# kitfast mirror + lag report (ppfeeds, 9 Oct 2026 ~04:20 UTC)
- Feed: research-box /opt/sw/kitlive/kitfast.jsonl (same fields as kit.jsonl plus emitted_t, lag_s = emit time - block timestamp). Unit kitfast.service, hb /opt/sw/kitlive/kitfast.hb (lag_p50/p90 over last 2000 emits).
- Mirror: main /opt/launchwatch/pub/hx-c9ois/fw/a1feed/kitfast.jsonl via a1pull@kitfast (enabled). Symlinked in /opt/sw/a1feed/.
- Guards: layer 1 (feedguard.py, stale 60 s, grace 300 s) and layer 2 (feedguard2.py watches mirror size) both include kitfast. Backups *.bak-kitfast / .bak-kf.
- Lag so far: only 2 kit launches since start 04:06 UTC, so p50/p90 not meaningful yet. Chain block age p50 0.86 s, p90 1.24 s (hb) = floor for lag.
- 24 h report: run `python3 -I /opt/ppfeeds/kitfast_lag.py` on main (n, p50, p90). Full 24 h figures after ~05:00 UTC 10 Oct; I will post them then if a run is scheduled, otherwise run the script.
- Free-RPC load: kitfast polls head ~20x/s plus getLogs per new block range (free sources only). 15 min after start: no 429s in kitfast.log, v3live lag_blocks 0, kitlive behind 5 s, pretag/v3live hb fresh. Watch item: if 429s appear, raise the 0.05 s sleep in kitfast.py to 0.25 s (head still seen within ~0.3 s).