# Data-quality issues in the funding dataset The dataset covers five months (Jan–May), with 446 rows in total: 115 in Jan, 96 in Feb, 98 in Mar, 95 in Apr and 42 in May. The evidence shows several real problems. Some counts differ slightly between my queries, so I flag where that happens. ## 1. Inconsistent amount formats (the biggest issue) The `Amount` column is text, not numeric, and it uses many conventions: | Format | Rows | Example | |---|---|---| | `$` with comma thousands | ~352–354 | $1,000,000 – $97,500,000 | | "Undisclosed" | 35 | | | Blank/null | 25 | | | No `$`, comma thousands | 11 | 1,000,000 – 76,000,000 | | `$` with mixed `.` and `,` separators | 11 | $11.300,000; $728.100,000 | | `$` short decimal | 5 | $165.5; $84.2 | | `$` no separators | 4 | $23; $60000000 | | Malformed separators | at least 1 | $1,40,000 | | Bare number | 1 | 6400000 | | Text in the amount field | 1 | "Vaibhav Anant" | - **Mixed separators in April:** values like `$728.100,000` and `$468.200,000` appear to be $728,100,000 and $468,200,000 with the wrong separator. That is my inference, not something the data confirms. - **Short decimals:** `$165.5` (MobiKwik) and `$84.2` (OkCredit) are ambiguous. They could be millions, but the unit isn't stated, so this is a mixed-units risk. - **Extreme values:** 10 rows use a four-group pattern, ranging from $1,100,000,000 to $5,000,000,000. These may be legitimate large rounds, but they could also be errors, and the evidence doesn't verify them. - **Tiny values:** `$23` sits beside values like `$60000000`, which suggests a unit mismatch. - **Amounts can't be summed or compared** without cleaning first. ## 2. Missing and placeholder values - "Undisclosed" is used as a placeholder for 35 amounts, and blank amounts are a separate issue (25 rows). - **May is badly incomplete:** of 42 rows, 21 have blank amounts, 19 have a blank company name, 21 are missing stage and 19 are missing location. Step 3 shows many fully empty rows, so May likely contains junk or empty rows. - **Stage:** the stage field is frequently missing: 29 in Feb, 43 in Jan and 31 in Mar by one query, plus 21 in May. Stage values are also inconsistent, mixing "undisclosed", "Unknown", blank, "Series-D" and "Series D", and "Pre-series A" and "Pre-seed". ## 3. Impossible or suspicious values - **Founded year:** April has 3 rows flagged as very old founding years: MTR Foods (1924), Rigi (1871) and Philips Electronics (1929). Hitachi (1959) also looks odd for a startup list. These are likely real old companies mixed into a startup dataset, or errors. - **May:** 19 rows have a non-numeric `Founded` value. - **Misaligned columns:** one row (FirstCure Health) has a description in the Sector column and blanks elsewhere. Another has a person's name in the amount field. Both suggest shifted or malformed rows. ## 4. Duplicates - May has 18 duplicate-company rows, which coincides with the blank-company rows, so these are probably empty rows rather than true duplicates. - Jan and Apr have 1 duplicate each. ## 5. Month-level skew Format problems are concentrated in April, which has 22 "other format" amounts (19 or 22 depending on the query) against 1–8 in other months. May is the most degraded. ## What remains open - I did not verify whether the large amounts ($1B–$5B) are real. - The unit of the short decimals is unconfirmed. - Currency is assumed to be USD because of the `$` prefix, and non-USD amounts weren't checked. - Some counts differ slightly across queries because the format categories were defined differently, so treat them as approximate. - Several steps failed to parse, so some checks, such as outlier statistics, were not completed.