# AN4.2 SKIP "operator holds float at entry" on the FW and SW books (DC Analyst 4)

## READ THIS FIRST
- **v1, 7 Oct 13:21 Bangkok. Verdict: KILL on both books, by the pre-registered bar** (dc-prereg.md "7 Oct 13:18 | A4 | AN4.2").
- **SW live book (71 trades, $1k):** skip when operator float ≥ 33% at entry.
  - Half A: 11 skipped, 0 of them a 3x. $ a trade +$935 → +$1,394, $ a day +$2,503 → +$2,893, beats random 99.6%. Passes.
  - Half B: only 1 trade skipped. $ a day +$10,400 → +$10,153, beats random 36%. Fails.
  - The live book already caps the operator "army" at 1.5% of supply, so little float risk is left for this skip to remove.
- **FW mc5 (177 trades, $1k):** the same skip removes 10 (A) and 54 (B) trades, including 3 and 6 that made 3x+.
  - $ a trade rises a little (+$1,851 → +$1,876; +$1,020 → +$1,199), but **$ a day falls**: +$4,473 → +$2,970 (A), +$13,130 → +$9,797 (B). Beats random only 48% / 69%. Fails.
- **Why the A2 lead does not carry over (first principles):** A2 saw it on coins that cannot be pulled, bought at $100k to $250k, where an operator still holding float has to sell into us. These books buy fast launch pumps (mc5 at +5 min, the SW gate at $1M within 3 h). There the operator's float is part of how the pump happens: the bundle buys first, then sells into the run. Skipping operator-held coins removes winners as often as losers.
- **The wider "launch bundle" version (OP-B: first 30 blocks + creator) is worse:** it would skip 72% to 95% of trades, and $ a day turns negative in 3 of 4 halves. Same direction as SW A1.26: on these books the bundle coins ARE the winners.
- **Not kept:** no change to either book. A2's lead stays a lead for the non-pullable lane only (DC2.4); it is not a book overlay.

## Table ($1k ticket; $ a trade, $ a day; "no top 2" = $ a day without the 2 best trades)
| Book | Half | Rule | Trades | Skipped | Skipped 3x+ | $ a trade full → kept | $ a day full → kept | Beats random | No top 2 full → kept |
|---|---|---|---|---|---|---|---|---|---|
| SW live | A | op float ≥ 33% (primary) | 49 | 11 | 0 | 935 → 1,394 | 2,503 → 2,893 | 99.6% | 1,655 → 2,045 |
| SW live | B | op float ≥ 33% (primary) | 22 | 1 | 0 | 2,633 → 2,693 | 10,400 → 10,153 | 35.6% | 5,675 → 5,428 |
| FW mc5 | A | op float ≥ 33% (primary) | 29 | 10 | 3 | 1,851 → 1,876 | 4,473 → 2,970 | 47.6% | 1,878 → 376 |
| FW mc5 | B | op float ≥ 33% (primary) | 148 | 54 | 6 | 1,020 → 1,199 | 13,130 → 9,797 | 68.9% | 7,225 → 3,892 |
| SW live | A / B | op float > 0 (sensitivity) | 49 / 22 | 40 / 19 | 11 / 7 | | 2,503 → 812 / 10,400 → 110 | 83% / 25% | |
| FW mc5 | A / B | op float > 0 (sensitivity) | 29 / 148 | 18 / 137 | 3 / 19 | | 4,473 → 3,280 / 13,130 → -711 | 91% / 0.1% | |
| SW live | A / B | bundle ≥ 33% (OP-B) | 49 / 22 | 37 / 18 | 12 / 7 | | 2,503 → 438 / 10,400 → -51 | 34% / 13% | |
| FW mc5 | A / B | bundle ≥ 33% (OP-B) | 29 / 148 | 21 / 140 | 6 / 19 | | 4,473 → -530 / 13,130 → -498 | 0.3% / 2.1% | |

## Method
- Trades: lab.py's own trade lists (research-box /root/fwa3/lab/out/sw_accept/live_book_trades.csv and out/sw/fw_mc5_trades.csv), x at $1k with tick-walk fills, unchanged. $ a day uses the lab's day counts (full book $ a day from its verdict).
- Universe source: harness gold (Uniswap v3 WETH pools, A3 raw rebuild, frozen 7 Oct 12:07), not an outcome-picked tape.
- Float at entry: every swap before the entry minute, by recipient (pool events in gold p/<pool>.pkl). Holder = positive net tokens bought; the router 0xcaf681a66d020601342297493863e78c959e5cb2 excluded. Share = operator holders' tokens ÷ all holders' tokens.
- OP-F (primary): token creator (gold idx.json) + SW's crew list for that creator (/opt/sw/crew_wallets.json) + recipients first funded via 0x84157d3d24630c297849c84d49ae7f1805c1b60f / 0x03802649e469ed5680c58dff9e65425e22d21cc5 or by the creator (crew.db and ffund tables).
- Random check: 1,000 random skips of the same count in the same half.

## Limits
- **Operator coverage is thin on these v3 coins:** none of the 248 creators is in SW's crew list, and funder tables cover only about 9% of holders. OP-F is mostly "the creator's own wallet holds float". A fuller funder trace (balance binary search, as in SW A1.26) could change the SW numbers, but the FW result (winners skipped, $ a day down 25 to 34%) points the same way as OP-B, which needs no funder data.
- Sells are booked to the swap recipient; a router-routed sell can sit on the router address, so some sellers' net is overstated.
- Prior work cited: FW-009 (operator flag skip on mc5) was KILLED for the same reason (flagged and unflagged coins earn the same per trade).

## Files
- Per-trade floats: http://159.223.88.168/hx-c9ois/dc/dc-an4-2-float.csv (md5 ccb6f6cee3be409ad8a510b4556a3765)
- Scores: http://159.223.88.168/hx-c9ois/dc/dc-an4-2-score.csv (md5 f7eeb87ab69e0a517b5fa5d776205a9f), script http://159.223.88.168/hx-c9ois/dc/dc-an4-2-score.py (md5 a51f04ef3aa4333eb36c09ae4cb3703f), float builder http://159.223.88.168/hx-c9ois/dc/dc-an4-2-float.py

## Cost
- 0 X credits, $0 API, 0 RPC, research-box about 10 CPU seconds (nice 10), 0.2 h.

First principles score: 80%. The book P&L is the lab's own honest fills; only the skip is new. Missing: a full funder trace for v3 holders, signer-based sells.

Version: v1, 7 Oct 2026 13:21 Bangkok, DC Analyst 4. Latest: http://159.223.88.168/hx-c9ois/dc/dc-an4-2-skip.md