Rust library · CLI · Python interface
Evidence you can build on.
Turn candidate SNVs or target intervals into exact read evidence. Reuse saved datasets for new questions, build your own bounded reducers, and verify how each artifact was produced.
Current source / unpublished evidence candidate. Exact evidence, panel QC,
portable datasets and the managed analyzer SDK are implemented and tested in
the development tree. Public stable v0.1.0 predates these capabilities.
Read implementation and release status.
One extraction, several useful outcomes
Review candidatesRead counts, strand support and allele quality for supplied SNVs.
Assess a panelTarget depth and breadth with uncovered bases in the denominator.
Reuse evidenceQuery saved positions without decoding the original alignments.
Add a reducerKeep your statistic; use the managed lifecycle and replay.
Scientific meaning stays separate from scheduling
- Count every eligible observation. Smaller admitted budgets can mean smaller tiles and more indexed I/O. Successful exact-evidence output does not change with budget, tile width or worker count.
- Know what a number means. Filters, sample scope, fields and counting rules are explicit. An unmeasured locus stays distinct from a stored zero-depth row.
- Reuse verifiable artifacts. Portable datasets, Arrow/TSV output, Parquet export and Python/R/SQL adapters let consumers build on saved evidence and its lineage.
Start from the current source
git clone https://github.com/logannye/rosalind.git
cd rosalind
cargo build --locked --bin rosalind
# Rust 1.83+ and native build prerequisites are required.
cd rosalind
cargo build --locked --bin rosalind
# Rust 1.83+ and native build prerequisites are required.
Follow the source quickstart or build prerequisites. Registry downloads and public releases should be evaluated against their own published feature set.
Understand and verify a result
Inspect receipt integrity in your browser Query and replay saved evidence Use Nextflow or Snakemake
The browser verifier inspects the receipt. Full artifact verification needs the recorded files; byte replay reruns the operation with its dependencies. A receipt is tamper-evident, not proof of authorship or biological accuracy.
Research scope. This profile counts aligned read bases, not molecules;
overlapping mates count separately. It is not a clinical interpretation system,
UMI consensus engine or validated production caller. Resource admission and
sampled checks are cooperative: native decoders can allocate before a checkpoint,
including during CRAM validation. Existing Linux cgroup limits provide a separate
OS assurance. Representative performance and independent adoption remain explicit
validation goals.