# AGENTS.md — The Hope Pub Quiz Dashboard

## Overview
Single-page Flask dashboard that reads `data.csv` (one row per quiz night) and renders summary statistics, a player cost table, and five Plotly charts via a Jinja2 template. No database — the CSV is the sole data store.

## Running the App
```bash
uv sync                                      # install deps from uv.lock / pyproject.toml
PYTHONPATH=src python src/app.py             # run from project root — resolves data.csv correctly
```
> **Path gotcha:** `app.py` reads `data.csv` with a bare relative path, so the CWD must be the **project root**. `PYTHONPATH=src` puts `src/` on the import path so local modules resolve.

## Data Shape (`data.csv`)
- **One row = one quiz night**
- Columns: `Date` (DD/MM/YYYY), `Absolute Position`, `Relative Position` (float 0–1, 0=1st/best, 1=last), `Number of Players`, `Number of Teams`, `Points on Scattergories`, then one binary column per player (1=attended, 0=absent).
- Date parsing uses `dayfirst=True`; the DataFrame is sorted ascending by date on every load.

## Module Responsibilities
| File | Role |
|---|---|
| `src/app.py` | Flask route, five Plotly chart builders, data loading |
| `src/stats.py` | `generate_stats(df)` → `(stats_dict, highlights_list)` tuple |
| `src/player_table.py` | `generate_player_table(df)` → flat list-of-lists; cost hard-coded at **£3/quiz** |
| `src/constants.py` | Player names, regression features, colour scheme, `ordinal(n)` helper |
| `src/templates/index.html` | Renders `highlights` list, `stats` dict, `player_table`, and `plots` dict |

## Key Conventions

### Adding/Removing a Player
1. Update `constants.PLAYER_NAME_COLUMNS` (ordered list — controls display order everywhere).
2. Update `constants.FEATURE_COLUMNS` (set — controls which columns feed the regression model).
3. Add the new binary column to `data.csv`.

### `generate_stats` Return Value
Returns a **tuple** `(stats, highlights)`:
- `highlights` — list of `{"label": str, "value": str, "detail": str}` dicts; rendered as 6 KPI cards.
- `stats` — plain `dict` of human-readable `label: value` pairs; rendered as a secondary list.

The `index()` route unpacks both: `stats, highlights = generate_stats(df)`.

### Plots Pipeline
1. Build a Plotly figure in `app.py` using `Relative Position` directly (or `Relative Position * 100` for percentile display), where **lower = better**.
2. Serialise: `json.dumps(fig, cls=plotly.utils.PlotlyJSONEncoder)`.
3. Store in the `plots` dict under a **snake_case key** (e.g. `"position_trend"`).
4. The template renders every entry automatically: `Plotly.newPlot("{{ key }}", ...)` — key is both the `<div id>` and JS target.

Current charts (in render order): `position_trend`, `player_impact`, `scattergories_vs_position`, `player_participation`, `calendar`.

### `player_table` Structure
`[0]` = header row, `[1:-1]` = data rows sorted by appearances descending, `[-1]` = totals footer. The template uses `player_table[1:-1]` for `<tbody>` and `player_table[-1]` for `<tfoot>`.

### `Relative Position` Convention
Raw data stores `Relative Position` (0=best). The dashboard keeps this convention everywhere: lower values are better in stats, tables, and chart labels. If a chart uses percentile text, it is `Relative Position * 100` (still lower = better).

### `ordinal(n)` Helper
Lives in `constants.py`. Returns e.g. `"1st"`, `"22nd"`, `"63rd"`. Import where needed: `from constants import ordinal`.

### Player Impact Chart
Shows average relative percentile when each player attends. Only players with **>= 3 appearances** are shown (`MIN_APPEARANCES = 3` in `generate_player_impact`). Green bar = below overall average (better); red = above (worse).

## Frontend
No build step. Tailwind CSS (`cdn.tailwindcss.com`) and Plotly (`cdn.plot.ly/plotly-3.0.1.min.js`) loaded from CDN. Charts use `{ responsive: true, displayModeBar: false }` config.