# Data-quality issues in the Jan–May 2022 funding tables The Amount column is the worst problem, and the May table has structural damage. Several other issues are smaller. All figures come from the profiling queries I ran (steps 1, 2, 5, 6 and 7). ## 1. Amount: inconsistent formats and mixed units `Amount` is stored as text in all five tables, in at least nine different formats (step 6): - **Clean `$1,000,000` style:** 354 rows. This is the only format that casts cleanly, and it appears in all 5 months. - **"Undisclosed":** 35 rows across 4 months. - **Blank:** 25 rows, mostly in May. - **No `$` prefix:** 11 rows such as `1,000,000` and `76,000,000`. - **Mixed `.` and `,` separators:** 11 rows, all in one month (April). Examples are `$11.300,000` and `$728.100,000`. These look like a European-style or garbled thousands format, and the intended value is ambiguous. - **Decimals with no unit:** 5 rows such as `$165.5`, `$84.2`, `$431.6` and `$240.2`. They are probably millions, but nothing in the data says so. - **No separators:** `$23000000` and `$60000000`. - **Bare number:** `6400000`. - **Malformed or non-numeric:** `$1,40,000` (Indian lakh-style grouping) and the text value "Vaibhav Anant", a founder name that landed in the Amount column. The `$ decimal` rows and the mixed-separator rows mean April mixes units, either raw dollars or millions. Summing Amount without cleaning would give badly wrong totals. April is the messiest month (step 7). Of its 95 rows, 61 have clean dollar amounts, 22 are in other formats, 10 are "Undisclosed" and 2 are blank. ## 2. Outliers and suspicious values - **Billion-scale amounts:** 10 rows range from `$1,100,000,000` to `$5,000,000,000`, all in one month. They are far above the typical `$1M–$97.5M` range and need checking. They could be genuine mega-rounds, but a unit error is also possible. - **April examples that look off:** YouKraft shows `76,000,000` at Seed stage, which is unusually large for a seed round. Rows such as Fivetran (`$728.100,000`) and Chargebee (`$468.200,000`) are plausible only if the separators are read as a typo. ## 3. Impossible or implausible `Founded` years - Three April rows have very old founding years (step 2). Step 5 shows them: Rigi (1871), MTR Foods (1924) and Philips Electronics (1929). Hitachi (1959) is also an old multinational, not a startup. - These look like established corporations, acquisition targets or mis-keyed entries rather than funded startups. Rigi at 1871 is almost certainly an error. - Some rows carry a 2022 founding year (for example Chhotastock in March). That is plausible for a company founded in the same year as its round, but worth confirming. ## 4. Structural damage in the May table - **Mistyped `Founded`:** May's `Founded` column is VARCHAR, while the other four are BIGINT. 19 of its 42 rows are non-numeric (step 2). - **Blank and shifted rows:** 19 rows have no company name or location (step 7). Step 3 shows fully blank rows. One row (FirstCure Health) is shifted: the `Founded` field holds "Noida" and the description sits in the wrong column. - **Counts:** only about 23 of May's 42 rows are real records. The card's note about "misaligned or blank" rows is confirmed. - **Duplicate names:** May shows 18 duplicate company names, which are the blank names. April and January have 1 duplicate each, which may be a company announcing two rounds or a real duplicate. - **Missing Amount and Stage:** 21 of May's rows have a blank Amount and 21 have a missing Stage. ## 5. Stage: missing and inconsistent labels - Missing Stage by month: Jan 43 of 115, Feb 29 of 96, Mar 31 of 98, May 21 of 42, and only 3 in April. - Labels are inconsistent. Step 5 shows "undisclosed", "Unknown" and blank all standing for the same thing. Others differ only in spelling or case: "Series-D" against "Series D", and "Pre-series A" and "Pre-series B" against the "Pre-seed" and "Seed" style. - April has an unusually high number of "undisclosed" stages (for example Swiggy, Myntra and Hitachi). ## 6. Other observations - Sector has roughly 21 to 55 distinct values per month, likely with overlapping or free-text categories. I did not profile this in depth. - There is no shared key across tables. Each month is a separate table, so combining them needs a UNION, and cleaning has to happen first. ## Recommended cleanup 1. Parse Amount into a numeric value with an explicit unit, and flag the ambiguous rows (mixed separators, decimals under 1,000, `$1,40,000`) for manual review. 2. Drop or repair the blank and shifted May rows. 3. Normalise the Stage labels and treat "undisclosed", "Unknown" and blank as one null value. 4. Review the `Founded` values below roughly 1960, and the billion-dollar amounts.