# Fantabeto Dashboard

**Project Al-Cihred** — Production-grade Streamlit dashboard for the 2026/27 Fantacalcio season.

```bash
streamlit run dashboard/app.py
```

---

## Architecture

```
data/warehouse/          ← Versioned Parquet (read by dashboard)
  players.parquet          505 rows — roster + 25/26 stats + per-90 metrics
  fixtures.parquet         10 rows — matchday 1 home/away + formations
  predictions.parquet      505 rows — FV projections with std + starter prob
  lineups.parquet          466 rows — probable lineups with starter %
  votes.parquet            12,049 rows — historical per-matchday votes (25/26)
  model_metrics.parquet    1 row — RMSE, R², sample count

dashboard/
  app.py                   ← Entry point: multi-page routing via sidebar
  warehouse.py             ← Read-only Parquet loader (@st.cache_data)
  viz/
    template.py            ← Plotly template "fantabeto_dark" + color palette
    charts.py              ← Pure functions: df → Figure (heatmap, radar, etc.)
    pitch.py               ← SVG pitch component with player badges
    components.py          ← KPI cards, role chips, section headers, CSS
  pages/
    01_matchday.py         ← Control Room: KPIs, fixture heatmap, start/sit
    02_players.py          ← Intelligence: radar, regression, card risk, news
    03_auction.py          ← War Room: waterfall, value scatter, grid heatmap
    04_lineup.py           ← Optimizer: SVG pitch, what-if toggles, opponent
    05_lab.py              ← Model Lab: error violins, calibration, backtest
  tests/
    test_dashboard.py      18 unit tests (warehouse, template, charts, pitch)

src/export/warehouse.py    ← Exporter: reads pipeline artifacts → Parquet
.streamlit/config.toml     ← Dark theme base, server config
```

---

## Design System — Stadium Night

| Property | Value |
|---|---|
| Background | `#0B0F17` |
| Card bg | `#111827` at 70% opacity, 14px radius, blur 8px |
| Border | `#1F2937` |
| Pitch Green | `#00D084` — positive, bonus |
| Gold | `#FFC94D` — captain, highlight |
| Red | `#FF4D5E` — malus, risk |
| Sky | `#38BDF8` — neutral data |
| Violet | `#A78BFA` — uncertainty |
| Type | Space Grotesk (headers), Inter (body), tabular numerals |
| Charts | All use `fantabeto_dark` template. No default Plotly palette anywhere. |

---

## Pages

### 1. Matchday Control Room
- **KPI row**: projected points, players at risk (P(start) < 70%), matchday count, FV trend sparkline
- **Fixture difficulty heatmap**: teams × gameweeks, colored by opponent xGA/FV strength
- **Start/Sit grid**: top 20 players with role chips, H/A indicator, projected FV ± std, risk status (🟢 START / 🟡 DOUBT / 🔴 RISK)
- **Bump chart**: projected rank trajectory across first 10 GWs

### 2. Player Intelligence
- **Search**: autocomplete on all 505 players, filtered by role/team
- **Player header**: FV avg, games played, GW1 start %, role chip, FVM, QI
- **Percentile radar**: 7 metrics normalized vs role average (goals, assists, xG, prog passes, tackles, interceptions, SoT%)
- **Goals vs Expected**: bar chart with divergence annotation
- **Bonus/malus donut**: goal bonus + assist bonus − card malus breakdown
- **Card risk gauge**: yellow/red card per-game risk indicators
- **News feed**: placeholder for RAG pipeline output (injuries, suspensions, tactical shifts)
- **Historical votes**: last 10 matchday votes table

### 3. Auction War Room
- **Budget controls**: sliders for budget (300–700 cr) and role quotas
- **KPI row**: players drafted, total spent, projected FV, avg price per player
- **Budget waterfall**: Sankey-style allocation per role (GK/DEF/MID/FWD)
- **Value scatter**: FV avg vs QI, bubble size = games played, top steals labeled, cost-per-FV isolines
- **Grid auction heatmap**: top players × bid multipliers, colored by value surplus
- **Recommended squad**: per-role player list with max bid, FV, and games played

### 4. Lineup Optimizer
- **What-if controls**: force IN/OUT a player, select formation (4-4-2, 4-3-3, 3-5-2, etc.)
- **KPI row**: expected points (with captain), captain name, avg start %, win probability
- **SVG pitch**: dark turf gradient, player badges sized by FV, gold captain ring, risk markers, bench strip, formation label
- **Player list**: role-chipped players with FV and risk flags
- **Opponent mirror**: per-duel edge arrows (🟢 advantage / 🔴 disadvantage / ⚪ neutral)

### 5. Model Lab
- **KPI row**: RMSE, R², training samples, feature count
- **Error violins**: prediction error distribution by role
- **Feature importance**: Pearson correlation with FV avg
- **Calibration curve**: predicted vs observed uncertainty
- **Backtest**: league average FV across 38 matchdays
- **Architecture notes**: model overview, key insights, limitations

---

## Running

```bash
# First time: export warehouse
python -m src.export.warehouse

# Install (if not already)
pip install streamlit

# Launch dashboard
streamlit run dashboard/app.py

# Run tests
python -m pytest dashboard/tests/ -v
```

## Data Refresh

The warehouse reads pre-computed Parquet files. To refresh after running the pipeline:

```bash
python -m src.export.warehouse
```

Then reload the dashboard — `@st.cache_data(ttl=3600)` will pick up new files after 1 hour or on manual cache clear.

## Dependencies

Added to `requirements.txt`:
- `streamlit>=1.35`

Already present:
- `plotly>=5.18`, `pandas>=2.1`, `numpy>=1.26`, `openpyxl>=3.1`