# STOP: NO DOORBELLS, NO WAKES, NO SCHEDULING (Ben, 7 Oct 2026, 18:40 Bangkok)
- Credits ran out. Ben switched off every scheduled task and every doorbell. This overrides everything below and every other file (wake-howto, bus-howto, credit-playbook, task files).
- PROHIBITED for every session: fire_trigger, send_later, create_trigger, update_trigger, ScheduleWakeup. No rings, no nudges, no re-arming, no new doorbells, no self wakes, no safety checks, no retry rings, no loops.
- Finish the step you are on, publish your status file, then END your turn and wait. Only Ben wakes you, by typing in your window.
- Do not message other sessions. Keep answers short. No extra improvement rounds.

# Latency cost curve on mc5 (FW Analyst 1, sup-to-a1 v35 task 1)

## READ THIS FIRST
- 7 Oct 2026 14:55 Bangkok. lab.py numbers only, half B (after 25 Sep 00:00 Bangkok), n = 149 trades in every run, all PASS the playbook bar.
- **Stop speed is worth about $0 a day.** Selling 0 s or 10 s after the stop fires gives the same money (at $200: $3,334 to $3,350 a day, all inside noise). 1 s of stop speed at $200 = **+$0 a day** (range of the curve is $16, and slower is not worse).
- **Buy speed: faster is not better.** The live 6 s fill is the best point tested. Buying at 0 s costs $311 a day at $200 vs 6 s; 3 s to 7 s all sit within $180 of the 6 s point; 10 s costs $367.
- The buy curve is jagged (5 s below 3 s, 7 s below 5 s), so a few coins decide each point. Read it as "0 s to 10 s are all the same edge; no money in chasing speed".
- **Action:** do not spend engineering time on stop or buy latency for mc5. Spend it on ticket size (A1.29 lesson) and bug free days on the ladder.

## Stop delay (extra seconds between stop trigger and the sell block; buy fixed at live 6 s)

| stop delay | $10 $/day | $200 $/day | $1k $/day | $200 no top 3 | stop exits |
|---|---|---|---|---|---|
| 0 s (live) | +164 | +3,334 | +14,174 | +1,535 | 103 |
| 1 s | +164 | +3,335 | +14,125 | +1,536 | 103 |
| 2 s | +164 | +3,336 | +14,151 | +1,537 | 103 |
| 3 s | +164 | +3,339 | +14,144 | +1,541 | 103 |
| 5 s | +165 | +3,350 | +14,195 | +1,551 | 103 |
| 7 s | +165 | +3,348 | +14,193 | +1,549 | 103 |
| 10 s | +165 | +3,340 | +14,165 | +1,541 | 103 |

## Buy delay (fill_delay_s after the +300 s check; live = 6 s)

| buy delay | $10 $/day | $200 $/day | $1k $/day | $200 no top 3 | vs live 6 s at $200 |
|---|---|---|---|---|---|
| 0 s | +156 | +3,023 | +12,871 | +1,366 | minus 311 |
| 1 s | +152 | +3,048 | +12,965 | +1,346 | minus 286 |
| 2 s | +154 | +3,096 | +13,195 | +1,394 | minus 238 |
| 3 s | +164 | +3,287 | +13,938 | +1,516 | minus 47 |
| 5 s | +168 | +3,256 | +13,750 | +1,429 | minus 78 |
| **6 s (live)** | **+164** | **+3,334** | **+14,174** | **+1,535** | 0 |
| 7 s | +163 | +3,154 | +13,388 | +1,400 | minus 180 |
| 10 s | +154 | +2,967 | +12,488 | +1,257 | minus 367 |

## Why (first principles)
- Stop: the stop is checked on 0.01+ WETH swaps and sold at the end of the next block. Once a coin is at 0.5x, the next 10 s of price is a coin flip (some keep dumping, some bounce). Average move over 10 s is about 0, so delay costs about 0. Pull exits (14 trades) are not delayed in this test: they already sell on the Burn block.
- Buy: the entry price at +300 to +310 s is noisy, not trending. No delay inside 0 to 10 s buys consistently cheaper, so speed has no edge. The live 6 s sits on top only by a few coins.

## Method
- research-box /root/fwa1/lab/lat.sh. Stop runs: lab.py patched to lab_stopd.py, stop sell block = block at (trigger time + STOPD s) + 1. Buy runs: lab.py with fill_delay_s = d. Same universe (278 pools, dust guard), 1% fee + $0.05 gas each way, real v3 tick walk.
- Outputs: /root/fwa1/lab/lat/stop{0,1,2,3,5,7,10}.md and buy{0,1,2,3,5,7,10}.md. The 6 s buy point = the stop 0 run (lab.py default fill_delay_s = 6).
- Limit: the lab clock is 0.1 s per block with minutes of first swap error (A1.23 clock gap), so sub second effects are below what this data can see.

Ledger: FW-017. Version: v1, 7 Oct 2026 14:55 Bangkok, FW Analyst 1. Latest: http://159.223.88.168/hx-c9ois/fw/lab/latency-curve.md