Help wanted, eventually -- evidence pack
bigbio.ai/articles/help-wanted-eventually

What this is: everything behind the article's numbers. Data as of October 20, 2026.

  prereg.md       the question, tests and decision rule, written and committed BEFORE any
                  data was pulled (2026-10-20 15:03 UTC), plus a dated amendment made after
                  a simulation test rejected the first interval method and before any real
                  result was computed
  numbers.md      every number the article may use, with the file it comes from
  results.json    the computed estimates and 95% intervals
  forecast.toml   the registered forecast (46%), scored April 10, 2027
  code/           the analysis (Python 3.12: pandas, numpy, scipy, statsmodels,
                  pymannkendall, ruptures) and the calibration test (pytest)
  data/           the raw pulls with fetch time and sha256 in each meta.json

To rerun:
  1. bash data/indeed/restore.sh   (fetches the two large Indeed files at the exact
     vintage from the publisher's repository and verifies their sha256)
  2. place code/*.py under src/hiring/, data/ beside it, then
     PYTHONPATH=src python src/hiring/analysis.py
  3. pytest code/test_recovery.py   (the calibration test, six cases)

Sources and licences: Indeed Hiring Lab job postings tracker (CC BY 4.0,
github.com/hiring-lab/job_postings_tracker); US Census Bureau Quarterly Workforce
Indicators and US Bureau of Labor Statistics QCEW and CES (US government works, public
domain); SEC EDGAR full-text search counts (public records). Press figures in
data/narrative/claims.csv are quoted with their URLs and belong to their publishers.

Removed for publication (personal data stays out; nothing removed entered a verdict):
- prereg.md: the question's personal framing reworded; one descriptive-only private series (a personal job-search crawler) removed; internal wiki links removed. Estimands, baselines, decision rule, amendment untouched.
- code/: fetch_scout.py omitted; load.scout() and analysis.seat_level() removed.
- results.json: the seat_level block (private crawler) removed.
- forecast.toml: the personal 'decision' line replaced with a note; everything else verbatim.