wild openfoodfacts sample

A categorical update_key field with only 9 distinct values across 50 rows, dominated by 'brands' at 56% (28/50) and 'sort' at 20%. The long tail mixes human-readable labels ('divinfood', 'nova-yogurts', 'germany2', 'france') with timestamp-style tokens ('key_1748337248', 'ingredients20240805'), suggesting inconsistent naming conventions for what appears to track update batches or jobs. Entropy ratio of 0.64 confirms the heavy concentration on a few keys. Treatment: Group rare keys into 'other' or normalize naming before using as a grouping dimension. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

9

top_value

brands

top_rate

0.56

cardinality

9

entropy

2.015

entropy_ratio

0.6357

Hierarchical product category strings (Open Food Facts style taxonomy paths), stored as comma-separated breadcrumbs. Near-unique with 45 distinct values across 50 rows and entropy ratio 0.99, and the strings appear in mixed languages (French, English, Polish, Bulgarian Cyrillic), so direct grouping will fragment. Top value covers only 4% of rows and one row is null. Treatment: Split on commas, normalise language, and keep only the top-level taxon as a categorical feature. high · anthropic:claude-opus-4-7

n

50

nulls

1 (2.0%)

unique

45

top_value

Snacks, Snacks sucrés, Biscuits et gâteaux, Biscuits, Biscuits secs

top_rate

0.04082

cardinality

45

entropy

5.451

entropy_ratio

0.9926

Numeric Eco-Score rating per item, ranging from 13 to 94 with a mean of 47.7 and median of 44. The distribution is mildly right-skewed (0.31) and platykurtic (-0.79), spanning a wide IQR of 36.5 with no outliers flagged. Notably, 14% of values are null and only 31 unique scores appear across 50 rows. Treatment: Impute or flag the 14% nulls, then use as a continuous feature without transformation. high · anthropic:claude-opus-4-7

n

50

nulls

7 (14.0%)

unique

31

min

13

max

94

mean

47.74

median

44

std

21.19

q1

27.5

q3

64

iqr

36.5

skew

0.3069

kurtosis

-0.7946

n_outliers

0

outlier_rate

0

zero_rate

0

This appears to be a categorical flag for environmental impact severity, but it carries no usable signal in this sample. 56% of the 50 rows are null, and the remaining 22 records all hold the empty string, leaving cardinality at 1 and entropy at 0. Treatment: Drop; the column is effectively constant and majority-null. high · anthropic:claude-opus-4-7

n

50

nulls

28 (56.0%)

unique

1

top_value

top_rate

1

cardinality

1

entropy

0

entropy_ratio

0

Finnish-language ingredient declarations, almost entirely absent: 90% of the 50 rows are null and only 4 distinct non-null values appear, two of which are empty strings. The few populated entries are verbose product ingredient lists (chocolate, wheat-based baked goods) with allergen markup, suggesting this is a localized free-text field rather than a categorical feature despite its low cardinality here. Treatment: Drop or set aside; null_rate 0.9 makes it unusable as a feature without a Finnish-text NLP pipeline. high · anthropic:claude-opus-4-7

n

50

nulls

45 (90.0%)

unique

4

top_value

top_rate

0.4

cardinality

4

entropy

1.922

entropy_ratio

0.961

This appears to be a flag indicating whether nutrition data was prepared, but it carries no information: only one unique value (an empty string) appears across all 48 non-null rows, with a 4% null rate. Entropy is 0 and top_rate is 1.0, so the column is constant. Treatment: Drop; constant column with no signal. high · anthropic:claude-opus-4-7

n

50

nulls

2 (4.0%)

unique

1

top_value

top_rate

1

cardinality

1

entropy

0

entropy_ratio

0

This column, packaging_shapes_tags, was skipped by the profiler so no descriptive statistics are available beyond a row count of 50 and a null rate of 0. The name suggests a tag-style field listing packaging shape descriptors, likely multi-valued per row, which is probably why it was bypassed. Without unique counts or value samples, nothing further can be confirmed. Treatment: Re-profile after splitting the tag list, then one-hot or multi-label encode. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

Column 'nutrient_levels_tags' was skipped by the profiler, so no statistics beyond a 50-row count and 0% null rate are available. The name suggests a list of nutrient classification tags (likely multi-valued strings like 'fat-in-low-quantity'), but uniqueness, cardinality, and value distribution are all unknown. Treat any downstream use cautiously until the column is re-profiled with list-aware parsing. Treatment: Re-profile with list/tag-aware parsing, then one-hot or multi-label encode the individual tags. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

The column packagings_materials was skipped by the profiler, so its kind is unknown and no descriptive statistics are available. We only know there are 50 rows and a 0.0 null rate; uniqueness, type, and value distribution are all missing. The name suggests structured packaging material data (likely nested or list-valued), which would explain why the profiler bailed out. Treatment: Inspect raw values manually and parse the nested structure before any downstream use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

This column is named `ingredients_without_ecobalyse_ids`, suggesting it lists ingredients that lack matching identifiers in the Ecobalyse reference system. Saturn skipped profiling, so type, uniqueness, and value distribution are unknown despite a populated null_rate of 0.0 across 50 rows. Treatment: Inspect raw values manually to determine structure (likely a list) before deciding on parsing or join strategy. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

This appears to be a Dutch-language generic product name field, likely from a food product catalog. It is largely unusable as-is: 76% of rows are null and among the 12 non-null entries, 9 are empty strings, leaving only 3 distinct real values (e.g., 'Extra fijne pure chocolade'). Cardinality is just 4 across 50 rows, so there is essentially no signal here. Treatment: Drop, or retain only as a descriptive label — too sparse to model. high · anthropic:claude-opus-4-7

n

50

nulls

38 (76.0%)

unique

4

top_value

top_rate

0.75

cardinality

4

entropy

1.208

entropy_ratio

0.6038

The column 'misc_tags' was skipped by the profiler, so no type inference, uniqueness count, or value statistics are available. The only confirmed signals are 50 rows with a 0.0 null rate. Without further stats, its content and structure cannot be characterized. Treatment: Re-profile with a parser suited to this column (e.g., list/JSON tags) before deciding on use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

Swedish-localised product name field, populated for only 4 of 50 rows (null_rate 0.92). The 4 present values are all unique, giving maximum entropy (entropy_ratio 1.0) but no repeated category to learn from. Values like "90% Cocoa" and "Dark 70%" look English rather than Swedish, suggesting localisation is incomplete or mislabelled. Treatment: Drop or defer until localisation coverage improves; not usable as a feature at 92% null. high · anthropic:claude-opus-4-7

n

50

nulls

46 (92.0%)

unique

4

top_value

90% Cocoa

top_rate

0.25

cardinality

4

entropy

2

entropy_ratio

1

A numeric count of scans per record, with 49 unique values across 50 rows and no nulls or zeros. The distribution is tightly clustered (median 492, IQR 217) but extremely right-skewed (skew 3.90, kurtosis 18.72) with a max of 2523 versus a Q3 of 604, producing 4 outliers (8%). The mean (577.94) sits well above the median, confirming a heavy upper tail. Treatment: Log-transform or winsorize before modelling to tame the heavy right tail. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

49

min

333

max

2,523

mean

577.9

median

492

std

343.9

q1

387

q3

604

iqr

217

skew

3.899

kurtosis

18.72

n_outliers

4

outlier_rate

0.08

zero_rate

0

Constant numeric column holding the value 996.0 across all 50 rows with no nulls. Despite being typed as numeric, the zero variance (std 0.0, iqr 0.0) and single unique value indicate this is a schema/version tag rather than a measurement. Carries no signal for modelling. Treatment: Drop before modelling; retain only as a provenance tag. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

1

min

996

max

996

mean

996

median

996

std

0

q1

996

q3

996

iqr

0

skew

0

kurtosis

0

n_outliers

0

outlier_rate

0

zero_rate

0

This column holds Open Food Facts product URLs, one per row, with every value unique across all 50 rows (cardinality 50, entropy_ratio 1.0). The URL path embeds a product barcode plus a slugified name, so it functions as a permalink/identifier rather than a feature. No nulls, but the long_tail alert simply reflects that every row is its own category. Treatment: Drop from modelling; keep as a join key or reference link to the source product page. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

50

top_value

https://world.openfoodfacts.org/product/6111242100992/perly

top_rate

0.02

cardinality

50

entropy

5.644

entropy_ratio

1

The column `vitamins_tags` was skipped by the profiler, so no type, uniqueness, or distribution statistics are available beyond a row count of 50 and a null rate of 0.0. The name suggests a list-valued field enumerating vitamin identifiers (e.g., tag-style strings), but this cannot be confirmed from the evidence. Without parsing, downstream use is blocked. Treatment: Re-profile after parsing as a list of tags, then one-hot or multi-hot encode. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

This column was skipped by the profiler (alert: "skipped"), so its kind is unknown and no statistics were computed beyond a row count of 50 with 0% nulls. The name suggests a debug artefact holding sorted language codes, likely a list or compound value the profiler couldn't classify. Without unique counts or value samples there is nothing further to infer. Treatment: Drop unless you can re-profile with list/struct support; it appears to be a debug field. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

Free-form packaging descriptions, likely from a food/product database (Open Food Facts style) given multilingual prefixes like 'en:', 'es:', 'pt:'. Cardinality is extreme: 41 unique values across 50 rows with entropy ratio 0.985, and the top value 'Plastique' covers only 9% — most entries are comma-separated multi-tag strings mixing languages. 12% are null, and the long_tail alert confirms there is no usable category structure as-is. Treatment: Split on commas, normalize language prefixes, and one-hot encode the resulting material tags rather than using the raw string. high · anthropic:claude-opus-4-7

n

50

nulls

6 (12.0%)

unique

41

top_value

Plastique

top_rate

0.09091

cardinality

41

entropy

5.278

entropy_ratio

0.9851

The column is named "grades" and contains 50 rows with no nulls, but saturn skipped profiling and could not infer a kind, so no distributional stats are available. Without n_unique or value summaries, it's impossible to tell whether this holds letter grades, numeric scores, or a nested structure. The "skipped" alert is the key signal: something about the storage type prevented standard analysis. Treatment: Manually inspect a sample to determine the underlying type before deciding on a downstream encoding. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

Values are Unix epoch seconds (min 1737907641, max 1768643720) so this column is a last-modified timestamp, likely covering early 2025 through late 2025. All 50 rows are unique with no nulls, but the distribution is heavily left-skewed (skew -1.96) with 6 outliers (12%) sitting far below the q1 of 1761612624, suggesting a small tail of much older edits while most records cluster within a ~6.1M second IQR. Treat as a timestamp, not a numeric feature. Treatment: Convert from epoch seconds to datetime and derive recency or bucketed features instead of using the raw integer. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

50

min

1.738e+09

max

1.769e+09

mean

1.763e+09

median

1.767e+09

std

8.093e+06

q1

1.762e+09

q3

1.768e+09

iqr

6.138e+06

skew

-1.961

kurtosis

2.972

n_outliers

6

outlier_rate

0.12

zero_rate

0

origin_nl appears to be a categorical attribute (likely a Dutch-language origin label) but is effectively empty in this sample. 76% of the 50 rows are null, and the remaining 12 non-null entries are all the empty string, giving a cardinality of 1 and entropy of 0. There is no usable signal here. Treatment: Drop; column has no variance and is mostly null. high · anthropic:claude-opus-4-7

n

50

nulls

38 (76.0%)

unique

1

top_value

top_rate

1

cardinality

1

entropy

0

entropy_ratio

0

Language code for the allergens text, with 6 distinct values across 50 rows and a 4% null rate. The distribution is nearly bimodal between 'en' (22) and 'fr' (21), with es/de/it/pl appearing once or twice each — a language mix worth flagging before any text processing. Treatment: Use as a language filter or routing key before tokenizing the allergens text. high · anthropic:claude-opus-4-7

n

50

nulls

2 (4.0%)

unique

6

top_value

en

top_rate

0.4583

cardinality

6

entropy

1.578

entropy_ratio

0.6104

The column 'states_hierarchy' was skipped by the profiler, so its kind is unknown and no descriptive statistics were computed. We can only confirm there are 50 rows with no nulls; uniqueness, type, and value distribution are unavailable. Treatment: Re-profile or inspect manually to determine type before any downstream use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

Japanese-language ingredients text, almost entirely absent from this sample. 98% of the 50 rows are null, and the single non-null value is an empty string, leaving cardinality at 1 and entropy at 0. Treatment: Drop; the column carries no usable signal in this sample. high · anthropic:claude-opus-4-7

n

50

nulls

49 (98.0%)

unique

1

top_value

top_rate

1

cardinality

1

entropy

0

entropy_ratio

0

The column `teams_tags` was skipped by the profiler, so its kind, uniqueness, and value distribution are unknown. Only two facts are available: 50 rows were seen and none were null. Without further stats, the content (e.g. whether it holds lists, delimited tags, or structured objects) cannot be characterised. Treatment: Re-profile with a parser that handles this column's type before deciding on downstream use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

This column appears to capture user-submitted allergen/ingredient traces, prefixed with a language code like '(en)' or '(fr)' followed by comma-separated tags such as 'en:milk,en:nuts'. With 35 unique values across 50 rows and entropy ratio 0.938, it is highly diverse; the top value '(en) ' (an empty tag list) covers only 14% and the distribution has a long tail. Notably, the language prefix is mixed (English and French) and many entries are blank tag lists, which complicates direct use as a category. Treatment: Parse the language prefix and split the tag list into a multi-hot allergen feature before modelling. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

35

top_value

(en)

top_rate

0.14

cardinality

35

entropy

4.811

entropy_ratio

0.9379

The column `origins_tags` was skipped by the profiler, so kind is unknown and no descriptive statistics were computed. The only confirmed signals are 50 rows present and a 0.0 null rate; uniqueness, value distribution, and data type are all unavailable. Treatment: Re-profile with an appropriate parser (likely a list/tag field) before deciding on downstream use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

This column records the unit of measurement for serving quantity, almost exclusively grams ('g' at 45 of 46 non-null rows, top_rate 0.978) with a single 'ml' entry. With only 2 unique values, an 8% null rate, and entropy_ratio of 0.151, it carries almost no information. Treatment: Drop or collapse to a binary flag; near-constant with negligible signal. high · anthropic:claude-opus-4-7

n

50

nulls

4 (8.0%)

unique

2

top_value

g

top_rate

0.9783

cardinality

2

entropy

0.1511

entropy_ratio

0.1511

The column "vitamins_prev_tags" was skipped by the profiler, so no type, uniqueness, or distribution stats are available. The only confirmed signals are 50 rows with a 0.0 null rate. Without further evidence the content (likely a list/array of prior tag values given the name) cannot be characterized. Treatment: Re-profile with a parser that handles nested/array values before deciding on use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

This column is labelled ingredients_hierarchy but saturn skipped profiling it, so no type, uniqueness, or value statistics are available. The only confirmed signals are that it has 50 rows and zero nulls. Without further evidence, the structure (likely nested or list-valued, given the name) cannot be verified. Treatment: Re-profile with a parser that handles nested or list-valued fields before deciding on use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

Numeric count of unique scans per row, with 48 distinct values across 50 records and no nulls or zeros. The distribution is heavily right-skewed (skew 3.91, kurtosis 18.71): median is 432 against a mean of 525.38, and the max of 2257 sits far beyond q3 of 560.75, producing 4 outliers (8% outlier rate). Std of 306.41 dwarfs the IQR of 198, confirming a long upper tail. Treatment: Log-transform or winsorize before modelling to tame the long upper tail. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

48

min

319

max

2,257

mean

525.4

median

432

std

306.4

q1

362.8

q3

560.8

iqr

198

skew

3.911

kurtosis

18.71

n_outliers

4

outlier_rate

0.08

zero_rate

0

Free-form labels/certifications column (e.g. organic, fair-trade, Triman, Green Dot) stored as comma-separated multi-label strings, often mixing English, French, Portuguese and Spanish tokens. Of 50 rows, 42 distinct values and entropy ratio 0.95 indicate near-unique combinations; the only repeated 'value' is the empty string (8 rows, 16%) on top of a 2% null rate, so roughly one in five records carries no label at all. The long_tail alert is well earned — almost every non-empty cell is its own bag of tags. Treatment: Split on commas into a multi-hot tag set (normalising language variants) before modelling. high · anthropic:claude-opus-4-7

n

50

nulls

1 (2.0%)

unique

42

top_value

top_rate

0.1633

cardinality

42

entropy

5.125

entropy_ratio

0.9504

Likely an English-language generic product name field, but it is essentially empty: the top value is the blank string at 83.7% of non-null rows, with a further 14% null. Only 7 actual product descriptions appear across 50 rows (e.g. 'Dark chocolate', 'Crackers'), all singletons, giving cardinality 8 and entropy ratio 0.37. The long_tail alert reflects that every real value occurs exactly once. Treatment: Drop or treat blanks as missing; too sparse and unique to use as a categorical feature. high · anthropic:claude-opus-4-7

n

50

nulls

7 (14.0%)

unique

8

top_value

top_rate

0.8372

cardinality

8

entropy

1.098

entropy_ratio

0.366

The column 'weighters_tags' was skipped by the profiler, so no type, uniqueness, or value statistics are available beyond a row count of 50 and a null rate of 0.0. Without kind detection or sample values, its content and structure cannot be characterised here. The name suggests it may hold tag-like annotations, but this is not confirmed by evidence. Treatment: Re-profile with parsing enabled to determine type and cardinality before deciding on use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

The column `popularity_tags` was skipped by the profiler, so its kind is unknown and no descriptive statistics were computed. The only signals available are that 50 rows were seen with a null rate of 0.0, meaning every row carries some value. Cardinality, type, and distribution are all missing, so the column's actual content cannot be characterized from this evidence. Treatment: Re-profile with the appropriate parser (likely list/JSON) before deciding on downstream use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

Likely a Finnish-localized product name field, but it is essentially empty: 90% nulls and the most frequent observed value is the empty string (top_rate 0.4 of the 5 non-null entries). Among the few populated rows, the names are in English (e.g., 'Excellence: 90% cocoa Dark Supreme', 'Arriba 85% Cacao Dark Chocolate'), contradicting the _fi suffix. With only 4 unique values across 50 rows, this column carries almost no usable signal. Treatment: Drop or defer until localization coverage improves; do not use as a feature. high · anthropic:claude-opus-4-7

n

50

nulls

45 (90.0%)

unique

4

top_value

top_rate

0.4

cardinality

4

entropy

1.922

entropy_ratio

0.961

This appears to be a French-language origin/provenance field describing where a product or its ingredients are made. The column is essentially empty: 40 of 50 rows hold the empty string and another 8% are null, leaving only 6 distinct non-blank descriptions ranging from a single country ('France') to multi-region ingredient breakdowns. Entropy ratio of 0.319 and a top_rate of 0.87 confirm the long-tail alert — almost no usable signal here. Treatment: Drop or defer; too sparse and unstructured to use without targeted NER on the few populated strings. high · anthropic:claude-opus-4-7

n

50

nulls

4 (8.0%)

unique

7

top_value

top_rate

0.8696

cardinality

7

entropy

0.8958

entropy_ratio

0.3191

Free-text generic product names, predominantly French with some English entries (e.g., "Compound Chocolate with MILK AND ALMONDS"). The dominant value is the empty string at 21/50 (top_rate 0.4375), and combined with a 0.04 null_rate this means most rows carry no usable name. The remaining 28 unique values are nearly all singletons, producing the flagged long tail. Treatment: Treat empty strings as missing, then tokenize/normalize language before embedding or matching. high · anthropic:claude-opus-4-7

n

50

nulls

2 (4.0%)

unique

28

top_value

top_rate

0.4375

cardinality

28

entropy

3.663

entropy_ratio

0.762

This column records the Nutri-Score version applied to each row, and every one of the 50 records carries the value "2023". With cardinality 1 and entropy 0, it offers no discriminative signal in this sample. Treatment: Drop, constant column. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

1

top_value

2023

top_rate

1

cardinality

1

entropy

0

entropy_ratio

0

This column, named ingredients_without_ciqual_codes, was skipped by the profiler so no descriptive statistics are available beyond a row count of 50 and a null rate of 0. The name suggests it holds ingredient entries that lack a matching CIQUAL food-database code, likely as a list or nested structure that the profiler could not introspect. Without unique counts or value samples, nothing further can be inferred. Treatment: Re-profile after parsing the nested structure, or explode to a list before downstream use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

This column was skipped by the profiler, so no statistics beyond row count and null rate are available. The name suggests it holds tags for manufacturing locations, likely a multi-valued or list-like field that the dissector could not classify. With 50 rows and 0% nulls reported but no uniqueness or value stats, nothing further can be inferred from the evidence. Treatment: Re-profile after parsing the tag list, then one-hot or multi-label encode top tags. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

The column `photographers_tags` was skipped by the profiler, so no kind, uniqueness, or value statistics are available beyond a row count of 50 and a null rate of 0.0. The name suggests it holds tag annotations associated with photographers, likely a list or delimited string, but this cannot be confirmed from the evidence. No further signal is present to characterise distribution, cardinality, or content. Treatment: Re-profile with list/string parsing enabled before deciding on downstream handling. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

Polish-language packaging text field that is effectively empty: 90% of the 50 rows are null and the remaining 10% are all the empty string, giving a single observed value and zero entropy. There is no usable signal here, only nulls and blanks. Treatment: Drop; the column carries no information. high · anthropic:claude-opus-4-7

n

50

nulls

45 (90.0%)

unique

1

top_value

top_rate

1

cardinality

1

entropy

0

entropy_ratio

0

The column `informers_tags` was skipped by the profiler, so no type, uniqueness, or value statistics were computed beyond a row count of 50 with 0% nulls. Without stats it's impossible to tell whether this holds scalar tags, delimited lists, or nested structures, though the plural name hints at a multi-valued tag field. Treat any interpretation as provisional until the column is re-profiled. Treatment: Re-run profiling with parsing for list/JSON values before deciding on downstream use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

English-language ingredient lists for food products, stored as free-form text rather than a controlled vocabulary. With 36 unique values across 50 rows and entropy ratio 0.93, values are nearly all distinct; the only repeated 'value' is the empty string (9 occurrences, top_rate 0.20), and 12% are null, so roughly a third of rows carry no usable ingredient text. Content is heterogeneous — multi-sentence allergen-tagged lists, percentages, punctuation noise, and at least one junk entry ('Hhhhh'). Treatment: Normalize, tokenize, and embed (or parse into ingredient lists) before modelling; treat empty strings as nulls. high · anthropic:claude-opus-4-7

n

50

nulls

6 (12.0%)

unique

36

top_value

top_rate

0.2045

cardinality

36

entropy

4.811

entropy_ratio

0.9306

Free-form Italian ingredient lists for food products, with 68% nulls and only 50 rows total. Of the 16 non-null entries, 5 are empty strings (top_rate 0.3125) and the remaining values are nearly all unique long product descriptions, yielding 12 distinct values and entropy_ratio 0.913. Effectively unstructured text rather than a categorical field, despite being typed as such. Treatment: Treat as free text: normalize empty strings to null, then tokenize/parse for allergen or ingredient extraction rather than one-hot encoding. high · anthropic:claude-opus-4-7

n

50

nulls

34 (68.0%)

unique

12

top_value

top_rate

0.3125

cardinality

12

entropy

3.274

entropy_ratio

0.9134

This appears to be a German-origin flag or label, but it carries no information in this sample: 60% of rows are null and the remaining 20 rows all hold the empty string, giving a single unique value and zero entropy. There is no signal to model on here. Treatment: Drop; constant column with majority nulls. high · anthropic:claude-opus-4-7

n

50

nulls

30 (60.0%)

unique

1

top_value

top_rate

1

cardinality

1

entropy

0

entropy_ratio

0

This is the NOVA food classification group (1-4 scale) indicating processing level, with 3 unique values present across 50 rows and a 4% null rate. The distribution is heavily skewed toward ultra-processed foods: median is 4.0, Q1-Q3 spans 3-4, and skew of -2.06 with kurtosis 5.65 confirms a long left tail with one outlier at the low end. Despite being numeric, only 3 of the 4 possible NOVA categories appear in this sample. Treatment: Treat as ordinal categorical rather than continuous; impute the 4% nulls with median (4) or a missing-indicator. high · anthropic:claude-opus-4-7

n

50

nulls

2 (4.0%)

unique

3

min

1

max

4

mean

3.646

median

4

std

0.601

q1

3

q3

4

iqr

1

skew

-2.062

kurtosis

5.651

n_outliers

1

outlier_rate

0.02083

zero_rate

0

Finnish-language packaging text field that is effectively empty in this sample. 90% of the 50 rows are null and the remaining 5 rows all hold the empty string, giving a single observed value and zero entropy. Treatment: Drop; column carries no information in this sample. high · anthropic:claude-opus-4-7

n

50

nulls

45 (90.0%)

unique

1

top_value

top_rate

1

cardinality

1

entropy

0

entropy_ratio

0

This column packs an Open Food Facts-style product completion checklist into a single comma-joined string of `en:*-completed` / `en:*-to-be-completed` tags covering nutrition, ingredients, photos, packaging, etc. With 26 unique combinations across just 50 rows (entropy ratio 0.91) and the most common state appearing only 8 times, it behaves like a long-tail composite status flag rather than a clean category. The values are clearly multi-valued — they should be split into individual status tags before any modelling. Treatment: Split on comma and one-hot encode each `en:*` tag instead of treating the concatenated string as a single category. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

26

top_value

en:to-be-completed, en:nutrition-facts-completed, en:ingredients-completed, en:expiration-date-completed, en:packaging-code-to-be-completed, en:characteristics-to-be-completed, en:origins-to-be-completed, en:categories-completed, en:brands-completed, en:packaging-completed, en:quantity-completed, en:product-name-completed, en:photos-validated, en:packaging-photo-selected, en:nutrition-photo-selected, en:ingredients-photo-selected, en:front-photo-selected, en:photos-uploaded

top_rate

0.16

cardinality

26

entropy

4.286

entropy_ratio

0.9119

This column appears to be a per-record sum of ingredient percentages where the precise share was not specified, expressed on a 0–100 scale (max 100.0, min 0.4). The distribution is heavily left-skewed (skew -1.18) with a median of 100.0 and Q3 also at 100.0, meaning at least half of the 50 rows have effectively all of their ingredient mass unspecified. Only 22 unique values across 50 rows and a mean of 79.4 confirm the concentration at the upper bound. Treatment: Treat as a data-quality indicator; consider binarizing (e.g. =100 vs <100) rather than using the raw left-skewed value. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

22

min

0.4

max

100

mean

79.42

median

100

std

31.64

q1

53.6

q3

100

iqr

46.4

skew

-1.183

kurtosis

-0.133

n_outliers

0

outlier_rate

0

zero_rate

0

The column 'added_countries_tags' was skipped by the profiler, so no type, uniqueness, or distributional statistics are available beyond a row count of 50 and a null rate of 0.0. The name suggests it holds country tags associated with record additions, likely a list-like or multi-valued field that the profiler could not classify. Without further stats, nothing can be said about cardinality, format, or content. Treatment: Inspect raw values manually and re-profile after parsing into a normalized list type. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

This column is a unique row identifier, with all 50 values distinct (n_unique=50, entropy_ratio=1.0) and no nulls. The values look like product barcodes (mostly 13-digit EAN/GTIN strings such as '6111242100992', with at least one shorter numeric like '20995553'), suggesting a product-level key rather than a sequential surrogate ID. The long_tail alert simply reflects that every value occurs exactly once (top_rate=0.02). Treatment: Drop from modelling; retain as a join key for linking to product metadata. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

50

top_value

6111242100992

top_rate

0.02

cardinality

50

entropy

5.644

entropy_ratio

1

The column 'nutrient_levels' was skipped by the profiler, so its kind is unknown and no descriptive statistics are available. We only know it has 50 rows with a 0.0 null rate; uniqueness, distribution, and value structure are all missing from the evidence. Treatment: Re-profile or manually inspect a sample to determine the underlying type before any downstream use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

Values range from 1,567,543,172 to 1,610,897,644 with a median of 1,608,147,866 — consistent with Unix epoch seconds spanning roughly 2019 to early 2021, masquerading as a numeric sort key. Distribution is heavily left-skewed (skew -2.78, kurtosis 8.09) with 4 outliers (9.1%) trailing toward older timestamps, and 12% of rows are null. The tight IQR of ~6.16M seconds (~71 days) versus a 43M-second range confirms most records cluster late in the window. Treatment: Convert from epoch seconds to datetime and use as a temporal feature rather than a raw numeric. high · anthropic:claude-opus-4-7

n

50

nulls

6 (12.0%)

unique

44

min

1.568e+09

max

1.611e+09

mean

1.605e+09

median

1.608e+09

std

8.692e+06

q1

1.604e+09

q3

1.61e+09

iqr

6.16e+06

skew

-2.782

kurtosis

8.091

n_outliers

4

outlier_rate

0.09091

zero_rate

0

Per-row URL pointing to a 200px product front image hosted on images.openfoodfacts.org, with French/English locale suffixes embedded in the filename. All 50 rows are unique with zero nulls, so this acts as a row-level asset reference rather than a feature. The path segments encode the product barcode (e.g. 6111242100992), making this effectively a derivable identifier. Treatment: Drop from modelling; retain as a fetch URL for image pipelines or extract the embedded barcode. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

50

top_value

https://images.openfoodfacts.org/images/products/611/124/210/0992/front_fr.172.200.jpg

top_rate

0.02

cardinality

50

entropy

5.644

entropy_ratio

1

The column 'packaging_recycling_tags' was skipped by the profiler, so no type, uniqueness, or value statistics are available beyond a row count of 50 and a null rate of 0.0. The name suggests it holds packaging or recyclability tags, likely a multi-valued or list-like field that the profiler could not categorise. Without parsed values, nothing can be said about cardinality, distribution, or label vocabulary. Treatment: Re-profile after parsing into a list or one-hot tag set before deciding on use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

This is a categorical food taxonomy field using Open Food Facts-style prefixed slugs (e.g., 'en:biscuits-and-cakes'). The distribution is heavily concentrated on sweets: 'en:biscuits-and-cakes' (17/49) and 'en:chocolate-products' (16/49) together account for roughly two-thirds of non-null rows, with 11 distinct categories across 50 records and a 2% null rate. Entropy ratio of 0.74 confirms moderate concentration rather than uniform spread. Treatment: Strip the 'en:' prefix and one-hot or target-encode; consider grouping the long tail of single-occurrence categories into 'other'. high · anthropic:claude-opus-4-7

n

50

nulls

1 (2.0%)

unique

11

top_value

en:biscuits-and-cakes

top_rate

0.3469

cardinality

11

entropy

2.549

entropy_ratio

0.7367

Column 'nova_groups_markers' was skipped by the profiler, so no type, uniqueness, or distribution stats are available beyond a row count of 50 and a null rate of 0.0. The name suggests it carries NOVA food-classification group markers, likely a list or structured field that the dissector could not parse. Without parsed values, nothing further can be said about its content. Treatment: Inspect raw values manually and reparse (likely a list/struct) before deciding on use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

German-language packaging description field, almost entirely unpopulated. With a 60% null rate and the empty string accounting for 19 of the 20 non-null rows (95% top_rate), only one row carries actual content ("1 Folie aus 22 PAP zum Recyclen"). Cardinality of 2 and entropy ratio of 0.29 confirm there is virtually no usable signal here. Treatment: Drop; effectively empty with only one informative value. high · anthropic:claude-opus-4-7

n

50

nulls

30 (60.0%)

unique

2

top_value

top_rate

0.95

cardinality

2

entropy

0.2864

entropy_ratio

0.2864

This column appears to hold lowercase ISO language codes, with 6 distinct values across 50 rows and no nulls. The distribution is dominated by 'fr' (25) and 'en' (19), together covering 44 of 50 rows, while 'es', 'de', 'it', and 'pl' form a thin long tail. Entropy ratio of 0.63 reflects this Franco-English skew rather than a balanced multilingual mix. Treatment: One-hot encode, optionally collapsing rare codes (it, pl, de, es) into an 'other' bucket. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

6

top_value

fr

top_rate

0.5

cardinality

6

entropy

1.628

entropy_ratio

0.6297

The column 'checkers' was skipped by the profiler, so its data type and value distribution are unknown. Only the row count (50) and null rate (0.0) are reported; n_unique and all other statistics are missing. Without further inspection there is no basis to infer what this column represents. Treatment: Re-profile or manually inspect the column before any downstream use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

Spanish-language packaging description, populated for almost none of the rows. 60% are null and of the 20 non-null entries, 19 are empty strings, leaving exactly one real value describing a recyclable cardboard box and plastic tray. Effective cardinality is 2 and entropy ratio is 0.29, so this column carries virtually no signal in this sample. Treatment: Drop unless a larger sample shows meaningful Spanish text coverage. high · anthropic:claude-opus-4-7

n

50

nulls

30 (60.0%)

unique

2

top_value

top_rate

0.95

cardinality

2

entropy

0.2864

entropy_ratio

0.2864

This column is labelled `unknown_nutrients_tags` and was skipped by the profiler, so no descriptive statistics, uniqueness count, or value samples are available. The only confirmed signals are that all 50 rows are non-null and the column kind is reported as 'unknown'. Without further evidence its content and structure cannot be characterised. Treatment: Re-profile with type inference enabled before deciding on use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

The column `editors_tags` was skipped by the profiler, so no type, cardinality, or value statistics are available beyond a row count of 50 and a null rate of 0.0. Without sample values or a detected kind, the content and structure are unknown — the name suggests editor-assigned tags, possibly a list or delimited string, but this is not confirmed by evidence. Treatment: Re-profile with list/string parsing enabled to determine structure before any downstream use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

This is a nutrition-score warning flag indicating whether fruit/vegetable/nut content was estimated from ingredients. Every one of the 45 non-null rows holds the value 1.0, and 10% of rows are null — so the column carries no discriminative signal in this sample, only a presence/absence distinction. Treatment: Drop as a feature; optionally retain a binary is_null indicator if the missingness itself is meaningful. high · anthropic:claude-opus-4-7

n

50

nulls

5 (10.0%)

unique

1

min

1

max

1

mean

1

median

1

std

0

q1

1

q3

1

iqr

0

skew

0

kurtosis

0

n_outliers

0

outlier_rate

0

zero_rate

0

This column appears to be a lowercase ISO language code label, with 6 distinct values across 50 rows and one null. English and French dominate at 22 occurrences each, leaving es, de, it, and pl with just 1-2 rows combined — a near-binary distribution despite the multilingual appearance. Entropy ratio of 0.61 confirms the imbalance. Treatment: Group rare codes (es/de/it/pl) into 'other' before stratifying or one-hot encoding. high · anthropic:claude-opus-4-7

n

50

nulls

1 (2.0%)

unique

6

top_value

en

top_rate

0.449

cardinality

6

entropy

1.57

entropy_ratio

0.6072

The column 'nutriscore_data' was skipped by the profiler, so its kind, uniqueness, and value distribution are unknown. The only confirmed signals are 50 rows with a 0.0 null rate. Without further stats, the contents (likely a nested Nutri-Score payload given the name) cannot be characterised. Treatment: Re-profile with nested/struct support enabled before deciding on use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

This column is flagged as skipped by the profiler, so no statistics beyond row count (50) and a null rate of 0.0 were computed. The name suggests it holds tag-style annotations for additional nutritional substances, likely a delimited or list-valued field that the dissector could not type. Without unique counts or value samples, its actual content and cardinality remain unverified. Treatment: Manually inspect raw values and re-profile as a multi-label tag list before deciding to encode or drop. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

Norwegian product name field (suffix _nb suggests Bokmål locale) that is almost entirely empty: 96% null across 50 rows, leaving only 2 non-null observations with one being an empty string and the other '99% mørk sjokolade'. With just two distinct values and effectively no signal, this column cannot support analysis as-is. Treatment: Drop unless joined to a richer localized catalog; null rate is too high to model. high · anthropic:claude-opus-4-7

n

50

nulls

48 (96.0%)

unique

2

top_value

top_rate

0.5

cardinality

2

entropy

1

entropy_ratio

1

This column records the basis on which nutrition data is reported, and every one of the 50 rows carries the single value "100g". With cardinality of 1, entropy of 0, and a top_rate of 1.0, the field provides no discriminating information whatsoever. Treatment: Drop; constant column carries no signal. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

1

top_value

100g

top_rate

1

cardinality

1

entropy

0

entropy_ratio

0

Numeric product quantities stored as strings, treated here as categorical with 27 distinct values across 50 rows. The mode '100' covers 23.4% of non-nulls, but entropy ratio of 0.90 confirms a long tail with most other values appearing only once or twice. Note 6% nulls and the presence of '0' as a quantity, which may indicate missing or placeholder stock entries. Treatment: Cast to numeric and treat as a quantitative feature; investigate zeros and nulls before modelling. high · anthropic:claude-opus-4-7

n

50

nulls

3 (6.0%)

unique

27

top_value

100

top_rate

0.234

cardinality

27

entropy

4.287

entropy_ratio

0.9017

This is a categorical column recording product type, but every one of the 50 rows holds the same value, "food". Cardinality is 1 and entropy is 0, so the column carries no information for modelling or segmentation. Treatment: Drop; constant column with zero entropy. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

1

top_value

food

top_rate

1

cardinality

1

entropy

0

entropy_ratio

0

The column `checkers_tags` was skipped by the profiler, so its kind is unknown and no statistics (uniqueness, value distribution, type) were computed. Only the row count (50) and null rate (0.0) are available; everything else is missing. The name suggests it may hold tag-like values associated with a checkers process, but this cannot be confirmed from the evidence. Treatment: Re-run profiling or manually inspect a sample before deciding how to use this column. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

The column 'nucleotides_tags' was skipped by the profiler, so no type, uniqueness, or distribution statistics are available beyond a row count of 50 and a null rate of 0.0. The name suggests it holds tag-style annotations related to nucleotides, likely a list or delimited string, but this cannot be confirmed from the evidence. No further signal is present to characterise its values. Treatment: Re-profile with list/string parsing enabled before deciding on use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

This column is named languages_tags, suggesting it holds language metadata (likely tag strings such as locale codes) for each record. Saturn skipped detailed profiling, so no cardinality, distribution, or value samples are available beyond a row count of 50 and a null rate of 0.0. Without uniqueness or value stats, no surprises can be flagged. Treatment: Re-profile or inspect raw values to determine structure before deciding whether to split tags and one-hot encode. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

This is a low-cardinality categorical column holding lowercase language codes (fr, en, es, de, it, pl), almost certainly a detected or declared language tag. The distribution is heavily concentrated on French (23/50) and English (20/50), with the top value covering 47.9% of non-null rows and entropy ratio of 0.61. Four percent of rows are null and three languages appear only once, so any per-language analysis will be unstable beyond fr/en. Treatment: Keep fr/en as-is and bucket de/es/it/pl into an 'other' category before encoding. high · anthropic:claude-opus-4-7

n

50

nulls

2 (4.0%)

unique

6

top_value

fr

top_rate

0.4792

cardinality

6

entropy

1.575

entropy_ratio

0.6093

The column `categories_hierarchy` was skipped by the profiler, so no type, uniqueness, or distribution stats are available. The name suggests a nested or path-like categorical structure (e.g., taxonomy levels), but this cannot be confirmed from the evidence. Only the row count (50) and null rate (0.0) are known. Treatment: Re-profile after parsing the hierarchy (e.g., split into level columns) before deciding on encoding. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

URLs pointing to small front-of-pack product images on the Open Food Facts CDN, one per row. Every one of 50 values is unique (entropy_ratio 1.0, top_rate 0.02) and there are no nulls, so this acts as a per-product asset link rather than a feature. URLs mix `front_fr` and `front_en` suffixes, hinting at a French/English language mix in the source catalogue. Treatment: Keep as a reference link; drop from modelling or fetch the images separately for vision features. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

50

top_value

https://images.openfoodfacts.org/images/products/611/124/210/0992/front_fr.172.200.jpg

top_rate

0.02

cardinality

50

entropy

5.644

entropy_ratio

1

This column was skipped by the profiler, so no statistics beyond row count and null rate are available. The name 'entry_dates_tags' suggests a composite field combining dates and tags, likely a nested or list-like structure that didn't fit a scalar type. With 50 rows and 0% nulls, every record has some value, but its shape is unknown from this evidence. Treatment: Inspect raw values and parse into separate date and tag columns before use. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

The column 'ecoscore_tags' was skipped by the profiler, so no statistics, uniqueness, or value samples are available beyond a row count of 50 and a null rate of 0.0. Based on the name alone, it likely holds Open Food Facts-style ecoscore category tags (e.g., 'en:b'), but this cannot be confirmed from the evidence. The 'skipped' alert is the key signal here and warrants a re-profile with appropriate parsing. Treatment: Re-profile with list/tag-aware parsing before deciding on encoding or drop. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

This appears to be a binary warning flag indicating that the fruits/vegetables/legumes share in a Nutri-Score calculation was estimated from ingredients. Every non-null value is 1.0 (n_unique=1, std=0), and 8% of rows are null, so the column carries no discriminative signal in this sample. Treatment: Drop; constant value provides no information. high · anthropic:claude-opus-4-7

n

50

nulls

4 (8.0%)

unique

1

min

1

max

1

mean

1

median

1

std

0

q1

1

q3

1

iqr

0

skew

0

kurtosis

0

n_outliers

0

outlier_rate

0

zero_rate

0

Counts the number of ingredients in a record that lack a CIQUAL code, so it's a data-quality feature describing how complete the ingredient mapping is. The distribution is right-skewed (skew 1.21) with a median of 3.5 but a max of 22 and one outlier; 18% of rows are already fully mapped (zero_rate 0.18). Only 15 unique values across 50 rows, so it behaves like a small ordinal count. Treatment: Treat as a count; consider log1p or a binary 'fully mapped' flag before modelling. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

15

min

0

max

22

mean

4.98

median

3.5

std

4.825

q1

1

q3

7.75

iqr

6.75

skew

1.208

kurtosis

1.491

n_outliers

1

outlier_rate

0.02

zero_rate

0.18

A numeric revenue feature spanning 19 to 674 with a mean of 230 and median of 233.5, suggesting per-record monetary or count values. Distribution is moderately right-skewed (0.71) with a wide IQR of 237.75 and only one outlier (2%), so spread is large but not pathological. All 50 rows are populated with no zeros and 46 unique values. Treatment: Consider a log or sqrt transform before regression to tame the right skew. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

46

min

19

max

674

mean

230

median

233.5

std

166.6

q1

72.75

q3

310.5

iqr

237.8

skew

0.7092

kurtosis

-0.02278

n_outliers

1

outlier_rate

0.02

zero_rate

0

This column appears to be a count of non-nutritive (artificial) sweeteners listed in a product's ingredients. Across all 50 rows it is exactly 0, with zero_rate of 1.0 and no nulls, so it carries no information in this sample. Treatment: Drop; constant zero provides no signal. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

1

min

0

max

0

mean

0

median

0

std

0

q1

0

q3

0

iqr

0

skew

0

kurtosis

0

n_outliers

0

outlier_rate

0

zero_rate

1

This is a count of ingredients on a product that lack Ecobalyse identifiers, ranging from 0 to 29 with a median of 6.5 and mean 8.16. The distribution is right-skewed (skew 1.28) with one outlier at the high end, suggesting most products have a handful of unmapped ingredients while a few have many. Only 2% are zero, meaning nearly every row has at least one ingredient missing an Ecobalyse ID — a notable data-coverage gap. Treatment: Use as-is or log-transform if feeding into a regression; treat as a coverage-quality signal. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

20

min

0

max

29

mean

8.16

median

6.5

std

5.898

q1

4

q3

11

iqr

7

skew

1.28

kurtosis

1.743

n_outliers

1

outlier_rate

0.02

zero_rate

0.02

This column was skipped by the profiler, so no type, uniqueness, or distribution stats are available beyond a row count of 50 and a null rate of 0.0. The name suggests it holds tags describing environmental impact levels, likely a list-valued or multi-label field that the profiler could not classify. Without parsed values there is no way to confirm cardinality, label vocabulary, or whether tags are single- or multi-valued. Treatment: Re-profile after parsing the tag structure (e.g., split lists) before deciding on encoding. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

This column is named `last_image_dates_tags`, suggesting it holds image-related dates and tags, but saturn skipped profiling so type and content cannot be confirmed. The only evidence available is 50 rows with no nulls; uniqueness, distribution, and value samples are all missing. Treatment: Inspect raw values manually to determine structure before deciding on parsing or modelling. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

This column was skipped by the profiler, so its contents are uncharacterised beyond a row count of 50 and a null rate of 0.0. The name suggests a nested or structured label taxonomy, which likely tripped the profiler's type detection. No uniqueness, value, or distribution statistics are available to confirm. Treatment: Inspect raw values manually and parse the hierarchy (e.g., split on delimiter or expand JSON) before profiling again. low · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

—

Free-text English product names with 34 unique values across 50 rows and high entropy ratio (0.91), indicating heavy diversity. Notable issues: 14% nulls plus an empty-string value taking the top slot at 23.3% (10 occurrences), so effective missingness is much higher than null_rate alone suggests. Values mix languages (e.g., 'Edelbitter-Schokolade', 'Chocolat noir', 'tonik') and include junk like 'Hhhhh', flagged as long_tail. Treatment: Normalise empties to null, language-detect, then tokenize/embed before modelling. high · anthropic:claude-opus-4-7

n

50

nulls

7 (14.0%)

unique

34

top_value

top_rate

0.2326

cardinality

34

entropy

4.654

entropy_ratio

0.9147

This appears to be a Nutri-Score warning value estimating fruit/vegetable/legume content from ingredients. The distribution is dominated by zeros (zero_rate 0.89, median and IQR both 0), with a handful of extreme values pushing the max to 50 and producing severe skew (5.93) and kurtosis (35.2). Five outliers (10.9% rate) drive the mean to 1.65 despite a std of 7.55, and 8% of rows are null. Treatment: Binarize (zero vs non-zero) or winsorize before modelling given the heavy zero mass and extreme skew. high · anthropic:claude-opus-4-7

n

50

nulls

4 (8.0%)

unique

6

min

0

max

50

mean

1.652

median

0

std

7.551

q1

0

q3

0

iqr

0

skew

5.932

kurtosis

35.23

n_outliers

5

outlier_rate

0.1087

zero_rate

0.8913

This column holds comma-separated allergen/ingredient trace tags with an `en:` language prefix (e.g. `en:milk,en:nuts`), so each cell is a multi-label set rather than a single category. Across 50 rows there are 23 distinct combinations and entropy ratio 0.87, indicating high diversity, and the most common value is the empty string at 22% (11 rows) — meaning missing-as-empty rather than a true null (null_rate 0.0). The long_tail alert reflects many combinations appearing only once or twice. Treatment: Split on commas and multi-hot encode the individual `en:` tags; treat empty string as missing. high · anthropic:claude-opus-4-7

n

50

nulls

0 (0.0%)

unique

23

top_value

top_rate

0.22

cardinality

23

entropy

3.922

entropy_ratio

0.8671

Finnish-language generic product name field, populated for only 5 of 50 rows (90% null). Among the 5 present values, all are unique with maximum entropy (2.32, ratio 1.0), and casing inconsistencies appear ("Tumma suklaa" vs "tumma suklaa") plus one empty string counted as a value. Treatment: Normalise case and treat empty strings as null; too sparse (90% missing) to use as a feature without imputation or dropping. high · anthropic:claude-opus-4-7

n

50

nulls

45 (90.0%)

unique

5

top_value

Hieno tumma suklaa jossa 90% kaakaota

top_rate

0.2

cardinality

5

entropy

2.322

entropy_ratio

1