Skip to content

INDEX.md

Latest commit

 

History

History
83 lines (67 loc) · 5.06 KB

INDEX.md

File metadata and controls

83 lines (67 loc) · 5.06 KB
Raw

Examples index

Lightweight map for agents and reviewers. Full policy: README.md. Pack file names: docs/replay.md.

Do not read every file. Start here, then open one anchor pack.

evidence_packs/ -- canonical anchors

Pack Skill Verdict Teaches
dicom_metadata_pass/ dicom_metadata_extract pass Baseline DICOM metadata + gates
dicom_metadata_trusted_warn/ dicom_metadata_extract + dicom_metadata_quality_v1 warn Trusted run with PHI-tag advisory warning
dicom_invalid_input_fail/ dicom_metadata_extract fail Invalid input / preflight
dicom_series_preflight_trusted_pass/ dicom_series_preflight + dicom_preflight_quality_v1 pass Trusted DICOM preflight run with verifier pack and trust summary
dicom_series_to_volume_pass/ dicom_series_to_volume pass DICOM series to NIfTI
dicom_series_to_volume_trusted_pass/ dicom_series_to_volume + dicom_volume_quality_v1 pass Trusted conversion with NIfTI artifact verifier
nv_segment_ct_pass/ nv_segment_ct pass Segmentation happy path
nv_segment_ct_trusted_pass/ nv_segment_ct + ct_segmentation_quality_v1 pass Trusted CT segmentation with anatomy plausibility verifier
nv_segment_ctmr_trusted_pass/ nv_segment_ctmr + ct_segmentation_quality_v1 pass Trusted NV-Segment-CTMR CT-body CUDA run with generated label map referenced but not bundled
nv_segment_ct_finetune_trusted_smoke_pass/ nv_segment_ct_finetune + ct_segmentation_finetune_quality_v1 pass Trusted MONAI bundle smoke finetune run with checkpoint-load, finite-loss, no-OOM, and trajectory verifier
nv_segment_ct_integrity_fail/ nv_segment_ct fail Integrity scan
nv_segment_ct_silent_failure_fail/ nv_segment_ct fail Silent failure gate
nv_generate_ct_rflow_pass/ nv_generate_ct_rflow pass CT synthesis wrapper happy path
nv_generate_ct_rflow_trusted_inventory_pass/ nv_generate_ct_rflow + ct_synthesis_quality_v1 pass Trusted CUDA rflow-ct synthesis inventory run with geometry, HU, label-set, and hash evidence
nv_generate_mr_trusted_inventory_pass/ nv_generate_mr + mr_synthesis_quality_v1 pass Trusted CUDA rflow-mr synthesis inventory run with image bytes, hashes, and numeric sanity
nv_generate_mr_brain_trusted_inventory_pass/ nv_generate_mr_brain + mr_synthesis_quality_v1 pass Trusted CUDA rflow-mr-brain synthesis inventory run with image bytes, hashes, and numeric sanity
nv_reason_cxr_trusted_mock_pass/ nv_reason_cxr + nv_reason_cxr_quality_v1 pass Trusted mock CXR reasoning path with image/hash, runtime identity, and forbidden-phrase verifier
benchmark_decathlon_spleen_clean/ nv_segment_ct pass Benchmark loop clean
benchmark_decathlon_with_corruption/ nv_segment_ct fail Benchmark corruption
benchmark_ct_segmentation_spleen_msd09_pass/ nv_segment_ct pass CT segmentation benchmark evidence
ct_segmentation_finetune_quality_v1_pass/ ct_segmentation_finetune_quality_v1 pass Verifier-only anchor for finetune checkpoint, trajectory, dataset-audit, and label-coverage checks

studies/ -- narrative

Study Topic
runs/with_vs_without_nv/studies/ Local generated with-vs-without study records, gitignored
docs/with-vs-without-skill-experiment.md Checked-in compact with-vs-without summary

dependency_locks/ -- tested resolutions

Lock Scope
model-skills-2026-07-08.json Exact tested dependency profiles and end-to-end outcomes for the current 12-skill catalog

workflows/

Workflow Steps Teaches
dicom_preflight_gate.yaml trusted preflight GPU-free pass/warn/fail preflight
ct_dicom_to_segmentation_evidence.yaml convert -> trusted segment DICOM to NIfTI to segment plus ct_segmentation_quality_v1 trust summary
orientation_safe_segmentation.yaml same as flagship Orientation gate halts before VISTA3D

drift/

Directory Lesson
baseline/ Anchor pack for drift comparison
environment_drift/ Dependency / env change
repeated_no_drift/ Repeat run stability

Reading order

  1. python tools/render_review_packet.py <pack>
  2. workflow_run_record.md
  3. validation_summary.json
  4. output.json if present
  5. manifest.json
  6. replay.sh for optional rerun

Commands

make review-packet PACK=examples/evidence_packs/dicom_series_preflight_trusted_pass
make review-packet PACK=examples/evidence_packs/nv_segment_ct_trusted_pass
python tools/render_review_packet.py examples/evidence_packs/dicom_metadata_trusted_warn
python tools/render_review_packet.py examples/evidence_packs/dicom_series_to_volume_trusted_pass
python tools/render_review_packet.py examples/evidence_packs/nv_generate_ct_rflow_trusted_inventory_pass
python tools/render_review_packet.py examples/evidence_packs/nv_reason_cxr_trusted_mock_pass
make diff RUN_A=examples/evidence_packs/dicom_metadata_pass RUN_B=runs/my_demo
cd examples/evidence_packs/dicom_metadata_pass && ./replay.sh