The data has real quality problems, mostly in the `Amount` text column and the May 2022 table. The five monthly tables have the same 13 columns, and `Amount` is stored as VARCHAR in all of them. The figures come from my amount-pattern profile (step 1), the per-month checks (step 2), and a look at the May rows (step 3). **1. `Amount` mixes formats and units (step 1)** - **Standard format:** 296 rows look like `$1,000,000` to `$97,500,000`, and 47 more are `$100,000` to `$800,000`. Both appear in all five months. - **Non-numeric placeholders:** 35 rows say "Undisclosed" (4 months), and 25 rows are empty strings (3 months). Step 2 counted no NULL amounts, so these blanks are empty strings, not NULLs. - **Malformed separators:** 11 rows such as `$11.300,000` and `$728.100,000` use a period where a comma belongs, and all fall in one month. - **Missing `$` sign:** 11 rows such as `1,000,000` to `76,000,000` have no currency symbol (3 months), and 1 row is a bare `6400000`. - **Unscaled values:** 5 rows such as `$165.5` and `$84.2` look like figures in millions rather than dollars. 4 rows such as `$23` and `$60000000` have no thousands separators and span wildly different scales. These are mixed units. - **Billion-scale outliers:** 10 rows run from `$1,100,000,000` to `$5,000,000,000`, all in a single month. They may be genuine mega-rounds, but they sit about 11× above the typical maximum of about $97.5M. They need verification, and they would swamp any sum or average. - **Misaligned value:** one `Amount` is a person's name, "Vaibhav Anant", which is a founder name shifted into the wrong column. All of this means `Amount` cannot be cast or summed without cleaning it first. **2. The May 2022 table is structurally broken (steps 2–3)** - **Mistyped `Founded`:** `Founded` is VARCHAR in May but BIGINT in the other months. 19 of its 42 rows are non-numeric. The card's temporal range even shows a city ("Noida") as a `Founded` value. - **Shifted columns:** in the sample I pulled (step 3), FirstCure Health has "Noida" in `Founded` and the sector text in `Location`, so the fields are shifted left or right. There are also fully blank rows. - **Duplicate keys:** 18 duplicate company names, which are mostly the blank rows. This breaks the "one row per company" grain. **3. Impossible or suspect values and duplicates** - **Implausible founding years:** April has 3 rows with a very old `Founded` year; the profile shows a minimum of 1871. That is possible for a legacy firm but is unusual for a startup list. - **Founding year after the funding date:** March's maximum `Founded` is 2022, which is the funding year itself. That is plausible, but check it against the funding date. - **Duplicate names:** April and January each have 1 duplicate company name. These may be repeat rounds or accidental duplicates. - **No other gaps:** apart from the May table, there are no null company names, `Founded`, `Stage` or `Location` values in any month. **Recommended cleaning** 1. Parse `Amount` into a numeric column and flag "Undisclosed" or blank rows as missing. 2. Fix the `.` versus `,` separators and rescale the sub-1,000 values, which are probably millions. 3. Manually confirm the billion-scale rows. 4. Realign or drop the misaligned and blank May rows, and cast `Founded` consistently. 5. Review the duplicate company names.