The data has real quality problems, mostly inconsistent formats and mixed units in `Amount`, plus a badly broken May table. Every figure below comes from the profiling queries I ran. ## 1. `Amount` is text in many formats, with mixed units `Amount` is VARCHAR in all five tables. Of 446 rows, only 352 (about 79%) follow the clean `$1,000,000` pattern (step 10). | Format | Rows | Example | |---|---|---| | `$` with comma thousands (clean) | 352 | `$1,000,000` … `$97,500,000` | | "Undisclosed" text | 35 | `Undisclosed` | | Blank | 25 | | | No `$`, comma thousands | 11 | `1,000,000` … `76,000,000` | | Mixed `.` and `,` separators | 11 | `$11.300,000`, `$728.100,000`, `$468.200,000` | | Short decimal with unclear unit | 5 | `$165.5`, `$84.2`, `$431.6` | | No separators | 4 | `$23`, `$60000000` | | Other numeric/mixed | 13 | `$1,40,000` (Indian-style grouping), `6400000` | | Plain text in an amount field | 1 | `Vaibhav Anant` | - **Mixed units:** the April rows with short decimals (`$431.6` for Livspace, `$165.5` for MobiKwik, `$240.2` for Amagi) look like millions. They would be read as a few hundred dollars if parsed literally. - **Mixed separators:** the `$728.100,000` style for Fivetran is neither US nor European formatting, so it can't be parsed safely. - **Magnitude outliers:** 10 rows in one month range from `$1,100,000,000` to `$5,000,000,000`, far above the usual `$1M`–`$97.5M` range. They could be genuine mega-rounds or a unit or format error (step 1). - **Implausible pairings:** YouKraft (a Seed round) shows `76,000,000`, and Mamaearth/Whitehar Jr have `Unknown` stages next to large amounts. These deserve a manual check. - **Misalignment:** one row has a founder's name (`Vaibhav Anant`) in `Amount`. ## 2. Missing or placeholder values - `Amount` is blank in 25 rows (21 of them in May, 2 each in February and April) and "Undisclosed" in 35 (step 12). - `Stage` is missing in 43 of 115 January rows, 29 of 96 February rows, 31 of 98 March rows and 21 of 42 May rows (step 7). - Even where `Stage` is filled, it uses several placeholders for "no stage": `undisclosed`, `Unknown` and empty. - Stage labels are also inconsistent, for example `Series-D` vs `Series D` and `Pre-series B` vs `Pre-seed`. ## 3. Impossible or atypical `Founded` values - April has three founding years that don't fit a startup list: Rigi (1871), MTR Foods (1924), Philips Electronics (1929). Hitachi (1959) also appears in the same table (step 5). These are established corporates or probable typos, and most carry "Undisclosed" amounts. - Some "Founded" values are in the future relative to the funding month, though none were flagged as impossible. For example, Chhotastock (2022) in March is plausible but at the edge. ## 4. The May table is structurally broken - `Founded` is VARCHAR in May only, and 19 of its 42 rows are non-numeric. One row, FirstCure Health, has a city (`Noida`) in `Founded` and values shifted across columns (step 3). - 19 rows have a blank `Company Name` and 19 have a blank `Location`. The 18 "duplicate" company names in May are almost all these blanks (step 2). - So May has only about 23 usable rows out of 42, and its counts are not comparable with other months. ## 5. Duplicates Outside May, duplicate names are minimal: one duplicate company row in January and one in April, and none in February or March (step 12). The cards say each table is keyed on `Company Name`, so these need checking, since a company could legitimately raise more than once. ## Recommendations - **Parse `Amount`:** normalise it to a numeric USD column and flag rows where units are ambiguous (short decimals and mixed separators). - **Standardise `Stage`:** map placeholders and variants to one set of labels. - **Repair or drop May:** fix the misaligned rows, or drop the blank ones. - **Review by hand:** the pre-1960 `Founded` rows, the billion-dollar rounds and the other magnitude outliers.