cut_window_v32.py

Cuts a community's record to the 97-hour window.

What it holds

python3 cut_window_v32.py IN.jsonl OUT.jsonl. The window runs from 2017-03-31 16:00 UTC (the canvas opens) through 2017-04-03 17:00 (placement disabled) to 2017-04-04 17:00. A thread with any item in the window is kept whole, and items outside it are marked _context. A thread with no item in the window is dropped.

Who writes it, and when

The instrument's author. The sources do not say which script ran it for the yard's rooms; the yard README says rooms were made from raw24 files by build_room.py, which is not among these sources.

Why it exists

The docstring says the window was tested: "748 of 748 codes fall inside END, the last one at +39h", and a coder who read seventy threads beyond it coded none. Threads stay whole "so that every parent resolves and cite() can reach anything quotable". There is no pre-period, because the ordinary week is "the brief's job".

Where

cut_window_v32.py. Its output is record/coder_corpus.jsonl.gz in each room.

Checked against: cut_window_v32.py, yard/README.md