R09.01.01.03 · TASK CLASS

Whole-slide pathology annotation

Navigate gigapixel slides and annotate regions to reference labels.

20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.

Typical applications: QuPath · ASAP · Aperio ImageScope
Class GUI-need label: G3 · GUI-only

Task proposals · 20

Expand to inspect the complete original wording and proposed checks. Original proposal text remains English in both interfaces.

R09.01.01.03-T01 Outline all metastatic foci in a lymph-node whole-slide image in QuPath

Original instruction

QuPath project input/ln_project/ contains LN_07.svs (H&E sentinel lymph node, public-dataset style). Find every metastatic focus, outline each with the brush or polygon tool, classify it 'Tumor', and export all annotations to output/LN_07.geojson.

Platforms: mac · macOS win · Windows
Applications: QuPath
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: path · Continuous path nav · Pan/zoom/scroll vis · Visual judgment
Work product: WP10 · Annotations
Conditions: None declared
Proposed difficulty: 4 / 5

Proposed verifier

Lesion-level detection F1>=0.8 (a lesion counts when a Tumor annotation centroid lies inside the gold polygon) and pixel Dice>=0.5 at level 4. QuPath script editor is disabled and scripting events in the project log fail the task.

Proposed partial credit

15% annotations exported with class; 45% lesion detection F1; 40% Dice

R09.01.01.03-T02 Find the mitotic hotspot and count mitoses in 10 HPFs in QuPath

Original instruction

In QuPath open breast carcinoma slide BRCA_112.svs. Find the area with the highest mitotic activity at the tumour periphery, place 10 contiguous high-power-field annotations (0.237 mm^2 each) classified 'HPF', mark each mitotic figure with a Points annotation 'Mitosis', and export output/mitoses.geojson plus total count in output/count.txt.

Platforms: mac · macOS win · Windows
Applications: QuPath
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: nav · Pan/zoom/scroll vis · Visual judgment click · Point & click
Work product: WP10 · Annotations
Conditions: long · Long-horizon
Proposed difficulty: 5 / 5

Proposed verifier

HPF set must overlap the gold hotspot (>=50% of fields inside); point-level F1>=0.7 vs pathologist mitoses within the chosen HPFs (8 um match radius); total count within +-3 of gold hotspot count; field areas within 2%. QuPath's script editor is disabled and the VM has no shell for the QuPath CLI, Python or OpenSlide, so annotations can only be made in the viewer; project writes by other processes are caught by file-access auditing.

Proposed partial credit

20% 10 HPFs correct size; 30% hotspot location; 35% point F1; 15% total count

R09.01.01.03-T03 Compute a Ki-67 index in a hotspot with positive cell detection in QuPath

Original instruction

In QuPath set image type Brightfield H-DAB for KI67_09.svs, estimate stain vectors from a representative region, draw a 1 mm^2 annotation 'Hotspot' over the highest proliferation area, run Positive cell detection with a DAB threshold you tune visually, and record the Ki-67 index in output/ki67.json. Save the project.

Platforms: mac · macOS win · Windows
Applications: QuPath
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: path · Continuous path vis · Visual judgment click · Point & click
Work product: WP12 · Answer/decision
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

Hotspot annotation area 0.95-1.05 mm^2 overlapping the gold hotspot; Ki-67 index within 3 percentage points of the pathologist reference for that region; detection settings saved in the project workflow. QuPath's script editor is disabled and the VM has no shell for the QuPath CLI or Python, so the analysis must be run in the GUI; project writes by other processes are caught by file-access auditing.

Proposed partial credit

15% image type and stain vectors; 25% hotspot placement; 60% Ki-67 index

R09.01.01.03-T04 Measure Breslow thickness on a melanoma slide in Aperio ImageScope

Original instruction

In Aperio ImageScope open MEL_21.svs. Measure Breslow thickness with the ruler tool from the top of the granular layer to the deepest invasive melanoma cell, save the annotation layer to output/MEL_21.xml and write the value in mm to output/breslow.txt.

Platforms: win · Windows
Applications: Aperio ImageScope
Interfaces: desk · Desktop GUI
GUI need · task setting: G3 · GUI-only
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: geom · Precise geometry nav · Pan/zoom/scroll vis · Visual judgment
Work product: WP12 · Answer/decision
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

Value within 0.1 mm of dermatopathologist gold; ruler endpoints within 50 um of the gold start and deepest-point annotations; XML layer contains the ruler. The VM has no shell, text editor, Python or OpenSlide, so the slide and annotation XML are reachable only in ImageScope; file writes by other processes are caught by file-access auditing.

Proposed partial credit

60% thickness value; 30% endpoints placement; 10% XML saved

R09.01.01.03-T05 Annotate tumour regions on three slides with polygons in ASAP

Original instruction

In ASAP open slides CRC_01..03.tif (colorectal resections). Outline invasive tumour regions with the polygon tool in group 'tumor', and stroma-free mucosa in group 'normal', then save each as output/CRC_0N.xml.

Platforms: win · Windows lin · Linux
Applications: ASAP
Interfaces: desk · Desktop GUI
GUI need · task setting: G3 · GUI-only
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: path · Continuous path nav · Pan/zoom/scroll vis · Visual judgment
Work product: WP10 · Annotations
Conditions: None declared
Proposed difficulty: 4 / 5

Proposed verifier

Per slide tumour Dice>=0.7 and normal Dice>=0.5 at level 3 vs gold; XML schema valid with correct group names; ASAP has no scripting and file edits are detected by missing ASAP XML metadata.

Proposed partial credit

60% tumour Dice (equal share per slide); 25% normal Dice; 15% valid XML groups

R09.01.01.03-T06 Detect tissue and remove pen-mark and fold artifacts in QuPath

Original instruction

In QuPath, run simple tissue detection on biopsy slide BX_44.svs (threshold of your choice), then edit the resulting annotation to remove the blue pen marks, the tissue fold and out-of-focus edge, and classify the final annotation 'Tissue'. Export output/BX_44_tissue.geojson.

Platforms: mac · macOS win · Windows
Applications: QuPath
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: path · Continuous path vis · Visual judgment click · Point & click
Work product: WP10 · Annotations
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

IoU>=0.9 vs gold artifact-free tissue mask; pen-mark and fold gold masks overlap the result by <5%; class 'Tissue'. QuPath's script editor is disabled and the VM has no shell for the QuPath CLI, Python or OpenSlide, so editing happens only in the viewer; project writes by other processes are caught by file-access auditing.

Proposed partial credit

40% tissue IoU; 25% pen marks removed; 25% fold removed; 10% class

R09.01.01.03-T07 Dearray a tissue microarray and fix misplaced cores in QuPath

Original instruction

In QuPath open TMA slide TMA_B2.svs (12 x 8 grid). Run the TMA dearrayer with the core diameter from input/tma_spec.txt, move the 5 cores that are misplaced, mark the 3 missing cores as missing, and add core IDs from input/tma_map.csv. Export the TMA map to output/tma_cores.csv.

Platforms: mac · macOS win · Windows
Applications: QuPath
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: drag · Drag & drop vis · Visual judgment click · Point & click
Work product: WP10 · Annotations
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

Exported core centres within 100 um of gold for all 96 positions; missing flags exactly the 3 gold cores; core IDs exact per grid position. QuPath's script editor is disabled and the VM has no shell for the QuPath CLI, Python or OpenSlide, so cores can only be fixed in the viewer; project writes by other processes are caught by file-access auditing.

Proposed partial credit

15% grid dearrayed; 45% core positions; 20% missing flags; 20% IDs

R09.01.01.03-T08 Annotate tumour, stroma and necrosis layers with holes in ImageScope

Original instruction

In Aperio ImageScope open GBM_05.svs and create three annotation layers named Tumor, Stroma and Necrosis. Outline each region with the pen tool, using negative regions to exclude embedded necrosis from tumour areas. Save output/GBM_05.xml.

Platforms: win · Windows
Applications: Aperio ImageScope
Interfaces: desk · Desktop GUI
GUI need · task setting: G3 · GUI-only
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: path · Continuous path nav · Pan/zoom/scroll vis · Visual judgment
Work product: WP10 · Annotations
Conditions: None declared
Proposed difficulty: 4 / 5

Proposed verifier

Rasterized at level 2 with negative regions subtracted: per-class Dice Tumor>=0.7, Stroma>=0.5, Necrosis>=0.6 vs gold; layer names exact. The VM has no shell, text editor, Python or OpenSlide, so the slide and annotation XML are reachable only in ImageScope; file writes by other processes are caught by file-access auditing.

Proposed partial credit

35% tumour; 25% stroma; 25% necrosis; 15% negative regions used correctly

R09.01.01.03-T09 Classify 40 pre-drawn tiles by tissue type in QuPath

Original instruction

QuPath project tiles_project contains slide LUNG_08.svs with 40 unclassified square annotations. Inspect each at high magnification and set its class to Tumor, Stroma, Immune or Other (necrosis, blood, mucus). Save the project and export output/tile_classes.csv.

Platforms: mac · macOS win · Windows
Applications: QuPath
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: nav · Pan/zoom/scroll vis · Visual judgment click · Point & click
Work product: WP10 · Annotations
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

Accuracy>=0.85 vs pathologist labels; annotation geometries unchanged; CSV matches the project's stored classes. QuPath's script editor is disabled and the VM has no shell for the QuPath CLI, Python or OpenSlide, so tiles can only be classified in the viewer; project writes by other processes are caught by file-access auditing.

Proposed partial credit

80% classification accuracy scaled to 0.85; 10% geometry unchanged; 10% CSV consistent

R09.01.01.03-T10 Point-annotate tumour-infiltrating lymphocytes in five ROIs in ASAP

Original instruction

In ASAP open BRCA_TIL_03.tif with 5 pre-drawn rectangular ROIs. Place a dot annotation on every stromal lymphocyte inside each ROI (group 'TIL'), then save output/TIL_03.xml and write per-ROI counts to output/til_counts.csv.

Platforms: win · Windows lin · Linux
Applications: ASAP
Interfaces: desk · Desktop GUI
GUI need · task setting: G3 · GUI-only
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: click · Point & click nav · Pan/zoom/scroll vis · Visual judgment
Work product: WP10 · Annotations
Conditions: long · Long-horizon
Proposed difficulty: 3 / 5

Proposed verifier

Per ROI count within 10% of gold; point-level F1>=0.75 (6 um match radius); counts in CSV equal saved points. The VM has no shell or Python, so ASAP's multiresolutionimageinterface bindings and OpenSlide are unusable; annotation XML writes by any process other than ASAP are caught by file-access auditing.

Proposed partial credit

50% point F1; 40% per-ROI counts; 10% CSV consistent

R09.01.01.03-T11 Train a tumour-stroma pixel classifier interactively in QuPath

Original instruction

In QuPath create a pixel classifier for slide HCC_02.svs: draw training scribbles for Tumor, Stroma and Ignore* (background/whitespace) at moderate resolution, iterate with the live preview until boundaries look right, save the classifier 'tumor_stroma', apply it inside the tissue annotation and report tumour area fraction in output/fraction.json.

Platforms: mac · macOS win · Windows
Applications: QuPath
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: path · Continuous path vis · Visual judgment click · Point & click
Work product: WP10 · Annotations
Conditions: None declared
Proposed difficulty: 4 / 5

Proposed verifier

Saved classifier exists in the project; checker applies it to 3 held-out regions: tumour pixel Dice>=0.75 vs gold; reported tumour fraction within 5 percentage points of gold. QuPath's script editor is disabled and the VM has no shell for the QuPath CLI or Python, so training happens only in the viewer; project writes by other processes are caught by file-access auditing.

Proposed partial credit

20% classifier saved with 3 classes; 40% held-out Dice; 40% tumour fraction

R09.01.01.03-T12 Navigate to slide coordinates and capture a 20x snapshot in ImageScope

Original instruction

In Aperio ImageScope open PROST_17.svs, navigate to x=41,200 um, y=18,750 um (top-left origin) at 20x magnification and save a 1000x1000 px snapshot of that location centred on the point to output/snap.png.

Platforms: win · Windows
Applications: Aperio ImageScope
Interfaces: desk · Desktop GUI
GUI need · task setting: G3 · GUI-only
GUI need · ordinary access: CL · CLI-sufficient
Interaction primitives: nav · Pan/zoom/scroll click · Point & click
Work product: WP09 · Files & folders
Conditions: None declared
Proposed difficulty: 2 / 5

Proposed verifier

Snapshot size 1000x1000; SSIM>=0.9 vs reference extracted with OpenSlide at the same centre and 20x; centre offset <20 um estimated by template matching. The VM has no shell, Python or OpenSlide, so slide pixels are reachable only in ImageScope; file writes by any process other than ImageScope are caught by file-access auditing.

Proposed partial credit

20% size; 40% location offset; 40% SSIM

R09.01.01.03-T13 Measure the closest tumour-to-margin distance in QuPath

Original instruction

In QuPath open excision slide BCC_30.svs, which has a pre-drawn polyline annotation 'Inked margin'. Outline the basal cell carcinoma as annotation 'Tumor', then measure the shortest distance from tumour to the inked margin with a line annotation and report it in mm in output/margin.json.

Platforms: mac · macOS win · Windows
Applications: QuPath
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: path · Continuous path geom · Precise geometry vis · Visual judgment
Work product: WP12 · Answer/decision
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

Reported distance within 0.2 mm of gold; line endpoints on the tumour boundary and margin polyline (within 30 um); tumour annotation Dice>=0.6. QuPath's script editor is disabled and the VM has no shell for the QuPath CLI, Python or OpenSlide, so measuring happens only in the viewer; project writes by other processes are caught by file-access auditing.

Proposed partial credit

25% tumour annotation; 25% line endpoints; 50% distance

R09.01.01.03-T14 Phenotype CD8 and PD-L1 cells on a multiplex IF slide in QuPath

Original instruction

In QuPath open 4-channel immunofluorescence slide MIF_12.ome.tiff (DAPI, CD8, PD-L1, panCK). Set channel display ranges, run cell detection on DAPI in the 'Tumor region' annotation, create single-measurement classifiers for CD8 and PD-L1 and a composite classifier, and export per-phenotype counts to output/phenotypes.csv.

Platforms: mac · macOS win · Windows
Applications: QuPath
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: CL · CLI-sufficient
Interaction primitives: vis · Visual judgment click · Point & click type · Text & forms
Work product: WP10 · Annotations
Conditions: tutorial · Follow docs/tutorial
Proposed difficulty: 5 / 5

Proposed verifier

Counts for CD8+, PD-L1+, double-positive and negative within 10% of gold; classifiers saved in the project; detected cell count within 10% of gold nuclei count. QuPath's script editor is disabled and the VM has no shell for the QuPath CLI or Python, so the analysis must be run in the GUI; project writes by other processes are caught by file-access auditing.

Proposed partial credit

20% cell detection; 20% classifiers saved; 60% phenotype counts (equal share each)

R09.01.01.03-T15 Accept or reject AI-proposed regions on a slide in ImageScope

Original instruction

Aperio ImageScope opens LN_15.svs with an annotation layer 'AI candidates' of 25 proposed metastasis regions. Review each at high magnification; move true metastases to a new layer 'Confirmed' and delete false positives (sinus histiocytes, germinal centres). Save output/LN_15_review.xml.

Platforms: win · Windows
Applications: Aperio ImageScope
Interfaces: desk · Desktop GUI
GUI need · task setting: G3 · GUI-only
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: nav · Pan/zoom/scroll vis · Visual judgment click · Point & click
Work product: WP10 · Annotations
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

Set of confirmed candidate IDs vs gold: precision>=0.85 and recall>=0.85; remaining candidates deleted; geometries of confirmed regions unchanged. The VM has no shell, text editor, Python or OpenSlide, so the slide and annotation XML are reachable only in ImageScope; file writes by other processes are caught by file-access auditing.

Proposed partial credit

45% precision; 45% recall; 10% geometries intact

R09.01.01.03-T16 Control pair of T12: extract the slide region with OpenSlide

Original instruction

Control pair of T12. Using Python and OpenSlide, extract a 1000x1000 px region of PROST_17.svs at 20x centred on x=41,200 um, y=18,750 um and save output/snap.png.

Platforms: lin · Linux mac · macOS win · Windows
Applications: Python · OpenSlide
Interfaces: term · Terminal
GUI need · task setting: CL · CLI-sufficient
GUI need · ordinary access: CL · CLI-sufficient
Interaction primitives: code · Code & commands
Work product: WP09 · Files & folders
Conditions: None declared
Proposed difficulty: 2 / 5

Proposed verifier

Same checks as T12: size, centre offset <20 um, SSIM>=0.9 vs reference; level/downsample selection correct for 20x.

Proposed partial credit

20% size; 40% location; 40% SSIM

R09.01.01.03-T17 Annotate Gleason patterns on prostate biopsy cores and assign grade group

Original instruction

In QuPath open PBX_06.svs (6 biopsy cores). Annotate Gleason pattern 3, 4 and 5 areas with the matching classes, compute the percentage of pattern 4 per core, and write output/gleason.json with per-core primary/secondary patterns, % pattern 4 and the overall ISUP grade group.

Platforms: mac · macOS win · Windows
Applications: QuPath
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: path · Continuous path nav · Pan/zoom/scroll vis · Visual judgment
Work product: WP10 · Annotations
Conditions: long · Long-horizon
Proposed difficulty: 5 / 5

Proposed verifier

Per-class Dice vs consensus gold (pattern 3>=0.5, 4>=0.5); per-core patterns exact; % pattern 4 within 10 points; grade group exact. Annotations must exist in the project (not only the JSON). QuPath's script editor is disabled and the VM has no shell for the QuPath CLI, Python or OpenSlide, so annotating happens only in the viewer; project writes by other processes are caught by file-access auditing.

Proposed partial credit

35% pattern annotations; 25% per-core patterns; 15% % pattern 4; 25% grade group

R09.01.01.03-T18 Correct a junior annotator's tumour outlines in ASAP

Original instruction

In ASAP open OV_14.tif with input/OV_14_junior.xml loaded. Fix the 6 boundary errors (tumour extending into stroma, missed nests) and relabel the 2 regions wrongly assigned to group 'tumor' that are endometriosis. Save output/OV_14_corrected.xml.

Platforms: win · Windows lin · Linux
Applications: ASAP
Interfaces: desk · Desktop GUI
GUI need · task setting: G3 · GUI-only
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: path · Continuous path drag · Drag & drop vis · Visual judgment
Work product: WP10 · Annotations
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

Tumour Dice vs gold improves by >=0.1 over the junior file and reaches >=0.75; the 2 regions now in group 'other'; untouched correct regions remain within 1% area. The VM has no shell or Python, so ASAP's multiresolutionimageinterface bindings and OpenSlide are unusable; annotation XML writes by any process other than ASAP are caught by file-access auditing.

Proposed partial credit

50% boundary fixes (Dice gain); 30% relabeled regions; 20% correct regions intact

R09.01.01.03-T19 Set up a QuPath project with image types and stain vectors

Original instruction

Create QuPath project output/her2_project/ with the 8 slides in input/her2_slides/, set image type Brightfield (H-DAB) for all, estimate stain vectors from a representative region on slide 1 and apply them to all slides, and add metadata 'block' and 'stain_batch' from input/slides.csv.

Platforms: mac · macOS win · Windows
Applications: QuPath
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: click · Point & click type · Text & forms drag · Drag & drop
Work product: WP08 · App/system state
Conditions: None declared
Proposed difficulty: 2 / 5

Proposed verifier

Project file lists 8 images; each image data has type BRIGHTFIELD_H_DAB and identical stain vectors (within 0.02 of reference estimate from the gold region); metadata values exact. QuPath's script editor is disabled and the VM has no shell for the QuPath CLI or Python, so the project must be set up in the GUI; project writes by other processes are caught by file-access auditing.

Proposed partial credit

25% 8 images added; 20% image types; 35% stain vectors; 20% metadata

R09.01.01.03-T20 Count tumour buds at the invasive front hotspot in QuPath

Original instruction

In QuPath open colorectal cancer slide CRC_BUD_04.svs. Scan the invasive front, choose the hotspot, place a 0.785 mm^2 circular annotation 'Hotspot', mark each tumour bud (single cells or clusters of up to 4 cells) with Points 'Bud', and write the count and ITBCC budding grade (Bd1-Bd3) to output/budding.json.

Platforms: mac · macOS win · Windows
Applications: QuPath
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: nav · Pan/zoom/scroll vis · Visual judgment click · Point & click
Work product: WP10 · Annotations
Conditions: None declared
Proposed difficulty: 4 / 5

Proposed verifier

Hotspot circle area within 2% and overlapping the gold hotspot; bud count within +-2 of gold in the chosen field; point F1>=0.7; grade exact. QuPath's script editor is disabled and the VM has no shell for the QuPath CLI, Python or OpenSlide, so counting happens only in the viewer; project writes by other processes are caught by file-access auditing.

Proposed partial credit

20% hotspot; 40% bud point F1; 20% count; 20% grade

Mapped research references

Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.

OSWorld-Science

Whole-slide pathology annotation

Task proposals · 20

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Original instruction

Proposed verifier

Proposed partial credit

Mapped research references

In this workflow family