The census builders

Authoritative source, reproduced unchanged. Exact Markdown · Agent copy with resolved links. These instructions govern research runs, not checkpoint interviews. They keep the run's own words: a room is a community, a mark an artwork, the analyst the aggregator, the case lead the community coder and the global model the population model.

The four scripts that made ../MASTER-census.csv on 27 August 2026, copied unchanged from the study's Drive (venue/ in the Layer A data folder). They are kept as the record of how the sheet was made. Run order:

build_venue_activity.py  <raw24> <census_740_mapping.csv> <outdir>   _progress.jsonl: venue counts per community
emit_tables.py                                   (in <outdir>)       venue_activity.csv, venue_summary.csv
screen_event_talk.py     <raw24> <census_740_mapping.csv> <outdir>   _screen_progress.jsonl: the lexical screen
build_master_census.py   <outdir>                                    MASTER-census.csv

What stops a rerun from here

• Paths are the study's. build_master_census.py reads its inputs beside itself (_screen_progress.jsonl, venue_summary.csv, validations.jsonl), the mapping from ../atlas/, and the raw folders from ../raw24*. To rerun, copy the four scripts and those files into one folder laid out that way.
• The coder verdicts are not shipped as a file. build_master_census.py reads them from Sandboxes/Gioia v28/case_facts.jsonl, the coders' output of an earlier run. Their result is the coded, coder_verdict and coder_model columns of MASTER-census.csv, which is the file this repository ships.
• raw24 comes from data/raw24/ (download or rebuild).

The counts (items_*, screen_class, hit_*) can be rebuilt this way. The judgements (validations.jsonl, the mapping, the verdicts) are inputs and cannot.

recount_during.py, also in venue/, is left out: its output (during_counts.jsonl) is read by none of the four.