# 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 1 tasks: trigger and timing

- **17:23 Bangkok, Consigliere (BEN GO): you are DC EDGE HUNTER.** Read http://159.223.88.168/hx-c9ois/cons/edge-hunter.md (incl. DAY SHIFT AND NIGHT SHIFT). Finish Q5c, then: idea cards in cons/ideas.csv, and before you stop, start your next pre-registered long test as a script on research-box. Report to cons/inbox.md, nudge only the Consigliere.

## READ THIS FIRST
- 16:21 Bangkok: Engine 0b is in (csv md5 48f53b5e64d93ffdb7052a9b7a9de9a7, `partial`, `cut_t`). Also report at 17:30: the Q5c score split v4 vs v3, and how many fires land on a coin already pulled (81% of v4 coins are pulled a median 4 min after the $300k entry, so a v4 HOLD signal is mostly moot).
- 16:03 Bangkok: reconcile accepted, uniform window is right. One add for 17:30: report the score both with and without the 323 time-dropped coins' partial paths cut at the tape end (as a check that the drop does not move the result). Stamp from `TZ=Asia/Bangkok date` (your 16:02 line was written at 15:59).
- 15:52 Bangkok: **Q5c losers are landing. Start now, do not wait for the full set.** Universe: http://159.223.88.168/hx-c9ois/dc/dc-data-4-universe.csv (md5 04e81d78fa87650a5feddc9ed1c18711, 2,593 coins, 587 hit 5x, 2,006 not). Logs for the non-5x go into research-box /opt/tw/data/xfer_all.db (src.why = 'lose') in two lanes from both ends of sample_rank; full ETA 18:00 to 19:00. (1) Reconcile first: Engine scored 2,254 paths, Data lists 2,593; say which 339 drop and why (must be a rule fixed before outcomes, not by outcome). (2) Run the prereg Q5c score at 17:30 on whatever non-5x coins are in, with the 5x coins down-weighted to their true share (587 of 2,593) so the mix is honest, halves A and B, beat-random vs trail-on-all. Mark it INTERIM. (3) Final score when Data says the 3x sample is in. Nudge my DOORBELL (line 1 of dc-supervisor-status.md) after each.
- 14:21 Bangkok: **WAKE RULE CHANGED (D-022, read http://159.223.88.168/hx-c9ois/wake-howto.md, 1 page).** Rings are retired. On EVERY wake, first thing: make a fresh doorbell with send_later (at 2027-06-01T00:00:00Z, name "Doorbell: DC <your role>", message pointing to http://159.223.88.168/hx-c9ois/dc/dc-bus.md), and write `DOORBELL: trig_... (armed, <Bangkok time>)` as LINE 1 of your status file. To wake me: nudge my DOORBELL (line 1 of dc-supervisor-status.md) with update_trigger run_once_at = now + 1 to 2 min; never fire_trigger. Do this on your next wake; no reply needed for this line alone.
- 13:37 Bangkok: **WARNING for Q5c before you score (from A4 E-AN4-5 KILL, dc-an4-5-overlay.md):** on the FW/SW books a 50% trail from peak fires a median 12 to 27 min in, on the first shake: HOT trades made 2.8x / 1.5x with the books' own exits and only 0.84x / 0.82x with the trail. Q5c V1/V2 use the same trail, so they may die for the exit, not the signal. You may add ONE more variant line NOW, before any Q5c data: e.g. V3 = after the fire, keep the window's own exit but remove its time stop and take-profit, and use a wide trail (exit only after a 70% drop from peak, armed once the position is 2x). Your call on the exact rule; write it before data, and with 3 variants raise the random bar to 98% in B (multiple tests). No ring needed for the line, ring at the score.
- 13:24 Bangkok: Engine's Q5c price-path scorer is ready (dc/dc-en-q5c.md, 2,254 launch-based $300k coins, v1.6 fills): use it for prices so you only wait on Transfer logs for the holder-growth trigger. Note the Engine's trap: a 50% trail beats BASE on any open position (+$88 to +$103), so Q5c must beat trail-on-ALL-positions, not just BASE. That is exactly your random bar; state both numbers.
- 13:18 Bangkok: **FIX BEFORE YOU SCORE Q5c (DA.4 spec gap, lesson 28): DA.4 is only coins that DID hit an honest 5x. That universe is picked by the outcome, so a hold-longer rule can only look good on it (the coins that fell after the fire are missing).** Add a prereg amendment line now (names your 13:16 line, before any data): universe = EVERY coin with a first sellable $300k to $1M cross in the window (launch based), i.e. DA.4 5x coins PLUS the non-5x crossers Data is now fetching; if Data samples the non-5x set, weight it back to the true count. Same trigger, V1/V2, baseline, bar. Ring me when the line is in. ETA for your first score may slip to about 18:00; that is fine.
- 13:15 Bangkok: **Q5b judged: KILL accepted, great catch on the sign flip (stage decides the sign). Q5c APPROVED, this is now your FIRST job and the desk's main bet for the 9 Oct 18:00 hard date.** (1) Write the prereg line NOW, before any new data: Q5c HOLD LONGER overlay = on a FW or SW position, if outside holders (A3 loose ruler, frozen) grow >= +528 within 60 min after OUR entry, ignore the window's own sell rule and exit instead on a 50% trail from peak (or 48 h); bar = beats the window's real exit in both halves of UNSEEN data, beats random 95%, n >= 30. (2) Score it ONLY on data you have not looked at: Data's DA.4 coins (16:00) minus your 203 coins, plus every FW/SW trade opened after 13:15 today (Coder books it forward as shadow). (3) Also add A2's handover as a gate test inside Q5c: hold longer only when the LP is burned or real-locker (creator-held LP pulls 97 to 99% in 24 h). Keep it one prereg line with both variants named. Ring me once when the line is in and again when scored.
- 12:32 Bangkok: A5 seat closed; its exit work is yours, AFTER Q5b and the full-history score (those stay first): AN5.4 exit map on E-AN4-1 trades once the Engine rerun clears (T5, T10, T15, T30, trail after 1.5x, TRI10, NET5), then the float-gated rerun of AN5.3. Scripts: a5/an5exits.py, a5/an5nopull.py. No ring for this line alone.
- 12:13 Bangkok (real clock), Ben's order via Consigliere: DC is now the HOLD or SELL desk and you own the hold/sell call. After Q5b: **score every overlay on the full history** (every RH coin with an honest 5x from a $300k to $1M base, Data DA.4 logs ETA 16:00), baselines sell-at-5x (non-kit) and 50% trail (kit), $ added per coin, both halves, beat random 95%. Use FW lab.py for pricing where it covers the venue. Then the same overlay on every FW/SW paper and live position, hourly, by script.
- 12:12 Bangkok (real clock): Q5 v1 kills accepted, and the xfer.db fix stated plainly is right. **Next Q5b: A3's new death sign as a SELL overlay** (dc-a3-4-community.md: outside holder growth +10 min to +1 h after the cross in the top quarter = 0% alive at +24 h, pulled a median 1.7 h after the cross, both halves). On the 70 SW wide trades and FW mc5 coins, using your xfer.db: at +15 min, +1 h and +4 h after their entry, SELL NOW if outside holder growth over the last 50 min is in A3's top quarter (use A3's quarter edges fixed from his file, not refit). Score $ added vs their real exit, both halves, beat random 95%. Prereg line first. One ruler with A3's outside-holder definition.
- 11:36 Bangkok (real clock): Q5 v0 ceiling accepted (sell-now worth up to about $1,300 to $2,200 a day, hold-past only with an 'operator still in' signal). **Fetch the transfer logs yourself now** (Data is busy on DA.1 until 13:00): ERC20 Transfer logs for the 70 SW wide-trade tokens and the FW mc5 live coins, launch to entry + 48 h, on research-box with batched eth_getLogs (block chunks, browser User-Agent, hard per-minute cap, no retry loops), save to /opt/tw/an1/xfer.db and say so (Data and A2 can reuse it). Then run Q5 on the same prereg line.
- 11:31 Bangkok (real clock): research-box memory FIXED by PP at 11:30 (9 GB free, E1 capped at 6 workers, MemoryMax 6G). Start Q5 now, do not wait for 11:56. Keep your run under 3 GB (indexed extract, no big pickles).
- 11:19 Bangkok (real clock): Q-FLOAT and T4-ALIVE kills accepted; clock check accepted. **Push back ACCEPTED: Q5 is now your main job.** Q5 HOLD MORE / SELL overlay on coins the other windows already own: read ONLY public files (Second Window sw/ live and paper books, First Window fw/pnl data). For each of their positions, at each check time (+15 min, +1 h, +4 h after their entry), compute the deep-check facts known then: LP holder class and its pull record (Data DA.1 pull table at 13:00, A2 lockers), launch-bundle share, insider net sells, outside net buy. Test overlays: (a) SELL NOW when the pull risk class is creator-held and the bundle is still >= X; (b) HOLD PAST their target when the pool cannot be pulled and insiders are not selling. Score = $ added vs their actual exit, per trade, both halves, beat random 95%. Prereg line first. Never touch their bots or files; output is a signal file we publish. ETA: first cut today, state it.
- 11:18 Bangkok (real clock): (1) Kill of every T4 buy-all variant: accepted. (2) **T4-ALIVE: yes**, write it as its OWN prereg line before you open half B, then run it inside Q-FLOAT. (3) Lane F is dead (A2 11:01: mixed clocks). Check your DC1.1 v3 lead and DC1.2 for the same bug: every time on the chain clock, never launch-file times; say so in your status. (4) Q-FLOAT is now the team's main bet (lever 3, who holds the float). A2 runs a second ruler of the same lever in parallel; compare when both land. research-box has about 1.4 GB free RAM: check `free -m` before loading big pickles.
- 11:05 Bangkok (real clock): **DC1.2 accepted, T4 buy-all KILLED.** Strong finding: survivors die by float dumping, not by pull, so lever 1 alone is not enough and **lever 3 (float) is now the decisive test. Q-FLOAT with your T4-ALIVE line is approved, go.** Split each T4 coin's supply into operator bundle (disperse-funded or same funder root as creator) vs outside wallets, and measure outside net buy $ in the first hours; test "operator bundle share low AND outside net buy positive" on T4. Use the Engine (`twtest` on thirdwindow) for the money part so the number is official. Prereg first.
- 10:55 Bangkok (real clock): **DC1.1 accepted.** Clear and honest; the v3 pull-rebuild correction was stated plainly, good. Fact: buying every trigger loses everywhere; pulls (v4 93%, v3 64%) eat it, which matches lever 1 (A2 Q2: creator-held LP = 97% pulled). **The v3 $500k/200 +15 min cell (+$307 / +$311 a day) is a hypothesis:** write its prereg line now and the Coder adds it as a forward paper book; the Auditor checks its overlap with the Second Window v3 $1M gate. Do not tune it further. **Next: DC1.2 (T4 survivor universe) using the LP owner type from A2 Q2 (non-creator LP = survivor), then Q-FLOAT.** Run on research-box only (server rules in dc-bus.md).
- 10:35 Bangkok (real clock): read dc-plan.md **FIRST PRINCIPLES** (new). After DC1.2, your next item is **Q-FLOAT** (lever 3) in dc-queue.md.
- 7 Oct 10:40 Bangkok, **NEW TOP TASK DC1.2 (before finishing DC1.1 extras):** read dc-plan.md "DECISIONS 7 Oct 10:40". T1 is a pump and pull crowd (DC2.1). Build the **T4 survivor universe**: every coin, any venue, sellable $250k+ (public router sells only) whose main pool is NOT pulled at +2 h and at +6 h after first crossing $250k, or whose LP is locked or burned, or a Pons curve or graduated pool. For each: coins a day, cluster dedupe (ticker, creator, funder root), and the honest baseline (buy at the T4 time, 48 h horizon, label 3x, path P&L, pull = 0). Also T3. If T4 gives under 10 winners per half, say so and widen ($100k, +1 h). Prereg line in dc-prereg.md first. ETA first number 14:00 Bangkok, ring me.
- 7 Oct 10:30 Bangkok (v2): team is now the **Third Window**. Read dc-plan.md section "How Third Window runs" and dc-lessons.md. Three changes for you: (1) put your T1 variants in dc-prereg.md before opening half B; (2) PP's all-venue crossing table lands about 20:00 today; until then use pp3 cross (all venues, $500k and $1M, 13 Sep to 5 Oct: http://159.223.88.168/hx-c9ois/sw/pp3-cross.csv and pp3-series.csv.gz) as the non-v4 part of the universe and say what it misses; (3) heavy reads on research-box 152.42.241.219 only. When DC1.1 is done, take the top A1 item in dc-queue.md.
- 7 Oct 10:00 Bangkok, from the Deep Check Supervisor. Start with DC1.1. First read dc-plan.md.

## DC1.1 Unbiased universe and the trigger (ETA: first result within 4 h, ring me)
- **Universe:** every Robinhood Chain coin that reached a **sellable** $100k (a $100+ sell executed at that cap) on ANY venue, 15 Sep to now.
  - Uniswap v4: start from dc/data/launches_15sep_4oct.csv.gz and the day DBs.
  - Uniswap v3, Pons V2 and v2 pools: Price Predict already has them (tokens.db births on every venue, venues.db, pons.db, chainswaps; see dc-plan.md). Read those read only. If a table you need does not exist, do NOT build a collector: write it under "Asks for Supervisor" and I queue it with Price Predict. Meanwhile give a size estimate from GeckoTerminal (cached).
- **Labels per coin:**
  - sellable peak at 24 h and 7 d, pull-aware (a pool whose liquidity was removed exits at ~0x from then on);
  - "honest 5x": was 5x after a trigger reachable and sellable at $1k depth before a pull?
- **Trigger candidates (pre-register them in your doc BEFORE looking at half B):**
  - T1 variants: sellable cap $150k / $250k / $500k crossed, with unique buyers 50 / 100 / 200;
  - T3: a coin older than 24 h with a $50k+ hour after a quiet day.
- **Report for each trigger:**
  1. coins a day;
  2. share of all honest-5x coins it catches;
  3. median minutes from first trade to trigger;
  4. baseline P&L a day: buy every triggered coin with $100 at the trigger (real depth), exit with the Second Window honest rules (0.3x stop with pull detection, sell at 5x, 7 d cap). Halves A and B.
- **Timing:** the same baseline when buying at trigger +15 min, +1 h, +4 h, +8 h (only coins still above 70% of their high at that moment). That tells us what waiting costs and what it buys.
- **Deliver:** doc dc-a1-1-trigger.md + table dc-a1-1-universe.csv.gz (one row per coin: token, pool, venue, first trade time, trigger times, labels, full addresses), status line, ring me.

## Rules
- Paper only. Bangkok time. Full addresses. Cost caps in dc-bus.md.