saturn

This is a 50-product sample from the Open Food Facts database, an open crowdsourced food product catalogue with 545 columns spanning multilingual product names, ingredient texts, allergen data, nutritional scores, packaging details, and community metadata. The most striking structural issue is extreme sparsity: the vast majority of language-specific columns (e.g. product_name_dz, ingredients_text_ja) have null rates of 96–98%, meaning content is concentrated in French and English fields. Two things most deserve a closer look: first, the Nutri-Score distribution is heavily skewed toward grade 'e' (54% of products), suggesting the sample leans toward nutritionally poor items; second, scan counts (scans_n, mean 578, max 2523) show a strong right-skewed tail with a few highly popular products dominating community attention.

This column is intended to capture an environmental impact level category, but it is effectively empty: 56% of the 50 rows are null and the remaining 44% (22 rows) contain only a blank string, yielding a single unique value and zero entropy. The column carries no usable information in its current state and is entirely uninformative for modelling or analysis.

This column appears to be a German-language origin/source label field ('origin_de'), but it contains effectively no usable data: the only observed value across all 50 rows is an empty string, appearing 20 times, with 60% of rows (30) being null. Cardinality is 1, entropy is 0, and top_rate is 1.0 — the column is entirely uninformative in its current state.

This column appears to be a Spanish-language origin/source label field ('origin_es'), but it is entirely devoid of meaningful content: the sole observed value is an empty string, appearing 20 times across 50 rows. With a 60% null rate and the remaining 40% being empty strings, the column carries zero informational entropy and is effectively blank across the entire dataset. This is a strong signal that the field was never populated.

This column appears to be an 'origin Italy' flag or similar origin/locale indicator, but it is effectively empty: 68% of its 50 rows are null, and the sole non-null value present is an empty string appearing 16 times. With cardinality of 1 and entropy of 0, the column carries zero information. The combination of high nulls and a blank-string-only value suggests the field was never populated in this dataset slice.

This column ('origin_nl') is a categorical field, likely a Dutch-language origin label or description, but it is effectively empty: 76% of the 50 rows are null, and the sole non-null value present is an empty string (''), appearing 12 times. With cardinality of 1, zero entropy, and a top_rate of 1.0 across only 12 non-null rows, the column carries no information whatsoever.

This column appears to hold Dutch-language packaging text for products, but is effectively empty: 76% of the 50 rows are null, and the sole non-null value is an empty string appearing 12 times, giving a cardinality of 1 and zero entropy. Every observed value is either missing or a blank string, meaning this column carries no usable information in this sample.

This column appears to be an Arabic-language origin field ('origin_ar') that is almost entirely empty. With an 80% null rate and cardinality of 1, the sole 'unique' value is itself an empty string appearing 10 times across 50 rows — meaning the column contains no actual data at all. This is a fully degenerate column with zero informational content.

This column, likely representing an origin point or location, is almost entirely empty: 80% of its 50 rows are null, and the only non-null value present is an empty string appearing 10 times — meaning the column contains no actual information whatsoever. With a cardinality of 1 and entropy of 0.0, it is completely invariant. The combination of high null rate and a sole value being an empty string suggests the field was never populated in this dataset.

This column appears to hold Arabic-language packaging text, but it is effectively empty: 80% of the 50 rows are null, and the remaining 10 non-null rows contain only an empty string — giving a single unique value with top_rate of 1.0 and zero entropy. The column carries no information whatsoever in this dataset snapshot.

This column appears to be a Portuguese-language packaging text field, almost certainly intended to carry product label or packaging descriptions. With an 80% null rate and the sole non-null value being an empty string appearing 10 times, the column contains zero usable information across all 50 rows. The effective data-present rate is 0%, making this column entirely empty in practice.

This column represents the unit basis for imported nutrition data, and every non-null value is identically '100g' — giving it a cardinality of 1 and an entropy of 0.0. With an 84% null rate across 50 rows, only 8 observations carry a value at all, making the column almost entirely absent. The combination of extreme nullity and zero variance means this column provides no discriminating information whatsoever.

This column appears to be a binary checkbox field (HTML-style 'on'/'off'), but only the value 'on' is ever recorded — cardinality is 1 with 'on' appearing in all 7 non-null rows. The 86% null rate is the dominant signal: nulls almost certainly represent unchecked state rather than missing data, meaning the column encodes a boolean with an unconventional null-as-false convention. Zero entropy confirms complete absence of variation among non-null values.

This column records the imported language of a record, and across the full 50-row dataset every non-null value is 'fr' (French) — a single unique value with zero entropy. With an 86% null rate, only 7 of 50 rows carry any value at all, making the column nearly empty and entirely constant where populated. Both the extreme null rate and perfect imbalance are flagged as alerts, suggesting this field may be partially populated metadata from an import pipeline rather than a reliable feature.

This column captures the unit basis for imported nutrition data (e.g., 'per 100g'), and is effectively a constant — the only observed value is '100g' across all 7 non-null rows. With an 86% null rate and cardinality of 1, it carries zero discriminative information. The combination of near-total missingness and zero entropy is a strong signal this field was either sparsely populated at ingestion or serves as a fixed schema placeholder.

This column appears to be a boolean or flag field (likely 'imported' status, now obsolete) that contains only the value '0' across all 7 non-null rows. With an 86% null rate and a cardinality of 1, the column carries zero information — entropy is exactly 0.0 and the single observed value covers 100% of non-null records. Both the near-total nulls and complete value imbalance are flagged as alerts.

This column, likely representing an origin financial institution or similar identifier, is almost entirely empty: 90% null rate with only 5 non-null rows across 50 records. Among those 5 non-null values, every single one is an empty string, meaning the column contains zero meaningful information—cardinality is 1, entropy is 0, and the sole 'value' is a blank.

This column appears to be an 'origin platform' or similar provenance field, but it is essentially empty: 90% of its 50 rows are null, and the only non-null value is an empty string appearing 5 times. With cardinality of 1 and entropy of 0.0, it carries zero information. The combination of high null rate and a single blank value strongly suggests this field was never populated in this dataset slice.

This column appears to be Finnish-language packaging text for a product dataset, but it is almost entirely empty: 90% of the 50 rows are null, and the sole non-null value across all 5 populated rows is an empty string. With cardinality of 1 and entropy of 0, the column carries zero information — it is effectively unpopulated.

This column appears to be a Polish-language packaging text field that is almost entirely empty: 90% of its 50 rows are null, and the sole non-null value present in 5 rows is an empty string. With cardinality of 1 and entropy of 0, the column carries zero information. The combination of a 90% null rate and a top_value of '' means not a single meaningful entry exists in this sample.

This column is a boolean flag indicating whether nutrition data was absent for a record. It has a 92% null rate across 50 rows, and the only 4 non-null values all carry the single value 'false', giving it zero entropy and cardinality of 1. The extreme null rate combined with complete value uniformity among non-nulls means this column carries no predictive signal whatsoever — it is effectively empty.

This column appears to be an origin or handling-unit identifier that is almost entirely unpopulated — 92% of its 50 rows are null, and the sole non-null value present is an empty string appearing 4 times. With cardinality of 1, zero entropy, and a top_rate of 1.0 across non-null values, the column carries no discriminative information whatsoever. This is a effectively a blank field in the current dataset snapshot.

This column, likely an origin or source indicator (possibly a survey or system variant field), is effectively empty: 92% of its 50 rows are null, and the sole non-null 'value' present in 4 rows is itself an empty string. With cardinality of 1 and entropy of 0, there is zero information content in this column. The combination of near-total nulls and a blank top value means the column carries no usable signal whatsoever.

This column contains Hungarian-language packaging text, but is almost entirely empty: 92% null rate across 50 rows, and the only non-null value observed is an empty string appearing 4 times. With cardinality of 1 and entropy of 0.0, the column carries zero information — it is effectively unpopulated.

This column appears to be a Swedish-language packaging text field ('sv' suffix indicating Swedish locale), but it is effectively empty in this dataset. A 92% null rate leaves only 4 non-null rows, and all 4 of those contain an empty string — meaning there is zero usable content across all 50 rows. The cardinality of 1 and entropy of 0.0 confirm complete absence of informational signal.

This column appears to be a Bulgarian-language generic name field (likely a pharmaceutical or product name localization), but it is almost entirely absent: 94% of rows are null and the remaining 3 non-null rows contain only an empty string. With cardinality of 1 and entropy of 0, the column carries zero information.

This column appears to be a Czech-language generic name field (indicated by the '_cs' suffix) that is almost entirely empty: 94% of its 50 rows are null, and the sole non-null value is an empty string appearing 3 times. With cardinality of 1 and entropy of 0, the column carries zero information — it is effectively unpopulated.

This column appears to be an Estonian-language generic name field ('et' locale suffix), but it is effectively empty: 94% of its 50 rows are null, and the sole non-null value is a blank string appearing 3 times, giving a cardinality of 1. The column carries zero information — entropy is 0.0 and top_rate is 1.0 across a single empty token.

This column is a Russian-language ingredients text field for food/product records, almost certainly a localized variant of a broader ingredients column. It is 94% null across 50 rows, and the only non-null value observed is an empty string (appearing 3 times), meaning there is effectively zero usable content in this column. Cardinality of 1 and entropy of 0.0 confirm complete absence of informational signal.

This column is intended to store Russian-language ingredients text with allergen information for food products. It is effectively empty: 94% of the 50 rows are null, and the sole non-null value present is an empty string (''), giving a cardinality of 1 and entropy of 0. There is no usable signal whatsoever in this column for the sampled data.

This column ('origin_bg') is a categorical field with 50 rows, but 94% of values are null and the sole non-null value is an empty string appearing 3 times — making it entirely devoid of usable information. Cardinality is 1, entropy is 0, and top_rate is 1.0, confirming complete uniformity across non-null entries. Both alerts (null_rate and imbalance) are triggered, which is expected given the near-total absence of data.

This column appears to be an origin or source tag in Amharic/Ethiopic script (indicated by the '_et' suffix), but it is effectively empty: 94% of the 50 rows are null, and the sole non-null value present is an empty string appearing 3 times. With cardinality of 1 and entropy of 0.0, the column carries zero information. This is likely an unfilled localization or metadata field.

This column appears to be a Russian-language origin/source field that is almost entirely unpopulated: 94% of the 50 rows are null, and the sole non-null value is an empty string appearing 3 times. With cardinality of 1, zero entropy, and a top_rate of 1.0, the column carries absolutely no information. It was likely intended to capture Russian-locale origin metadata but was never populated.

This column contains Bulgarian-language packaging text for products, but it is almost entirely empty: 94% of the 50 rows are null, and the sole non-null value observed is an empty string appearing 3 times (top_rate 1.0). With cardinality of 1 and entropy of 0.0, the column carries zero information in its current state.

This column appears to be Czech-language packaging text (`_cs` locale suffix), but it is almost entirely empty: 94% null rate across 50 rows, and the only observed non-null value is an empty string appearing 3 times. With cardinality of 1 and entropy of 0, the column carries zero information — it is effectively unpopulated for this dataset slice.

This column contains Estonian-language packaging text (`_et` locale suffix), but is effectively empty: 94% of its 50 rows are null, and the sole non-null value across all 3 populated rows is an empty string. With cardinality of 1 and entropy of 0.0, the column carries zero information — it has never been populated in this dataset.

This column holds Russian-language packaging text, but is almost entirely empty: 94% of the 50 rows are null, and the sole non-null value appearing 3 times is an empty string — giving a cardinality of 1 and zero entropy. In practice the column carries no information whatsoever across the observed sample.

This column appears to be a binary flag or pass/fail indicator for ingredient percentage analysis, taking only two distinct values across all 50 rows: 1.0 (present in the vast majority) and -1.0 (a minority case). With Q1, median, and Q3 all equal to 1.0 and a mean of 0.84, roughly 84% of records are coded 1.0 while the remaining ~16% are -1.0, which are flagged as the 4 outliers (8% outlier rate). The extreme skew (−3.10) and kurtosis (7.59) are entirely explained by this near-constant binary distribution, not by a continuous numeric spread.

This column appears to be a synthetic or encoded identifier rather than a true popularity metric — values cluster tightly in the 23.9–24.0 billion range, with a median of ~23,999,500,422 and a max of ~23,999,992,269, suggesting a fixed-prefix integer key scheme. The strong negative skew (−2.67) and high kurtosis (5.11) are driven by 5 outlier values that fall far below the cluster, near the minimum of ~22,999,500,355, which is about 1 billion lower than the bulk of records. Despite the name 'popularity_key', the distribution is inconsistent with any organic popularity signal and is almost certainly a generated or composite key.

This column likely represents a count of scans per record (e.g., barcode or document scans), with 50 records and no nulls. The bulk of values sit in a moderate range (Q1=387, median=492, Q3=604), but extreme positive skew (3.90) and very high kurtosis (18.72) are driven by 4 outliers (8% of rows) reaching up to 2523 — more than 4× the median. The min of 333 suggests a natural floor, possibly a minimum scan threshold or truncation artefact.

This column represents a count of unique scans (likely QR-code or barcode scan events) per record, with 50 observations and no nulls. The bulk of values cluster between 362.75 (Q1) and 560.75 (Q3), yet a right-skewed tail (skew=3.91, kurtosis=18.71) driven by 4 outliers pulls the mean (525.38) well above the median (432.0), with a maximum of 2257.0 — nearly 4× the Q3 value. The outlier rate of 8% in just 50 rows is a strong signal that a small number of records see dramatically higher scan volumes than the rest.