**The most counterintuitive finding is that the "typical" deal and the dollar totals describe two different datasets, and the biggest single number is probably a data error.** **1. A tiny share of deals carries most of the money.** - Of the 2,065 deals with a usable USD amount, the median deal is only about **$1.7M**. The total is about **$38.06B**, which works out to a mean of roughly $18M, over 10× the median. - Just **73 deals of $100M or more (about 3.5% of valued deals) account for 63.6% of all dollars** (steps 1, 8 and 12). - Any average-ticket or "how much should I raise" benchmark built on this data is therefore dominated by a few mega-rounds. The median is a far better guide. **2. The largest "deal" looks implausible.** - The top row is **Rapido Bike Taxi, Series B, 27/08/2019, $3.9 billion**. That is larger than Flipkart's $2.5B private-equity round (11/08/2017) and Paytm's $1.4B (step 0). - A Series B at that size for a bike-taxi startup is very unlikely. It is probably a unit or entry error, such as millions recorded as dollars. - That single row is about **10.2% of all dollars** (step 1). Treat it as a suspect outlier rather than a real deal. - I did not independently verify the true Rapido figure. **3. Missing and corrupt amounts hide a lot of the picture.** - About **978 of 3,044 rows (about 32%) have no usable amount** (steps 12 and 13). - At least one row (Drums Food, 21/07/2016) has a non-finite "nan" amount. It sorts to the top of a descending ranking and silently breaks naive sums and maxima (steps 0 and 8). - Because the amount column is stored as text, many aggregate queries on this dataset can return `nan` unless unparseable values are filtered out first. **Takeaway:** the headline of about $38B in funding rests on roughly 70 large rounds, one of which is likely mis-entered, and on only about two-thirds of the deal rows. Any trend or sector ranking based on dollar totals will mostly reflect a few outliers and which deals happen to report an amount. Deal counts and median ticket sizes are the more reliable measures here. The figures come from the query results in steps 0, 1, 8, 12 and 13. Small differences in row counts across steps (2,065 vs 2,066 valued) come from how the single non-finite row was handled.