# Audit Committee Transparency Barometer — 2026 research files

This note lists the files in the September 2026 research package. Download them from the Downloads section of the Audit Committee Transparency Barometer showcase. Question numbers on the showcase and in these files run Q1 through Q13.

## Question numbers

`question_id` is the public number, Q1–Q13. `historical_question_id` is the number used when the proxies were scored. For Q1–Q10 the two numbers are the same. Three later questions keep their original numbers in `historical_question_id` and appear in public as follows:

- historical Q13 is public Q11 (board skills matrix)
- historical Q14 is public Q12 (technology and AI skills)
- historical Q15 is public Q13 (audit committee oversight of AI)

`data/exact_model_prompt.md` keeps the original Q13–Q15 headings in the prompt itself. A short note at the top of that file records the public numbers. The scoring instructions are unchanged.

## What is included

### Evidence

`long/readings_long.csv` is included. It is one row per company and question for 2010, 2015, and 2020–2026, with the quoted evidence. The same table is also split by year as `long/by_year/readings_long_YYYY.csv`. Both are linked from the showcase Downloads section. `question_id` in these files is already Q1–Q13.

### Figures

Charts of the share of scored proxies at Rating 2 or Rating 3, for 2010, 2015, and 2020–2026.

- `figures/three_index/` — S&P 500, S&P MidCap 400, and S&P SmallCap 600
- `figures/sp1500/` — the combined S&P 1500

The chart files use the public question numbers Q1–Q13.

### Data tables

| File | What it contains |
|------|------------------|
| `data/chart_ready_rating_shares.csv` | Rating counts and shares by year, index, and question. This is the table behind the charts. |
| `data/question_results.csv` | The same counts and shares, with the population and unscored counts. `question_id` is Q1–Q13 and `historical_question_id` is the scored number. |
| `data/question_codebook.csv` | Each question’s title and the Rating 1, 2, and 3 criteria, with both question numbers. |
| `data/data_dictionary.csv` | What each column means. It describes the files in this package. |
| `data/population_coverage.csv` | Which companies are in the S&P 1500 population each year, and which filing was scored. |
| `data/exact_model_prompt.md` | The instructions used to score each proxy. |
| `data/schema_mapping.csv` | How package columns relate to the source fields. |
| `data/q_series.json` | The same rating shares in the form used to draw the charts. |
| `data/build_receipt.json` and `data/run_manifest.json` | When the package was built, and which source run it came from. |

### Examples

`exemplars/selected_examples_2026.csv` is the 26 examples shown on the showcase: the filing passage, where it appears, and the rationale for the rating. `question_id` is the public Q1–Q13 number.

## How to read a share

Shares are the count at that rating divided by the number of scored proxies. A proxy whose skills matrix could not be read is left out of that question’s count. It is not treated as Rating 1. In `readings_long.csv`, `assessment_status` is blank when the row was not assessed (the filing was omitted, or the reading did not return a usable result). The outcome is then in `reading_status`.