# Data dictionary

## `data/sample_metadata.csv`

One row per retained TCGA expression sample used in the early-stage versus late-stage demonstration.

| Variable | Description |
|---|---|
| `sample_id` | TCGA aliquot/sample identifier carried from the source `SummarizedExperiment`. |
| `ajcc_pathologic_stage` | AJCC pathologic stage from the associated clinical metadata. |
| `group` | Analysis group: `Early` for Stage I/IA/IB and `Late` for Stage III/IIIA/IIIB/IIIC. |

**Important:** the row unit is an expression sample, not necessarily a unique participant. Participant IDs and sample-type codes can be derived from the TCGA barcode. The released metadata contain 441 samples representing 391 unique participant IDs.

## `data/independent_testing_results.csv`

One row per transcript feature from the original independent feature-wise testing workflow.

| Variable | Description |
|---|---|
| `feature_id` | Versioned Ensembl transcript/gene feature identifier from the source expression matrix. |
| `effect_log2_mean_difference` | Difference in mean `log2(count + 1)` expression, Late minus Early. |
| `p_value_raw` | Raw p-value from the feature-wise two-sample t-test performed on the raw STAR count values. |
| `p_value_bh` | Benjamini-Hochberg adjusted p-value. |
| `p_value_bonferroni` | Bonferroni-adjusted p-value. |

## `data/limma_results.csv`

One row per transcript feature from the limma empirical Bayes analysis of `log2(count + 1)` expression values.

| Variable | Description |
|---|---|
| `feature_id` | Versioned Ensembl feature identifier. |
| `logFC` | Estimated Late-minus-Early coefficient on the log2(count + 1) scale. |
| `AveExpr` | Average log-scale expression reported by limma. |
| `t` | Moderated t-statistic. |
| `P.Value` | Raw limma p-value. |
| `adj.P.Val` | Benjamini-Hochberg adjusted limma p-value. |
| `B` | limma B-statistic (log-odds of differential expression under the fitted empirical Bayes model). |

## `data/multiplicity_summary.csv`

Compact set of numerical results used to validate the public analysis.

| Variable | Description |
|---|---|
| `metric` | Name of the summary statistic. |
| `value` | Numerical value for the statistic. |

## `output/validation_checks.csv`

Comparison of supplied summary values with results recalculated directly from the released CSV files. `PASS` indicates agreement within the tolerance used by the public analysis.

## `output/sample_structure.csv`

Diagnostic counts derived from the TCGA sample barcodes, including unique participant IDs and sample-type-code counts. These diagnostics are included for transparency and do not alter the published multiplicity calculations.

## `output/top100_overlap_features.csv`

Feature identifiers appearing in both the top 100 limma features ranked by FDR-adjusted p-value and the top 100 ranked by absolute `logFC`.