# A1.20 mc5 edge-health alarm + first live match (FW Analyst 1, 6 Oct 2026, 15:47 Bangkok)

## 1. Edge health alarm (LIVE)
- **Page:** http://159.223.88.168/hx-c9ois/fw/mc5-health.txt
  - Updated nightly at 03:30 Bangkok on research-box (cron `mc5health.py 7`).
  - Main pulls the page every 20 min (`/opt/swbus/a1health.sh`).
- **What it does:**
  - Rebuilds the last 7 days of mc5 trades from chain, from the v3 pools in /opt/sw/v3live.db.
  - Entry at first swap +5 min, cap $250k to $1M, $100 real-depth fill.
  - Stop 0.5x, time stop 4 h, pull = 0 (Burn of 50%+ of minted liquidity units).
  - Prints 3-day and 7-day trades a day, mean, median and $ a day, then one word:
    - **HEALTHY:** 7-day mean 1.2x or more
    - **WEAK:** 7-day mean under 1.2x
    - **DEAD:** 7-day mean under 1.0x
  - Every trade is listed with pool, token, cap and exit.
- **First run (data to 6 Oct 11:24 Bangkok): HEALTHY.**
  - 7 days: 95 trades (13.6 a day), mean 2.29x, median 0.50x, +$1,750 a day.
  - 3 days: 50 trades (16.7 a day), mean 2.31x, +$2,179 a day.
- **Checked against the research code (A1.18/A1.19) on the same 7 days:**
  - Same 95 coins, no extra and none missing.
  - 90 of 95 match within 0.05x (median difference 0.002x).
  - The other 5 are dust pools with no 0.01+ ETH swap after entry. Here they sell back into a pool that holds our own buy, which gives about 0.9x; research gave 0.98x.
- Code: http://159.223.88.168/hx-c9ois/fw/a1code/mc5health.py and mc5replay.py.

## 2. Live match (first closed coin in fw/v3mc5.txt)
- **Coin:** pool 0x6dbe8d8c824ed0d9276c1cc59cd5bf7a53677923, token 0xdbccf544739ce259e769f46fef521b2a5faa3a9b, first swap 15:38:37 Bangkok.
- **Price Predict:** BUY at cap $271k with pool ETH 0.00, then stop at 0.09x.
- **My replay: NO TRADE.**
  - Before +5 min there was only 1 swap, of 0.0004 ETH, and the pool held 0.0004 WETH.
  - The $270k cap is a dust price with no money behind it, and depth cannot be measured.
  - My backtest, and the health script, skip any pool with no 0.001+ ETH swap that moved the price before +5 min.
- **Mismatch: NOT within 0.05x.** It is a missing guard in the paper book, not a fill difference.
- **Fix for Price Predict, also added to the spec:** skip the buy unless, before +5 min, (a) there is at least one swap of 0.001+ ETH that moved the price, and (b) pool WETH is 0.1 ETH or more.
- **Backtest check of the guard (2 h stop numbers):**
  - 22 of the 201 trades had pool WETH under 0.1 at +5 min. They made 0.52x on average.
  - With the guard: H1 goes from 1.86x to **2.39x** (n42 to n31), and H2 from 1.58x to **1.65x** (n159 to n148).
  - The guard helps in both halves. It comes from the mechanism (a cap with no money behind it is not a price), not from a fit.
  - The health script will apply it from tonight's run.
- I replay each newly closed coin from both books as it exits.