**The most surprising finding is how much of the dataset's money comes from a handful of rows, and how likely it is that the largest of them is a data error.** - **Heavy concentration.** Of 3,044 rows, 2,066 have an amount. One of those has a non-numeric value (Drums Food parses to `nan`), so the stats below use the 2,065 clean rows (step-1). Together they sum to about **$38.06B**, but the **median deal is only about $1.7M**. - **Dollars sit in a few mega-deals.** The 73 deals of $100M or more are about 3.5% of valued deals. They account for roughly **63.6% of all dollars**. - **One row is 10% of everything.** The single largest entry is **Rapido Bike Taxi, Series B, 27/08/2019, listed at $3.9B**. That is about **10.2% of all funding in the dataset** (step-0, step-1). A bike-taxi startup out-raising Flipkart ($2.5B) and Paytm ($1.4B) looks implausible. It may be a unit or entry error, such as $3.9M recorded as $3.9B. I haven't verified it against an outside source. - **Totals and averages mislead.** If that one row is wrong, headline totals shift by about 10%. If you ignore the missing and malformed amounts, "average funding" tells you little about a typical deal. - **Many deals have no amount.** 978 of 3,044 deals (about 32%) have no usable amount (step-5, step-6). Any "total funding" figure covers only two-thirds of the deals. - **Amounts are stored as text.** The amount column is VARCHAR, and at least one value (`nan`) turns the sums into NaN unless you filter it out. In my later aggregate queries (steps 2 and 4–6), the total, mean and max came back as `nan` for exactly this reason. That is why I relied on the filtered query from step-1 for dollar figures. **Implication:** for benchmarking raise sizes by vertical or round, use medians and exclude or verify the top few outliers. Sums and means are dominated by a few deals, at least one of which looks suspect.