Five chronologically-ordered task streams mined from the merge histories of major
LLM-infrastructure repos. Each stream is a strongly-coupled chain (uses-symbol WCC)
of functional-capability tasks — the continual-learning core of the repo.
Each task provides:
statement — a detailed functional requirement (background + typed signatures +
numbered behavioral spec + worked-examples truth table), not "apply this diff".
test_code… See the full description on the dataset page:
https://huggingface.co/datasets/lwaekfjlk/codingcl-streams.
