v7.4.1

Documentation

• Plainer README wording (8d79d47)
• Break up README walls of text (d55cadc)

Other Changes

• README's see-also lists the graph repos and pangenome tutorials (a3edd8d)
• Point the build badge at the push workflow (83bf666)
• Document the library and CLI output shapes in the README (b83f741)

Tests

• Fail when docs/api.md misses a public member or export (378f09b)

Full changelog: v7.4.0...v7.4.1

v7.4.0

Other Changes

• PairAlignment and lcs are subpath exports, so a bundle can leave the database reader out (ce037e3)

Full changelog: v7.3.0...v7.4.0

v7.3.0

Other Changes

• Export weightedLcs, the heaviest common subsequence of two walks (3f2bb5a)

Full changelog: v7.2.0...v7.3.0

v7.2.0

Other Changes

• Triage agent-docs to the ideas/handoffs convention, add agent-docs/CLAUDE.md (cb83841)
• Export pairAlignments and pairCigar, which align two walks from node lengths alone (ddada98)

Full changelog: v7.1.0...v7.2.0

v7.1.0

Other Changes

• HaplotypeOverview takes an abort signal (0a869b7)
• The keep threshold stays at 32: on HPRC v2.1 its crossover moves with the window (5cefee1)
• Each named walk the cut holds and its span on its own path (486ff07)

Full changelog: v7.0.0...v7.1.0

v7.0.0

Other Changes

• The fuzzer's keep route calls subgraphInInterval with keep (7fa3c2d)
• A hairpin walk two haplotypes share counts once (e972f94)
• SubgraphInInterval rejects a reversed or NaN interval (d233fe5)
• Distinct output no longer depends on whether a haplotype index is attached (0633500)
• Keep stats count the reference among the pieces (df30429)
• Overlapping snarls: count distinct node ids, as upstream does (b25ecee)
• Range-check a shared run's predecessor before looking up its seed (aeda927)
• Chain toward the best chain that matches minMatch, not just the best chain (f4dab73)
• Harden the pager: no unhandled prefetch rejections, no self-evicting prefetch (a691810)
• Retry the path cache after a failure; check the schema version first (043f24a)
• Validate CLI arguments; exit quietly on a closed pipe (4ddf283)
• Hash GFA names without crypto.subtle on insecure pages (2db167b)
• Limit caps the nodes in the subgraph, not per fragment (472729c)
• Ship dual ESM/CJS like the other gmod parsers (e929013)
• GetPathFragments throws on a NaN window instead of returning [] (1a6f4bc)
• The 6.0.1 bug review and its fixes (f6315e9)
• MergeDistinct and alignToRef settle deferred twins instead of merging them for good (e5ef545)
• Queries reject offsets, context, limit and node ids that are not whole numbers (13342e1)
• The second Fable review and its fixes (c3fe56e)
• Budget a walk's prefetch runs together; prefetch only leaves in a scan (1261e4d)
• The CLI takes whole numbers; handoff records the prefetch fixes (0d32289)
• Request study: harness and interim results for HTTP cost per view (d13d509)
• Request study: full sweep, large windows, prefetch on and off (625d740)
• Read haplotype index format 3, and fetch the keep route's rows in parallel with read-ahead (2fef7a2)
• Test that formats 2 and 3 answer alike; a per-request breakdown tool (8154a14)
• Keep route: read the subgraph's samples while the walks run (54503d0)
• Request study: chr22 before and after format 3, with the local Range server and windows (e631f29)
• Read the haplotype index's overview: every haplotype classed per bin at the level the zoom asks for (c05730a)
• The overview query and flag; the study's final chr22 numbers (5390b45)
• Clamp the level, return undefined for a path without one; fixtures rebuilt with the corrected overview (b4585a6)
• Overview renderers: a PNG from Node and a page in the browser, with the requests each view costs (e25017d)
• Point the overview section at the renderers (6eff627)
• Overview renderers: whole-genome costs with the hosted graph (4814a29)
• Overview renderers: costs from local disk with the production index (438c692)
• Validation sweep: open and query through the current API (fcdae2c)
• Read format 3 alone (39fe18b)
• CI fuzz installs gbz-haplotype-index 0.3.0, which writes format 3 (310e350)
• Export the overview's types and class constants from the package entry (4e7498c)
• HPRC tests and request tools read the hosted format 3 index (5663b5e)
• ESLint skips the overview page's build output (7df06f1)
• Prettier on the files that missed it (1392647)

Full changelog: v6.0.1...v7.0.0

v6.0.1

Other Changes

• Close the holes a review of 6.0.0 found (89acd08)
• The data-flow diagram names the 6.0 methods; only interval queries set the limit error's window fields (a6f55b1)

Full changelog: v6.0.0...v6.0.1

v6.0.0

Other Changes

• Released, and the hosted index carries stray rows (f126008)
• Move the indexer crate and the haplotype index docs to GMOD/gbz-haplotype-index (ebf8c19)
• Keep-bounds fixture puts anchors and visits on each walk boundary (d5316b6)
• Every reader boundary mutant is killed but the equivalent one (9b91f25)
• Update README.md (f5530bc)
• Take one options object everywhere, and read the file size from the SQLite header (4e1d1ef)
• The options-object API (e0d6909)

Full changelog: v5.0.0...v6.0.0

v5.0.0

Other Changes

• Genome-wide keep sweep on 4.1.0, and a stray-section table measured (9a9099d)
• Keep route: walk adjacent anchor sections and stray rows when the index lists them (bfa1c05)
• Store stray rows WITHOUT ROWID, route diagram for the stray route, --stats line (12d3b64)
• Stray rows: keep loci within 4 kb of one another as one, and test the 4 kb around each (43f9a45)
• Fixture for a contig with no anchor visit and no sample in the window (1e3012e)
• Stray rows by reference bin, checked against the nodes the haplotype index lists per bin (cfd462f)
• Keep with distinct walks merges after choosing, and keepHaplotypes refuses unnamed or merged walks (2209c00)
• One keep route: a haplotype index without stray rows identifies every walk (72fa932)
• Twins from reverse samples in the window, crate 0.2.0, fixture rebuild script (29798dd)
• Bins, stray rows and the keep route's checks (fc5cfde)
• Run the sampled route once per window and narrow a copy per keep set (8a35d4d)
• Fixes from the fuzzer: a walked stretch ends before the next node, and a reference walk on a reverse handle (9eb3436)
• Export the stray row types, hosted-index test for both index formats, handoff draft (c15666b)
• What the sweeps, the GBZ truth and the fuzzer measured (3be4f8c)
• Indexer README: give graph.gbz.db with the GBZ (861ff44)
• A filled snarl can add nodes outside the bins (0d232f7)
• Keep-all on the targeted windows (7d54634)
• GetSubgraphForRange refuses a window that spans several fragments of the path (e7752ca)
• The index to host, the 32 kb-anchor build, and what the plugin session did (947cb6c)
• HPRC tests default to the hosted index; AMY1 expects the 1,362 records that joined detours give; README: bump the stray format tag with the rule (fa083e6)
• What this session verified on ada, the paper edits, and the audit left running (e400b78)
• Keep route: read the anchor past one that starts exactly bound outside the bins (4c7f8f8)
• --edge-windows aims windows at the stray bound around each anchor (05dc15e)
• Handoff and docs: the audit's finding, its fuzz runs on the fixed code, and what remains (47e6500)
• The audit paragraph without the colon (40d45de)
• The longest HPRC node is 1,024 bp, not 342; the build's peak memory in the doc's own units; the large stratum ran two settings (55936ca)
• Drop a walk's twin once identification shows it is the same stretch read the other way (d8e36f2)
• The evening's cross-validation, the twin fix, the plugin and BandageJS state, the chains on ada (0d9894a)
• List walks in upstream's order, by start handle then offset (b1f737c)
• What the GBWT and gbz-base each name, and the terms upstream uses (aafb2bc)
• The review against upstream and the chain on the final main (eb7ba8c)
• Fuzz against GFA truth on every push, so the indexer's stray-row rule and the reader's walks cannot drift apart unnoticed (a6a46b2)
• The CI fuzz job and the plugin's stray-row fixtures (c187528)
• The walk-rows gap run, the 4.0.29 release, BandageJS already fixed on its main (77afc46)
• Build every database with chains from a vg distance index, and count the keep queries that fill a snarl (3ff9d6c)
• Snarl fills under fuzz, with chains in every database (255d553)
• A large scale, 150-400 kb references with 16-40 haplotypes and six sites of each structural kind (0820dfe)
• The large fuzz scale and its two batches (c752b0e)
• Chains 3 and 4 complete and identical, chain5 under way, two more large fuzz batches (059b7a3)
• Handoff closed: chain5 matched on every pass; docs record the rerun and the snarl-fill and large fuzz batches (20141e5)
• Proof that the keep route finds every piece, from the code on both sides (10f23ab)
• Figures for the keep-route proof, the walks that reach a bin and a filled snarl (0ddb077)
• Keep-route proof states the assumptions an adversarial review found unstated (f83696e)
• The low anchor loop never moves, so mutating its comparison changes nothing (fcd347c)

Full changelog: v4.1.0...v5.0.0

v4.1.0

Other Changes

• Write the bin path as npm normalizes it (9e0e975)
• --reference-interval samples the reference paths more densely (4e9eec4)
• Test the far-sample fallback on a reference stretch the context reaches (485f9da)
• --reference-interval, and what puts a sample far from its anchors (00e98c3)
• The keep route walks as far as its sample check allows, and falls back on two copies (09c5bd5)
• Walk a chosen path's extra anchor visit near its pairs, and correct the keep docs (665e4bb)
• Keep-route omission modes, what is fixed, and the plan to close the rest (3c33200)
• Alignment records join a walk that leaves the subgraph and comes back (4aed986)
• Joined records count aligned bases, and a detour piece joins between two ends (61f3125)

Full changelog: v4.0.0...v4.1.0