# aipaper_scores_30d peak_mc fill (ppdata, 27 Oct)
File (research-box): /opt/crew/work/ppdata/aipfill/aipaper_scores_30d_filled.csv  (same columns as /tmp/aipaper_scores_30d.csv; entry_mc, peak_mc, sell_mc filled)
Source: mcap30.py recompute from /opt/hist/rh day DBs (ETH/WETH-quoted pools only; peak = highest confirmed cap, 6h swaps then 1-min bars).
Coverage: 30,990 rows; blank 26,864 -> 21,277.
- 4 Oct - 14 Oct: filled for ETH/WETH-quoted coins (remaining blanks there = non-ETH quote pools, e.g. 0x0bd7d3, pons 0x5fc5, skipped by mcap rules).
- 15 Oct - 27 Oct: STILL BLANK (~99%). Cause: rh day DBs from 15 Oct are the new thin rebuild format (no `complete` flag, bars=0, swaps partial), so mcap30 cannot compute peaks. Not a stale-data issue, a gap in the rh rebuild for those days.
Fix options: (a) rebuild bars/complete for rh 15-26 Oct (dataeng), or (b) take peaks from runtrack / gold_idx for those tokens (join on token). Tell me which; (b) is cheaper.
Query: python3 -c "import pandas as pd; d=pd.read_csv('<file>'); print(d.groupby(d.t//86400).peak_mc.count())"