**Short answer:** the biggest problems are the `Amount` column, which is text in at least eight formats, and the May 2022 table, which is partly blank or misaligned. There are also a few doubtful `Founded` values and messy `Stage` labels. ### 1. `Amount` is text in many formats, with mixed units `Amount` is VARCHAR in all five tables. Across all rows it falls into these formats (step 10): | Format | Rows | |---|---| | `$1,000,000`-style (the clean form) | 352 | | "Undisclosed" | 35 | | Blank | 25 | | Other numeric or mixed separators, e.g. `$1,40,000` | 13 | | No `$`, commas only | 11 | | Short decimal, e.g. `$165.5` | 5 | | `$` with no separators, e.g. `$23`, `$60000000` | 4 | | Plain text | 1 | - **Mixed units (April only).** Values like `$165.5` (MobiKwik), `$84.2` (OkCredit), `$431.6` (Livspace), `$221.5` (WayCool) and `$240.2` (Amagi) look like millions written as decimals. Others look like millions with the decimal and thousands separators swapped: `$728.100,000` (Fivetran), `$468.200,000` (Chargebee), `$624.800,000` (DealShare), `$11.300,000` (Whitehat Jr). Read literally, these are wrong by orders of magnitude. April also has `$23`, which is almost certainly not a literal $23. - **Absolute outliers.** 10 rows, all in one month, are in the $1.1B–$5B range (`$1,100,000,000` to `$5,000,000,000`, step 1). I have not checked these against the company names, so I cannot say whether they are genuine mega-rounds or a unit error. The largest clean value is `$97,500,000`, and none of the 10 appears in the other months. - **Doubtful values.** `76,000,000` for YouKraft (Seed stage) and `$1,40,000`, which uses Indian-style grouping, look suspicious. - **Misalignment.** One `Amount` value is a person's name, "Vaibhav Anant", which points to a shifted row. - **Missing amounts.** Blank amounts are 21 of 42 in May, plus 2 each in February and April. "Undisclosed" appears in 6 January rows, 12 March rows, 10 April rows and 7 February rows. Blank and "Undisclosed" mean different things but are both non-numeric. - **Currency.** Amounts carry a `$` or nothing, and no currency column exists, so USD is an assumption. ### 2. The May 2022 table is partly broken - 21 of 42 rows have a blank `Amount`. - 19 rows have no company name, 19 have no location, and 21 have no stage (step 7). - 19 `Founded` values are non-numeric, which is why that column is VARCHAR here and BIGINT elsewhere. - Duplicate company names: 18 in May, which are probably the blank rows. - A sample shows a row for FirstCure Health with "Noida" in the `Founded` column and the other fields shifted. The next 11 rows are completely empty (step 3). ### 3. `Founded` has implausible values for startups April contains Rigi (1871), MTR Foods (1924), Philips Electronics (1929) and Hitachi (1959). These are established corporations, not startups, and 1871 is likely an error (step 5). The other tables' `Founded` ranges are plausible: 1991–2021 for February, 1998–2021 for January, and 1994–2022 for March. ### 4. `Stage` is missing and inconsistently labelled - **Missing stage:** January 43 of 115, February 29 of 96, March 31 of 98 and May 21 of 42. April is only 3 of 95, but it uses "undisclosed" and "Unknown" as placeholder stage values (step 5). - **Inconsistent labels:** "Series-D" next to "Series D", and "Pre-series A" and "Pre-series B" next to "Pre-seed". April has about 16 distinct stage labels. - **Stage and amount disagree:** MoEngage has $133M with stage "undisclosed", and Nykaa has `$148.500,000` with stage "undisclosed". ### 5. Minor issues - One April row has a missing location. - One duplicated company name each in January and April. - The card says the grain is one row per company, but a company could have more than one round in a month, so the duplicates need a manual check. ### Recommended cleaning 1. Parse `Amount` into a numeric USD column. Map "Undisclosed" and blanks to NULL, and flag the short-decimal and swapped-separator rows as probable millions. 2. Review the 10 rows of $1B or more. 3. Drop or repair the blank and misaligned May rows. 4. Normalise the `Stage` labels. 5. Review `Founded` values before 1960. The final cleaning rules above are my interpretation. The unit-mix findings rest on pattern evidence, not on external verification of individual deals.