We release CratesBump, the first public dataset for predicting version bumps in the Rust ecosystem, together with semver-extract, the tool that builds it. CratesBump is built from every crate on crates.io with at least two releases before 14 September 2026 (about 220,000 crates), giving over two million consecutive release pairs. Each pair has 41 features (code changes from AST diffs, dependency changes, complexity, release cadence, GitHub commit and issue activity), its source repository where listed, and two labels: the literal reading of MAJOR.MINOR.PATCH, and Cargo's rule, under which a 0.y bump is a breaking change. semver-extract regenerates the dataset from public crates.io data and the GitHub API. To illustrate its use we (i) benchmark bump predictors on held-out repositories, finding the task learnable but far from solved, with lower scores on later releases; (ii) measure which feature groups carry the signal; (iii) check, against declared dependency ranges, how often Cargo's rule shields dependents from pre-1.0 breaking releases; and (iv) audit a sample of releases with cargo-semver-checks, finding minor and patch releases that break the public API despite their version number. Dataset and tool are released under open licenses.
Figures & tables
Fig. 1: How CratesBump is built (1–4) and used in Section V (5).
Group
Cols
Type
What the columns hold
Pair key
3
text
package , prev_version , version
Repository
1
text
repository , e.g. github.com/serde-rs/serde
Labels
2
label
major, minor or patch, by Cargo’s rule and by the literal reading
Publish time
1
time
when version was published (UTC)
Code (AST)
20
count
files, functions, blocks added/deleted
Dependencies
4
count
dependencies added/removed/updated
TABLE I: The 48 columns of CratesBump, one row per release pair ( prev_version → version ). The 41 feature columns (lower part) describe what changed between the two releases.
Version change: literal → Cargo label
Pairs
Share
0.y.z → 0.y.(z+1) : patch → minor
1,033,349
60.3 %
0.y.z → 0.(y+1).0 : minor → major
555,772
32.5 %
0.0.z → 0.0.(z+1) : patch → major
109,239
6.4 %
0.0.z → 0.y.z : minor → major
14,225
0.8 %
TABLE II: Pairs on which the literal and Cargo labels disagree, by kind of version change: all pre-1.0, 1,712,585 pairs (82.3%).
Model
Macro-F1
MCC
Always “minor” (baseline)
0.236 ± 0.007
0.000 ± 0.000
Random by label share (baseline)
0.333 ± 0.001
0.000 ± 0.001
Random Forest
0.448 ± 0.007
0.216 ± 0.012
LightGBM
0.447 ± 0.007
0.223 ± 0.012
XGBoost
0.457 ± 0.004
0.239 ± 0.011
Hist Gradient Boosting
0.459 ± 0.005
0.241 ± 0.011
TABLE III: Predicting the Cargo bump label (major, minor or patch) of pairs from repositories not seen in training: mean ± s.d. over five repository splits. Macro-F1 averages the three labels’ F1; MCC is 0 for random guessing and 1 for perfect prediction. The first two rows ignore the features; the best score is in bold.