Turn a CSV export into an interactive, privacy-safe dashboard

Build a data dashboard in Bolt.new that turns a CSV file into interactive charts entirely in the browser — no upload to any server, no backend, since the data in this file is internal payroll data and should never leave the user's machine.

INPUT
The user drops or selects a CSV file with these columns: date, department, employee_count, total_hours, overtime_hours. Parse it client-side with a CSV parsing library, not a hand-rolled split(','), — real CSVs have quoted fields containing commas, and a naive split silently corrupts those rows instead of erroring on them.

DATA HANDLING RULES
Validate the file before charting anything: if a required column from the list above is missing, show a clear error naming exactly which column is missing rather than crashing or silently charting nothing. If a row has a malformed value in a numeric column — text where a number is expected, an empty cell — exclude that row from calculations and show a count of how many rows were skipped and why, rather than letting one bad row silently turn a sum into NaN across the whole dashboard. Assume the file could be large — around 15,000 rows is the realistic upper bound — so parse and aggregate in a way that doesn't freeze the browser tab; if the parsing library supports streaming or chunked parsing, use it rather than loading the whole file into one array before any processing starts.

CHARTS TO BUILD
1. total_hours over time as a line chart, bucketed by week.
2. A breakdown of total_hours by department as a bar chart, sorted descending, with a "show top 10, group the rest as Other" behavior if there are more than 10 distinct categories.
3. A summary row of stat cards above the charts: total total_hours, average per week, and the single highest week by value.
4. A raw data table below the charts, sortable by column, with the skipped-row count from above shown as a small note, not hidden.

INTERACTIVITY
Let the user filter by a date range and by one category value from department; every chart and stat card above must update together when a filter is applied — never let the line chart reflect the filter while the bar chart still shows the unfiltered set.

CONSTRAINTS
This is entirely client-side — no fetch call to any external endpoint with the parsed data, ever. If the file is dropped, do not persist it anywhere beyond the current browser session; a page reload can reasonably clear the dashboard and ask for the file again. clean and neutral, one accent blue, no illustrations