fix(metrics): align TSV metric output with fgbio Metric format (#498) by nh13 · Pull Request #504 · fulcrumgenomics/fgumi

nh13 · 2026-07-09T00:54:52Z

Closes #498 (metrics-format items) — makes fgumi's metrics .txt outputs readable by fgbio's Metric.read and matches fgbio's rows, column names/order, and value tokens on a shared input.

What changed

#	Item	Fix
MET-02	Non-finite floats unreadable	New shared `float` serde module writes f64 fields as fgbio's `Infinity`/`-Infinity`/`NaN` tokens (integral values without a trailing `.0`); applied to every serialized f64 field across `correct`/`shared`/`consensus`/`simplex`/`duplex`/`group`. Previously only `correct.rs` was inf-safe; the rest emitted ryu `inf`/`-inf`, which fgbio's `Double.parseDouble` cannot read.
MET-03	Empty metrics file has no header	`write_metrics` now always emits the header, even for a zero-row slice; the duplex-metrics / simplex-metrics per-family/UMI/yield writes route through it.
MET-04	`UmiGroupingMetrics` wrong columns	Serializes exactly fgbio's 5-column `UmiGroupingMetric` (`accepted_sam_records`, `discarded_non_pf`, `discarded_poor_alignment`, `discarded_ns_in_umi`, `discarded_umis_to_short`). The fgumi-internal summary fields stay in memory for logging but are `#[serde(skip)]`-ed (derivable; the family distribution already lives in `.family_sizes.txt`).
—	Duplex `family_sizes` dense→sparse	Emits one row per observed family size (matching fgbio `familySizeMetrics`) instead of a dense `1..=max` range.
—	`duplex_umi_counts` nondeterministic	Adds `umi` as a deterministic secondary sort key so ties no longer fall back to HashMap iteration order.
MET-05	Consensus KV rejection rows	Partial: matches fgbio's exact rejection description text ("Reads has a different, and minority, set of indels"; "Only Generating One Strand of Duplex Consensus"). See "Deferred" below.
MET-01	Float representation (accepted, not a bug)	Doc-only: corrects the `format_float` doc-comment that falsely claimed byte-format parity with fgbio. Numeric equality is the contract.

CorrectUmis zero-count seeding (issue item 5) already reproduces as fixed on main — both correct execution paths pre-seed a row per expected UMI plus the all-N sentinel — so no change was needed there.

fgbio-oracle evidence

reports/fgbio-parity-fixtures/met-02-03-04-fgbio-read.sh feeds fgumi's exact output bytes (proven by the unit tests) to fgbio 4.1.0's real Metric.read:

[PASS] MET-04 read UmiGroupingMetric (populated) -> rows=1 accepted_sam_records=90 discarded_umis_to_short=1
[PASS] MET-03 read UmiGroupingMetric (header-only empty) -> rows=0 (header-only parsed cleanly)
[PASS] MET-02 read UmiCorrectionMetrics with representation=Infinity -> representation.isPosInfinity=true fraction=0.5
[FAIL] MET-02 negative-control read representation='inf' (expect FAIL) -> MetricBuilderException

The negative control confirms the old ryu inf token was genuinely unreadable by fgbio.

Deferred (out of scope here)

MET-05's caller-type-aware row-set seeding — always emitting raw_reads_rejected_for_non_paired_reads, ..._potential_umi_collision, and the codec-specific reasons per fgbio's usedByVanilla/usedByDuplex/usedByCodec model — is deferred to the consensus PR. It requires adding RejectionReason variants and wiring them into the duplex/codec callers (this is the same behavioral change as R2-UCC-01, "fgbio rejects fragments as NonPairedReads"), which belongs with the consensus work rather than a metrics-format PR. Only the caller-independent description-text parity is done here.

Testing

New unit tests: inf/empty-header round-trips, exact 5-column UmiGroupingMetric header, sparse family sizes across a gap, deterministic duplex-UMI tie order, rejection-text parity.
cargo ci-fmt / cargo ci-lint clean; full suite 2215 passed.

Summary by CodeRabbit

New Features
- fgbio-compatible float serialization for metric TSVs now applies across exported metrics, including correct round-tripping of NaN, Infinity, and -Infinity.
- Empty metric distributions now write header-only TSVs (no data rows).
Bug Fixes
- Family-size histograms now emit only observed sizes (no unobserved zero rows).
- Duplex UMI metrics use deterministic tie-breaking when values are equal.
- Rejection-reason text output and grouping-metrics TSV columns align with fgbio.
Tests
- Updated and extended TSV round-trip, exact header/schema, and non-finite float assertions.

coderabbitai · 2026-07-09T00:54:59Z

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

@coderabbitai resume to resume automatic reviews.
@coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

▶️ Resume reviews
🔍 Trigger review

Walkthrough

Metric TSV output now uses shared f64 serialization, matches fgbio’s grouping schema, preserves headers for empty outputs, emits sparse duplex family sizes, and orders tied duplex UMI counts deterministically. Rejection descriptions match fgbio wording.

Changes

Metrics output compatibility

Layer / File(s)	Summary
Shared float serialization `crates/fgumi-metrics/src/{float,lib,correct,consensus,shared,simplex,duplex,group}.rs`	Adds shared finite and non-finite f64 TSV handling and applies it to metric fields.
Grouping metric schema `crates/fgumi-metrics/src/group.rs`, `src/lib/commands/group.rs`, `tests/integration/test_group_command.rs`	Serializes five fgbio grouping columns, skips internal summaries, and updates validation assertions.
Header-preserving metric writer `crates/fgumi-metrics/src/writer.rs`, `src/lib/metrics/mod.rs`, `src/lib/commands/{simplex_metrics,duplex_metrics,clip}.rs`	Writes headers for empty metric slices and routes metric outputs through the shared writer.
Sparse and deterministic duplex metrics `crates/fgumi-metrics/src/duplex.rs`	Emits observed family sizes only and adds deterministic UMI tie ordering with regression tests.
Rejection label parity `crates/fgumi-metrics/src/rejection.rs`	Updates rejection descriptions and exact-string tests to fgbio wording.
Metrics test dependencies `crates/fgumi-metrics/Cargo.toml`	Adds temporary-file, rstest, and proptest dependencies.

Estimated code review effort: 4 (Complex) | ~45 minutes

Possibly related PRs

fulcrumgenomics/fgumi#130: Overlaps on duplex metric computation and ordering.
fulcrumgenomics/fgumi#195: Overlaps on simplex metric structures and output handling.
fulcrumgenomics/fgumi#232: Overlaps on grouping distribution metric fields.

Suggested labels: enhancement

Sequence Diagram(s)

sequenceDiagram
  participant MetricsCommand
  participant MetricsWriter
  participant FloatSerde
  participant TSVFile
  MetricsCommand->>MetricsWriter: write metric collection
  MetricsWriter->>FloatSerde: serialize metric f64 fields
  FloatSerde->>TSVFile: emit fgbio-compatible values
  MetricsWriter->>TSVFile: preserve header for empty collections

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Linked Issues check	⚠️ Warning	`#498` is only partially satisfied because consensus rejection row seeding/parity remains deferred; `#39` is unrelated to this metrics PR.	Implement the remaining consensus rejection row seeding and text parity so fgumi matches fgbio's rejection rows for `#498`.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title is concise and accurately summarizes the main change: aligning metrics TSV output with fgbio's format.
Out of Scope Changes check	✅ Passed	No unrelated code changes stand out; the edits stay on metrics TSV serialization, writer plumbing, and supporting tests.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests
Commit unit tests in branch nh/fix-metrics-fgbio-parity

_{Comment @coderabbitai help to get the list of available commands.}

codecov · 2026-07-09T01:09:55Z

Codecov Report

❌ Patch coverage is 99.18033% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 92.61%. Comparing base (97b85c9) to head (a3e7ff5).

Files with missing lines	Patch %	Lines
crates/fgumi-metrics/src/writer.rs	97.56%	1 Missing ⚠️
src/lib/commands/duplex_metrics.rs	88.88%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #504      +/-   ##
==========================================
+ Coverage   92.60%   92.61%   +0.01%     
==========================================
  Files         165      166       +1     
  Lines       99636    99803     +167     
==========================================
+ Hits        92265    92431     +166     
- Misses       7371     7372       +1

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

…2-MET-05, R2-UCC-01) Bring the duplex caller's KV consensus statistics in line with fgbio's per-caller rejection-row model. R2-UCC-01 (S4): fgbio's DuplexConsensusCaller partitions unpaired/fragment reads out and rejects them as NonPairedReads (`val (pairs, frags) = recs.partition(_.paired); rejectRecords(frags, NonPairedReads)`); fgumi had no such reason and folded fragments into the strand groups, where they were ignored by strand grouping but mis-accounted. Add a NonPairedReads rejection reason (metrics + ConsensusMetrics field), map the caller-side FragmentRead reason to it (it previously mis-mapped to SameStrandOnly), and partition + reject fragments in DuplexConsensusCaller::consensus_reads. R2-MET-05 (S4): fgbio seeds and always emits per-caller rejection rows via `initializeRejectCounts` (`UmiConsensusCaller.scala:223-224`) — the duplex caller emits `raw_reads_rejected_for_{non_paired_reads,single_strand_only,potential_umi_collision}` even at zero, in addition to the four vanilla rows. fgumi emitted only the four vanilla rows and put the duplex reasons in a non-zero-only "optional" set. Replace the fixed core/optional split in ConsensusMetrics::to_kv_metrics with per-caller seeding driven by a new ConsensusCallerKind {Vanilla, Duplex, Codec}: seeded rows are always emitted (in fgbio order), and fgumi's finer-grained reasons are emitted only when non-zero. Re-key the collision reason from `duplicate_umi` to fgbio's `potential_umi_collision` and align the three newly seeded rows' descriptions with fgbio. (This is the row-set half of R2-MET-05; #504 owns the remaining description-text parity. Codec does not currently emit KV stats, so codec seeding is defined but unexercised; simplex already matched fgbio's four vanilla rows.) Verified fgbio<->fgumi end to end (reports/fgbio-parity-fixtures/r2-met-05-ucc-01-duplex-rejection-rows.sh): - R2-MET-05: fgbio duplex emits 7 rejection rows; fgumi BEFORE emitted 4; fgumi AFTER emits the same 7 rows (keys match fgbio exactly). - R2-UCC-01: with one fragment read, fgbio reports non_paired_reads=1; fgumi BEFORE emitted no such row; fgumi AFTER reports non_paired_reads=1.

…ex rejection rows Two fgbio parity fixes in the duplex caller and its rejection metrics. R2-UCC-01 — duplex fragment accounting. fgbio partitions unpaired/fragment reads out and rejects them as `non_paired_reads` (DuplexConsensusCaller.scala:204-205). fgumi dropped them from the R1/R2 strand groups but folded them into the input total; they are now counted as `FragmentRead` (→ `non_paired_reads`). R2-MET-05 — always-emit duplex rejection rows. fgbio seeds each caller's `usedBy*` rejection reasons to zero so their rows always appear (initializeRejectCounts). Add `ConsensusMetrics::to_kv_metrics_seeded` + `DUPLEX_SEEDED_REJECTIONS`; the duplex command seeds `non_paired_reads`, `single_strand_only`, and `potential_umi_collision` so those KV rows are always present. R2-UCC-02 (fgbio's distinct cell-barcode `require`) is deliberately NOT ported: fgumi's grouper keys on `(MI, cell)`, so a molecule spanning multiple cell barcodes is split into per-cell groups and consensus-called per cell (lossless), whereas fgbio's `require(barcodes.length <= 1)` fails fast. The caller therefore never sees more than one cell barcode via the CLI (verified: a same-MI/two-CB molecule yields two per-cell consensus reads), so a caller-level guard would be unreachable dead code. This is a deliberate, lossless architectural divergence. Tests: `to_kv_metrics_seeded` always-emits-at-zero and is not duplicated when non-zero. Full suite passes; ci-fmt/ci-lint clean. Branch stacked on #504.

…2-MET-05, R2-UCC-01) Bring the duplex caller's KV consensus statistics in line with fgbio's per-caller rejection-row model. R2-UCC-01 (S4): fgbio's DuplexConsensusCaller partitions unpaired/fragment reads out and rejects them as NonPairedReads (`val (pairs, frags) = recs.partition(_.paired); rejectRecords(frags, NonPairedReads)`); fgumi had no such reason and folded fragments into the strand groups, where they were ignored by strand grouping but mis-accounted. Add a NonPairedReads rejection reason (metrics + ConsensusMetrics field), map the caller-side FragmentRead reason to it (it previously mis-mapped to SameStrandOnly), and partition + reject fragments in DuplexConsensusCaller::consensus_reads. R2-MET-05 (S4): fgbio seeds and always emits per-caller rejection rows via `initializeRejectCounts` (`UmiConsensusCaller.scala:223-224`) — the duplex caller emits `raw_reads_rejected_for_{non_paired_reads,single_strand_only,potential_umi_collision}` even at zero, in addition to the four vanilla rows. fgumi emitted only the four vanilla rows and put the duplex reasons in a non-zero-only "optional" set. Replace the fixed core/optional split in ConsensusMetrics::to_kv_metrics with per-caller seeding driven by a new ConsensusCallerKind {Vanilla, Duplex, Codec}: seeded rows are always emitted (in fgbio order), and fgumi's finer-grained reasons are emitted only when non-zero. Re-key the collision reason from `duplicate_umi` to fgbio's `potential_umi_collision` and align the three newly seeded rows' descriptions with fgbio. (This is the row-set half of R2-MET-05; #504 owns the remaining description-text parity. Codec does not currently emit KV stats, so codec seeding is defined but unexercised; simplex already matched fgbio's four vanilla rows.) Verified fgbio<->fgumi end to end (reports/fgbio-parity-fixtures/r2-met-05-ucc-01-duplex-rejection-rows.sh): - R2-MET-05: fgbio duplex emits 7 rejection rows; fgumi BEFORE emitted 4; fgumi AFTER emits the same 7 rows (keys match fgbio exactly). - R2-UCC-01: with one fragment read, fgbio reports non_paired_reads=1; fgumi BEFORE emitted no such row; fgumi AFTER reports non_paired_reads=1.

nh13 · 2026-07-10T12:18:36Z

@coderabbitai review

coderabbitai · 2026-07-10T12:18:42Z

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

nh13 · 2026-07-10T13:18:22Z

@coderabbitai review

coderabbitai · 2026-07-10T13:18:28Z

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

nh13 · 2026-07-10T13:18:52Z

@coderabbitai review

coderabbitai · 2026-07-10T13:19:00Z

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

coderabbitai

Actionable comments posted: 3

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@crates/fgumi-metrics/src/duplex.rs`:
- Around line 330-353: Add a fixture-backed golden test for
DuplexMetrics::family_size_metrics comparing the emitted sparse rows and
fractions against an fgbio-generated family_sizes.txt baseline. Extend the
existing tests near test_family_size_metrics_are_sparse and
test_family_size_metrics_union_disjoint_cs_ss_ds_keys, loading representative
fixture input and asserting the generated output matches the golden file
exactly.

In `@crates/fgumi-metrics/src/rejection.rs`:
- Around line 59-60: Update the test_to_kv_metrics_* tests to assert exact
description strings for raw_reads_rejected_for_minority_alignment and
raw_reads_rejected_for_single_strand_only, matching the descriptions returned by
MinorityAlignment and SameStrandOnly and preserving fgbio parity.

In `@src/lib/commands/duplex_metrics.rs`:
- Around line 224-231: Reuse the UMI metrics already computed before the duplex
block instead of calling main_collector.umi_metrics() again. Update the
duplex_umi_metrics calculation in the duplex_umi_counts branch to pass the
existing metrics value into main_collector.duplex_umi_metrics, preserving the
current output and writing behavior.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

Push a commit to this branch (recommended)
Create a new PR with the fixes

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 7ba74c13-0704-4429-804e-dc1c82b1e62b

📥 Commits

Reviewing files that changed from the base of the PR and between f638abc and c5e61ce.

📒 Files selected for processing (16)

crates/fgumi-metrics/src/consensus.rs
crates/fgumi-metrics/src/correct.rs
crates/fgumi-metrics/src/duplex.rs
crates/fgumi-metrics/src/float.rs
crates/fgumi-metrics/src/group.rs
crates/fgumi-metrics/src/lib.rs
crates/fgumi-metrics/src/rejection.rs
crates/fgumi-metrics/src/shared.rs
crates/fgumi-metrics/src/simplex.rs
crates/fgumi-metrics/src/writer.rs
src/lib/commands/clip.rs
src/lib/commands/duplex_metrics.rs
src/lib/commands/group.rs
src/lib/commands/simplex_metrics.rs
src/lib/metrics/mod.rs
tests/integration/test_group_command.rs

…2-MET-05, R2-UCC-01) Bring the duplex caller's KV consensus statistics in line with fgbio's per-caller rejection-row model. R2-UCC-01 (S4): fgbio's DuplexConsensusCaller partitions unpaired/fragment reads out and rejects them as NonPairedReads (`val (pairs, frags) = recs.partition(_.paired); rejectRecords(frags, NonPairedReads)`); fgumi had no such reason and folded fragments into the strand groups, where they were ignored by strand grouping but mis-accounted. Add a NonPairedReads rejection reason (metrics + ConsensusMetrics field), map the caller-side FragmentRead reason to it (it previously mis-mapped to SameStrandOnly), and partition + reject fragments in DuplexConsensusCaller::consensus_reads. R2-MET-05 (S4): fgbio seeds and always emits per-caller rejection rows via `initializeRejectCounts` (`UmiConsensusCaller.scala:223-224`) — the duplex caller emits `raw_reads_rejected_for_{non_paired_reads,single_strand_only,potential_umi_collision}` even at zero, in addition to the four vanilla rows. fgumi emitted only the four vanilla rows and put the duplex reasons in a non-zero-only "optional" set. Replace the fixed core/optional split in ConsensusMetrics::to_kv_metrics with per-caller seeding driven by a new ConsensusCallerKind {Vanilla, Duplex, Codec}: seeded rows are always emitted (in fgbio order), and fgumi's finer-grained reasons are emitted only when non-zero. Re-key the collision reason from `duplicate_umi` to fgbio's `potential_umi_collision` and align the three newly seeded rows' descriptions with fgbio. (This is the row-set half of R2-MET-05; #504 owns the remaining description-text parity. Codec does not currently emit KV stats, so codec seeding is defined but unexercised; simplex already matched fgbio's four vanilla rows.) Verified fgbio<->fgumi end to end (reports/fgbio-parity-fixtures/r2-met-05-ucc-01-duplex-rejection-rows.sh): - R2-MET-05: fgbio duplex emits 7 rejection rows; fgumi BEFORE emitted 4; fgumi AFTER emits the same 7 rows (keys match fgbio exactly). - R2-UCC-01: with one fragment read, fgbio reports non_paired_reads=1; fgumi BEFORE emitted no such row; fgumi AFTER reports non_paired_reads=1.

…3-04, DXM3-06, SIMM3-02) Completes the metrics-parity work #504 started. - DXM3-04: #504 wired the fgbio-style float formatter onto the fgumi-metrics crate's metric structs but left `DedupMetricsOutput.duplicate_rate` (in fgumi_lib) on raw serde f64. Make `fgumi_metrics::float` public and wire it there so dedup's one fraction column formats like the rest (a whole-number rate writes `1`, not `1.0`). (Verified in re-scoping that the audit's "clip" struct has no f64 fields, and group's `avg_reads_per_molecule` is `#[serde(skip)]`, so dedup is the only serialized gap.) - DXM3-06: `fgumi duplex-metrics` accepted `--min-ab-reads 0`, silently labeling every tag family a duplex. fgbio's CollectDuplexSeqMetrics validates `minAbReads >= 1`; add the same guard (message matches fgbio's wording). - SIMM3-02: `fgumi simplex-metrics` accepted `--min-reads 0` (every family of size >= 0 labeled a consensus family). Add the `min_reads >= 1` guard. Real-tool parity: fgbio source (CollectDuplexSeqMetrics.scala:288) confirms `validate(minAbReads >= 1, ...)`; fgumi now errors identically on 0. Tests: dedup float-format serialization + duplex/simplex zero-threshold reject. Stacked on #504. Tracker: reports/2026-07-09-fgumi-final-audit-burndown-tracker.md (W3).

nh13 · 2026-07-12T00:23:43Z

@coderabbitai review

coderabbitai · 2026-07-12T00:23:52Z

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

nh13 · 2026-07-12T05:01:21Z

@coderabbitai review

coderabbitai · 2026-07-12T05:01:26Z

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

nh13 · 2026-07-12T06:01:32Z

@coderabbitai review

coderabbitai · 2026-07-12T06:01:37Z

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

coderabbitai

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@crates/fgumi-metrics/src/consensus.rs`:
- Around line 60-65: Add a regression test for the metric type containing
avg_input_reads_per_consensus and avg_raw_read_depth, exercising NaN and
Infinity round-trips through write_metrics and read_metrics. Assert both fields
preserve their values, consistent with
correct.rs::test_nan_infinity_serialization. Also add the required
programmatically generated identity-or-documented-divergence test against the
fgbio baseline for this metric-format change.

In `@crates/fgumi-metrics/src/writer.rs`:
- Around line 79-90: Update write_header_only to set the temporary file’s
permissions to match those used by the direct populated-file write path before
calling tmp.persist(path_ref). Preserve the existing atomic temp-file workflow
and apply the mode normalization after rewriting the header but before renaming.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

Push a commit to this branch (recommended)
Create a new PR with the fixes

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 48037dc2-8c1f-42ff-b824-1d7d0069b7ee

📥 Commits

Reviewing files that changed from the base of the PR and between ee10699 and ebc51e1.

⛔ Files ignored due to path filters (1)

Cargo.lock is excluded by !**/*.lock, !**/*.lock

📒 Files selected for processing (17)

crates/fgumi-metrics/Cargo.toml
crates/fgumi-metrics/src/consensus.rs
crates/fgumi-metrics/src/correct.rs
crates/fgumi-metrics/src/duplex.rs
crates/fgumi-metrics/src/float.rs
crates/fgumi-metrics/src/group.rs
crates/fgumi-metrics/src/lib.rs
crates/fgumi-metrics/src/rejection.rs
crates/fgumi-metrics/src/shared.rs
crates/fgumi-metrics/src/simplex.rs
crates/fgumi-metrics/src/writer.rs
src/lib/commands/clip.rs
src/lib/commands/duplex_metrics.rs
src/lib/commands/group.rs
src/lib/commands/simplex_metrics.rs
src/lib/metrics/mod.rs
tests/integration/test_group_command.rs

…ows (CONS-01, DUP-01, R2-MET-05, R2-UCC-01) (#508) * fix(consensus): guard sort order and duplex error rates (CONS-01, DUP-01) Two missing input guards on the standalone consensus callers, both matching fgbio's UmiConsensusCaller / CallDuplexConsensusReads behavior. CONS-01 (S3): fgbio's UmiConsensusCaller.checkSortOrder (UmiConsensusCaller.scala:165-173) errors unless the input is template-coordinate sorted; simplex/duplex/codec had no sort-order guard (only group/dedup did), so coordinate-sorted standalone input silently interleaved molecules and every molecule was split — corrupting the entire output. Add check_consensus_sort_order (commands/common.rs) mirroring checkSortOrder's three cases exactly (UmiConsensusCallerTest.scala:34-60): template-coordinate => accept; SO:unsorted + GO:query (no SS) => warn; anything else => error. It reads only the already-read header and never re-opens the stream, so it is safe on stdin. Wired into both the threaded and single-threaded header-read sites of simplex/duplex/codec. New fgumi-sam helpers is_query_grouped_unsorted and header_as_template_coordinate back the guard and the test-input builder. DUP-01 (S3): fgbio's CallDuplexConsensusReads validates errorRate{Pre,Post}Umi > 0 (CallDuplexConsensusReads.scala:131-132) — Q0 means P(err)=1, a nonsensical prior. fgumi duplex accepted a raw u8 with no check. Add Duplex::validate() rejecting a zero pre/post-UMI error rate. (codec already guarded this; simplex correctly allows >=0 per fgbio, and its u8 is never negative.) Consensus command tests now stamp template-coordinate order on their input BAMs (a new SamBuilder::set_template_coordinate_sort_order helper, plus the codec / duplex input-writer helpers), matching fgbio's command tests which build input with SamOrder.TemplateCoordinate. Verified fgbio<->fgumi end to end (reports/fgbio-parity-fixtures/ cons-01-dup-01-consensus-guards.sh): - CONS-01, coordinate input: fgbio exit 1 (reject); fgumi BEFORE exit 0 (silently accepted); fgumi AFTER exit 1 (rejects, same message class); template-coordinate input AFTER exit 0. - DUP-01, --error-rate-pre/post-umi 0: fgbio exit 1; fgumi BEFORE exit 0; fgumi AFTER exit 1; defaults AFTER exit 0. * fix(consensus): seed duplex KV rejection rows and reject fragments (R2-MET-05, R2-UCC-01) Bring the duplex caller's KV consensus statistics in line with fgbio's per-caller rejection-row model. R2-UCC-01 (S4): fgbio's DuplexConsensusCaller partitions unpaired/fragment reads out and rejects them as NonPairedReads (`val (pairs, frags) = recs.partition(_.paired); rejectRecords(frags, NonPairedReads)`); fgumi had no such reason and folded fragments into the strand groups, where they were ignored by strand grouping but mis-accounted. Add a NonPairedReads rejection reason (metrics + ConsensusMetrics field), map the caller-side FragmentRead reason to it (it previously mis-mapped to SameStrandOnly), and partition + reject fragments in DuplexConsensusCaller::consensus_reads. R2-MET-05 (S4): fgbio seeds and always emits per-caller rejection rows via `initializeRejectCounts` (`UmiConsensusCaller.scala:223-224`) — the duplex caller emits `raw_reads_rejected_for_{non_paired_reads,single_strand_only,potential_umi_collision}` even at zero, in addition to the four vanilla rows. fgumi emitted only the four vanilla rows and put the duplex reasons in a non-zero-only "optional" set. Replace the fixed core/optional split in ConsensusMetrics::to_kv_metrics with per-caller seeding driven by a new ConsensusCallerKind {Vanilla, Duplex, Codec}: seeded rows are always emitted (in fgbio order), and fgumi's finer-grained reasons are emitted only when non-zero. Re-key the collision reason from `duplicate_umi` to fgbio's `potential_umi_collision` and align the three newly seeded rows' descriptions with fgbio. (This is the row-set half of R2-MET-05; #504 owns the remaining description-text parity. Codec does not currently emit KV stats, so codec seeding is defined but unexercised; simplex already matched fgbio's four vanilla rows.) Verified fgbio<->fgumi end to end (reports/fgbio-parity-fixtures/r2-met-05-ucc-01-duplex-rejection-rows.sh): - R2-MET-05: fgbio duplex emits 7 rejection rows; fgumi BEFORE emitted 4; fgumi AFTER emits the same 7 rows (keys match fgbio exactly). - R2-UCC-01: with one fragment read, fgbio reports non_paired_reads=1; fgumi BEFORE emitted no such row; fgumi AFTER reports non_paired_reads=1.

…3-04, DXM3-06, SIMM3-02) Completes the metrics-parity work #504 started. - DXM3-04: #504 wired the fgbio-style float formatter onto the fgumi-metrics crate's metric structs but left `DedupMetricsOutput.duplicate_rate` (in fgumi_lib) on raw serde f64. Make `fgumi_metrics::float` public and wire it there so dedup's one fraction column formats like the rest (a whole-number rate writes `1`, not `1.0`). (Verified in re-scoping that the audit's "clip" struct has no f64 fields, and group's `avg_reads_per_molecule` is `#[serde(skip)]`, so dedup is the only serialized gap.) - DXM3-06: `fgumi duplex-metrics` accepted `--min-ab-reads 0`, silently labeling every tag family a duplex. fgbio's CollectDuplexSeqMetrics validates `minAbReads >= 1`; add the same guard (message matches fgbio's wording). - SIMM3-02: `fgumi simplex-metrics` accepted `--min-reads 0` (every family of size >= 0 labeled a consensus family). Add the `min_reads >= 1` guard. Real-tool parity: fgbio source (CollectDuplexSeqMetrics.scala:288) confirms `validate(minAbReads >= 1, ...)`; fgumi now errors identically on 0. Tests: dedup float-format serialization + duplex/simplex zero-threshold reject. Stacked on #504. Tracker: reports/2026-07-09-fgumi-final-audit-burndown-tracker.md (W3).

nh13 · 2026-07-12T15:43:16Z

@coderabbitai review

coderabbitai · 2026-07-12T15:43:27Z

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

nh13 · 2026-07-12T16:44:11Z

@coderabbitai review

coderabbitai · 2026-07-12T16:44:16Z

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

coderabbitai

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@crates/fgumi-metrics/src/writer.rs`:
- Around line 53-96: Update write_metrics so populated metrics also serialize
through a temporary file in the destination directory and atomically persist it
to path_ref, matching write_header_only’s temp-file workflow. Preserve the
existing empty-metrics header behavior and contextual error messages, and ensure
the temporary file uses the same permissions approach as the current direct
write.

In `@src/lib/commands/clip.rs`:
- Around line 305-333: The per-template clipping and mate-repair workflow is
duplicated between execute_single_threaded and the process_fn path. Extract the
shared primary-pair and lone-fragment logic—including clip thresholds/flags,
overlap and mate-extension metrics, set_mate_info_raw, and
fix_supplemental_mate_info—into a free function accepting plain configuration
values rather than &self, then call it from both paths while preserving current
record selection and outcomes.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

Push a commit to this branch (recommended)
Create a new PR with the fixes

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 499a2567-2367-4db7-b3d6-162c05e145f1

📥 Commits

Reviewing files that changed from the base of the PR and between ee10699 and 67423d1.

⛔ Files ignored due to path filters (1)

Cargo.lock is excluded by !**/*.lock, !**/*.lock

📒 Files selected for processing (17)

crates/fgumi-metrics/Cargo.toml
crates/fgumi-metrics/src/consensus.rs
crates/fgumi-metrics/src/correct.rs
crates/fgumi-metrics/src/duplex.rs
crates/fgumi-metrics/src/float.rs
crates/fgumi-metrics/src/group.rs
crates/fgumi-metrics/src/lib.rs
crates/fgumi-metrics/src/rejection.rs
crates/fgumi-metrics/src/shared.rs
crates/fgumi-metrics/src/simplex.rs
crates/fgumi-metrics/src/writer.rs
src/lib/commands/clip.rs
src/lib/commands/duplex_metrics.rs
src/lib/commands/group.rs
src/lib/commands/simplex_metrics.rs
src/lib/metrics/mod.rs
tests/integration/test_group_command.rs

coderabbitai

Caution

Inline review comments failed to post. This is likely due to GitHub's internal server error or limits when posting large numbers of comments. If you are seeing this consistently it is likely a permissions issue. Please check "Moderation" -> "Code review limits" under your organization settings.

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@crates/fgumi-metrics/src/writer.rs`:
- Around line 53-96: Update write_metrics so populated metrics also serialize
through a temporary file in the destination directory and atomically persist it
to path_ref, matching write_header_only’s temp-file workflow. Preserve the
existing empty-metrics header behavior and contextual error messages, and ensure
the temporary file uses the same permissions approach as the current direct
write.

In `@src/lib/commands/clip.rs`:
- Around line 305-333: The per-template clipping and mate-repair workflow is
duplicated between execute_single_threaded and the process_fn path. Extract the
shared primary-pair and lone-fragment logic—including clip thresholds/flags,
overlap and mate-extension metrics, set_mate_info_raw, and
fix_supplemental_mate_info—into a free function accepting plain configuration
values rather than &self, then call it from both paths while preserving current
record selection and outcomes.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

Push a commit to this branch (recommended)
Create a new PR with the fixes

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 499a2567-2367-4db7-b3d6-162c05e145f1

📥 Commits

Reviewing files that changed from the base of the PR and between ee10699 and 67423d1.

⛔ Files ignored due to path filters (1)

Cargo.lock is excluded by !**/*.lock, !**/*.lock

📒 Files selected for processing (17)

crates/fgumi-metrics/Cargo.toml
crates/fgumi-metrics/src/consensus.rs
crates/fgumi-metrics/src/correct.rs
crates/fgumi-metrics/src/duplex.rs
crates/fgumi-metrics/src/float.rs
crates/fgumi-metrics/src/group.rs
crates/fgumi-metrics/src/lib.rs
crates/fgumi-metrics/src/rejection.rs
crates/fgumi-metrics/src/shared.rs
crates/fgumi-metrics/src/simplex.rs
crates/fgumi-metrics/src/writer.rs
src/lib/commands/clip.rs
src/lib/commands/duplex_metrics.rs
src/lib/commands/group.rs
src/lib/commands/simplex_metrics.rs
src/lib/metrics/mod.rs
tests/integration/test_group_command.rs

🛑 Comments failed to post (2)

crates/fgumi-metrics/src/writer.rs (1)
53-96: 🗄️ Data Integrity & Integration | 🔵 Trivial | ⚡ Quick win

Only the empty-metrics path got atomic-write hardening; the populated path (the common case) is still a direct in-place write.

A crash mid-write_tsv on a non-empty metrics file leaves a truncated/corrupt TSV on disk — the same class of problem write_header_only's temp-file+persist was built to avoid, just for the branch that matters more in practice.
♻️ Sketch: reuse the temp-file+persist pattern for both branches
 pub fn write_metrics<P: AsRef<Path>, T: Serialize + Default>(
     path: P,
     metrics: &[T],
     description: &str,
 ) -> Result<()> {
     let path_ref = path.as_ref();
-    if metrics.is_empty() {
-        return write_header_only::<_, T>(path_ref).with_context(|| {
-            format!("Failed to write {} metrics header: {}", description, path_ref.display())
-        });
-    }
-    DelimFile::default()
-        .write_tsv(path_ref, metrics)
-        .with_context(|| format!("Failed to write {} metrics: {}", description, path_ref.display()))
+    let dir = path_ref.parent().unwrap_or_else(|| Path::new("."));
+    let tmp = tempfile::Builder::new().make_in(dir, std::fs::File::create)?;
+    if metrics.is_empty() {
+        write_header_only_into(tmp.path())?;
+    } else {
+        DelimFile::default().write_tsv(tmp.path(), metrics)?;
+    }
+    tmp.persist(path_ref)
+        .with_context(|| format!("Failed to write {} metrics: {}", description, path_ref.display()))
 }
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@crates/fgumi-metrics/src/writer.rs` around lines 53 - 96, Update
write_metrics so populated metrics also serialize through a temporary file in
the destination directory and atomically persist it to path_ref, matching
write_header_only’s temp-file workflow. Preserve the existing empty-metrics
header behavior and contextual error messages, and ensure the temporary file
uses the same permissions approach as the current direct write.
src/lib/commands/clip.rs (1)
305-333: 📐 Maintainability & Code Quality | 🔵 Trivial | 🏗️ Heavy lift

Clip+mate-repair logic duplicated between single- and multi-threaded paths.

The primary-pair clipping + set_mate_info_raw/fix_supplemental_mate_info sequence is implemented twice (here and in Clip::clip_pair/clip_fragment), and this very diff had to add the identical mate-info-fixing step to both copies. Any future correctness fix to pair clipping or mate-info repair now needs to be applied in two places, risking silent divergence between single- and multi-threaded output.

Consider extracting the shared per-template body (threshold clipping + overlap/extend clipping + set_mate_info_raw + fix_supplemental_mate_info) into a free function taking plain values (clipper, thresholds, flags) instead of &self, callable from both execute_single_threaded and the process_fn closure.

Also applies to: 635-709
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/lib/commands/clip.rs` around lines 305 - 333, The per-template clipping
and mate-repair workflow is duplicated between execute_single_threaded and the
process_fn path. Extract the shared primary-pair and lone-fragment
logic—including clip thresholds/flags, overlap and mate-extension metrics,
set_mate_info_raw, and fix_supplemental_mate_info—into a free function accepting
plain configuration values rather than &self, then call it from both paths while
preserving current record selection and outcomes.

Make fgumi's metrics `.txt` outputs readable by fgbio's `Metric.read` and match fgbio's rows, column names/order, and value tokens on a shared input. Verified end-to-end against fgbio 4.1.0's real `Metric.read` (see fixture below). - Non-finite floats (MET-02): add a shared serde helper (`float` module) that writes f64 fields as fgbio's `Infinity`/`-Infinity`/`NaN` tokens (integral values without a trailing `.0`), and apply it to every serialized f64 field across correct/shared/consensus/simplex/duplex/group. Previously only `correct.rs` was inf-safe; the rest emitted ryu `inf`/`-inf`, which fgbio's `Double.parseDouble` cannot read. - Empty-metrics header (MET-03): `write_metrics` now always emits the column header, even for a zero-row slice (the csv writer's lazy headers otherwise produce a 0-byte file that fgbio rejects with "No header found"). Route the duplex-metrics and simplex-metrics per-family/UMI/yield writes through it. - UmiGroupingMetrics (MET-04): serialize exactly fgbio's 5-column `UmiGroupingMetric` (`accepted_sam_records`, `discarded_non_pf`, `discarded_poor_alignment`, `discarded_ns_in_umi`, `discarded_umis_to_short`). The fgumi-internal summary fields (`total_records`, family aggregates) are retained in memory for logging but `#[serde(skip)]`-ed; they're derivable and the family distribution already lives in `.family_sizes.txt`. - Duplex family sizes sparse: emit one row per observed family size (matching fgbio `familySizeMetrics`) instead of a dense `1..=max` range. - Duplex UMI counts determinism: add `umi` as a secondary sort key so ties no longer fall back to nondeterministic HashMap order. - Consensus KV rejection text (MET-05, partial): match fgbio's exact rejection descriptions ("Reads has a different, and minority, set of indels"; "Only Generating One Strand of Duplex Consensus"). The caller-type-aware row-set seeding (non_paired_reads / potential_umi_collision, etc.) is deferred to the consensus PR (fgbio's usedByVanilla/Duplex/Codec seeding + R2-UCC-01), as it requires new RejectionReason variants wired into the duplex/codec callers. - format_float doc (MET-01): correct the doc-comment that falsely claimed it matches fgbio; numeric equality (not byte format) is the metric contract. CorrectUmis zero-count seeding (#498 item 5) already reproduces as fixed on main: both correct execution paths pre-seed a row per expected UMI plus the all-N sentinel. No change needed; kept as-is. fgbio-oracle evidence (reports/fgbio-parity-fixtures/met-02-03-04-fgbio-read.sh): fgbio 4.1.0 `Metric.read[UmiGroupingMetric]` parses fgumi's 5-column file and its header-only empty file (0 rows, no error); `Metric.read[UmiCorrectionMetrics]` parses `representation=Infinity` — while the old ryu `inf` token is rejected. CI: cargo ci-fmt / ci-lint clean; full suite 2215 passed.

nh13 · 2026-07-12T17:28:00Z

Addressed the two actionable findings from the latest CodeRabbit review (they came through only in the review-body "Prompt for AI agents" block, since the inline comments failed to post):

crates/fgumi-metrics/src/writer.rs — atomic populated write. Done. write_metrics now routes both the empty (header-only) and populated cases through a single write_metrics_atomic helper that serializes to a temp file in the destination directory and atomically renames it into place, so an interrupted write never leaves a partially-written metrics file. The temp file is created via File::create (through Builder::make_in) so it inherits the umask-applied mode rather than NamedTempFile's owner-only 0o600. Added an rstest that pins atomicity (no stray temp files) for both the empty and populated paths.
src/lib/commands/clip.rs — deduplicate the per-template clip/mate-repair logic. Deferred to a focused follow-on PR, refactor(clip): unify per-template clip/mate-repair across threading paths #582, rather than mixing it into this metrics PR — this PR's only change to clip.rs is a one-line metrics adaptation, and the duplication is a pre-existing structural issue in a correctness-critical hot path. refactor(clip): unify per-template clip/mate-repair across threading paths #582 extracts the shared logic into a ClipParams::clip_template used by both the single- and multi-threaded paths and adds an end-to-end parity test asserting byte-identical output across the two modes.

…3-04, DXM3-06, SIMM3-02) Completes the metrics-parity work #504 started. - DXM3-04: #504 wired the fgbio-style float formatter onto the fgumi-metrics crate's metric structs but left `DedupMetricsOutput.duplicate_rate` (in fgumi_lib) on raw serde f64. Make `fgumi_metrics::float` public and wire it there so dedup's one fraction column formats like the rest (a whole-number rate writes `1`, not `1.0`). (Verified in re-scoping that the audit's "clip" struct has no f64 fields, and group's `avg_reads_per_molecule` is `#[serde(skip)]`, so dedup is the only serialized gap.) - DXM3-06: `fgumi duplex-metrics` accepted `--min-ab-reads 0`, silently labeling every tag family a duplex. fgbio's CollectDuplexSeqMetrics validates `minAbReads >= 1`; add the same guard (message matches fgbio's wording). - SIMM3-02: `fgumi simplex-metrics` accepted `--min-reads 0` (every family of size >= 0 labeled a consensus family). Add the `min_reads >= 1` guard. Real-tool parity: fgbio source (CollectDuplexSeqMetrics.scala:288) confirms `validate(minAbReads >= 1, ...)`; fgumi now errors identically on 0. Tests: dedup float-format serialization + duplex/simplex zero-threshold reject. Stacked on #504. Tracker: reports/2026-07-09-fgumi-final-audit-burndown-tracker.md (W3).

…3-04, DXM3-06, SIMM3-02) (#520) Completes the metrics-parity work #504 started. - DXM3-04: #504 wired the fgbio-style float formatter onto the fgumi-metrics crate's metric structs but left `DedupMetricsOutput.duplicate_rate` (in fgumi_lib) on raw serde f64. Make `fgumi_metrics::float` public and wire it there so dedup's one fraction column formats like the rest (a whole-number rate writes `1`, not `1.0`). (Verified in re-scoping that the audit's "clip" struct has no f64 fields, and group's `avg_reads_per_molecule` is `#[serde(skip)]`, so dedup is the only serialized gap.) - DXM3-06: `fgumi duplex-metrics` accepted `--min-ab-reads 0`, silently labeling every tag family a duplex. fgbio's CollectDuplexSeqMetrics validates `minAbReads >= 1`; add the same guard (message matches fgbio's wording). - SIMM3-02: `fgumi simplex-metrics` accepted `--min-reads 0` (every family of size >= 0 labeled a consensus family). Add the `min_reads >= 1` guard. Real-tool parity: fgbio source (CollectDuplexSeqMetrics.scala:288) confirms `validate(minAbReads >= 1, ...)`; fgumi now errors identically on 0. Tests: dedup float-format serialization + duplex/simplex zero-threshold reject. Stacked on #504. Tracker: reports/2026-07-09-fgumi-final-audit-burndown-tracker.md (W3).

nh13 had a problem deploying to github-actions July 9, 2026 00:54 — with GitHub Actions Failure

nh13 temporarily deployed to github-actions July 9, 2026 01:07 — with GitHub Actions Inactive

This was referenced Jul 9, 2026

metrics: TSV output divergences from fgbio Metric format (tracking) #498

Closed

fix(consensus): sort-order + error-rate guards and duplex rejection rows (CONS-01, DUP-01, R2-MET-05, R2-UCC-01) #508

Merged

nh13 mentioned this pull request Jul 9, 2026

fix(consensus): fgbio rejection-reason parity + thread-independent duplex metrics #511

Closed

nh13 mentioned this pull request Jul 9, 2026

fix(codec): emit fgbio KV metrics format (thread-consistent, seeded rows) #513

Merged

nh13 force-pushed the nh/fix-metrics-fgbio-parity branch from b0fd52c to c5e61ce Compare July 9, 2026 15:10

nh13 temporarily deployed to github-actions July 9, 2026 15:10 — with GitHub Actions Inactive

This was referenced Jul 9, 2026

docs(metrics): correct format_float doc-comment (R2-MET-01) #515

Closed

fix(codec): model codec-specific fgbio rejection reasons + seed KV rows #516

Merged

This was referenced Jul 9, 2026

fix(metrics): finish fgbio float format + reads-threshold guards (DXM3-04, DXM3-06, SIMM3-02) #520

Merged

feat(compare): harden fgumi compare into a sound, faithful fgbio-parity oracle #530

Merged

coderabbitai Bot reviewed Jul 10, 2026

View reviewed changes

Comment thread crates/fgumi-metrics/src/duplex.rs

Comment thread crates/fgumi-metrics/src/rejection.rs

Comment thread src/lib/commands/duplex_metrics.rs

nh13 force-pushed the nh/fix-metrics-fgbio-parity branch from c5e61ce to dd984c8 Compare July 10, 2026 18:32

nh13 temporarily deployed to github-actions July 10, 2026 18:32 — with GitHub Actions Inactive

coderabbitai Bot reviewed Jul 12, 2026

View reviewed changes

Comment thread crates/fgumi-metrics/src/consensus.rs

Comment thread crates/fgumi-metrics/src/writer.rs Outdated

nh13 force-pushed the nh/fix-metrics-fgbio-parity branch from ebc51e1 to fda0a20 Compare July 12, 2026 15:13

nh13 temporarily deployed to github-actions July 12, 2026 15:13 — with GitHub Actions Inactive

nh13 force-pushed the nh/fix-metrics-fgbio-parity branch from fda0a20 to 67423d1 Compare July 12, 2026 15:19

nh13 temporarily deployed to github-actions July 12, 2026 15:19 — with GitHub Actions Inactive

coderabbitai Bot reviewed Jul 12, 2026

View reviewed changes

nh13 force-pushed the nh/fix-metrics-fgbio-parity branch from 67423d1 to a3e7ff5 Compare July 12, 2026 17:15

nh13 temporarily deployed to github-actions July 12, 2026 17:15 — with GitHub Actions Inactive

nh13 mentioned this pull request Jul 12, 2026

refactor(clip): unify per-template clip/mate-repair across threading paths #582

Merged

nh13 merged commit 3e400a8 into main Jul 12, 2026
10 checks passed

nh13 deleted the nh/fix-metrics-fgbio-parity branch July 12, 2026 17:29

nh13 mentioned this pull request Jul 12, 2026

chore: release v0.5.0 #454

Merged

coderabbitai Bot mentioned this pull request Aug 12, 2026

fix(dedup)!: report per-reason filtered template counts #740

Merged

Conversation

nh13 commented Jul 9, 2026 • edited by coderabbitai Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What changed

fgbio-oracle evidence

Deferred (out of scope here)

Testing

Summary by CodeRabbit

Uh oh!

coderabbitai Bot commented Jul 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Reviews paused

Walkthrough

Changes

Sequence Diagram(s)

❌ Failed checks (1 warning)

Uh oh!

codecov Bot commented Jul 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

nh13 commented Jul 10, 2026

Uh oh!

coderabbitai Bot commented Jul 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

nh13 commented Jul 10, 2026

Uh oh!

coderabbitai Bot commented Jul 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

nh13 commented Jul 10, 2026

Uh oh!

coderabbitai Bot commented Jul 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

nh13 commented Jul 12, 2026

Uh oh!

coderabbitai Bot commented Jul 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

nh13 commented Jul 12, 2026

Uh oh!

coderabbitai Bot commented Jul 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

nh13 commented Jul 12, 2026

Uh oh!

coderabbitai Bot commented Jul 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

nh13 commented Jul 12, 2026

Uh oh!

coderabbitai Bot commented Jul 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

nh13 commented Jul 12, 2026

Uh oh!

coderabbitai Bot commented Jul 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

nh13 commented Jul 12, 2026

Uh oh!

nh13 commented Jul 9, 2026 •

edited by coderabbitai Bot

Loading

coderabbitai Bot commented Jul 9, 2026 •

edited

Loading

codecov Bot commented Jul 9, 2026 •

edited

Loading

coderabbitai Bot commented Jul 10, 2026 •

edited

Loading

coderabbitai Bot commented Jul 10, 2026 •

edited

Loading

coderabbitai Bot commented Jul 10, 2026 •

edited

Loading

coderabbitai Bot commented Jul 12, 2026 •

edited

Loading

coderabbitai Bot commented Jul 12, 2026 •

edited

Loading

coderabbitai Bot commented Jul 12, 2026 •

edited

Loading

coderabbitai Bot commented Jul 12, 2026 •

edited

Loading

coderabbitai Bot commented Jul 12, 2026 •

edited

Loading