BLLU DISTRIBUTION — SUPPLEMENTARY MATERIAL FINAL CONSOLIDATED VERSION
============================================================

This package contains Supplementary Notes S1–S5, the corresponding
reproducibility code and numerical outputs, the Comment 7 audit
materials, and the input data and R scripts for the empirical discrete
goodness-of-fit analysis.

PACKAGE CONTENTS

01_Notes/ 01_Supplementary_Note_S1_PMF_numerical_checks.pdf Numerical
comparison of the explicit finite-sum PMF representation with the
defining mixture integral.

02_Supplementary_Note_S2_Identifiability.pdf Numerical assessment of the
two-moment mapping. The note explicitly treats these as numerical
diagnostics, not a proof of global identifiability.

03_Supplementary_Note_S3_Quantile_estimator_sensitivity.pdf Mean-square
approximation identity and sensitivity analysis for the large-n quantile
estimator, together with the small-sample diagnostic.

04_Supplementary_Note_S4_Estimation_diagnostics.pdf MLE optimization
settings, convergence diagnostics, MoM checks, likelihood-surface
checks, and replication-count sensitivity. The PDF includes the
likelihood-surface figures for Scenarios I and II.

05_Supplementary_Note_S5_Computational_reproducibility.pdf
Computational details for numerical evaluation of the BLLU PMF and
moments, limiting cases and numerical checks, and the empirical discrete
goodness-of-fit analysis with Monte Carlo calibration.

02_Reproducibility_Code/ S1_PMF_checks/ Python implementation and
compact CSV summaries for the PMF checks.

S2_Identifiability/ Base-R implementation and numerical summary tables
for the two-moment identifiability assessment.

S3_Quantile_sensitivity/ Python implementation, replication-level output
for the targeted quantile-estimator experiment, and
summary/failure/bound tables. The separate R-based small-sample
diagnostic in Note S3, Section 3, is documented in the note, but its R
script and raw replication-level data are not included in this archive.

S4_Estimation_diagnostics/ Main R Monte Carlo simulation and MLE
multi-start diagnostic scripts; full Monte Carlo replication-level
results; full MLE diagnostics (2,400 fit-level and 7,200 start-level
records); convergence and likelihood-surface summaries; and
likelihood-surface grids and PNG plots for representative N = 1,500
samples in Scenarios I and II.

S5_Computational_reproducibility/ 01_BLL_PMF_and_moments.R R
implementation for numerical evaluation of the BLLU PMF and moments.

    02_BLLU_limit_cases_and_checks.R
      R implementation of limiting-case calculations and numerical
      checks of selected distributional properties.

    07_Empirical_GOF_and_MC_calibration.R
      R implementation of the empirical discrete CvM (W2) and AD (A2)
      statistics and their Monte Carlo calibration (50,000 replications).
      The BLLU PMF is evaluated using the LLU quantile representation
      and midpoint quadrature with K = 2,000 points. The script also
      records the R session information.

    BLLU_Application_Data.xlsx
      Input workbook containing the grouped frequency tables and
      observation-level data used in the SO2 and traffic applications.

    README.txt
      Instructions for running the S5 scripts and reproducing the
      associated numerical analyses.

Comment7_audit/ R audit script and summary outputs for the common-sample
and estimator-validity checks associated with Comment 7. The audit uses
the raw Monte Carlo results and full MLE replication diagnostics
supplied in S4. Prefix-sensitivity checks use nested prefixes of the
same 300 replications, not independent Monte Carlo experiments.

REPRODUCTION NOTES

1.  The Notes are ordered S1–S5 to match the order in which the
    numerical checks and supplementary analyses are discussed in the
    manuscript.

2.  The main Monte Carlo study and the Comment 7 audit were implemented
    in R. The targeted PMF and quantile sensitivity checks were
    independently implemented in Python, as described in the Notes. The
    computational checks and empirical goodness-of-fit workflow
    documented in S5 were implemented in R.

3.  Run the S5 scripts from the
    02_Reproducibility_Code/S5_Computational_reproducibility/ folder.
    The empirical goodness-of-fit script reads the included
    BLLU_Application_Data.xlsx and writes bllu_04_CvM_AD_MC_results.csv
    to the working directory.

4.  For the empirical goodness-of-fit analysis, the recorded software
    environment is R version 4.5.2 (Windows 11 x64), with dgof 1.5.1 and
    readxl 1.6.0. The script uses set.seed(123), B = 50,000, and K =
    2,000. The fitted model parameters are held fixed during Monte Carlo
    calibration. The script checks that grouped frequencies agree with
    the observation-level data before testing. The complete
    sessionInfo() output is recorded in the empirical GOF script.

5.  The S3 replication-level output is supplied for the targeted
    quantile experiment in Section 2. The separate R-based small-sample
    diagnostic in Section 3 is documented in Note S3, but its R script
    and raw replication-level data are not included.

6.  S4 includes the full 300-replication Monte Carlo output and the
    complete MLE multi-start diagnostics: 2,400 fit-level records and
    7,200 start-level records. The likelihood-surface grids and plots
    for representative N = 1,500 samples in Scenarios I and II are also
    included.

7.  The Comment 7 audit script uses the raw Monte Carlo results and full
    MLE replication diagnostics supplied in S4. Its prefix-sensitivity
    checks use nested prefixes of the same 300 replications, not
    independent Monte Carlo experiments.

8.  CSV files are supporting numerical outputs; scripts are included to
    permit regeneration of the corresponding analyses where the required
    code and input data are supplied.

FILE-SPECIFIC NOTES

-   The S1 folder contains the executable Python implementation and
    numerical CSV outputs; an editable TeX source is not included.
-   The S2 folder contains the Base-R implementation and supplied
    numerical CSV outputs. A separate Python implementation and a
    heatmap file are not included.
-   The S3 Section 3 R diagnostic is described in the note but is not
    included as executable code in this archive.
-   The S4 folder contains the main simulation and diagnostic scripts,
    complete replication-level outputs, convergence summaries, and
    likelihood-surface data/plots.
-   The S5 folder contains the R scripts, input workbook, and
    computational documentation for the PMF/moment calculations,
    limiting-case checks, and empirical GOF/Monte Carlo calibration.
-   The five Supplementary Note PDFs are the submission-ready documents.

MANUSCRIPT CROSS-REFERENCES

-   Section 2.1: Supplementary Note S1.
-   Section 3.1: Supplementary Note S2.
-   Section 3.3: Supplementary Note S3.
-   Section 3.4: Supplementary Note S4.
-   Empirical CvM/AD goodness-of-fit analysis (Sections 4.1 and 4.2):
    Supplementary Note S5 and
    02_Reproducibility_Code/S5_Computational_reproducibility/.
-   Comment 7 audit: see 02_Reproducibility_Code/Comment7_audit/.

The scientific content and claims of Supplementary Notes S1–S4 have not
been altered in preparing this consolidated archive. Supplementary Note
S5 documents the computational workflow and numerical checks described
above.
