saturn·

web accessibility data top100

source /tmp/saturn-uploads/f55be675ada5/web_accessibility_data_top100.csv 92 rows 6 columns profiled 2026-05-01 raw JSON static .html .ipynb Report Notebook

Reading

dataset summary · high confidence anthropic:claude-opus-4-7

This dataset profiles 92 popular websites with accessibility metrics, including error counts, error density, and two ranking signals (popularity and wave rank). The error metrics are highly skewed: errors range from 0 to 364 with a median of just 5, and 8 sites (about 10%) qualify as outliers — worth flagging as the worst offenders. The notes field is the richest qualitative signal, with 'Low contrast text' (12 sites) and 'Missing form input label' (9 sites) dominating the issue mix. Popularity_rank is evenly spread across 1–100, so it works well as a control axis when comparing error patterns across the popularity spectrum.

citing: errors · error_density · notes · popularity_rank · wave_rank

Schema

6 columns
Per-column summary. Click column name to jump to its detail.
Alerts
domain categorical 0.0% 92
long_tail
wave_rank numeric 9.8% 83
popularity_rank numeric 9.8% 82
errors numeric 9.8% 36
high_skew outliers
error_density numeric 9.8% 64
high_skew outliers
notes categorical 0.0% 38
long_tail

domain

categorical identifier long_tail
This column holds web domain names (e.g., google.com, youtube.com, wikipedia.org), with one row per domain. Every one of the 92 values is unique (n_unique equals n, entropy_ratio = 1.0, top_rate ≈ 0.0109), so it functions as a row identifier rather than a categorical feature. No nulls are present, but the long_tail alert reflects the fully flat distribution. Treatment: Use as a row key or join key; do not one-hot encode. high · anthropic:claude-opus-4-7
n
92
nulls
0 (0.0%)
unique
92
top_value
google.com
top_rate
0.01087
cardinality
92
entropy
6.524
entropy_ratio
1

wave_rank

numeric feature
A numeric ranking-style field spanning 1,183 to 991,094 with mean 301,136 and median 203,569, suggesting position or score values rather than counts. The distribution is right-skewed (skew 0.96) with a wide IQR of 395,414, but no outliers were flagged. Notably, 9.78% of values are null and only 83 of 92 rows are unique, so there are a few repeated ranks. Treatment: Impute or flag the ~10% nulls and consider a log or rank transform before modelling given the right skew. medium · anthropic:claude-opus-4-7
n
92
nulls
9 (9.8%)
unique
83
min
1,183
max
991,094
mean
3.011e+05
median
203,569
std
2.746e+05
q1
8.221e+04
q3
477,628
iqr
3.954e+05
skew
0.9633
kurtosis
-0.1876
n_outliers
0
outlier_rate
0
zero_rate
0

popularity_rank

numeric feature
Almost certainly a 1-to-100 popularity ranking, with min 1.0, max 100.0, and a near-symmetric distribution (mean 51.23, median 52.0, skew -0.05). The spread is essentially uniform across the range (std 29.39, IQR 49.0, kurtosis -1.19, no outliers), which is consistent with rank data rather than a measured score. Note 9.78% of rows are null and 82 unique values across 92 rows means a handful of ties or repeats. Treatment: Impute or flag the ~10% nulls; treat as ordinal and avoid log transforms given the uniform 1-100 spread. high · anthropic:claude-opus-4-7
n
92
nulls
9 (9.8%)
unique
82
min
1
max
100
mean
51.23
median
52
std
29.39
q1
27.5
q3
76.5
iqr
49
skew
-0.05046
kurtosis
-1.192
n_outliers
0
outlier_rate
0
zero_rate
0

errors

numeric feature high_skew outliers
Numeric counter of errors per record, ranging from 0 to 364 with a median of just 5 but a mean of 27.17 — a classic long-tail count. Distribution is severely right-skewed (skew 3.80, kurtosis 16.12) with 8 outliers (9.6% of values) and a std of 57.20 dwarfing the IQR of 24.5. Note also a 9.78% null rate and 6.02% exact zeros, so missingness and zero-inflation both need handling. Treatment: Impute nulls and apply a log1p transform before modelling to tame the heavy right tail. high · anthropic:claude-opus-4-7
n
92
nulls
9 (9.8%)
unique
36
min
0
max
364
mean
27.17
median
5
std
57.2
q1
2
q3
26.5
iqr
24.5
skew
3.803
kurtosis
16.12
n_outliers
8
outlier_rate
0.09639
zero_rate
0.06024

error_density

numeric feature high_skew outliers
Likely a per-record error density (errors per unit), bounded at 0 with a long right tail: median 0.0057 vs mean 0.0272 and max 0.303, with skew 3.23 and kurtosis 10.95. Roughly 9.6% of values are flagged outliers, 6% are exactly zero, and 9.8% are null, so a small set of high-error records dominates the distribution. Treatment: Log1p- or winsorise before modelling, and impute or flag the ~10% nulls. high · anthropic:claude-opus-4-7
n
92
nulls
9 (9.8%)
unique
64
min
0
max
0.303
mean
0.02725
median
0.0057
std
0.05376
q1
0.0025
q3
0.02545
iqr
0.02295
skew
3.229
kurtosis
10.95
n_outliers
8
outlier_rate
0.09639
zero_rate
0.06024

notes

categorical free_text long_tail
Free-form QA notes describing accessibility issues found on items, dominated by recurring phrases like "Low contrast text" (12/92) and "Missing form input label" (9/92). High entropy ratio (0.89) and 38 unique values across only 92 rows indicate a long tail of compound descriptions (e.g., "Low contrast text, missing alt text, empty link, empty button"). Notable signals include sentinel-like entries "No detected errors" and "No data", plus an off-topic "Asia-based" tag that suggests inconsistent note conventions. Treatment: Split on commas into multi-label issue tags before aggregating; isolate sentinel values like "No data". high · anthropic:claude-opus-4-7
n
92
nulls
0 (0.0%)
unique
38
top_value
Low contrast text
top_rate
0.1304
cardinality
38
entropy
4.663
entropy_ratio
0.8885