import streamlit as st

st.markdown(
    """
# KPI Anomaly Detection Documentation

## 1. Objective
Detect KPI trend breaks (change points) per cell and export anomaly-focused datasets for investigation.

## 2. When to use this tool
Use this page when you need to:
- detect abrupt KPI behavior changes
- isolate affected cells per KPI
- visualize change points over time
- export anomaly evidence to Excel

## 3. Input files and accepted formats
- Required: KPI file in `.csv`
- Expected delimiter in current implementation: `;`

## 4. Required columns
The first 5 columns are interpreted as:
1. date/time
2. controller
3. BTS
4. cell
5. DN
All remaining columns are treated as KPI series.

## 5. Step-by-step usage
1. Open `KPI Analysis > KPIs Anomaly Detection`.
2. Upload the KPI CSV file.
3. Set `Penalty` (lower = more sensitive, higher = less sensitive).
4. Review detected KPI list and affected cells.
5. Inspect plot with detected change points.
6. Generate and download Excel anomalies report.

## 6. Outputs generated
- interactive anomaly plot for selected KPI/cell
- per-KPI anomaly datasets
- downloadable Excel file (one sheet per KPI)

## 7. Frequent errors and fixes
- No anomalies detected.
  - Fix: reduce `Penalty` or verify KPI variability.
- Parsing/date issues.
  - Fix: check first column date format and CSV delimiter (`;`).
- Too many false positives.
  - Fix: increase `Penalty` and inspect noisy KPIs.

## 8. Minimal reproducible example
- Input: CSV with at least 30 points per cell and one KPI showing a level shift.
- Action: upload file, set penalty to `2.5`, inspect one KPI/cell.
- Expected result: chart with change-point markers and exportable Excel output.

## 9. Known limitations
- Cells with fewer than 30 points are ignored.
- Detection quality is sensitive to noise and seasonality.
- Current uploader is configured for CSV input only.

## 10. Version and update date
- Documentation version: 1.0
- Last update: 2026-02-23
"""
)