The `Amount in USD` column is the main problem. Dates have a smaller set of format errors. The card also shows the ingestion artifacts below. Not every column was checked for deeper problems such as duplicates or city spellings. **1. Mixed formats in `Amount in USD` (stored as text)** - Only 2,066 of 3,044 rows (68%) hold a clean number. The other 978 (32%) are non-numeric (step-0). - Missing values appear under many labels: `N/A` (959 rows), `\xc2\xa0N/A` (4), `Undisclosed` (3), `undisclosed` (3) and `unknown` (1). A single null check would miss most of them (step-1). - Some amounts carry a stray non-breaking space prefix (`\xc2\xa0`), such as `\xc2\xa016,200,000` and `\xc2\xa05,000,000`. This is an encoding artifact, and these values fail a plain numeric cast. - One amount has a trailing plus sign, `14,342,000+`. It is a lower bound rather than an exact figure. - Digit grouping is inconsistent. Some values use Western commas (`16,200,000`) and others use Indian lakh/crore grouping (`3,90,00,00,000`, `62,50,000`). If the commas are stripped, both parse correctly, but the grouping shows the values came from different sources. - The column is named "USD", but nothing confirms every value is in dollars. See the next point. **2. Outliers and possible mixed units** - Amounts that parse range from 16,000 to 3.9 billion, with a median of 1,725,000 (step-0). - The top values (step-2) are: - Rapido Bike Taxi, 27/08/2019: 3,900,000,000. - Flipkart, 11/08/2017: 2,500,000,000. - Paytm, 18/05/2017: 1,400,000,000. - Flipkart, 21/03/2017: 1,400,000,000. - Paytm, 25/11/2019: 1,000,000,000. - Flipkart.com, 28/07/2015: 700,000,000. - The Rapido figure stands out. It is the single largest amount, well above Flipkart and Paytm, and I judge it implausible for a bike-taxi startup. It is a prime candidate for a unit error, such as rupees or a mis-keyed figure recorded as dollars. I could not confirm this from the data. - The Flipkart and Paytm entries are large, but they look like genuine mega-rounds. Check them rather than drop them. - Flipkart appears under two names, `Flipkart` and `Flipkart.com`, so entity names are not standardized. **3. Impossible or non-finite values** - Drums Food (21/07/2016) has an amount that parses as `nan`. That is why the column's maximum shows `nan`, and any `MAX` or `SUM` over it is polluted until it is cleaned. **4. Inconsistent and invalid date formats** - The column is declared `dd/mm/yyyy`, but 3,036 of 3,044 rows match that pattern. Eight do not (step-0 and step-2): - `05/072018` (Loan Tap), a missing separator. - `01/07/015` (HandyHome), a truncated year. - `12/05.2015` (Mobiefit, Plancess), a dot instead of a slash. - `13/04.2015` (The Porter), a dot instead of a slash. - `15/01.2015` (Wishberry), a dot instead of a slash. - `22/01//2015` (Corporate360), a double slash. - `\xc2\xa010/7/2015` (Infinity Assurance), a non-breaking space and an unpadded month. - Dates are text, so these rows will fail or be dropped in any time-series parse until they are repaired. **5. Practical implications** - Treat all the N/A, Undisclosed and unknown labels as NULL. Strip `\xc2\xa0`, commas and the trailing `+` before casting. - Review the Rapido value separately before using any sum or average. A few values in the billions can distort totals, so prefer medians. - Repair the eight malformed dates by hand or with a rule. - Treat any amount average as based on only about two-thirds of the deals.