Experiment registry
Experiments
Every experiment run for this study writes a manifest recording its exact configuration, environment, and runtime — so results can be traced back to a reproducible run, never a guess.
exp-tfidf-001
tfidf
- Split
- test
- Device
- cpu
- Seed
- 42
- Runtime
- 1.49s
- Python
- 3.11.16
- Git commit
- 3e0d3492a8
- Timestamp (UTC)
- 2026-09-09 16:10:37
- Dataset
- BeIR/nfcorpus
exp-bm25-001
bm25
- Split
- test
- Device
- cpu
- Seed
- 42
- Runtime
- 0.94s
- Python
- 3.11.16
- Git commit
- 3e0d3492a8
- Timestamp (UTC)
- 2026-09-09 16:10:45
- Dataset
- BeIR/nfcorpus
exp-bge-001
BAAI/bge-base-en-v1.5
- Split
- test
- Device
- mps
- Seed
- 42
- Runtime
- 145.73s
- Python
- 3.11.16
- Git commit
- 17abfc141e
- Timestamp (UTC)
- 2026-09-09 16:24:17
- Dataset
- BeIR/nfcorpus
exp-medcpt-001
ncbi/MedCPT-Query-Encoder + ncbi/MedCPT-Article-Encoder
- Split
- test
- Device
- mps
- Seed
- 42
- Runtime
- 194.89s
- Python
- 3.11.16
- Git commit
- 7fdb8f335e
- Timestamp (UTC)
- 2026-09-09 16:32:09
- Dataset
- BeIR/nfcorpus
exp-hybrid-rrf-001
bm25+medcpt-rrf
- Split
- test
- Device
- cpu
- Seed
- 42
- Runtime
- 0.01s
- Python
- 3.11.16
- Git commit
- 90e4f0a0a5
- Timestamp (UTC)
- 2026-09-09 16:36:11
- Dataset
- BeIR/nfcorpus
exp-reranker-001
ncbi/MedCPT-Cross-Encoder
- Split
- test
- Device
- mps
- Seed
- 42
- Runtime
- 2176.78s
- Python
- 3.11.16
- Git commit
- eab2847f60
- Timestamp (UTC)
- 2026-09-09 17:21:35
- Dataset
- BeIR/nfcorpus
exp-evaluate-all-001
statistical_analysis
- Split
- test
- Device
- cpu
- Seed
- 42
- Runtime
- 1.15s
- Python
- 3.11.16
- Git commit
- a6b61fe17b
- Timestamp (UTC)
- 2026-09-09 17:29:04
- Dataset
- BeIR/nfcorpus
exp-error-analysis-001
error_analysis
- Split
- test
- Device
- cpu
- Seed
- 42
- Runtime
- 0.93s
- Python
- 3.11.16
- Git commit
- 3bdf4c444e
- Timestamp (UTC)
- 2026-09-09 18:54:59
- Dataset
- BeIR/nfcorpus
Manifest schema
Every manifest is validated against a required schema (src/biomedical_ir/manifests.py) before being written: experiment_id, timestamp, git_commit, dataset, split, model, parameters, seed, device, python_version, library_versions, runtime_seconds, and status (pending / running / complete / failed — a failed run must preserve its error, never be silently replaced with a guessed value).