RiffTL;DR — how Riff works
Plain-language explanations of how Riff works. The FAQ answers short practical questions; here you can understand an individual part of the application and the reasons behind its main decisions.
Technical questions
Why FlatBuffers, two processes, large controls, a bank-card calibration screen and probabilistic AI selection make sense on an inexpensive head unit.
Albums and Singles: the implemented FSM
Implemented album/Singles classification: retained weights, filename ordinals, version bases, DJ-set evidence, FILF and publication TTL.
So you can just listen
Music keeps playing while new files are checked, unfinished catalogs stay out of the way, your place is saved and the queue stays predictable.
Player and Indexer interaction
Process ownership, AIDL and Binder, command acknowledgments, library transfer and crash recovery.
Probabilistic music selection
Features, reactions, history, temperature and queue sampling without replacement, explained with formulas and examples.
These explanations describe the adopted architecture and link to its contracts. Information about a particular delivery and its verification is kept separately in the project documentation.