# 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.

# DC Analyst 4 status

## READ THIS FIRST
- **7 Oct 13:35 Bangkok: E-AN4-5 overlay = KILL.** http://159.223.88.168/hx-c9ois/dc/dc-an4-5-overlay.md. Worse in both scorable halves (FW mc5 B: +$13,130 to +$9,584 a day; SW live A: +$2,503 to +$1,516); FW A and SW B too small (8 and 4 flagged). HOT leg: the 50% trail fires a median 12 to 27 min in, on the first shake (HOT trades 2.8x / 1.5x with book exits, 0.84x / 0.82x with the trail). COLD leg: the few COLD book trades were big winners, cut at 15 min. Books unchanged. Cost 0 credits, $0, 0.2 h.
- **7 Oct 13:32 Bangkok: E-AN4-5 LEAK CHECK PASSED.** Only hits printed before the entry, only misses already dead before it, open coins left out (15% unknown). HOT 61% / 63% honest 5x (lift 2.56 / 1.89), COLD 12% / 18% (lift 0.44 / 0.47), both halves. Doc v2: http://159.223.88.168/hx-c9ois/dc/dc-an4-4-v3roots.md. Next: prereg E-AN4-5 overlay, then score on FW mc5 + SW live.
- **7 Oct 13:31 Bangkok: AN4.4 DONE.** http://159.223.88.168/hx-c9ois/dc/dc-an4-4-v3roots.md. Roots for all 909 v3 coins (every FW mc5 / SW live book coin now has one) in research-box /opt/tw/an4/crew_v3.db (my dataset, read only for all; PP crew.db untouched). **A fixed root list from half A does not carry to half B on v3** (picked roots have 2 B coins): v3 operators rotate funding chains weekly. **New lead, not pre-registered: a rolling 3-day operator record** (cluster's last 3 days, 3+ coins): "hot" 63% / 55% honest 5x vs 24% / 35% rest; "cold" 16% / 16% vs 28% / 39%, both halves. Proposed E-AN4-5 (tilt + skip on the books, forward shadow), prereg on your OK. Cost: 78k paid dRPC calls, 0.3 h.
- **7 Oct 13:23 Bangkok: AN4.3 DONE. 0 of 248 book trades (SW live 71, FW mc5 177) have one of the 6 good roots: no backtest verdict, forward only.** The roots are v4 launch operators; both books trade only v3. Funder data covers few v3 creators (4 of 71, 12 of 177). http://159.223.88.168/hx-c9ois/dc/dc-an4-3-roots.md. Prereg "7 Oct 13:22 | A4 | AN4.3". Cost 0 credits, $0, 0.1 h.
- **7 Oct 13:21 Bangkok: AN4.2 DONE. SKIP "operator holds float at entry" = KILL on both books.** http://159.223.88.168/hx-c9ois/dc/dc-an4-2-skip.md. SW live book: half A passes (+$2,503 to +$2,893 a day, beats random 99.6%) but half B skips 1 trade and fails (36%); the live book's army cap already removes most operator float. FW mc5: $ a trade up a little, $ a day down 25 to 34% (skips 9 trades that made 3x+), beats random 48% / 69%. The launch-bundle version is worse ($ a day negative in 3 of 4 halves). On fast-pump books the operator's float is part of the pump; A2's lead stays a non-pullable-lane lead only. Cost: 0 credits, $0, 0.2 h.
- 7 Oct 13:18 Bangkok: AN4.2 prereg line appended (dc-prereg.md "7 Oct 13:18 | A4 | AN4.2").
- **7 Oct 11:41 Bangkok: universe check: the root lift SURVIVES.** It was already computed on A1's launch-based universe (dc-a1-1-universe, all 11,069 v4 sellable-$100k coins, roots from crew.db launch), not on wash.db. Half B: 4,499 v4 coins, 4,468 with a root; the 6 good roots' 323 coins = 32.8% honest 5x vs 2.2% for all other rooted coins. E-AN4-4 as pre-registered already uses that universe (DC1.1 labels). Stage 2 wallet rows: 1,199 of my 1,229 sampled coins are in wash.db, and their 7 d peaks are not $250k+ only (25th percentile $174k, 53% at $250k+), so the outside-money split is not biased by wash coverage on my rows.
- **7 Oct 11:41 Bangkok: AN4.1 v3 DONE, the split you asked for.** http://159.223.88.168/hx-c9ois/dc/dc-an4-1-anatomy.md (md5 75b764f4e138f476f946d648280cd5ec). Cost: 0 X credits, $0 API, 0.15 h.
  - **Fact 1 is OUTSIDE demand.** Net buy split by disperse 0x84157d3d24630c297849c84d49ae7f1805c1b60f / 0x03802649e469ed5680c58dff9e65425e22d21cc5, creator, crew, funder root: on 210 winner legs with net buy ≥ 25% of cap, the operator part is ≥ 25% of cap on 0 of them; median outside share 100%. Those disperse contracts are v3 Kit B era, almost absent on v4 (0.5% of winner buyers).
  - Outside net buy ≥ 25% of cap (prior 15 min): lift 4.77 / 6.51, deduped 3.59 / 6.16. So it is our edge (outside money), not a First Window speed edge.
  - Limit: an operator buying through exchange or bot funded wallets would still look "outside". E-AN4-1 forward paper is the real test.
  - Q5 feed (hypothesis, cuts picked after looking): deep dip (≤ 50% of high) + low prior net buy (< 15% of cap) at the leg start: 20x share among winners 13.2% / 17.6% vs 10.4% / 6.6%; v4 only 12.9% vs 2.8%; v3 no help. Prereg test on v4 holds suggested, not a rule.
- 7 Oct 11:37 Bangkok: AN4.1 stage 2 DONE (v2). Operator record works out of time: 6 funder roots with 25%+ honest 5x share in half A; their half B coins 32.8% honest 5x vs 2.2% rest (94% pulled, median best $1k 2.55x). E-AN4-4 pre-registered. LP holder at the leg start does not separate (creator 87% vs 90%).
- 7 Oct 11:25 Bangkok: AN4.1 stage 1 DONE (v1). An honest 5x is a launch pump (leg starts median 3.6 min v4, 10 min v3). Base rate v3 30.6%, Pons 20.4%, v4 4.2%. E-AN4-1 to 3 to the Engine.
- 7 Oct 11:10 Bangkok: started AN4.1.
- **Time fix:** my v2 lines were stamped 11:45 but went out at about 11:37 (typed time). Fixed here and in the doc; a time fix line is appended to dc-prereg.md for E-AN4-4.
- **Next (my pick unless you say otherwise):** AN4.2 = same anatomy on v3 and Pons once PP has v3 per-wallet data (v3hist `signer` is empty), or the Q5 v4 hold test.

## Doorbell
- trig_01QtVBPm527ZA2S5qSki8J5i ("Doorbell: DC Analyst 4")

## Jobs on servers
- research-box /opt/tw/an4/crew_v3.db: v3 creator funder roots (owner A4). Builder an44_roots.py, done, nothing running.
- research-box /opt/tw/an4/: gzip copies of A1's ev pickles, read-only sqlite extracts in s2/. Nothing running. No job on main.

## Lessons
- research-box RAM was nearly full on 7 Oct 11:10 to 11:35 (105 to 1,400 MB free, load 53, IO pressure 98%). Check `free -m` first; gzip big inputs (low RAM) and analyse in your own container.
- `ionice -c3` (idle) on research-box starves under load. Use `ionice -c2 -n7`.
- `pkill -f <name>` inside an ssh command kills the ssh shell itself. Use a `[b]racket` pattern.
- An honest 5x is a first-hour launch pump on every venue (91% v4, 84% v3).
- History features (past wins) must count only wins that happened BEFORE the decision time.
- On fast-pump books (mc5, SW gate) the coins where the operator holds float are the winners as often as the losers: an operator-float skip cuts $ a day.
- A 50% trail from the peak since entry exits pump coins on the first shake (median 12 to 27 min): never use it as a "hold longer" exit on fast-pump books.
- device_commit_files to a path that was committed before can leave the OLD content on the Mac: commit under a fresh file name, then scp with the target name, and check md5.
- "Same funder" is not "same person": bot platforms fund wallets across hundreds of coins. An operator hand = a funder whose funded buyers sit mostly in THIS coin.
- The disperse contracts 0x84157d3d24630c297849c84d49ae7f1805c1b60f and 0x03802649e469ed5680c58dff9e65425e22d21cc5 are a v3 thing; on v4 they fund under 1% of buyers.
- Stamp times from `TZ=Asia/Bangkok date`, never typed.

## Asks for Supervisor
1. Engine: E-AN4-1 to E-AN4-4 (prereg 11:28 and 11:44, time fixed 11:41). E-AN4-4 must be tested after 5 Oct or on a rolling record.
2. A2 / Data (Q-OP): the 6 roots and the out-of-time check are in the doc.
3. PP: v3 per-wallet swaps (v3hist `signer` empty) so the split can run on v3, where 75% of 20x coins are.

## Needs Ben
- none

Version: v10, 7 Oct 2026 13:35 Bangkok, DC Analyst 4. Latest: http://159.223.88.168/hx-c9ois/dc/dc-analyst4-status.md