The dataset has no true NULLs, but missing and malformed values are hidden inside text columns. The main problems are in `Amount in USD` and `Date dd/mm/yyyy`, which are both stored as VARCHAR. **1. Missing amounts disguised as text (the biggest issue)** - All 3,044 rows have a non-null `Amount in USD`, but only 2,066 parse as numbers. The other 978 (32%) do not. - 959 of those are the plain string `N/A`. Another 4 are `\xc2\xa0N/A` (a stray non-breaking-space artifact) and 7 are `Undisclosed`, `undisclosed` or `unknown`. - One value is `14,342,000+`, an approximate figure with a trailing plus. - Seven values carry the `\xc2\xa0` prefix but are otherwise valid numbers, such as `\xc2\xa05,000,000` and `\xc2\xa020,000,000`. A naive numeric cast drops them. This is why my regex check for the space character reported 0 prefixed values: the artifact is stored as literal escaped text, not as the real character. - Any average or sum over the raw column silently excludes about a third of deals unless these are cleaned first. Cleaning needs the same treatment for `N/A`, `Undisclosed` and `unknown` casing and spelling variants. **2. Outliers and impossible values** - One amount is the literal `nan` (Drums Food, 21/07/2016). Because it parses as NaN, the column's max also returns NaN and breaks aggregates. - The parseable amounts run from 16,000 to 3,900,000,000, with a median of 1,725,000. - The largest values are Rapido Bike Taxi at 3.9B (27/08/2019), Flipkart at 2.5B (11/08/2017), two Paytm deals at 1.4B and 1.0B, another Flipkart deal at 1.4B, and Flipkart.com at 700M (28/07/2015). - The Flipkart and Paytm figures are plausible mega-rounds. The Rapido 3.9B is roughly 2,000 times the median and looks like a probable entry or unit error, but I could not verify it from this data. **3. Inconsistent formats and mixed units** - Amounts use Indian digit grouping in some rows (for example `3,90,00,00,000` and `62,50,000`) and Western grouping in others (`5,000,000`). The two styles parse the same once commas are removed, but they show the data came from mixed sources. - The column is labelled USD, but the Indian grouping suggests some values may originally have been in rupees. There is no currency column, so I cannot confirm or rule out mixed currencies. - Dates are `dd/mm/yyyy` text, and 3,036 rows follow the standard pattern (the other 8 show up as non-standard in the first check). The non-standard ones are: - `05/072018` (Loan Tap) - `01/07/015` (HandyHome) - `12/05.2015` (Mobiefit, Plancess) - `13/04.2015` (The Porter) - `15/01.2015` (Wishberry) - `22/01//2015` (Corporate360) - `\xc2\xa010/7/2015` (Infinity Assurance, with the space artifact and an unpadded month) - Later checks counted 7 unparseable dates, so expect 7 to 8 depending on whether the space-prefixed one is repaired. **4. Other signals** - `Sr No` is a clean unique key: 3,044 distinct values across 3,044 rows. - Startup name, industry vertical, sub-vertical, city, investors, type and remarks are all non-null. This reflects empty-string or placeholder values rather than real completeness. I did not profile those placeholders. - The card lists about 903 distinct industry verticals and about 66 distinct investment types. That is a high count for such fields and suggests spelling variants and free-text categories, which I did not examine. - Names also carry artifacts, such as `\xc2\xa0Infinity Assurance`, and the same company appears under variants (`Flipkart` and `Flipkart.com`). **Recommended cleanup before analysis:** strip the `\xc2\xa0` artifacts and the trailing `+`, map `N/A`, `Undisclosed` and `unknown` to NULL, handle `nan`, cast amounts to numbers, repair or drop the 7 to 8 bad dates, and review the Rapido 3.9B record.