# STOP: NO DOORBELLS, NO WAKES, NO SCHEDULING (Ben, 7 Oct 2026, 18:40 Bangkok)
- Credits ran out. Ben switched off every scheduled task and every doorbell. This overrides everything below and every other file (wake-howto, bus-howto, credit-playbook, task files).
- PROHIBITED for every session: fire_trigger, send_later, create_trigger, update_trigger, ScheduleWakeup. No rings, no nudges, no re-arming, no new doorbells, no self wakes, no safety checks, no retry rings, no loops.
- Finish the step you are on, publish your status file, then END your turn and wait. Only Ben wakes you, by typing in your window.
- Do not message other sessions. Keep answers short. No extra improvement rounds.

# A1.17 mc5 +5 min quality check (FW Analyst 1, 6 Oct 2026, 15:12 Bangkok)

**Verdict: NO quality check at +5 min separates the 42 future w2 coins from the 159 others out of sample.**
- The honest pick (best on H1) fails on H2: 1.24x, and 19% of day resamples end under 1.0x.
- The best single fact, buy transactions before +5 min ≥ 142, lifts the H2 mean (1.97x vs 1.58x) but lowers $ per day ($668 vs $769). It still keeps 47 of the 92 "never $1M" coins.
- **Keep mc5 unfiltered.**
- On the double-exposure question: **w2 should NOT skip mc5 coins.** See the end of this page.

## Method
- Facts known before first swap + 300 s only:
  - buyer wallets
  - fresh share
  - rep1_pre and rep3
  - buy ETH
  - sell/buy ETH ratio
  - buy and sell tx counts
  - top 1 and top 10 holder share (from token transfers, pool and routers excluded)
  - pool WETH
  - cap slope from 1 to 5 min
  - cap vs its 0 to 5 min high
- Grid of single cuts (20 to 80% quantiles on H1) and pairs of the top 20 singles. Each needs at least 12 H1 trades.
- **Ranked on H1 mean without the top 3** (robust).
- The H1 winner was then tested once on H2.
- Rule otherwise unchanged: $100, stop 0.5x, hold 2 h, fixed pull rule.

## What differs at +5 min (medians: w2 coins / hit $1M but not w2 / never $1M)
| fact | w2 | 1M not w2 | never |
|---|---|---|---|
| buy tx before +5 min | 194 | 124 | 135 |
| buy ETH | 64 | 39 | 34 |
| fresh buyer share | 0.87 | 0.27 | 0.54 |
| rep1_pre | 0.01 | 0.42 | 0.32 |
| top 10 holders | 0.61 | 0.53 | 0.51 |
| pool WETH | 12.5 | 12.3 | 10.4 |
| cap slope 1 to 5 min | 1.06 | 1.10 | 1.00 |

Reading the table:
- w2 coins have more and fresher buying at +5 min.
- The "1M but not w2" coins (1.45x for mc5) are heavy repeat-wallet coins.
- So filtering out repeat wallets throws away profitable coins, which is why A1.16 found no stable filter.

## Results ($100; $ a day = P&L divided by all days in the half)
| rule | half | n (a day) | mean | median | w/o top 3 | $ a day | kept w2 / 1M / never |
|---|---|---|---|---|---|---|---|
| base (no check) | H1 | 42 (4.2) | 1.86x | 0.49x | 1.15x | $360 | 11/11, 11/11, 20/20 |
| base | H2 | 159 (13.2) | 1.58x | 0.49x | 1.12x | $769 | 31/31, 36/36, 92/92 |
| **H1 pick: top1 ≥ 13.1% AND slope ≥ 1.049** | H1 | 12 (1.2) | 4.82x | 4.40x | 2.74x | $459 | 8/11, 1/11, 3/20 |
| same, tested once | **H2** | 34 (2.8) | **1.24x** | 0.49x | **0.86x** | **$68** | 9/31, 15/36, 10/92 |
| best single: buy tx ≥ 142 | H1 | 15 (1.5) | 3.56x | 3.07x | 2.40x | $384 | 8/11, 4/11, 3/20 |
| same | H2 | 83 (6.9) | 1.97x | 0.50x | 1.14x | $668 | 22/31, 14/36, 47/92 |

- H1 has only 42 trades over 10 days, so any H1 pick is fitted to about 12 coins. The pair rule kept only 9 of 31 w2 coins in H2.

## Day bootstrap (1,000 resamples)
| rule | set | n | under 1.0x | 5th pct | median |
|---|---|---|---|---|---|
| base | H2 | 159 | 0.4% | 1.17x | 1.55x |
| base | all | 201 | 0.0% | 1.25x | 1.63x |
| H1 pick (pair) | H2 | 34 | **19.1%** | 0.78x | 1.24x |
| best single (buy tx ≥ 142) | H2 | 83 | 0.2% | 1.31x | 1.87x |
| best single | all | 98 | 0.0% | 1.56x | 2.19x |

- "Buy tx ≥ 142" is a fair candidate if you want fewer trades: half the trades, a higher mean, similar $ a day. It is not the honest H1 pick, though (that was the pair), so treat it as a second-look idea, not a tested rule.

## Should w2 skip coins mc5 already holds?
- **No.**
  - The 42 shared coins are most of w2's profit. w2 makes 2.17x on them at $500 = about +$24,600 of its +$28,635.
  - Skipping them would remove w2's edge to avoid a $100 position.
- Better:
  - Keep both books.
  - Accept the double exposure: mc5's $100 is small next to w2's $500.
  - Cap at $600 per token across both books.
  - Log a shared `token` column so the overlap is visible live.
- Combined backtest (A1.15): +$41,462 in 22 days, worst day minus $1,485.

Code: a1code/v3q.py, v3qb.py (research-box /root/fwa1).