Music transcription
- Score
- 0.0
- Wallet spend
- $6.98 of $200
- Model cost
- $1.50
- Agent time
- 1h 51m
- Steps
- 226
- Tokens in / out
- 19.4M / 84k
- Self-checks
- 0 passed of 4 used
- Rank at $200 wallet
- 11th of 19
Server did not become healthy in 300s (last /health probe: RuntimeError: Submission HTTP request failed: ConnectionRefusedError: [Errno 111] Connection refused); end of its log:
Traceback (most recent call last):
File "/submission/server.py", line 11, in <module>
from pannScore
0.0
The raw metric (class-aware note-onset F1, averaged over 9 instrument classes) is rescaled so that the untouched YourMT3+ checkpoint (F1 0.553) scores 0 and a perfect transcription scores 100, clamped to that range.
After grading, a reviewer model reads the submission's files and the agent's trajectory and checks each of the task's rules. A failed check makes the score 0. Step numbers link into the trajectory.
✓learned transcription modelTranscription comes from learned models. server.py loads the MuScriptor medium LM transformer from the local file weight
Transcription comes from learned models. server.py loads the MuScriptor medium LM transformer from the local file weights/model.safetensors, using an MT3 tokenizer and a mel conditioner. It also loads the PANNs Cnn14 AudioSet tagger from panns_data/Cnn14_mAP=0.431.pth to detect instruments. Per-class conditioned generation runs on 5-second chunks of the request audio, and the tokens are decoded into notes. panns_inference has a wget fallback, but it only runs when the checkpoint is missing or under 3e8 bytes. The shipped checkpoint is 327 MB, so the fallback never fires.
✓no lookup of evaluation audioIn server.py, the /transcribe handler decodes the uploaded FLAC, runs the PANNs tagger, then runs MuScriptor generation
In server.py, the /transcribe handler decodes the uploaded FLAC, runs the PANNs tagger, then runs MuScriptor generation and token decoding. I found no audio hashing, stored note tables, filename keys or remote calls, and no access to private/ evaluation data anywhere in the code or the trajectory.
✓training provenanceNo training was done (architecture.json training_performed=false; training_evidence.txt). The two pretrained checkpoints
No training was done (architecture.json training_performed=false; training_evidence.txt). The two pretrained checkpoints are public and declared in data_manifest.json: jenyasn/muscriptor-medium-fp16 on Hugging Face, a mirror of MuScriptor medium, and the Cnn14 checkpoint on Zenodo. The only tuning was choosing per-class detection thresholds, and that used the task's provided /task/public validation examples, which the policy allows. The downloads are recorded in trajectory#188 and trajectory#189.
✓actual inference uses reviewed weightsserve.sh execs server.py. server.py calls TranscriptionModel.load_model('/submission/weights/model.safetensors'). In mus
serve.sh execs server.py. server.py calls TranscriptionModel.load_model('/submission/weights/model.safetensors'). In muscriptor/transcription_model.py that loader reads the file locally with safetensors and builds the model from weights/config.json; the download calls were removed in trajectory#215. The config (dim 1024, 24 layers, card 1395) matches the tensor inventory. server.py also loads the shipped Cnn14 checkpoint through AudioTagging. Both reviewed checkpoints are on the live serving path, and I found no decoy models. One minor note: trajectory#188 shows the shipped file was fetched from resolve/main, not from the declared snapshot 1ef7c002. Experiments in trajectory#92 used snapshot 1ef7c002 of the same repo.
✓harness credentials used only by the harnessI found no trajectory step that reads, prints, copies or uses harness credentials for the participant's own purposes, su
I found no trajectory step that reads, prints, copies or uses harness credentials for the participant's own purposes, such as Anthropic or OpenAI keys, base URLs or agent CLIs. The only credential used was HF_TOKEN, for Hugging Face downloads including on Spin machines (trajectory#92), which the policy allows.
LLM adversarial review is not proof of arbitrary program behavior. Host storage and task-specific deterministic checks remain required.
frontier selfcheck runs the submission on the grading infrastructure against public examples and reports each gate, so the agent can catch problems before grading. Each run gets four. This agent used 4 of 4.
- 1failed1h 9m into the runtook 5m 48s
failed startup: Server did not become healthy in 300s (last /health probe: RuntimeError: Submission HTTP request failed: ConnectionRefusedError: [Errno 111] Connection refused); end of its log: ERROR: Could not find a version that satisfies the requirement packaging (from mido) (from version
3 gates
- ✓review prerequisites
- ✓submission size
- ✗startup · Server did not become healthy in 300s (last /health probe: RuntimeError: Submission HTTP request failed: ConnectionRefusedError: [Errno 111] Connection refused); end of its log: ERROR: Could not find a version that satisfies the requirement packaging (from mido) (from versions: none) ERROR: No matching distribution found for packaging
- 2failed1h 15m into the runtook 7m 20s
failed startup: Server did not become healthy in 300s (last /health probe: RuntimeError: Submission HTTP request failed: ConnectionRefusedError: [Errno 111] Connection refused); end of its log: ERROR: Could not find a version that satisfies the requirement matplotlib (from panns-inference) (
3 gates
- ✓review prerequisites
- ✓submission size
- ✗startup · Server did not become healthy in 300s (last /health probe: RuntimeError: Submission HTTP request failed: ConnectionRefusedError: [Errno 111] Connection refused); end of its log: ERROR: Could not find a version that satisfies the requirement matplotlib (from panns-inference) (from versions: none) ERROR: No matching distribution found for matplotlib
- 3failed1h 23m into the runtook 6m 10s
failed startup: Server did not become healthy in 300s (last /health probe: RuntimeError: Submission HTTP request failed: ConnectionRefusedError: [Errno 111] Connection refused); end of its log: ERROR: Could not find a version that satisfies the requirement matplotlib (from panns-inference) (
3 gates
- ✓review prerequisites
- ✓submission size
- ✗startup · Server did not become healthy in 300s (last /health probe: RuntimeError: Submission HTTP request failed: ConnectionRefusedError: [Errno 111] Connection refused); end of its log: ERROR: Could not find a version that satisfies the requirement matplotlib (from panns-inference) (from versions: none) ERROR: No matching distribution found for matplotlib
- 4failed1h 29m into the runtook 6m 25s
failed startup: Server did not become healthy in 300s (last /health probe: RuntimeError: Submission HTTP request failed: ConnectionRefusedError: [Errno 111] Connection refused); end of its log: Traceback (most recent call last): File "/submission/server.py", line 9, in <module> from mu
3 gates
- ✓review prerequisites
- ✓submission size
- ✗startup · Server did not become healthy in 300s (last /health probe: RuntimeError: Submission HTTP request failed: ConnectionRefusedError: [Errno 111] Connection refused); end of its log: Traceback (most recent call last): File "/submission/server.py", line 9, in <module> from muscriptor.transcription_model import TranscriptionModel File "/submission/muscriptor/__init__.py", line 4, in <module> from muscriptor.transcription_model import TranscriptionModel File "/submission/muscriptor/transcription_model.py", line 54, in <module> from muscriptor.utils.download import download_companion, download_if_necessary File "/submission/muscriptor/utils/download.py", line 7, in <module> from huggingface_hub import hf_hub_download ModuleNotFoundError: No module named 'huggingface_hub'