{"claim":{"id":"b8b48d0a-cf5f-470c-af8b-d2b40387ee16","submission_id":"2c61663b-6131-4ed2-8f0d-eeb4af70d6ef","submission_revision_id":"fbfba837-87aa-4206-903e-7d6f89fdd91e","builder":"builder-01","status":"active","claimed_at":"2026-10-15T14:44:28.741529Z","delivered_at":null,"ended_at":null,"delivery_note":null,"end_reason":null},"task":{"submission_id":"2c61663b-6131-4ed2-8f0d-eeb4af70d6ef","submission_revision_id":"fbfba837-87aa-4206-903e-7d6f89fdd91e","revision_version":5,"submission_status":"under_review","submitted_at":"2026-08-02T20:11:51.965513Z","submitter_name":"Ari Nemet","submitter_email":"ari.m.nemet@gmail.com","task":{"status":"submitted","version":5,"extra_fields":{"difficulty_self_test":{"model":"Gemini 3.1 Pro Extended thinking","harness":"gemini.google.com","estimated_score":"<1%"}},"submitted_at":"2026-08-02T20:11:51.965513+00:00","industry_domain":"Music","task_short_name":"audio_to_measure_timestamps","operating_system":"Windows","task_description":"Objective: Determine, by ear, the exact second every measure begins in a solo piano recording, and produce a timestamp file plus a synced video of the score.\n\nStarting State:\ninput/recording.wav: An original recording of the 1st movement of Robert Schumann's Faschingsschwank aus Wien, Op. 26, recorded specifically for this task. No score, tempo, or measure count is given.\n\nInstructions & Rules: Use recording.wav to determine the exact second each measure begins (where its downbeat lands). You may open and play recording.wav in a media player (e.g. Windows Media Player) to listen to it. Do not use any audio-analysis tools, scripts, or libraries to compute or estimate timestamps. Write the result to measures.txt in the format:\n\n<measure> <second, to 3 decimal places>\n...\n\nIn Shotcut: import recording.wav as audio, find the score on imslp.org (confirm it's public domain worldwide and a legible high-resolution scan) as a expert would need to find their own score, and would need to account for any discrepancies that arise, import it as video, and add a red vertical line that appears at each measure's start time per measures.txt, synced to the audio. Build the red line by hand in Shotcut's timeline, do not generate or edit the project's XML file directly. Export as output/score_sync.mp4 and include output/measures.txt.","software_licensing":"Free/Open Source","evaluation_criteria":null,"verification_method":"Hard gates (any leads to 0%):\nmeasures.txt or the video file is missing\nVideo is corrupted, or its duration is not within 5 seconds of the audio's length\nmeasures.txt or the video is in the wrong format (not .txt / not .mp4)\nNumber of rows in measures.txt does not match the expected measure count for the piece\n\nOtherwise, weighted scoring. Measure timing accuracy (90%): measures.txt, matched to the expert reference row by position. Within 50ms, full credit. Beyond 50ms, credit tapers to 0% at 500ms off. Average across all rows. Video sync (10%): an AI vision judge checks a few frames of the video at times pulled from the reference measures.txt, and checks if the red line is roughly in the right place and moving forward with the audio, not stuck or jumping around. Score is the percent of frames that pass.","software_and_version":"Shotcut 26.8","specific_requirements":null,"expected_completion_time":null},"files":[{"file_id":"a67ca5d4-2bcf-4224-ab87-0a858fd0b119","filename":"recording.wav","category":"input_materials","content_type":"audio/x-wav","size_bytes":101535356},{"file_id":"28b54bd7-edba-4948-a86c-d42c011439ba","filename":"reference.txt","category":"reference_output","content_type":"text/plain","size_bytes":8069}],"deep_review":{"overall_verdict":"Borderline Accept","review_text":"**Overall Verdict:** Borderline Accept — the timing task and primary metric are viable, but the score-video cursor behavior is insufficiently specified for reproducible visual evaluation across solver-selected scans.\n\n**Completeness Check:**\n- Instruction: Unclear — timestamp requirements are clear, but the red line’s duration, movement, page transitions, and placement across differing score editions are underdefined.\n- Inputs: Complete — the supplied recording is a readable, complete 528.829-second stereo PCM WAV.\n- Toolchain: Complete — Shotcut on Windows provides a viable production path, and the score-acquisition workflow is stated.\n- Reference output: Complete — the protected reference contains 553 sequential, strictly increasing measure timestamps within the recording duration.\n- Rubric: Unclear — the 90% numerical timing metric is objective, but the 10% vision criterion lacks reproducible cursor and score-layout acceptance rules.\n\n**File and Task Reachability:**\nThe attachment set is `reachable`, while the end-to-end task is `conditionally_reachable` pending clarification of the visual target. Both manifest attachments were inspected: `recording.wav` is a structurally coherent 101,535,356-byte RIFF/WAVE file containing 48 kHz, 16-bit stereo PCM audio, and `reference.txt` is an 8,069-byte ASCII file containing 553 well-formed rows numbered 1 through 553. The reference uses six decimal places rather than the solver-facing three, but numerical comparison with a 50 ms tolerance makes this nonmaterial. No manifest attachment was omitted. The recording’s claimed musical identity and the availability and worldwide public-domain status of a particular score scan could not be independently verified within the permitted evidence boundary. The workflow otherwise supports listening and annotation, acquiring a qualifying score, authoring the synchronized video in Shotcut, exporting both required artifacts, and evaluating row count, duration, timestamps, and sampled video frames. License keys, accounts, tokens, and other credentials are not required from the contributor. Confidential, proprietary, or internal material may be evaluated as submitted evidence without requiring additional disclosure.\n\n**Special Screening:**\nNone\n\n**Recommended Changes:**\nThe audio input, protected timestamp reference, hard gates, and primary numerical scoring method are coherent. Define one exact score-video convention: whether the red line is instantaneous, held over the current measure, or continuously interpolated; how it advances within systems and across pages; whether any qualifying score edition is accepted; and the spatial and temporal tolerances used by the vision judge.","reviewed_at":"2026-09-01T19:52:59.312735Z"},"claim":{"id":"b8b48d0a-cf5f-470c-af8b-d2b40387ee16","submission_id":"2c61663b-6131-4ed2-8f0d-eeb4af70d6ef","submission_revision_id":"fbfba837-87aa-4206-903e-7d6f89fdd91e","builder":"builder-01","status":"active","claimed_at":"2026-10-15T14:44:28.741529Z","delivered_at":null,"ended_at":null,"delivery_note":null,"end_reason":null}},"revision_is_current":true}