[TimeToBuildBob]

Summary

• Preserve both untrimmed minute-resolution CSVs collected at Erik’s request: 8,733 aggregate rows and 18,055 asset rows, October 6–19; SHA-256 hashes and provenance in data/hf/README.md.
• Sampler exited successfully at its October 20 cutoff; Bob’s service is now disabled. Original files remain preserved on the sampling host.
• Include a bounded reach readout: +38,292 non-updater asset downloads and +211 net stars. Platform increments reconcile exactly with the aggregate delta.

Attribution limits

The only two committed launch-event timestamps precede sampling. Later planned posts have no verified timestamp/URL in launch-events.csv, so there is no defensible per-post lift or channel ranking yet. Downloads are requests, not unique people, installs or paid conversions. No conversion rate is inferred.

Verification

• Byte-for-byte SHA-256 match against both source CSVs; no trimming or rewriting.
• Validated schemas, timestamp ordering, unique series/timestamp keys and nonnegative counters.
• Existing analysis helpers reconstruct per-asset counters; updater assets excluded from reach, retained in raw data.
• pytest test_sample_hf.py analyze_stats.py: 7 passed.

This delivers the requested raw data; Erik retains the decision on resolution/coarsening.