# Price Predict review: honest 10x and scams (5 Oct 2026, Bangkok)

## Read this first (for the Price Predict session)
- Written by a separate session that only reads your code and data. Nothing was changed. These are proposals, with receipts.
- Ben's goal: catch 10x from the moment a coin can actually be bought, and only on coins that aren't scams.
- Checked: every paper pick in aipaper5.db, aipaper30.db and aipaper.db (784 picks, 312 coins, 4 to 5 Oct). Each one was re-scored against the chain: every swap and every liquidity add or remove on each pool, for 24 h, from the public RPC.
- Scam test used here: the coin must be sellable through a public router, and our 10x must happen before the creator pulls the liquidity.

## Main findings
1. **Almost every launch is a planned rug.**
   - 97% of the 14,158 launches from 21 Sep to 3 Oct have their liquidity added by the creator through the Uniswap v4 position manager (0x58daec3116aae6d93017baaea7749052e8a04fa7). The creator keeps the right to pull it.
   - 87% are pulled within 24 h. In the pond (0.01% fee, 10 ETH start), 100% are pulled.
   - The pulls are real: 96% take 90% or more of the liquidity in one go, it's never added back, and trading stops (median 0 trades in the 10 min after vs 139 before).
   - 97% of pulls are done by the same wallet that launched the coin.
   - There is no warning: the pull comes a median 1.8 s after the last trade, and creators don't collect fees first.
2. **The real 10x does exist, before the pull.**
   - Of the coins that reached a sellable $500k and went 10x from the 5 s price, 100% peaked before the pull. The pull came a median 10 min after the peak.
   - Launches with locked liquidity (3%) almost never run (0.9%).
   - So "no rug coins at all" would kill the lane. The right test is: sellable, and sold before the pull.
3. **Price Predict's paper books are far too optimistic.** Per coin, picks 6 h or older (226 coins):

| | Booked by Price Predict | Honest (chain) |
|---|---|---|
| Reached 10x | 11 to 24% per book | 11% of coins (8 to 18% per book) |
| Pulled before 10x or the stop | not counted | 82% of coins, median 7.5 min after our buy |
| 24 h exits | average +202% (29 exits) | 25 of 29 were pulled pools, worth 0 |
| Open picks | 57% of all picks | 371 are on pools that were already pulled |
| Average result per pick | +1.1 to +4.1 (closed only) | -0.4 to +0.8 |

4. **One scam type is booked as a win: the private-router honeypot.**
   - It's the fee 0.3% WETH family.
   - 5 of the 24 "honest 10x" coins had public-router buys (13 to 52 each) but zero public-router sells. More than 99.5% of all trades went through one private router per coin.
   - Examples: 0x06d2d9229a9491969ecfb4c94d2e45c31b463f59, 0x53d025fbaee3057622d7ea0817b0b81b6645af66, 0xe0477657e3c374ed63a42ae6949e3e4475de4dd0, 0x5d2f94a1b1260e9e1fb88a21a8c22bc91c094791, 0x97d4c3c1b72aebb599b70002e9109b88248d51e3.
   - Honest clean 10x after removing them: 11.3% per coin.
5. **Which picks give clean 10x:**
   - Pond coins: 22% (50 picks). fee 0.05%: 11% (98). fee 1%: 0% (29 picks, none).
   - Pool liquidity over 10 ETH at our buy: 19% vs 8%, and the pull comes later (16 min vs 8 min). Holds in both halves (24% vs 13%, 14% vs 4%).
6. **The entry price is realistic.** A buy landing 3 s after the decision fills a median 0.8% above the booked entry (75th percentile +5%).

## Proposals (ordered)
1. **Honest scoreboard.** A pulled pool is worth 0 after the pull. A 10x only counts if it happened before the pull. Open picks on pulled pools get closed at 0. (The sell session found the same: claude/sell-system-v1.md.)
2. **Sellability gate at the decision.** Skip when the public routers have bought but never sold, and most trades go through a router that isn't public. At 30 s this flags 4 coins, 3 of them honeypots.
3. **Pond first.** Book every 0.01% fee, 10 ETH start launch at first sight (rule_fee001_first already does this) and score inside the pond. It beats every model book on clean 10x (22% vs 8 to 18%).
4. **Skip fee 1% launches.** 0 of 29 picks hit a clean 10x. History: 1.2%.
5. **Add pool liquidity at the buy** (the quote reserve from the Swap event's liquidity and price) as a feature or tiebreak.
6. **Schema fixes:**
   - aipaper.db and aipaper30.db have no dec_b column, and aipaper.db has an empty model column, so their fills can't be audited.
   - 85 catch-up picks were booked on pools that had already been pulled before the decision.

## Data needed to do this better
- **Archive RPC or a Blockscout API key.** It would let us find who funded each creator wallet and group rugs by operator (some operators may let coins run, others pull at once). The public RPC has no old state, and Blockscout blocks the container with Cloudflare.
- **ModifyLiquidity events, plus the swap sender (router), in the day DBs.** Right now they need separate RPC pulls.
- **tx.from for liquidity adds and removes** (who launched and who pulled).

## Data warnings
- The live book window is short: 4 Oct 14:00 to 5 Oct 09:00 BKK for matured picks. Small numbers.
- The sellable exit is estimated from the pool's liquidity at the 10x swap; a real sell could come out a little worse.
- The history study (14,158 launches) uses mcap.db sellable caps and the 6 h price peak.

Scripts: container /root/pp (fetch.py, parse.py, evalpp.py, pullchk.py, rmtx.py, mlpools.py, hist1.py). Exports: http://159.223.88.168/hx-c9ois/brx/pp_picks.csv.gz and http://159.223.88.168/hx-c9ois/brx/pools_map.csv.gz

Version 1, 5 Oct 2026. Latest: http://159.223.88.168/hx-c9ois/brx/pp-review-scams.md