# Data dictionary

One row per decision (`ftc_decisions_metadata.csv`, `.jsonl.gz`, `.parquet`). Full text is in `fulltext/ftc_fulltext_volNNN.jsonl.gz` (one JSON object per decision: decision_id, citation, respondent, text), joinable on `decision_id`.

| column | meaning |
|---|---|
| decision_id | Stable id: v<volume>-<sequence>. Decisions added by later splits end in s1, s2. |
| volume |  |
| citation | Official reporter cite, e.g. '40 F.T.C. 662'. |
| printed_start_page |  |
| printed_end_page |  |
| n_pages |  |
| respondent | Cleaned respondent name(s): manual override > vision read of the page image > Muse extraction > regex. |
| docket | FTC docket number (C- / D- / numeric). Source in docket_source. |
| complaint_date | ISO date of the complaint, when printed. |
| decision_date | ISO date of the decision/order. Blank for ~2.5% (mostly 1915-1925 volumes that print no date). |
| document_type | consent_order, final_order, modifying_order, interlocutory_order, dismissal, opinion, ... (Muse, else regex). |
| is_consent |  |
| case_type | consumer_protection / antitrust / both / other / procedural (Muse; blank = unclassified). |
| statutes | Statutes named in the caption (regex), '|'-separated. |
| practice_tags | Practice tags (Muse; else keyword tags). |
| industry |  |
| product |  |
| outcome | Muse-extracted outcome. |
| relief | Muse-extracted relief terms. |
| order_term_years |  |
| money_usd | Money amount stated (civil penalty/redress), USD, Muse. |
| money_kind |  |
| hearing_examiner |  |
| commission_counsel |  |
| respondent_counsel |  |
| has_dissent |  |
| respondent_location |  |
| respondent_type |  |
| alleged_conduct |  |
| split_confidence | Confidence of the decision boundary in the volume split: high / medium / low. |
| decision_date_source | Where the date came from: manual, vision (page image), muse, regex, with +agreement flags. |
| docket_source |  |
| respondent_source |  |
| quality_flags | Pipeline flags, e.g. respondent_garbled, no_decision_date. Blank = none. |
| merged_from | Continuation fragments merged into this decision (page-split artifacts). |
| year |  |
| year_estimated |  |
| source_pdf_url |  |
| year / year_estimated | Year used for browsing; year_estimated=1 when no date was found and the volume's median year is used. |
| source_pdf_url | Link to the source volume PDF (page anchor where available). |

## Penalty Offense Notices

These CC0 1.0 files contain FTC notice-recipient records, citations to FTC administrative decisions, and later enforcement action matches. `npo_recipients.csv` includes raw names, addresses when published, source PDF/page, normalized company IDs. `npo_cited_decisions.csv` includes reporter citation, case name, page resolution, matching decision ID, external flag, and evidence quote. `npo_followon.csv` includes FTC action title/date/URL and matching evidence.

## Order Status

`orders_status.csv` has one row per decision with the extracted disposition, rule-based status, action evidence basis, link to a set-aside/modifying decision when matched, and the court-complaint sunset caveat. Status classifications are derived from the 1995 FTC policy statement and indexed FTC orders; verify the original decision and action.

## Internal Citations

`citations.csv` contains each internal FTC reporter citation found in a decision page, the page context (up to 600 characters on each side), page id, page-range resolution, and quote-validated treatment label when available. Self-cites are omitted. A `volume_not_in_library` resolution means no decision volume is present in the current database.