Changelog

Install or upgrade with python -m pip install --upgrade osteosarc, and check your version with osteosarc --version. Pin both the package version and your snapshot (by name or download date) for reproducible analyses. Full release notes are on GitHub.

Unreleased

  • Link FASTQs from published filename prefixes and the Tempus tumor/normal row aliases, count only raw-read files, and generate working download commands. FASTQ library metadata supplies assay/provider filters even for shared folders such as CeGaT RNA delivered inside a WES directory. Refresh the sample guide against all 31 samples in the October 7 metadata. Data-page provider notes and mixed DNA/RNA analysis directories no longer introduce false assay or provider conflicts.
  • Supplementary-depth probes stream record counts and cache small receipts instead of entire BAM extracts. Existing verified probes and overflow results are reused; the depth cap and selected fixture records are unchanged. Reused BAM counts are checked against their original records, including compact acquisition receipts.
  • Recognize the reviewed 2026-10-05 upstream allele, ID, specimen and timeline corrections while preserving historical snapshots and future drift checks. A public CeGaT normal-DNA call supplies OTUD4's newly published literal allele; the original Natera report and somatic status remain unverified.

0.15.4

  • Bounded extraction reads BAM pipes without constructing a pysam AlignmentFile on a nonseekable stream (#106). A timeout partway through the header retains its timeout and stderr without a secondary Illegal seek destructor error. Original header and record blocks are copied without changing stored values; overflow, truncation and producer failure still prevent cache publication.

0.15.3

  • Stream ordinary seed context during mate/SA recovery instead of retaining every alignment and unrelated query name in memory (#109). Every original seed record, duplicate and diagnostic survives. Unchanged recovered BAMs share the seed cache file; changed outputs are checked against the full record multiset.
  • Partner queries select pointer-bearing seed names in every round, retaining candidates for later recursive links. The changed query scope uses recovery request schema 3; prior cache entries remain intact.

0.15.2

  • Store repeated query-name filters as verified shared assets and pin upstream acquisition receipts without copying their full contents (#107).

0.15.1

  • Optional bounded batches for mate/SA recovery preserve available partners after another query times out. Separate partner-query count and timeout limits retain explicit missing/truncated outcomes (#104). Default acquisition requests and cache identities are unchanged.

0.15.0 (2026-09-29)

  • subset_bundle and test-data make --from carve out members offline while preserving original source headers, acquisition receipts and selection results (#90).
  • New sv-regressions-v2 panel supplies the three Isovar fusions' retained sides; newly requested fusion bundles use it. Published v1 panels and bundles stay unchanged (#96).
  • Historical retrieval cases have verified native coordinates, exact BAM contig names, mitochondrial reference identity and a guard against incompatible merges (#99).
  • Reviewed Tempus evidence overlap is available on files and file selections, with explicit preference required to exclude one product. Both original labels and acquisition receipts remain intact; the true sample identity is still unresolved (#100).
  • Python 3.14 support: the tests run on Python 3.9 to 3.14.
  • Comprehensive vaccine/RNA test corpus: 51 vaccine-associated variants from the union of membership sources, a complete per-BAM support matrix, four DNA-supported SNVs with zero reported RNA alternate reads, a historical RNA disagreement case, frameshifts and SV evidence. Its openvax-v3 read specification is an unpublished candidate.
  • Shared read builds can checkpoint completed source selections and set their supplementary-alignment depth and region limits explicitly in the spec.
  • Compact carry-forward preserves query positions for placed-unmapped records. Comprehensive control audits verify source URLs and BAM/index checksums against acquisition receipts; packaged RNA-support claims follow the actual recounts.

0.14.4 (2026-09-28)

openvax-v2, the next bundle of reads the OpenVax libraries share (#71, #88). See shared test data.

  • It has openvax-v1's 993 members, the libraries' 429 record for record, and leaves out mates with no position of their own (116 records, none of them the libraries'), so it rebuilds without reading any BAM's unplaced reads.
  • The reads it keeps bring their split alignments: 14 more records that their SA tags name. An alignment in a pileup, with more than 10,000 records at its first base, is left out.
  • SV targets say which side of each breakend they keep, where that's known, and any sequence inserted at the junction. openvax-v1's recipe dropped the sides the SV panel and catalogue give; ATP5MG--KMT2A, FOXO3--STRADA-CCDC47 and TPST1--CRCP now give theirs, and DLG5's deletion is at the junction esvee resolved, with its 24-base insertion.
  • Carrying a bundle's library records forward keeps a source's mates with no position only if one of those records lacks a position.
  • Reading past max_records, or a recovery policy's, raises RecordLimitError (an IntegrityError). For regional reads the cache remembers it: asking again, even offline, gives the same answer without reading.

openvax-v1 stays published, and the libraries move to openvax-v2 at their own pace.

0.14.3 (2026-09-27)

Anyone can rebuild a bundle from its recipe (#89).

  • A recipe's BAMs can be read through any snapshot that lists each one unchanged (the same URL, key, size and modification time as the recipe pins), not only through the snapshot it was made from, which only its builder had. osteosarc test-data make --recipe uses --snapshot if you give one, else the recipe's own snapshot if you have it, else your newest.

0.14.2 (2026-09-27)

A tidier command line, from running every command and reading what it printed, and one place that decides where the OpenVax libraries' shared cache is.

  • The shared cache is datacache's. osteosarc finds its cache with datacache.get_cache_root("openvax", "OSTEOSARC_CACHE", "OPENVAX_DATA_CACHE"), as the other OpenVax libraries can, so test data they share is downloaded once. It needs datacache 1.12.0, which no longer imports pandas.
  • Regions in a BAM's own genome build. A Region's assembly can be None, meaning the build of the BAM it's read from, and osteosarc reads FILE REGION uses that when --assembly is left out (a sample, whose BAMs may differ, still needs it).
  • osteosarc test-data on its own shows its actions and examples.
  • Tables.
  • A cut cell ends with "…" instead of mid-word or mid-number.
  • The line under the headers sits under its columns and stops at the terminal's edge.
  • Command-line lists leave out columns empty in every row.
  • When even that is too wide, osteosarc variants leaves out its vaccines, found_by and corrections columns, and says so.
  • Wrapping. A sample's notes and corrections wrap without splitting a word, and so do the one-week timeline's entries, which follow --width. osteosarc files shows each kind's two commonest formats, counts whole.
  • osteosarc downloads lists each file's size and key, then its whole path. It no longer lists snapshots' metadata or pages of the bucket's listing.
  • --min-mapq and --exclude-flags say what they do; "1 record" and "1 slide" are singular.

0.14.1 (2026-09-27)

Downloads survive a passing server error (#86).

  • The checks of a remote object's identity, made before and after a download and when reading reads from a remote BAM, are tried again after a server error (500, 502, 503, 504), too many requests (429) or a dropped connection: up to five tries, about 1, 2, 4 and 8 seconds apart, or as long as the server asks, up to 30 s. A timeout, a TLS error or a refusal (other 4xx) still fails at once. The download itself was already tried again by datacache.
  • A download pinned by checksum, such as a published bundle, no longer fails when those checks do: the checksum vouches for the bytes.

0.14.0 (2026-09-27)

Fast test data with no options (#84).

  • Mates with no position are left out by default. make_bundle, osteosarc test-data make and osteosarc reads --fetch-pairs leave out mates that didn't align and have no position of their own, as STAR stores them. Finding those means reading every unplaced read of each BAM, most of the time an RNA-seq BAM takes: test-data make for one variant from T2_tumor's three RNA-seq BAMs took 15 s instead of 91 s, and for another 12 s instead of 19 minutes. make_bundle(..., unplaced_mates=True), or --unplaced-mates on either command, keeps them (it replaces 0.13.2's --placed-mates). In Python, extract_reads(fetch_pairs=True) still fetches every mate, so existing recipes, such as openvax-v1's, rebuild the same. Bundles made with 0.13 read their BAMs again the first time they're made with 0.14 (their extractions are cached under the old setting), unless given --unplaced-mates.
  • Removed osteosarc.legacy_fixtures (#72), which no OpenVax library imports any more. osteosarc.cohort_bundle stays.

0.13.3 (2026-09-26)

Faster bundle builds (#82). Rebuilding openvax-v1 from cached reads takes 2.7 minutes and 1.4 GB instead of 31 minutes and 21 GB, and gives the same bundle, byte for byte.

  • Selection reads only what it can choose from. For each BAM, osteosarc first reads every target's window from the extract's index, then keeps the templates found there, whole, and any a library names or pins. Before, it held all of the T2 tumor RNA-seq extract's 13.5 million records.
  • Each source is counted once. Making a bundle copied and counted every record of a source once for each of its members; it now counts them once, and keeps only the records the members pin (it still reads the whole extract to find them).
  • Record checksums are about twice as fast.
  • select_fixtures(...).records holds, for a source whose members all pin exact records, only those records.

0.13.2 (2026-09-26)

Faster read extraction (#80).

  • Leave out mates with no position. extract_reads(..., fetch_pairs=True, unplaced_mates=False), or osteosarc reads --placed-mates, gives what fetch_pairs=True does less the mates that didn't align and have no position of their own, as STAR stores them: finding those means reading every such read in the BAM. Six extractions from BG003082's RNA-seq BAM took 40 s instead of 68 s, the same reads less 28 unplaced mates. Where no read has such a mate it's a second or two slower. Bundles can leave them out too: make_bundle(..., unplaced_mates=False) or osteosarc test-data make --placed-mates. By default bundles keep every mate, as before, and openvax-v1 rebuilds the same.
  • osteosarc reads SAMPLE reads a sample's BAMs four at a time, and the first error stops those not yet begun.
  • An extraction's time limit covers all its samtools runs, however many.

0.13.1 (2026-09-26)

Faster commands (#78): three to eight times faster once a snapshot has been used. On the current snapshot, osteosarc reads on a cached result goes from 7.9 s to 2.8 s, files from 6.1 s to 1.6 s, samples T1_tumor from 7.5 s to 1.8 s, variants from 1.8 s to 0.2 s and --version from 0.5 s to 0.2 s.

  • A snapshot's catalogues are built once. The file catalogue (some 400,000 bucket objects), its header and the variant catalogue are saved in the cache the first time, for that snapshot, osteosarc version and set of corrections. Loading one warns of the stale corrections to its sources, as building it does. Each user keeps their own copies, in a folder only they can write in, and a copy can hold only osteosarc's catalogue classes. Each kind keeps its three most recently used copies, and copies unused for 30 days are removed when another is saved.
  • Looking files up by key, URL or sample, and looping over any collection, no longer goes through every file in Python.
  • Corrections find their records through an index instead of scanning every record, and variants and the timeline list exactly the corrections that touched them.
  • osteosarc imports pandas only to download a file. Checksums of local BAMs are remembered in records each user keeps for themselves; those from 0.13.0 (in the cache's digests folder) are no longer used and can be deleted. A read-only cache still works.

0.13.0 (2026-09-26)

A simpler site, and less code (#75).

  • The home page is the overview. Concepts is folded into a short "How the data fits together" there, linking to each topic's page.
  • Removed, now that the OpenVax libraries take their reads from openvax-v1:
  • osteosarc.regional_corpus;
  • the generators in legacy_fixtures (generate, pack, verify, export_fixture and their helpers) and cohort_bundle (generate_cohort_bundle, retrieval_regions); the functions Isovar and Vaxrank import stay;
  • osteosarc.bundles.generate_panel; osteosarc test-data make --recipe, or generate_bundle, does the same;
  • osteosarc.records.sam_digest, and SAM-text record identities in recipes;
  • osteosarc.reads.subset_templates, a random sample that make_bundle's balanced sets replace;
  • reading bundles made by osteosarc 0.2 or earlier, which now says to make them again.
  • The frozen SV test recipe moves from the package into tests/data.

0.12.0 (2026-09-26)

Test data in a call or two (#73).

  • Make a bundle without a recipe. osteosarc test-data make DIR SAMPLE|FILE --variant ID --sv ID and data.make_bundle(dir, variants=..., files=...) write a verified bundle with openvax-v1's selection: a balanced set of templates at each variant in each BAM, and templates joining each SV's breakends, every record pinned. Files can be BAMs or samples. They say which BAMs they skip and why, and stop, before streaming any reads, if a requested variant or SV can't be read from the BAMs given. test-data make DIR --recipe FILE replaces test-data generate, and fetches the recipe's sources from the snapshot it names, so a bundle's own recipe.json rebuilds it.
  • Use a member in one call. osteosarc.bundle_file(bundle, member) returns it as a local indexed BAM (or SAM), exported once into the cache with the other members from its source BAM, read-only, and reused offline: the helper each OpenVax library had written for itself. An unknown member's error suggests close names.
  • One rule for bundles. list_bundle, export_bundle, verify_bundle, check_fixtures and bundle_file all take a bundle folder or a published bundle's bare name, such as openvax-v1; a Path, or text with a slash, is always a folder. Offline, a bundle that isn't cached yet says how to get it.
  • The package docstring, the data overview, osteosarc --help and the docs lead with these.

0.11.3 (2026-09-26)

Follow-ups from moving Isovar, Topiary, Vaxrank and Varcode onto openvax-v1, which all four now use.

  • Receipts and bundles hold no local paths (#65). Read receipts record a file by its place in the cache (objects/sha256/..., named by checksum) or else by its name, beside its checksum, and bundles leave out download times. So the same reads, extracted with the same tools from any cache, make the same bundle, and Isovar's and Vaxrank's recorded receipts lose their local paths too. Cache keys no longer depend on where the cache is, so reads extracted by earlier versions are fetched once more. openvax-v1 keeps its published bytes; the next version will be clean.
  • fetch_bundle(name, offline=True) never downloads, and raises OfflineError if the bundle isn't cached (#66).
  • osteosarc test-data check --help says fixture paths are relative to the fixtures file (#67).
  • check_fixtures is importable from osteosarc and takes a published bundle's name, and the Python API page shows it and osteosarc.shared.published (#68).
  • Export keeps the name of a member already named after its file: topiary/reads.bam is written as reads.bam, not reads.bam.bam; two members that would share a file are an error (#69).
  • The cross-library builder check is gone, with the builders it compared.

0.11.2 (2026-09-26)

  • Rebuilding shared test data no longer needs the libraries' own copies (#63). The build script's --carry openvax-v1 keeps each library's members from the previous bundle, pinned by checksum, so the next version builds after a library deletes its test files; a --required list replaces all a library had. Fixture targets now record the regions they were planned from, so later carries plan them the same way.
  • osteosarc test-data treats a published bundle's name as that bundle, even when a folder of the same name exists; write ./NAME for the folder.
  • The required_*.py scripts take the library revision to read; openvax-v1 means the commits openvax-v1 was built from, and reproduces its inputs exactly. A rebuild never replaces a published release record.
  • The cross-library builder check compares whichever libraries you give it.

0.11.1 (2026-09-26)

  • osteosarc test-data check counts every SAM line under a JSON pointer, whatever fields hold it, so records kept as sam and partner_sam (Isovar's fusion corpus) no longer look half missing, and names or versions beside them aren't counted (#61). Pointers follow the JSON Pointer standard, a pointer to nothing says where it stopped, and a fixture that can't be read is reported without stopping the check of the others.

0.11.0 (2026-09-26)

The OpenVax libraries can take their test reads from one place: openvax-v1, chosen once here (#56). See shared test data.

  • What it holds: reads at every variant on the site plus other alleles the libraries test, reads joining the breakends of 7 RNA fusions and 8 DNA SVs, and every record in Isovar's, Topiary's, Vaxrank's and Varcode's current test files, exactly as they are.
  • Using it: osteosarc test-data list openvax-v1 downloads it the first time, and export and verify take its name too; check compares a library's own copies with it. In Python, fetch_bundle("openvax-v1") returns its folder.
  • How reads are chosen: a new allele classifier sorts each read at a variant into alt, ref, other or uncallable, even for indels in repeats, and openvax-v1 keeps up to 20 alt, 10 ref, 5 other and 2 uncallable templates per variant and BAM, whole.
  • Export writes just the members. osteosarc test-data export BUNDLE DIR --member NAME writes NAME.bam, with its index, into DIR, which may already exist; it no longer copies the whole bundle, and never replaces a different file. Manifests no longer record exports.
  • Recipes: an exact member can give each record its own reason, and a target can be one of a library's test files.
  • Docs: the example file names for reads saved with --to now match what it writes.

0.10.0 (2026-09-26)

A smaller, more consistent osteosarc, centered on exploring the data and making test data (#58).

  • Test data in one command. osteosarc reads T1_tumor --assay rna-seq --variant ID --padding 100 --to tests/data streams just those reads from each of a sample's BAMs, and saves each as a small indexed BAM named for its source and variant. data.extract_reads(..., to="tests/data") does the same in Python.
  • Command line. osteosarc timeline --around DATE replaces osteosarc on, and osteosarc test-data (generate, list, verify, export) replaces fixtures; its select, pack and panel actions are gone (use the Python functions). table and discover are gone.
  • No silent overwrites. Saving a download or reads into a folder never replaces a file with different contents; it stops and says so.
  • Python.
  • Dataset loses table (use parse), claims, timepoints, annotations, vaccine_names, pipeline_names, receipts, open_variants, and the generate_bundle and select_fixtures shortcuts (use the functions).
  • The package's top-level names are the core API. Helpers such as list_bucket, read_records and SNAPSHOT_SOURCES now come from their modules.
  • Names. The SV interest catalogue is now SV candidates: load_sv_candidates() and the sv-candidates-v1 panel. FASTQ folder tables name assays and platforms as everywhere else, with a library column (gene expression, TCR, BCR, antibody tags, long reads).
  • Docs.
  • Page addresses match their titles (command-line, python-api, snapshots, corrections, testing, map2, sv-candidates, test-data, openvax).
  • The three pages about moving code onto osteosarc are merged into two.
  • Every page has much less code formatting.

0.9.0 (2026-09-26)

The command line and the Python API now use the same two nouns, samples and files, and every command prints something a person can read (#46–#55).

Samples. data.samples is the list of samples (tumor, organoid and blood), each a Sample with its sequencing, BAMs and FASTQ folders. data.samples["T1_tumor"] and osteosarc samples T1_tumor show one sample's files with their sizes and whether they're downloaded, followed by the commands that fetch them; osteosarc samples --files lists every sample's files. Sequencing now also comes from the site's FASTQ table, so the four 2026 blood draws, which the registry leaves blank, show single-cell and CITE-seq instead of "unknown".

Files. data.assets, data.asset(), Asset and Assets are now data.files, data.file(), File and Files, and osteosarc assets is osteosarc files. With no filters it summarizes the bucket by kind and folder; lists put BAMs first and name each file's sample. files.select(sample=...) and file.samples link files to samples, and scans and slides are a new image kind.

Downloads. osteosarc downloads and data.downloads() list what's on this computer, with local paths, and data.local_path(file) finds one file's copy without the network. download(file, to=DIR) and osteosarc download FILE --to DIR put a file, and its index, in a folder under its own name.

Timeline. The chart has a row per treatment, named and grouped by class, over a month axis, with rows for time points, samples, procedures, scans and each MRD assay. Lab draws, DICOM studies and other frequent records are left out unless you pass --all (everything=True), and the legend says how many.

Command line.

  • Commands are grouped (browse, get data, snapshots, more) in osteosarc and osteosarc --help, and osteosarc on its own says which snapshot it's using.
  • variants, vaccines, corrections, sync, table and discover print text; --json gives the old output. variants ID and corrections ID show one record.
  • curation is now corrections. specimens and timepoints are gone: use samples.
  • download and reads print plain paths (reads --json for the receipt).
  • osteosarc repl replaces the explore mini-shell: it opens Python with the snapshot loaded as data.

Python. repr(data) lists what's there and how to get it; collections end with how to narrow them. The old data.samples (what each source says about a file) is now data.claims, with file_ids. data.specimens, data.describe_samples(), data.assets_for_sample() and data.explore() are gone.

0.8.0 (2026-09-25)

A first look at the data is easier from both the command line and Python (#44):

  • osteosarc on its own suggests where to start.
  • Without a snapshot, commands say which cache they looked in and to run osteosarc sync. In a terminal, osteosarc explore offers to download one.
  • In Python, data, variants, files, tables and the timeline show readable previews instead of <object at 0x…>. Text views such as describe_samples() display without quotes, and single variants, files and events no longer print their full source records.
  • data.summary() shows what's in a snapshot and what to try next; data.explore() opens the interactive explorer.
  • The README and docs home start with "Explore the data".

0.7.0 (2026-09-24)

Every correction was re-checked against the reference genome and the site's data. None only rewrites an equivalent form of a value, and these fixes follow from the review:

  • The Tempus tumor's variants match T0, so the T1 labels on the Tempus files are probably what's wrong. tempus-timepoint now says so, and a new flag, tempus-file-labels, marks the files.
  • provider-IPISRC044-T1-rna also fixes the site's read-count rows, which switched to UCLA on 2026-09-21.
  • transcript-DCHS2 fixes only the typo, to the versionless NM_001142552 the site now uses; it no longer adds a version.
  • New: transcript-COL4A2 and transcript-GTF3C5 fix two more accession typos.
  • MAP2 keeps its published read counts as an approximation: the site counts any large deletion there, which in the T0 tumor exome is always the real change.
  • USH2A's retired entry is recorded as a confirmed duplicate, and the kept entry's sequence context is fixed. The moved Tempus alleles lose sequence context taken from the wrong position.
  • fam157a-withdrawn-protein shows as fixed upstream where the site has the same note.
  • The corrections page is rewritten in plain language.

0.6.1 (2026-09-24)

  • The documentation is rewritten in plain language. Key concepts now says where every piece of data comes from: osteosarc.com, its S3 bucket, and the website's source repository on GitLab (#41).

0.6.0 (2026-09-24)

  • RecoveryPolicy(on_timeout="incomplete") keeps the verified seed reads and partners already matched when a later partner query times out, instead of failing. The receipt says status="incomplete" and names the failed query; a later call retries it and resumes from verified extractions. Fixture recipes can request this through a source's acquisition, and bundles record the incomplete acquisition (#28).

0.5.0 (2026-09-24)

  • The CLI is easier to explore with (#38):
  • samples and specimens show sequencing with the filter names, such as rna-seq; wes; wgs; scrna-seq (ont, pacbio).
  • samples --assay scrna-seq --platform ont lists the samples with that sequencing; describe_samples() takes the same filters.
  • assets prints a table with complete file keys and a hint when rows are cut off; --json prints the full records.
  • Explorer tables never truncate file keys, and sizes read as KB, MB or GB.
  • An assay, platform or tissue filter that no file uses raises an error listing the valid values, instead of an empty selection. A registry label such as scRNA_ONT names the filters to use instead.

0.4.0 (2026-09-24)

  • Removed the CLI's positional snapshot argument (osteosarc samples baseline); choose a snapshot with --snapshot. Removed the empty [reads] install extra.
  • The explorer's assets command accepts sample=, as in assets sample=T1_tumor.

0.3.0 (2026-09-24)

  • Snapshots are indexed by download date, and the most recent is the default (#35). Dataset.sync() and osteosarc sync need no name: they save a snapshot named by the UTC date and reopen it for the rest of that day. Dataset.open() and every CLI command use the newest snapshot. Dataset.snapshots() and osteosarc snapshots list them. Dataset.open(date="2026-09") or --snapshot 2026-09 picks the newest from a download date, month or year, and Dataset.open(name) or --snapshot NAME picks one exactly by name or ID prefix. Offline, sync() builds a snapshot from sources already in the cache.
  • Named snapshots keep working. The CLI's leading snapshot argument (osteosarc samples baseline) is deprecated in favor of --snapshot.
  • The documentation no longer uses a baseline snapshot name.

0.2.6 (2026-09-24)

  • Reorganized documentation with key concepts, a command-line guide, a complete API reference and this changelog. The README is a shorter landing page that lists the main features (#32).
  • The documentation example checker covers every published page and the strict build validates anchors (#31).

0.2.5 (2026-09-24)

  • The SV interest catalogue: 637 structural-variant nominations with original calls, breakend geometry, Ensembl 115 annotation, T2 expression and scoped RNA evidence. It loads offline with load_sv_interest() or load_panel("sv-interest-v1") (#29).
  • The command line can now run the whole quickstart: assets --sample, reads --variant ID --padding N and --version (#33).

0.2.4 (2026-09-23)

  • Partner recovery selects seed query names with samtools view -N before applying its record cap, so unrelated reads at dense loci can't exhaust it. Adds ReadFilter(query_names=...) (#26).

0.2.3 (2026-09-23)

  • Shared historical fixture adapters for Isovar, Topiary and Vaxrank, and the additional SV research panel. See OpenVax fixture adoption (#15).
  • Python 3.9 support.

0.2.2 (2026-09-23)

  • Portable fixture bundles: generate, pack, verify, list and export (#14).

0.2.1 (2026-09-23)

  • Bounded mate and SA partner recovery with explicit receipts (#13).

0.2.0 (2026-09-23)

  • Versioned fixture recipes and one selection executor shared by the Python API, Dataset and CLI (#12).

0.1.4 (2026-09-22)

  • Reconciled eight corrections after upstream source changes; historical snapshots keep their original IDs and corrections (#10).

0.1.3 (2026-09-21)

  • to_varcode() converts chrM and M to Ensembl's MT and keeps the original contig name (#8).

0.1.2 (2026-09-21)

  • Verified GRCh38 alleles for FAM157A and COL3A1, with explicit outcomes for MUC3A, OTUD4 and USH2A (#5).
  • Clearer sample, timepoint and assay documentation.

0.1.1 (2026-09-20)

  • Downloads use datacache; pysam is a standard dependency.
  • Adds describe_samples(), assets_for_sample() and extract_reads(variants=...).
  • SAMtools capabilities are checked before acquisition (#3); malformed count rows no longer hide valid entries (#4).

0.1.0 (2026-09-19)

  • First release: pinned metadata snapshots, file discovery, variants and vaccines, indexed read extraction, source corrections, the clinical timeline and the interactive explorer.