Appendix 14C: Chapter 14 artifact dictionary (what each output is for)

Chapter 14 is designed to produce a small “analysis pack” of artifacts that are:

easy to inspect (CSV / Markdown),
easy to automate (JSON),
easy to review visually (figures + manifest),
and reproducible (regenerate from the same dataset + seed).

This appendix lists each artifact, its purpose, and what to look at first.

Where the Chapter 14 artifacts live

Chapter 14 is run via:

make business-ch14

The Makefile passes an output root like --outdir outputs/track_d. The Chapter 14 script then writes inside a Track D subfolder, so the final location is:

outputs/track_d/track_d/

Typical directory layout:

outputs/track_d/track_d/
  ch14_driver_table.csv
  ch14_regression_design.json
  ch14_regression_summary.json
  ch14_regression_memo.md
  ch14_figures_manifest.csv
  figures/
    ch14_fig01_cogs_vs_units.png
    ch14_fig02_revenue_vs_units.png
    ch14_fig03_actual_vs_predicted_revenue_m3.png

Suggested review order (fastest to insight)

ch14_driver_table.csv (sanity check the monthly driver table)
ch14_regression_memo.md (read the story in plain English)
figures/*.png (does the relationship look stable / linear-ish?)
ch14_regression_summary.json (exact coefficients, R², and structured values)
ch14_regression_design.json + ch14_figures_manifest.csv (repro + reporting metadata)

Artifact dictionary

Chapter 14 outputs (artifact → purpose → what to look for first)
Artifact	Purpose	What to look for first
`ch14_driver_table.csv`	The monthly driver table used for all models (units sold, invoice count, revenue, COGS). This is the “data contract” for Chapter 14.	Check month alignment and magnitude: - Are months continuous? - Are `units_sold` positive and plausible? - Do revenue/COGS values look reasonable and consistent with your NSO story?
`ch14_regression_memo.md`	Human-readable summary (the “starter executive memo”). This is what you would paste into a planning email or a short doc.	Read it top-to-bottom: - Are the slopes/intercepts interpreted correctly (baseline vs rate)? - Do the guardrails make sense (driver lens, not causation)?
`ch14_regression_summary.json`	Machine-readable results: coefficients, R², key diagnostics, and structured values for downstream automation.	Look for: - Coefficients (intercept + slopes) - R² per model - Any “notes”/flags the script includes for interpretation
`ch14_regression_design.json`	Reproducibility “design contract”: expected inputs, driver definitions, and model formulas. Useful for reviewers and for future chapters that want to rely on the same structure.	Confirm: - Model formulas match the chapter narrative (m1/m2/m3) - Driver definitions match the dataset lineage (inventory movements, AR events, IS lines)
`ch14_figures_manifest.csv`	A reporting manifest listing each figure file plus metadata (chart type, title, axis labels, guardrail note, data source). This supports consistent documentation and future “report assembly.”	Open it and confirm: - All figure filenames exist under `figures/` - Titles/labels read well (this matters for end users) - Guardrail notes match the chapter’s interpretation discipline
`figures/ch14_fig01_cogs_vs_units.png`	Visual check for Model 1: COGS vs units sold with a fitted line.	Look for: - Is the relationship roughly linear? - Do you see major outliers dominating the fit? - Does the intercept look plausible (near-zero vs baseline cost)?
`figures/ch14_fig02_revenue_vs_units.png`	Visual check for Model 2: Revenue vs units sold with a fitted line.	Look for: - Is slope stable (implied blended price-per-unit)? - Any clusters suggesting mix/price shifts? - Does a non-zero intercept suggest timing or non-unit revenue?
`figures/ch14_fig03_actual_vs_predicted_revenue_m3.png`	Visual check for Model 3: multi-driver revenue model (actual vs predicted / fit quality view).	Look for: - Are predictions systematically high/low in certain months (patterned residuals)? - Does adding invoice_count appear to improve fit meaningfully? - Any “regime change” behavior (model works early, breaks later)?

How to regenerate the artifacts (and why they are not committed)

Artifacts under outputs/ are generated and not committed to git. This keeps the repo small and makes your results reproducible.

Standard regeneration:

# Ensure the dataset exists locally (gitignored)
make business-nso-sim
make business-validate

# Rebuild Chapter 14 artifacts
make business-ch14

If you want to run Chapter 14 on a different dataset folder:

python -m scripts.business_ch14_regression_driver_analysis \
  --datadir data/synthetic/nso_v1 \
  --outdir outputs/track_d \
  --seed 123

Common “first checks” when something looks wrong

If slopes look odd, first verify ch14_driver_table.csv: a sign convention mistake (e.g., units sold negative) can flip interpretations.
If a figure is missing, open ch14_figures_manifest.csv and confirm the filename is listed and matches the file on disk.
If revenue/COGS appear inconsistent, re-run:
```
make business-validate
```
Track D assumes measurement quality before modeling.

Closing note

Chapter 14 artifacts are intentionally designed to make regression “accounting-grade”:

a driver table you can audit,
models you can explain,
visuals you can sanity-check,
and structured JSON you can automate.

Appendix 14B (NSO v1 data dictionary) explains the tables. Appendix 14C explains the artifacts produced from those tables.