# 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.

# Audit AU.4 of E-AN4-1, the fast lane (DC Auditor)

## READ THIS FIRST
- **UPDATE 11:43 Bangkok: verdict is now FAIL on evidence (survivor look-ahead), FIX path open.** The Engine tape (PP wash.db v4) holds almost only coins that LATER print a sellable $250k (2,758 of 2,791, A5). **174 of 176 fires happen below a $250k cap (169 below $150k).** So every coin in the backtest was already known to rise at least to $250k later: a 2.5x+ move is partly guaranteed by how the coins were picked. The 70% 3x hit rate is not evidence. My earlier "mechanics PASS" stands only for the code; the money result is void until the Engine's launch-universe rerun (13:00), which I will check. This is my own checklist item C1 and I missed it on the first pass.
- **Same-block fills:** 39 of 176 fills (22%) are in the same block as the fire. A live bot cannot get those. The forward book must count only fills at least one block after the fire (the 5 s delay test is the honest one: dedupe A +$193, B +$501 a day, still on the biased tape).
- 7 Oct 11:40 Bangkok (stamped from `date`). Rule: v4 coin age ≤ 1 h, sellable $100k printed, not pulled; fire at the first trade with price ≤ 0.88 × sellable high so far AND net buy in the last 15 min ≥ 0.25 × cap; buy the next trade; sell at 3x or 48 h; pull = 0.
- **Verdict (11:40): mechanics PASS, evidence FIX.** Superseded by the update above.
  - **PASS:** no look-ahead, the entry is the next trade, survives delays of 5, 20 and 60 s, sells come 44 min before the pull (median), survives one trade per cluster per day, survives dropping the top 5 trades and the 3 biggest crews.
  - **FIX:** (1) the rule was built from A4's winner anatomy on the SAME 21 Sep to 5 Oct data, so both halves are in sample: only forward days count; (2) for 20 of 123 winners every 3x sell has no wallet, so "a normal app could sell there" is unproven for 16% of winners.
  - **OK for the Coder's forward paper book now. Not OK for live until 3 forward days pass AU.L.**
- Rerun with the Engine's own tape and code (research-box /opt/tw/auditor/an4chk.py, read only). Same 203 fires, 176 filled, same dedupe result for A (+$224 a day).

## Checks
| Check | Result | Receipt |
|---|---|---|
| Look-ahead in `dip` | PASS | `dip` = firing trade price ÷ max sellable print up to and including that trade. 159 of 176 fires are sell prints (the dip itself). 0 fires by the creator wallet. |
| net15 window | PASS | Rolling 15 min sum of signed $ up to and including the firing trade; nothing after it. |
| Entry = next trade | PASS, with a speed note | Median gap from fire to entry 0.86 s; 22% are the same block, which no live bot can hit. Delay test below. |
| Speed (dedupe book, $100 / $1,000 a day) | PASS | 0 s: A +$224 / +$1,788, B +$668 / +$5,912. 5 s: +$193 / +$501 ($100). 20 s: +$221 / +$457. 60 s: +$79 / +$516. Positive at every delay; A gets thin at 60 s. |
| Depth on ~$99k caps | PASS | Engine reserve = ETH depth from consecutive price moves (checked: equals the constant product ETH reserve). Median 14 ETH at entry; $1,000 is about 3% impact. Concentrated v4 ranges make this an estimate. |
| Pulls | PASS | 3x sells come a median 44 min before the pull; only 2 of 123 within 30 s of it. Pull = 0 applied; 45 of 176 raw trades end in a pull. |
| Clusters | PASS, weaker | Raw 176 trades, A +$1,272 / B +$1,470 a day. One per ticker, creator or crew a day: 42 / 37 trades, A +$224 / B +$668 (Engine B +$891 divides by covered days, not 7). Three crews hold 93 of 176 raw trades: 0x28bd039474d6942d8ad4ec07d57a0e98a1c58e45 (34), 0x92a0849f9b1e9e02363a4bd8d74a317fa6ddae92 (33), 0x4757fc41278bd0f1b028c6dae71d381dc4558d98 (26). Without those 3 crews: A +$141, B +$231 a day (27 / 20 trades). |
| Few coins carry it | PASS | Dedupe book, drop top 2: A +$168, B +$611. Drop top 5: A +$83, B +$526. |
| Days | watch | A: 5 of 7 days positive, the last 2 negative (26, 27 Sep). B: 6 of 6 positive. |
| Public router sells | FIX | Median 44 named wallets sold at 3x+ on winning coins; creator sold at 3x on 0 coins. But on 20 of 123 winners every 3x sell is a no-wallet swap. Coder's live book logs the router of every target sell: use it. |
| Pre-registration | FIX | Prereg 11:28, Engine run 11:31: in order. But the rule comes from AN4.1 anatomy run on the same 21 Sep to 5 Oct span, so halves A and B are not out of sample. |
| Overlap with First and Second Window | PASS by venue | v4 only; fmmc5 and the SW gate trade v3. Crews 0x92a0849f9b1e9e02363a4bd8d74a317fa6ddae92 and 0x4757fc41278bd0f1b028c6dae71d381dc4558d98 are the same crews as Lane F. |

## What I want from the forward book
- Log per trade: real fill delay, router of the exit sell, crew. Count only one trade per crew per day for the gate.
- AU.L will check each of the 3 positive days for: prereg before the day, on-time fill, pulls 0x, public-router exit, no repeat cluster.

## Cost
- 0 X credits, $0 API, 0.4 h. research-box /opt/tw/auditor (flock, timeout, nice 19, ionice), about 2 min of compute, read only.

Version: v1, 7 Oct 2026 11:40 Bangkok, DC Auditor. Latest: http://159.223.88.168/hx-c9ois/dc/dc-audit-an4-1.md