Blame for README.md

---
configs:
- config_name: tournaments
  data_files:
  - split: '2023'
    path: tournaments/2023-*
  - split: '2024'
    path: tournaments/2024-*
  - split: '2025'
    path: tournaments/2025-*
  - split: full
    path: tournaments/full-*
- config_name: teams
  data_files:
  - split: '2023'
    path: teams/2023-*
  - split: '2024'
    path: teams/2024-*
  - split: '2025'
    path: teams/2025-*
  - split: full
    path: teams/full-*
- config_name: players
  data_files:
  - split: '2023'
    path: players/2023-*
  - split: '2024'
    path: players/2024-*
  - split: '2025'
    path: players/2025-*
  - split: full
    path: players/full-*
- config_name: games
  data_files:
  - split: '2023'
    path: games/2023-*
  - split: '2024'
    path: games/2024-*
  - split: '2025'
    path: games/2025-*
  - split: full
    path: games/full-*
- config_name: tossup_questions
  data_files:
  - split: '2023'
    path: tossup_questions/2023-*
  - split: '2024'
    path: tossup_questions/2024-*
  - split: '2025'
    path: tossup_questions/2025-*
  - split: full
    path: tossup_questions/full-*
- config_name: bonus_questions
  data_files:
  - split: '2023'
    path: bonus_questions/2023-*
  - split: '2024'
    path: bonus_questions/2024-*
  - split: '2025'
    path: bonus_questions/2025-*
  - split: full
    path: bonus_questions/full-*
- config_name: question_placements
  data_files:
  - split: '2023'
    path: question_placements/2023-*
  - split: '2024'
    path: question_placements/2024-*
  - split: '2025'
    path: question_placements/2025-*
  - split: full
    path: question_placements/full-*
- config_name: tossup_responses
  data_files:
  - split: '2023'
    path: tossup_responses/2023-*
  - split: '2024'
    path: tossup_responses/2024-*
  - split: '2025'
    path: tossup_responses/2025-*
  - split: full
    path: tossup_responses/full-*
- config_name: bonus_responses
  data_files:
  - split: '2023'
    path: bonus_responses/2023-*
  - split: '2024'
    path: bonus_responses/2024-*
  - split: '2025'
    path: bonus_responses/2025-*
  - split: full
    path: bonus_responses/full-*
---
# SCOUT: ACF 2023–2025 quizbowl play, anonymized

SCOUT (Skill Composition Of Unseen Teams) pairs quizbowl questions with
recorded tournament play. This repository is an anonymized replica of
`mgor/acf-co-23-25@1156b66`: ACF Fall, Winter, Regionals, Nationals, and Chicago Open
question sets from 2023 through 2025, with the games, player buzzes, and team
bonus answers played on them. It has the same nine linked configs, splits,
rows, and IDs as the source; only the fields listed under *Anonymization*
differ.

```python
from datasets import load_dataset

games = load_dataset("mgor/SCOUT-anon", "games", split="2024")
tossups = load_dataset("mgor/SCOUT-anon", "tossup_questions", split="full")
buzzes = load_dataset("mgor/SCOUT-anon", "tossup_responses", split="2024")
```

Each config has `2023`, `2024`, `2025`, and `full` splits. `full` has every
row. Tournament, team, game, placement, and response rows use the tournament
or edition year; a player or a canonical question appears in every year it
occurs, so year splits of `players`, `tossup_questions`, `bonus_questions`
overlap.

## Anonymization

| Field | Treatment |
| --- | --- |
| `players` | `player_id`, an opaque `p-NNNNN` numbered in random order, and `name`, set to the same ID. Real names, slugs, person IDs, and source record IDs are removed. |
| `teams.players`, `tossup_responses.player_id` | The same opaque player IDs. |
| `teams.name`, `teams.slug` | A random club pseudonym plus the source suffix (`A`, `B`, `2`, ...). Entries of one club share the pseudonym, and a team name keeps one pseudonym across tournaments. Novelty team names count as their own club. |
| `teams.hash` | MD5 (first 12 hex) of the sorted anonymized roster. |
| `games.packet_name`, `question_placements.packet_name`, question `metadata.packet` | A neutral label such as `Packet C`, `Prelims 3`, or `Packet J (Playoffs 1)`. Source packet names list writers and submitting teams. |

The mapping uses a secret seed and is not published. Tournaments (names, host
sites, dates), question text, and all other IDs are unchanged: they are public
event and packet information and name no players.

## Configs

| Config | Grain | Key fields |
| --- | --- | --- |
| `tournaments` | tournament site | `tournament_id`, `level` (fall/winter/regionals/nationals/open), `field_level`, `dots` (1 to 4 difficulty tier), dates, `question_set` |
| `teams` | team entry at a tournament | `team_id`, `tournament_id`, `name`, `players` (roster of `player_id`) |
| `players` | person | `player_id`, `name` (= `player_id`) |
| `games` | game | `game_id`, `tournament_id`, `team_a`, `team_b`, `round`, `type`, `packet_id`, `tossups_read` |
| `tossup_questions` | canonical tossup | `qid`, `question`, answers, `clue_spans` (character `[start, end]` per clue), `metadata` (category, subcategory, difficulty, question set, packet, `human_buzz_positions`) |
| `bonus_questions` | canonical bonus | `qid`, `leadin`, `parts` (question, answers, `value`, `difficulty_modifier` e/m/h), `metadata` |
| `question_placements` | question x packet edition | `qid`, `question_type`, `packet_id`, `question_number`, `is_canonical_qid_placement` |
| `tossup_responses` | recorded buzz | `game_id`, `qid`, `team_id`, `player_id`, `buzz_number`, `correctness`, `value`, `token_position`, `n_clues_heard` |
| `bonus_responses` | team x bonus part | `game_id`, `team_id`, `bonus_qid`, `bonus_part_id`, `correctness`, `value`, `is_scored` |

Join teams and games to `tournaments` by `tournament_id`, games to teams by
`team_a` / `team_b`, buzzes to tossups by `qid`, bonus responses to bonuses by
`bonus_qid`, and placements to questions by `question_type` and `qid`. A game's
tossups are the placements of its `packet_id` with `question_number` up to
`tossups_read`; tossups with no buzz were read and went dead.

## Interpretation

- `teams.players` is the tournament roster, not the lineup of a game.
- Recorded buzzes are observations, not every opportunity to buzz.
  `token_position` counts whitespace tokens of the source question, including a
  leading moderator note that sanitizing removed for some tossups;
  `n_clues_heard` is the 1-based clue a buzz is credited to (a buzz at most 5
  tokens past a clue's end counts for that clue).
- Buzz `value` is normally 15, 10, 0, or −5; `correctness` is `value > 0`.
- 114 bonus-part rows from 2023 Nationals have `is_scored: false` (source `NA`);
  exclude them from conversion rates.
- `games.type` is set only when the source packet name had an explicit stage
  label, so null does not imply a stage.