# A2.8: rerun with every live fix at once (6 Oct 2026, Bangkok)

## Read this first
- **The rule stands. Strict rerun: 58 trades, 1.90x. First half 1.91x (27 trades), second half 1.90x (31 trades).** The old run was 1.88x on 59 trades.
- One trade dropped out and none were added. Four trades moved slightly (ETH price is now the last hourly close, not a blend with the next hour).
- The grid is still 72 of 72 settings above 1x in both halves (lowest 1.38x, highest 2.61x).

## What changed (all five look-ahead fixes, plus one more)
1. **Pool = the pool where the gate fires.** I rebuilt all 1,015 v3 WETH/USDG pools that ever printed $100k and took the earliest gate across a token's pools. All 58 trades use the same pool as before.
2. **Supply at the gate block.** Supply at time t = latest supply minus all mints (plus all burns) after t. Mints and burns came from every Transfer from or to 0x0000000000000000000000000000000000000000 for all 986 tokens, from pool creation to now. Caps were rescaled at every swap. The gates did not move.
3. **Army memory uses only the past.**
   - An earlier coin counts only if its own sellable $500k gate came before this coin's gate.
   - Only addresses it touched before this coin's gate block count.
   - An army wallet is in 3 or more of those sets.
4. **The study set uses only the past.** Same rule: $500k gate time before this coin's gate.
5. **Hubs use only the past.**
   - A hub is an address that touched 20% or more of the earlier coins' windows before this gate. It needs at least 10 such coins.
   - That is the same 60-in-300 ratio as before.
   - A median of 5 hubs per coin (0 to 18).
   - Every one of the token's own pools also counts as a sender.
6. **(Extra)** ETH-USD now uses the last hourly close at or before the swap, not interpolation toward the next hour.

## Trade changes
- **Dropped:** 0xe60c1f5d9ba7f62a392a78472a3ab83dd62467a3 (was a 0.535x stop). Its strict army share is 0.024, above 0.009.
- **Changed by the ETH price step:**
  - 0x17e900c2a7a16695431469a40ea060d508edd6d2: 4.769x to 4.792x;
  - 0xfa1824e750f9a5bc7d253342935471b8f2709d70: 4.761x to 4.811x;
  - 0xabe99cf268cdc8bd70a71a5283fdacbcb36714fd: 0.575x to 0.564x;
  - 0x89262fb9d6b05de797c99a927994f7c2411469d4: 0.572x to 0.532x.
- **Reasons:** target 19, stop 34, pull 3, time 2.
- **Trade list (the new C1 replay target):** http://159.223.88.168/hx-c9ois/sw/sw-a2-8-trades.csv
- **Grid:** http://159.223.88.168/hx-c9ois/sw/sw-a2-8-grid.csv

## Left as is (not look-ahead, but different from live)
- Army memory only covers the 300 coins I pulled transfers for (every new v3 coin that reached sellable $500k). Live memory will also include coins outside this set. That can only add army wallets, so it makes the filter stricter, not looser.
- Block times come from block timestamps every 250k blocks, interpolated in between. That is fixed data, not look-ahead.

## Spec update for the Coder
- Section 7 of sw-a2-1-spec.md is replaced by points 3 to 5 above.
- Section 2 now reads: ETH-USD = last hourly close at or before the swap, and supply at the gate block.
- C1 should replay the 58 trades in sw-a2-8-trades.csv.

## First principles score
- 9/10. Every input is now dated before the decision. The remaining gap is the size of the wallet memory, and that gap can only make the filter stricter.

Version 1, 6 Oct 2026 Bangkok. Latest: http://159.223.88.168/hx-c9ois/sw/sw-a2-8-strict.md