# A1.13 Money map in $ a day (Analyst 1, 6 Oct 2026, 15:30 Bangkok)

## Read this first
- **Only two cells make real money in both halves after $500 real fills, and both are v3.** No v4 cell is a big edge. Most v4 cells lose money once a $500 buy pays for the pool's depth.
- **Top 5 by $ a day** (one $500 entry per coin, at real depth, using real outsiders' own exits, pulls = 0, 1% fee each side):

| # | Cell | Coins | $ a day | Half A | Half B | Days positive | Median coin | Rule a bot can follow |
|---|---|---|---|---|---|---|---|---|
| 1 | **v3 rule, its own exits (A2.8)** | 57 | **+$1,670** | +$1,759 | +$1,601 | 81% | 0.57x | the current v3 rule; add `rep1_pre` 5% skip (A1.11) |
| 2 | **v3, buy in the 30 min after the $1M gate, outsiders' real exits** | 54 | **+$1,291** | +$2,013 | +$730 | 73% | 0.96x | same gate; shows the money is real for outsiders too, not just in our sim |
| 3 | v4 no-hook "other" fee classes (mostly fee 3000 / spacing 60), buy after $500k | 119 | +$1,776 | +$3,266 | +$150 | 57% | 0.63x | **watch only**: half B nearly flat, median coin loses 37%, 5 coins give 91% of the gain |
| 4 | v4 dynfee, buy 2 to 6 h after launch | 24 | +$52 | +$26 | +$80 | 50% | | too small to matter |
| 5 | nothing else is positive in both halves | | | | | | | |

- **v3 sits at the top, as expected.** Its money comes from the rule's own exits (target at about 4.8x, stop at about 0.56x). Outsiders who just held from the gate to 4 h made less in half B (+$730 a day), so the exit rule adds money.
- **Pre-gate v3 entries look huge (+$4,586 a day) but that is hindsight.** We only see coins that later crossed the gate. A bot can't know that before the gate.
- **Why v4 fails in dollars, from first principles:**
  - Pools are shallow. Median virtual ETH depth: Pons 4.3, doppler 7.8, c500 4.5, pond 30. A $500 buy is 0.19 ETH, so in a Pons pool it costs about 5% in and about 5% out.
  - Real outsiders lose at the median in every v4 cell. The winners are a few coins.
  - In % terms, some v4 cells looked good (Pons buys at 2 to 6 h: +32%). That was tokens still held, marked at the 6 h price. Counting only ETH actually sold, those same buyers were at −19%.
- **The fast cells are where the money is lost:** first 2 minutes in c10k (−$52k a day), other (−$41k), c500 (−$39k), pond 2 to 30 min (−$9.7k). That matches A1.6, A1.9 and A1.12: crews and wash farms eat early outside money.

## Quick fix (v1.7): eth_getCode on my wallet lists
- Checked 43 addresses named in A1.9, A1.11 and A1.12, plus the A1.9 farm list (589) and every A1.11 repeat candidate (2,695).
- **Labels that changed: 6, all in A1.11.**
  - The 6 "farm wallets" I listed in A1.11 are token contracts (3,554 bytes each). They are the coins themselves: `0xcefbfccd83443e7aa1c493f4df0036013542f301`, `0xcc4cb99ef15e9799e1d7e45058683a72daf5b022`, `0xabe99cf268cdc8bd70a71a5283fdacbcb36714fd`, `0x55683371036401f8cae676c260efcde9be36a7f4`, `0xc3ba8a6f8d011881ceca8ad6e2b524232a40aa26`, `0x5cee66125e28aec539365c7a34fd25a2210c3966`. I printed coin addresses where wallet addresses belonged. My mistake.
  - **The filter result does not change.** Rerun with every address that has contract code removed: keep 49 coins at 2.08x (A 2.18x, B 2.01x), drop 8 at 1.08x. Same as before.
  - The real overlap is hundreds of wallets that each bought pre-gate in 2 of the 7 flagged 21 to 24 Sep coins. Many put in a near-equal 1.4 to 1.5 ETH each (for example `0x3c840e6b5a595fe9a81b6515ab25e482963af73e`, `0xf78643b5c13bc1977ea6fa41ef000174d04f0ff0`, `0xbd9a1b4843d49d3f96e36485bab986c5b4c1380f`, `0xc4ba47f68e4c5318c69283c0f21a8225bfad9813`). Equal sizes point to a farm, but this is not proven.
- **No other label changed:**
  - A1.9: 20 of the repeat wallets have 23 bytes of code. That is an EIP-7702 delegated wallet, a normal wallet with a smart account switched on, so they stay wallets. One real contract sits in the farm list: `0x2be87ad70cf11ea294d7c42044b5b8277a3e4874` (13,587 bytes, in 1,228 pools, tiny ETH). It looks like a bot or router contract and was already excluded, so the effect is nil. `0x34ba704793312bc1fac17a16e3b56ae93667be0f` was already described as a contract.
  - A1.12: the hook and the quote token are contracts, as written. The named wallets have no code.

## Method
- **Data:**
  - v4: 4,424 pools sampled from 63,331 (per day and class: up to 25 big plus 10 small), 13 Sep to 5 Oct, weighted back to the full count.
  - Every token transfer in a pool's first 6 h (10.8M), followed through routers to the final wallet.
  - v3: the 57 rule coins, transfers and swaps up to gate + 4 h.
- **Outsiders:** I removed wallets in 5+ sampled coins of a class on the same day, bots in 30+ coins, wallets in 20%+ of a crew's coins, buys at the launch block, creators, and contracts.
- **Real result per coin:** the outsiders' own ETH back ÷ ETH in. Each position is capped at 20x. Tokens are 0 if the pool was pulled; otherwise unsold tokens are valued at the last price in the window. A coin needs 3+ outside positions in the cell to count.
- **$500 fill:** depth from the pool's own swaps (L = ETH moved ÷ change in price root, virtual ETH reserve R).
  - Buy factor R ÷ (R + 0.19 ETH). Sell factor at the exit reserve R × √(multiple). 1% fee each side.
  - 16% of positions had broken depth reads (hook pools that rewrite swap amounts). For those, R = 0.25 × market cap in ETH, the middle of the classes that read cleanly.
- **$ a day** = sum over coins that day of $500 × (net multiple − 1), averaged over all 23 days (v3: 16 days). Halves: A to 24 Sep, B after.

## Limits
- The 6 h window (v4) and 4 h window (v3) value unsold tokens at the last price. That is still paper for whatever is held.
- "3+ outside positions" is a mild look-ahead: it picks coins that drew outsiders.
- The fallback depth for hook pools is a model, not a read.
- The v3 rule line uses Analyst 2's simulated exits on depth, not live fills.

## Asks for Supervisor
- **Keep all sizing effort on v3.** It is the only cell with real money a day in both halves. Window 1 and all fast v4 cells stay off.
- PP / Analyst 2: test cell 3 (v4 no-hook fee 3000 / spacing 60, entry after $500k) on PP's fair crossing list `pp3-cross.csv`, forward from 25 Sep only. If half B stays flat, drop it.
- Coder: rename the A1.11 farm examples in any book notes. The 6 addresses are coins, not wallets. The `rep1_pre` filter itself is unchanged and still valid.

First principles score: 80%. Each cell's money comes from real outsiders' own buys and sells, with a $500 fill priced on the pool's own depth curve. Missing: real exits past 6 h, live fills, and a direct depth read for hook pools.

Scripts: /root/a1/a113/groups3.py, real.py, depth.py, dollars.py, v3/v3pos.py, v3/v3usd.py; getCode: /root/a1/gc/gc.py, /root/a1/a111/redo.py (review container). Latest: http://159.223.88.168/hx-c9ois/sw/a1-13-money-map.md