R10.01.01.04 · TASK CLASS

Transcript editing synced to audio

Correct transcripts against recordings with accurate speaker labels and timestamps.

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

Typical applications: Eclipse CAT · Case CATalyst · oTranscribe
Class GUI-need label: G2 · GUI-essential

Task proposals · 20

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

R10.01.01.04-T01 Correct a five-minute interview transcript against its audio in oTranscribe

Original instruction

oTranscribe is open in the browser with input/interview_05min.mp3 (synthetic witness interview) and the ASR draft input/interview_draft.otr loaded. Correct the wording against the audio, keep the existing timestamps, label speakers 'INVESTIGATOR:' and 'WITNESS:', and export the result as output/interview_final.otr and output/interview_final.txt.

Platforms: web · Any OS (web)
Applications: oTranscribe
Interfaces: brow · Browser
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: av · Audio/video over time type · Text & forms key · Shortcuts
Work product: WP05 · Time-based media
Conditions: None declared
Proposed difficulty: 2 / 5

Proposed verifier

jiwer WER vs gold transcript <= 3% (draft starts at 11%); speaker label accuracy per turn >= 95%; the 22 original timestamps preserved within 0.2 s. The 12 seeded errors are audio-only (homophones, numbers), so text-only edits cannot reach the threshold. ASR engines are absent from the VM and outbound network is firewalled, so those errors can be fixed only by listening in oTranscribe.

Proposed partial credit

60% WER threshold (scaled); 25% speaker labels; 15% timestamps preserved.

R10.01.01.04-T02 Edit a 20-minute deposition draft to certified accuracy with audio sync

Original instruction

Case CATalyst job 'Marlow_Depo_0915' (synthetic) has a rough steno translation synced to input/marlow_audio.wav. Using the audio sync, fix untranslated steno, misstrokes and wrong words, make Q/A and colloquy formatting correct and keep the audio links intact. Save the job and export output/marlow.txt (ASCII, 25 lines per page).

Platforms: win · Windows
Applications: Case CATalyst
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G3 · GUI-only
Interaction primitives: av · Audio/video over time type · Text & forms key · Shortcuts
Work product: WP05 · Time-based media
Conditions: licensed · Licensed pro app
Proposed difficulty: 3 / 5

Proposed verifier

ASCII export: WER vs gold <= 2%; Q/A/colloquy line-type accuracy >= 98%; page:line count within 1 page of gold; untranslates remaining = 0. ASR engines are absent from the VM, and the job file must retain audio-sync marks, which only Case CATalyst writes; an export not matching the saved job is detected and fails.

Proposed partial credit

50% WER; 25% Q/A formatting; 15% no untranslates; 10% audio sync intact.

R10.01.01.04-T03 Reassign speaker turns in a four-party hearing transcript

Original instruction

Descript project 'Zoning_Hearing_0822' contains a recorded hearing with 4 speakers (chair, applicant, counsel, resident) and an auto-transcript with wrong speaker labels and merged turns. Using the audio, split merged turns at the right points and assign each turn to the correct speaker from input/appearance_sheet.pdf. Export output/hearing_transcript.docx with timestamps.

Platforms: mac · macOS win · Windows
Applications: Descript
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G3 · GUI-only
Interaction primitives: av · Audio/video over time click · Point & click type · Text & forms
Work product: WP05 · Time-based media
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

Word-level speaker attribution accuracy >= 97% vs gold (aligned by word timing); turn boundaries within 0.5 s; text WER not worse than the draft. Speaker identities are only distinguishable by voice; names appear only on the appearance sheet. Diarization and ASR tools are not installed on the VM, and the docx must match the speaker assignments saved in the Descript project.

Proposed partial credit

60% speaker attribution; 30% turn splits; 10% export with timestamps.

R10.01.01.04-T04 Insert exhibit-marking parentheticals at the audio-confirmed points

Original instruction

Eclipse CAT job 'Chen_Depo_0903' (synthetic) with synced audio. input/exhibits.txt lists 9 exhibits with approximate times. Listen around each time and insert the parenthetical '(Exhibit N marked for identification.)' at the exact transcript line where marking occurs, using the job's parenthetical style, and build the exhibit index page.

Platforms: win · Windows
Applications: Eclipse CAT
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G3 · GUI-only
Interaction primitives: av · Audio/video over time nav · Pan/zoom/scroll type · Text & forms
Work product: WP05 · Time-based media
Conditions: licensed · Licensed pro app
Proposed difficulty: 4 / 5

Proposed verifier

Export: 9 parentheticals with exact text, each placed within 1 line of the gold position (approximate times are off by up to 40 s); exhibit index lists page:line matching placements; no other text changes (diff outside parentheticals empty). ASR engines are absent from the VM, and the export must be generated by Eclipse from the saved job, whose audio-sync marks only the app writes.

Proposed partial credit

55% parenthetical placement; 25% index; 20% no collateral edits.

R10.01.01.04-T05 Timestamp every speaker change in an interview transcript

Original instruction

InqScribe document input/agency_interview.inqscr with the audio input/agency_interview.m4a and a plain transcript lacking timecodes. Insert a timecode at the start of every speaker turn (format [hh:mm:ss.ff]) aligned to when the speaker begins talking, then export output/agency_interview_timed.txt.

Platforms: mac · macOS win · Windows
Applications: InqScribe
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: CL · CLI-sufficient
Interaction primitives: av · Audio/video over time key · Shortcuts type · Text & forms
Work product: WP05 · Time-based media
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

Parsed timecodes for 84 turns vs gold onsets: mean absolute error <= 0.5 s and 95% within 1.0 s; transcript text unchanged except timecodes. ASR and forced-alignment tools are not installed on the VM, so onsets must be set by listening in InqScribe; the export must match the saved .inqscr document.

Proposed partial credit

70% timecode accuracy; 20% all turns stamped; 10% text unchanged.

R10.01.01.04-T06 Produce a certified deposition transcript from a one-hour rough draft

Original instruction

Eclipse CAT job 'Okafor_Depo_1002' (60 min audio, synthetic) has a rough draft. Produce the certified transcript: corrected text, speaker and examination headers, Q/A and colloquy formatting per input/state_format_rules.pdf (25 lines, 1-inch margins), index of examinations and exhibits, reporter certificate, word index. Export PDF and ASCII to output/.

Platforms: win · Windows
Applications: Eclipse CAT
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G3 · GUI-only
Interaction primitives: av · Audio/video over time type · Text & forms key · Shortcuts nav · Pan/zoom/scroll
Work product: WP05 · Time-based media
Conditions: licensed · Licensed pro app long · Long-horizon
Proposed difficulty: 5 / 5

Proposed verifier

ASCII: WER <= 2%, line-type accuracy >= 98%, 25 numbered lines/page; examination and exhibit indexes match gold page:line; certificate present with case caption; word index entries for the 30 gold terms. PDF text equals ASCII. ASR engines are not installed on the VM, and both exports must be generated by Eclipse from the saved job (audio-sync marks retained).

Proposed partial credit

45% text accuracy; 20% formatting rules; 20% indexes; 15% certificate and word index.

R10.01.01.04-T07 Resolve inaudible markers by listening to each spot

Original instruction

oTranscribe with input/call_0417.mp3 and input/call_0417.otr, which has 8 '[inaudible hh:mm:ss]' markers. Listen to each spot (slow playback allowed); replace the marker with the words where they can be made out, and keep '[inaudible]' where they cannot. Export output/call_0417_final.otr.

Platforms: web · Any OS (web)
Applications: oTranscribe
Interfaces: brow · Browser
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: av · Audio/video over time type · Text & forms
Work product: WP05 · Time-based media
Conditions: None declared
Proposed difficulty: 2 / 5

Proposed verifier

Gold: 5 resolvable spots with exact words, 3 truly masked by noise. Score: resolvable spots matched (word error <= 1 each); masked spots must stay '[inaudible]' (guessing counts as an error); rest of transcript unchanged. ASR engines are absent from the VM and outbound network is firewalled, so spots can be resolved only by listening in oTranscribe.

Proposed partial credit

60% resolvable spots; 30% masked spots left inaudible; 10% no collateral edits.

R10.01.01.04-T08 Fix drifting caption timing on a recorded hearing video

Original instruction

input/council_hearing.mp4 (32 min) has captions input/council_hearing.srt that start correct but drift 1.2 s late by the end, and one 40-second section after a recess break is offset by 6 s. In Subtitle Edit use the waveform and visual sync to correct timing so every cue starts within 150 ms of speech onset; keep cue text unchanged. Save output/council_hearing_fixed.srt.

Platforms: win · Windows
Applications: Subtitle Edit
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: CL · CLI-sufficient
Interaction primitives: av · Audio/video over time drag · Drag & drop vis · Visual judgment
Work product: WP05 · Time-based media
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

Cue start times vs gold: >= 95% within 150 ms, none off by > 500 ms; cue count and text unchanged; no overlapping cues. Linear shift alone fails the recess segment. Python and forced-alignment tools (ffsubsync, aeneas) are not installed on the VM, so timing must be fixed against the waveform in Subtitle Edit.

Proposed partial credit

50% global drift fixed; 35% recess segment fixed; 15% text and count unchanged.

R10.01.01.04-T09 Re-link transcript audio sync after the recording was trimmed

Original instruction

Case CATalyst job 'Ito_Depo_0828' was synced to the original audio. The reporter replaced it with input/ito_audio_trimmed.wav, which drops the first 37 s and a 2-minute off-record break at 00:41:10. Re-link the new audio so every line's sync point plays the right words, and confirm by spot-checking the 10 lines listed in input/spotcheck.txt.

Platforms: win · Windows
Applications: Case CATalyst
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G3 · GUI-only
Interaction primitives: av · Audio/video over time nav · Pan/zoom/scroll click · Point & click
Work product: WP05 · Time-based media
Conditions: licensed · Licensed pro app hidden · Hidden/stale state
Proposed difficulty: 4 / 5

Proposed verifier

Job file sync table vs gold offsets: >= 98% of lines within 0.5 s, including lines after the removed break; the 10 spot-check lines within 0.3 s; text unchanged. The sync table is written only by Case CATalyst and the job format is undocumented, so hand-edited job files are detected by failing to open in the app.

Proposed partial credit

30% leading offset fixed; 50% post-break offset fixed; 20% spot-check lines.

R10.01.01.04-T10 Locate testimony by audio and cite page, line and time

Original instruction

Eclipse CAT job 'Chen_Depo_0903' with synced audio. The attorney needs the passage where the witness first discusses the March 3 surcharge invoice and the later passage where she corrects the amount. Find both, and write output/cites.txt with each passage's start and end page:line and audio timestamps.

Platforms: win · Windows
Applications: Eclipse CAT
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G3 · GUI-only
Interaction primitives: nav · Pan/zoom/scroll av · Audio/video over time type · Text & forms
Work product: WP12 · Answer/decision
Conditions: licensed · Licensed pro app
Proposed difficulty: 3 / 5

Proposed verifier

cites.txt: both passages' page:line ranges within 1 line of gold and timestamps within 2 s. The correction is phrased without the keyword 'March' (seeded), so naive search misses it. ASR engines are not installed on the VM, and timestamps can be read only from Eclipse's audio sync.

Proposed partial credit

40% first passage; 40% correction passage; 20% timestamps.

R10.01.01.04-T11 Apply a witness errata sheet as appended corrections

Original instruction

Case CATalyst job 'Marlow_Depo_0915' (certified). The witness returned input/errata_signed.pdf with 7 changes. Per input/state_format_rules.pdf, do not alter the certified text; add an errata page listing page, line, change and reason, update the reporter's certificate addendum noting receipt of errata, and export output/marlow_with_errata.pdf.

Platforms: win · Windows
Applications: Case CATalyst
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G3 · GUI-only
Interaction primitives: type · Text & forms nav · Pan/zoom/scroll
Work product: WP01 · Document
Conditions: licensed · Licensed pro app
Proposed difficulty: 3 / 5

Proposed verifier

Certified body text byte-equal to the original export; errata page rows equal the 7 gold entries; addendum present with signature date; PDF page count = original + errata pages. PDF libraries are not installed on the VM, and the PDF must be exported by Case CATalyst from the saved job (errata page present in the job file).

Proposed partial credit

40% body untouched; 40% errata rows; 20% addendum.

R10.01.01.04-T12 Edit an interpreted deposition with bilingual segments

Original instruction

Eclipse CAT job 'Ruiz_Depo_0910' (synthetic, Spanish-speaking witness through an interpreter). Correct the English record against the audio, mark interpreter turns with 'THE INTERPRETER:', add '(Through the interpreter)' after the oath per input/style_guide.pdf, flag every untranslated Spanish response with a scopist note, and fix speaker labels where the witness answered in English directly.

Platforms: win · Windows
Applications: Eclipse CAT
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G3 · GUI-only
Interaction primitives: av · Audio/video over time type · Text & forms nav · Pan/zoom/scroll
Work product: WP05 · Time-based media
Conditions: licensed · Licensed pro app long · Long-horizon
Proposed difficulty: 5 / 5

Proposed verifier

English WER <= 3%; speaker attribution (attorney/interpreter/witness) accuracy >= 97%; parenthetical placement per style guide; scopist notes on the 6 gold untranslated responses (F1 >= 0.9). ASR engines are absent from the VM, and notes must exist as Eclipse scopist notes in the saved job, which only the app writes.

Proposed partial credit

35% text accuracy; 35% speaker attribution; 15% parentheticals; 15% untranslated flags.

R10.01.01.04-T13 Segment an oral-argument recording by speaker in ELAN

Original instruction

ELAN with input/oral_argument.wav (18 min, 3 judges and 2 counsel, synthetic). Create a tier per speaker and annotate each speaking turn with start/end boundaries and the speaker's role, including interruptions where judges overlap counsel. Save output/oral_argument.eaf.

Platforms: mac · macOS win · Windows lin · Linux
Applications: ELAN
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: CL · CLI-sufficient
Interaction primitives: av · Audio/video over time drag · Drag & drop vis · Visual judgment
Work product: WP10 · Annotations
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

EAF parsed: segment boundaries vs gold with boundary error <= 250 ms for >= 90% of turns; speaker-role assignment accuracy >= 95%; overlaps represented on separate tiers. Diarization and ASR tools are not installed on the VM, so turns must be segmented by listening in ELAN.

Proposed partial credit

50% boundaries; 35% speaker roles; 15% overlaps captured.

R10.01.01.04-T14 Transcribe a short voicemail exhibit verbatim with time marks

Original instruction

Using Express Scribe with foot-pedal hotkeys (or keyboard shortcuts) and Word, transcribe input/voicemail_ex14.wav (2 min) verbatim, including fillers and false starts, with a time mark every 30 s and the header block from input/transcript_header.docx. Save output/voicemail_ex14.docx.

Platforms: mac · macOS win · Windows
Applications: Express Scribe · Word
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: CL · CLI-sufficient
Interaction primitives: av · Audio/video over time type · Text & forms key · Shortcuts
Work product: WP05 · Time-based media
Conditions: multiapp · Multi-app
Proposed difficulty: 2 / 5

Proposed verifier

WER vs verbatim gold <= 5% (fillers counted); 4 time marks within 1 s of 0:30 multiples; header block fields exact. ASR engines and OOXML libraries are absent from the VM, so the audio must be transcribed by listening in Express Scribe and the file saved by Word (app metadata and rsids checked).

Proposed partial credit

70% verbatim accuracy; 15% time marks; 15% header.

R10.01.01.04-T15 Redact identifiers in a transcript and bleep the matching audio

Original instruction

input/911_call.wav and its transcript input/911_call.docx (synthetic) must be released with the caller's phone number, home address and date of birth removed. Replace them in the transcript with '[REDACTED]' and, in Audacity, replace each corresponding audio span with a 1 kHz tone of equal length. Export output/911_call_redacted.wav and .docx.

Platforms: mac · macOS win · Windows lin · Linux
Applications: Audacity · Word
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: CL · CLI-sufficient
Interaction primitives: av · Audio/video over time drag · Drag & drop type · Text & forms
Work product: WP05 · Time-based media
Conditions: multiapp · Multi-app
Proposed difficulty: 4 / 5

Proposed verifier

Audio: in each of 5 gold spans, 1 kHz tone energy dominates (>= 90%) and ASR of the output finds none of the identifiers; total duration unchanged within 10 ms; non-span audio correlation >= 0.99. Transcript has 5 '[REDACTED]' at the right places. sox, ffmpeg and Python audio libraries are absent from the VM and Audacity's mod-script-pipe is disabled, so tones must be applied in Audacity.

Proposed partial credit

45% audio spans toned; 20% duration and rest intact; 35% transcript redactions.

R10.01.01.04-T16 Cut video deposition clips from page-line designations

Original instruction

Descript project with the synced video deposition input/marlow_video.mp4 and its transcript. input/video_designations.xlsx lists 11 page:line ranges for trial playback. Create one clip per designation trimmed to the first and last designated words (no extra question fragments), title each 'Marlow <page:line>', and export output/clips/ as MP4 plus a clip report CSV with in/out timecodes.

Platforms: mac · macOS win · Windows
Applications: Descript · Excel
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: CL · CLI-sufficient
Interaction primitives: av · Audio/video over time drag · Drag & drop nav · Pan/zoom/scroll
Work product: WP05 · Time-based media
Conditions: multiapp · Multi-app
Proposed difficulty: 4 / 5

Proposed verifier

11 clips; in/out points within 0.3 s of gold word boundaries; ASR of each clip matches the designated text (WER <= 5%); report CSV timecodes equal clip boundaries. ffmpeg and ASR tools are not installed on the VM, so clips must be cut and exported in Descript (encoder metadata checked).

Proposed partial credit

60% clip boundaries; 25% clip content; 15% report.

R10.01.01.04-T17 Export a page-line to timecode sync table for a video deposition

Original instruction

Case CATalyst job 'Okafor_Video_1002' with synced video. Verify sync at the start of every page by playback, correct any line whose sync point is off by more than 1 s (seeded drift after a camera restart), and export a sync table output/okafor_sync.csv (page, line, timecode) for all lines.

Platforms: win · Windows
Applications: Case CATalyst
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G3 · GUI-only
Interaction primitives: av · Audio/video over time nav · Pan/zoom/scroll drag · Drag & drop
Work product: WP05 · Time-based media
Conditions: licensed · Licensed pro app hidden · Hidden/stale state
Proposed difficulty: 4 / 5

Proposed verifier

Sync CSV vs gold: >= 97% of lines within 0.5 s, all lines after the camera restart within 1 s; row count equals transcript lines. ASR and alignment tools are not installed on the VM, so drift can be found only by playback in Case CATalyst; the CSV must match the saved job's sync table.

Proposed partial credit

40% pre-restart lines; 45% post-restart lines corrected; 15% complete table.

R10.01.01.04-T18 Transcribe a contentious multi-party hearing with overlapping speech

Original instruction

Descript project 'Custody_Hearing_0711' (45 min, synthetic; judge, two counsel, two witnesses, frequent crosstalk). Starting from the ASR draft, produce a transcript with speaker names from input/appearances.pdf, overlapping speech marked with '--' per input/style_guide.pdf, '(Pause.)' and '(Recess.)' parentheticals at the gold points, and export output/custody_hearing.docx.

Platforms: mac · macOS win · Windows
Applications: Descript
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G3 · GUI-only
Interaction primitives: av · Audio/video over time type · Text & forms click · Point & click
Work product: WP05 · Time-based media
Conditions: long · Long-horizon
Proposed difficulty: 5 / 5

Proposed verifier

WER <= 3%; word-level speaker attribution >= 97%; overlap markers at >= 80% of gold crosstalk points; parentheticals within 2 s of gold positions. ASR and diarization tools are not installed on the VM, and the docx must match the transcript saved in the Descript project.

Proposed partial credit

40% text accuracy; 35% speaker attribution; 15% overlap marks; 10% parentheticals.

R10.01.01.04-T19 Correct the same deposition draft with ASR and scripts only (control)

Original instruction

Control pair of T02. Given input/marlow_audio.wav and input/marlow_rough.txt, produce output/marlow_cli.txt using command-line tools only (e.g. whisper.cpp alignment plus scripted edits) with Q/A formatting and 25 lines per page.

Platforms: lin · Linux
Applications: whisper.cpp · Python
Interfaces: term · Terminal
GUI need · task setting: CL · CLI-sufficient
GUI need · ordinary access: CL · CLI-sufficient
Interaction primitives: code · Code & commands
Work product: WP05 · Time-based media
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

Same metrics as T02 (WER <= 2%, Q/A line-type accuracy >= 98%). Measures whether the CAT audio-sync GUI adds accuracy over a scripted pipeline.

Proposed partial credit

50% WER; 30% Q/A formatting; 20% pagination.

R10.01.01.04-T20 Resolve a scopist's thirty check marks against the audio

Original instruction

Eclipse CAT job 'Chen_Depo_0903' carries 30 scopist 'check' marks on doubtful words. For each, play the synced audio, correct the word or confirm it, then remove the mark. Leave a note only on spots that remain truly unclear after listening.

Platforms: win · Windows
Applications: Eclipse CAT
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G3 · GUI-only
Interaction primitives: av · Audio/video over time nav · Pan/zoom/scroll type · Text & forms
Work product: WP05 · Time-based media
Conditions: licensed · Licensed pro app
Proposed difficulty: 3 / 5

Proposed verifier

At the 30 marked spots, word accuracy vs gold >= 93%; 0 check marks remain; notes only on the 2 gold-unclear spots; diff outside marked spots empty. ASR engines are not installed on the VM, so doubtful words can be resolved only by listening via Eclipse's audio sync.

Transcript editing synced to audio

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