# F1-dev-01 — Analytic firm dataset and revenue aggregates

## Question
Build a documented one-row-per-firm analytic dataset from the synthetic file `data/firms.csv` and report: the number of distinct firms, the mean and median revenue in millions of USD over firms with known revenue, and the mean head count over all distinct firms.

## Data and exclusions
`data/firms.csv` has 445 data rows (`n_input_rows`), 5 columns. Detected problems (`data_issues`):
1. **Duplicate rows** — 45 rows repeat an earlier row exactly (same firm_id, identical values); none conflict. Kept first occurrence, giving 400 distinct firms (`n_unique_firms`).
2. **Mixed revenue units** — 60 distinct firms report revenue in `USD_thousands`, the rest in `USD_millions`. All values were converted to millions USD (thousands ÷ 1000). Converted magnitudes match those of the millions rows, so the unit labels look correct.
3. **Missing revenue** — 7 raw rows, i.e. 6 distinct firms (F0048, F0121, F0129, F0269, F0348, F0399), have no revenue (`n_missing_revenue`). They are left out of the revenue aggregates and `out/analytic.csv`, but they still count toward `n_unique_firms` and `mean_employees`.

Exclusion ledger (`out/exclusions.csv`): 445 → (exact duplicates, −45) → 400 → (unit conversion, −0) → 400 → (missing revenue, −6) → 394 = `n_analysed`.

## Method
This is a deterministic descriptive computation in `out/analysis.py`, with no randomness (`random_seed` = null). The baseline `mean_revenue_raw` is the plain mean of the non-empty `revenue` values as read, before any cleaning. Mean and median revenue are computed over the 394 analytic firms in millions USD; with an even n, the median is the average of the two middle values. Mean employees is computed over all 400 distinct firms.

## Results
- Distinct firms: 400 (`n_unique_firms`); firms with missing revenue: 6 (`n_missing_revenue`); firms analysed: 394 (`n_analysed`).
- Mean revenue: **249.96 M USD** (`mean_revenue_musd` = 249.9620761421).
- Median revenue: **162.306 M USD** (`median_revenue_musd`).
- Mean employees: **1101.435** (`mean_employees`).
- Baseline raw mean: 30695.88 in mixed file units (`mean_revenue_raw`). It is not interpretable, because it mixes thousands and millions and counts duplicate rows twice. Comparing it with `mean_revenue_musd` shows how large the unit problem is.
- Feasibility: complete (`feasibility`).

## Uncertainty
These figures are exact statistics of the file as provided, not estimates of a wider population, so no sampling standard errors or confidence intervals are reported. The remaining uncertainty is about how the data were handled, not about sampling. It comes from three sources: (i) trusting the row-level `unit` labels; (ii) the 6 firms with missing revenue — if their revenue is not missing at random, the revenue aggregates could be biased; (iii) treating exact duplicates as assembly errors rather than as separate firms.

## Limitations and open issues
- The data are simulated. Unit labels were checked only for plausible magnitudes; mislabelled units within the plausible range cannot be detected.
- No imputation was done for missing revenue, so the revenue aggregates cover 394 of 400 firms.
- Revenue is right-skewed: the mean (249.96) is well above the median (162.306).