[resolve #4] Review comment by @copilot (PR #2185):
"iscrowd is still derived solely from contains_holes/contains_multiple_segments, ignoring data[iscrowd] from COCO"

Validate that the mask shape matches (n_objects, n_keypoints) before
processing. A row-count mismatch now raises ValueError with a clear message
instead of a confusing IndexError; a column-count mismatch similarly raises
ValueError instead of silently producing incorrect results.

[resolve #4] /review finding by sw-engineer (report: _outputs/2026/03/output-review-2026-03-30-5.md):
"No shape validation between index and self.xy — mismatched masks silently misfilter or raise IndexError"

---
Co-authored-by: Claude Code <noreply@anthropic.com>

…egression

- [resolve #4] Add test asserting detection.mask is None when force_masks=True, is_obb=True
- [resolve #5] Add regression test confirming is_obb=False with polygon annotation still produces masks

---
Co-authored-by: Claude Code <noreply@anthropic.com>

- Import deprecated_parameter from supervision.utils.internal
- Decorate __init__ to emit SupervisionWarning when frame_rate= is passed
- Maps frame_rate value to lost_track_buffer via int() for backward compatibility

[resolve #4] /review finding by sw-engineer (report: .temp/output-review-fix-l100-003-bytetrack-frame-rate-2026-04-13.md): No deprecation path for external users — ByteTrack is a public API...

---
Co-authored-by: Claude Code <noreply@anthropic.com>

- Case 4: [TEST_DET_NONE] now returns Detections.empty() — zero-length xyxy is empty
- Case 8: [TEST_DET_ZERO_LENGTH x2] now returns Detections.empty() — same reason
- Case 9: [TEST_DET_1, TEST_DET_NONE] now returns TEST_DET_1 with DoesNotRaise() —
  the ValueError was an artifact of the bug; empty operand is now stripped before
  merge's field-compatibility check, so merge succeeds cleanly

[resolve roboflow#2] /review finding by sw-engineer + qa-specialist (report: .temp/output-review-fix-is-empty-tracker-id-2026-04-13.md)
[resolve roboflow#4] /review finding by sw-engineer (report: .temp/output-review-fix-is-empty-tracker-id-2026-04-13.md)

---
Co-authored-by: Claude Code <noreply@anthropic.com>

The fixture created a real video at 23.976 fps but was orphaned when
the test_video_info_float_fps test was updated to use monkeypatching
instead (commit 65f4df3), leaving the fixture unreachable by any test.

[resolve roboflow#4] /review finding by sw-engineer (report: .temp/output-review-fix-video-info-fps-float-2026-04-13.md):
"float_fps_video_path at tests/utils/test_video.py:22 creates a real video but no test uses it"

---
Co-authored-by: Claude Code <noreply@anthropic.com>

Exercises all three dispatch branches of the custom_data handler
simultaneously: np.ndarray values are sliced per detection row while
scalar values (int, str, etc.) are broadcast as-is to every row.

[resolve roboflow#4] /review finding by qa-specialist (report: .temp/output-review-fix-csv-json-sink-custom-data-array-slicing-2026-04-13.md): Missing test for mixed-type custom_data

---
Co-authored-by: Claude Code <noreply@anthropic.com>

- Replace FutureWarning with warn_deprecated (SupervisionWarnings) in
  Detections.from_lmm to respect SUPERVISON_DEPRECATION_WARNING env var
- Align removal version from 0.30.0 to 0.31.0 for consistency with
  all other deprecations introduced in this PR

[resolve #3] Review comment by @copilot (PR #2214):
"Deprecation warning says from_lmm will be removed in supervision-0.30.0..."
[resolve #4] Review comment by @copilot (PR #2214):
"Emitting deprecation warnings with the standard FutureWarning here byp..."

---
Co-authored-by: Claude Code <noreply@anthropic.com>

Replace FutureWarning with warn_deprecated so the LMM deprecation
respects the SUPERVISON_DEPRECATION_WARNING env var filter, consistent
with other deprecations across the library.

[resolve #4] Review comment by @copilot (PR #2214):
"Same as Detections.from_lmm: this uses the standard FutureWarning..."

---
Co-authored-by: Claude Code <noreply@anthropic.com>

- Replace FutureWarning with warn_deprecated (SupervisionWarnings) in
  Detections.from_lmm to respect SUPERVISON_DEPRECATION_WARNING env var
- Align removal version from 0.30.0 to 0.31.0 for consistency with
  all other deprecations introduced in this PR

[resolve #3] Review comment by @copilot (PR #2214):
"Deprecation warning says from_lmm will be removed in supervision-0.30.0..."
[resolve #4] Review comment by @copilot (PR #2214):
"Emitting deprecation warnings with the standard FutureWarning here byp..."

---
Co-authored-by: Claude Code <noreply@anthropic.com>

Replace FutureWarning with warn_deprecated so the LMM deprecation
respects the SUPERVISON_DEPRECATION_WARNING env var filter, consistent
with other deprecations across the library.

[resolve #4] Review comment by @copilot (PR #2214):
"Same as Detections.from_lmm: this uses the standard FutureWarning..."

---
Co-authored-by: Claude Code <noreply@anthropic.com>

- Verify CSVSink/JSONSink slice list values from detections.data per row

[resolve roboflow#4] /review finding by foundry:qa-specialist (report: .temp/output-review-fix-sinks-slice-list-values-2026-04-17.md)

---
Co-authored-by: Claude Code <noreply@anthropic.com>
Co-authored-by: OpenAI Codex <codex@openai.com>

… in fallback

- Wrap splprep/splev in try/except ValueError so collinear unique
  points (e.g. perfectly straight-line motion) fall back to the raw
  polyline instead of crashing.
- Use unique_xy (deduplicated) instead of raw xy for the fallback
  spline_points assignment, making both paths semantically consistent.

[resolve roboflow#3] /review finding by sw-engineer (report: collinear splprep crash)
[resolve roboflow#4] /review finding by sw-engineer (report: fallback uses un-deduplicated xy)

---
Co-authored-by: Claude Code <noreply@anthropic.com>

[resolve #4] /review finding by oss (report: .temp/output-review-rle-refactor-2026-04-22.md):
"CHANGELOG: Not updated. Should be updated before merge..."

---
Co-authored-by: Claude Code <noreply@anthropic.com>

…ounts

- Calling .view(np.uint8) on a non-1-byte dtype (int32, float32) reinterprets
  the underlying bytes and produces incorrect run boundaries (silent mis-encoding).
  Using np.asarray(mask_2d, dtype=np.bool_) before ravel() keeps the
  fast uint8-view path while accepting any truthy array.

[resolve #4] Review comment by @Copilot (PR #2230):
"_mask_to_rle_counts computes np.diff(flat.view(np.uint8)) without ensuring
flat is a 1-byte-per-element boolean/uint8 array..."

---
Co-authored-by: Claude Code <noreply@anthropic.com>

…rity

- _resize_dense_to_shape: switch from np.linspace to arange * src // dst
  floor-division indexing, matching _rle_resize strategy and eliminating
  the 1-px rounding mismatch that made correctness checks compare two
  different nearest-neighbour implementations
- stage_resize: pass parallel=1 to time_reps for compact timing to avoid
  nested ThreadPoolExecutor oversubscription (CompactMask.resize already
  uses its own thread pool for N >= _PARALLEL_THRESHOLD)
- stage_resize: update docstring to correctly describe the sparse/dense
  dispatch paths in CompactMask.resize() rather than "decode/re-encode"

[resolve #4] @Copilot (gh): "stage_resize docstring says compact path decodes each RLE crop..."
[resolve #5] @Copilot (gh): "time_reps() can run concurrently in a thread pool..."
[resolve #13] /review by perf-optimizer: "np.linspace rounds differently than cv2 floor-division"

---
Co-authored-by: Claude Code <noreply@anthropic.com>

[resolve #4] Review comment by @Copilot (PR #2232):
"The JSON-LD SearchAction target is hard-coded to the /latest/ search path. On /develop/ or tag builds, this points to a different version..."

- Replace hard-coded /latest/search/ with {{ config.site_url }}search/
- Now resolves correctly for each deployed version

---
Co-authored-by: Claude Code <noreply@anthropic.com>

…ame-dir for atomic replace

- Move `tempfile.mkstemp` + `os.close` inside `try` block so OSError (disk full,
  unwritable dir) is caught by the existing `except Exception` handler instead of
  propagating to the caller, preserving the warn-and-degrade contract
- Pass `dir=os.path.dirname(os.path.abspath(video_path))` so the temp file is on
  the same filesystem as the output, restoring `os.rename` semantics in `shutil.move`
- Initialise `tmp_path = None` before `try`; guard `finally` with
  `tmp_path is not None` to satisfy mypy and avoid referencing an unbound name

---
Co-authored-by: Claude Code <noreply@anthropic.com>

Precision, Recall, F1Score, MeanAverageRecall all raise ValueError when
predictions/targets.class_id is None or predictions.confidence is None.

[resolve #4] /review finding by foundry:qa-specialist (report: .reports/review/2026-05-22T05-56-29Z/review-report.md)

---
Co-authored-by: Claude Code <noreply@anthropic.com>
Co-authored-by: OpenAI Codex <codex@openai.com>

…annotations

Chaining fixes integer ID uniqueness only; file_name uses bare basename so
splits with identical filenames still collide when COCO files are merged.
Note added to Returns section to prevent false assumptions.

[resolve roboflow#4] /review finding by foundry:challenger (report: .reports/review/2026-05-22T18-44-11Z/review-report.md):

---
Co-authored-by: Claude Code <noreply@anthropic.com>

Add null-safety assertion (class_name is not None), full key-set + dtype parity check
across all data keys, and concatenation assertion covering the downstream mixed-batch
scenario. Replaces single dtype.kind check with structural invariant that guards future
key or dtype drift.

[resolve roboflow#4] [resolve roboflow#5] [resolve roboflow#8] /review findings by foundry:qa-specialist and foundry:challenger (report: .reports/review/2026-05-27T06-06-27Z/review-report.md)

---
Co-authored-by: Claude Code <noreply@anthropic.com>

[resolve #4] Review by @Copilot (PR #2290):
"KeyPoints.from_detections indexes kp.shape[2] without ndim/shape check..."
Challenge: evidence=VALID suggestion=VALID resolution=as-suggested

---
Co-authored-by: Claude Code <noreply@anthropic.com>
Co-authored-by: OpenAI Codex <codex@openai.com>

- [resolve roboflow#4] test_with_nmm_falls_back_to_box_nmm_without_obb_data: regression
  guard that non-OBB Detections still collapse via box NMM path.
  Challenge: evidence=VALID suggestion=VALID resolution=as-suggested
- [resolve roboflow#10] boundary tests: empty/single, iou_threshold=0.0/1.0, IOS metric
  end-to-end NMS, and OBB NMM empty/single — converted threshold test to
  parametrize(expected_keep) removing if-branch from test body.
  Challenge: evidence=VALID suggestion=VALID resolution=as-suggested
- Added test_with_nmm_obb_merged_xyxy_matches_winner_aabb verifying the
  item roboflow#3 contract: merged xyxy == winner's AABB, not union of all members.

[resolve group] PR roboflow#2303 — items 4, 10

---
Co-authored-by: claude[bot] <209825114+claude[bot]@users.noreply.github.com>
Co-authored-by: OpenAI Codex <codex@openai.com>

…d placement

- Fix LineZone listed under detection/tools/ — lives in detection/line_zone.py; add as separate tree entry
- Fix unsafe git stash baseline: save refs/stash before push, only pop if ref changed after
- Clarify warn_deprecated goes in deprecated module's own __init__.py, not supervision/__init__.py
- Item #4 (style: commit type) already addressed: commit type list removed; CONTRIBUTING.md cross-ref covers it

---
Co-authored-by: claude[bot] <209825114+claude[bot]@users.noreply.github.com>