Demystifying the Lucene's search relevance scoring function by replicating it - valuable if we continue with the ES inverted index approach. Good for hyper-parameter optimization too. Can now setup testing against a more stable interface.
Clean up model eval scripts. Generalize eval scripts to handle other datasets.
Progress toward
Getty -> blockchain -> indexer ingest -> indexer API output
test:Indexer model updates: