a. (constant), including their
diagonal. Correlation does not establish causation. No p-values are added here.

SIMPLE LINEAR REGRESSION
Unweighted ordinary least squares with intercept. FIRST selected column is Y,
SECOND is X. Require at least 3 complete pairs and variation in X.
Results include slope/intercept, SEs, t statistics, two-sided p-values, 95% CIs,
R squared, Pearson r, residual degrees of freedom, SSE and residual standard
error. Complete X/Y pairs, fitted values and residuals are exported as snapshots
along with an observed scatter and fitted line.
Inference assumes a correctly specified linear relationship and independent,
normal, constant-variance errors. It is not a diagnostic confirmation of those
assumptions. No prediction intervals or multiple regression are included.
Constant Y gives undefined R squared and Pearson correlation; zero residual
variance gives undefined t/p statistics rather than fabricated p-values.
Excel T.INV.2T and T.DIST.2T supply the t-distribution calculations.

DATA AND REPORTING
Select one rectangle, at most 20 columns and 100000 observations, up to 2 million
cells. Current calculated values are read once; no automatic source recalculation
is performed. Recalculate the workbook yourself before selecting data if needed.
Hidden/filtered rows are INCLUDED. Dates are Excel serial numbers; don't include
date/category columns in numeric analyses unless that is your intention.
Results are separate ST9_... sheets in the SOURCE workbook. Input data is not
changed. Re-run after edits; results do not automatically update from the source.
Descriptive/correlation reports record the source address and summaries, but do
not copy the entire input dataset. Regression includes the complete paired data.
Large correlation ranges may take time; the pilot runs on Excel's main thread.
An interrupted/failed report build can leave a partial ST9_ sheet; do not treat
that partial output as complete. An error is reported.

RETAINED EXAMPLE
examples\ICC_Sensitivity_VS8_Worked_Examples.xlsx remains included and is compatible
with the VS9 Sensitivity commands. Its VS8 title identifies the feature version,
not an instruction to install the old add-in alongside VS9.

CURRENT MODULE STATUS
RiskSim, Sensitivity, StatLab: supplied pilot code. DecisionTree, EvoOpt, NeuralLab,
RiskOpt, AI Assist: planned. StatLab is an initial subset, not a complete StatTools
replacement. No third-party branding or proprietary analysis code is included.

VALIDATION
Independent benchmark arithmetic and Student-t references were checked with
SciPy. Workbook formula outputs and all three sheets were visually reviewed.
Source checks cover callbacks, image references, imports, guards and report code.
Windows VBA compilation, native worksheet functions, plot rendering and complete
Excel operation remain acceptance tests on your machine. Source checks alone do
not establish runtime correctness. Every upgrade includes a worked workbook.
