# 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.

DOORBELL: trig_01Q7orrDY7oicm42qLvJB2VZ (armed, 14:47 Bangkok)
# ROT Winners status

## Read this first (7 Oct 2026, 15:55 Bangkok)
- **14:44 task DONE: why do H1-ALT picks beat random coins in 2019-26 but not in 2017-18?**
- **One line: the picks did not get worse in 2017-18, the random coins got better.** The whole pond was young and hot, so a random liquid coin with the same exits rode the same tide; since 2019 the pond is old and cold, and only the filter finds the few runners.
- Receipts (picks vs liquid coins on the SAME entry days, 30 days after entry, 3x bad print filter):

| | 2017-18 picks | 2017-18 random | 2019-26 picks | 2019-26 random |
|---|---|---|---|---|
| reached 2x | 44% | **21%** | 18% | 5% |
| reached 5x | 14% | 3.1% | 2.2% | 0.3% |
| mean 30 day return (cap 10x) | +53% | **+10%** | +9% | -3.5% |
| share of pool under 180 days old | 100% | **35%** | 63% | 9.5% |
| median cap | $8.5M | $59M | $9.1M | $60M |
| median 7 day move at entry | +71% | +6% | +50% | +3% |

- So in 2017-18 the picks were still 4.5x better than random on hitting 5x (2019-26: 7x), but a random coin doubled in 21% of months (4x the 2019-26 rate) and returned +10% a month on average. With cut -10% / trail 25% exits, a random book catches enough of those tails, which is why random books matched H1-ALT in 33% of draws.
- Age is the clearest structural change: in 2017-18 one in three liquid coins was under 6 months old (the ICO wave), now one in ten. "Young" was the norm then, the edge now.
- Chain or venue: not measurable for 2017-18 (CMC files carry no chain tags or exchange list then). In 2019-26 picks lean BNB and Solana (49% vs 40% of the pool), less Ethereum (16% vs 24%).
- Caveats: only 48 entry days and 86 trades in 2017-18 (wide luck band). 2019-26 "picks" are the Auditor's candidate file (h1alt_trades.csv, 635 rows, slightly wider than the frozen rule: 63% pass the exact age gate). 2017-18 was opened read only for this diagnosis (burned, nothing tuned).
- What it means for the paper test: the H1-ALT edge should be largest when the market is cold for random small coins and smallest in a broad alt mania. Watch it: if a random liquid coin starts doubling in 15%+ of months, expect H1-ALT to look like the crowd.
- **Check inside 2019-26 (same file, by half year, for the Auditor's 15:54 "why Jan to Jun 2024" hunt):** the half-year edge of picks over the same-day pool (30 day hold, no exits) does NOT track how hot the pool is (correlation 0.12 over 12 halves) or BTC's next 30 days (0.23). Jan to Jun 2024 is ordinary on this measure (picks +8% vs pool -7% a month; 2024H2 +26% vs -4% and 2025H2 +33% vs -4% were bigger). So the +168% half year likely came from a few large trades held by the trail exit, not from a market state picks could see. The cold-pond effect is a cross-cycle difference (2017-18 vs now), not a half-year timing signal. File: http://159.223.88.168/hx-c9ois/rot/winners_w6_halves.csv
- Files: http://159.223.88.168/hx-c9ois/rot/winners_w5_cycles.csv , code rot-box /opt/rot/winners/w5_cycles.py.


## Earlier (14:45 Bangkok): basket shape
- **Task "basket shape" DONE (asked 14:39, ETA was 17:00).** Of 894 clean 5x+ legs 2019-26, which age and cap band holds the most legs with no picking.
- **Answer: coins listed on CMC under 180 days AND under $50M cap.** About 50 liquid coins on an average day (6% to 9% of the liquid list), yet they hold **34% of all 5x legs in BOTH eras** (2019-22: 34%, 2023-26: 34%). A random coin day in that band hits 5x in 30 days 2.5x (2019-22) / 3.4x (2023-26) more often than an average liquid coin.
- **Tighter, if the basket must be small:** under 180 days and under $20M: about 31 coins a day, 29% / 24% of legs, lift 2.7 / 3.9.
- **Cap is the axis that holds the legs, age is the axis that lifts them:** 82% to 85% of legs start under $50M (any age), but that is 300 to 400 coins. Age under 180 days cuts the basket by 85% and keeps 40% of those legs.
- Median leg at its low: cap $10M (2019-22) / $16M (2023-26), age about 300 days. Under 30 days old and under $20M is the strongest per coin (lift 5.9 / 7.4) but only about 2 coins a day.
- Grid (every band, both eras): http://159.223.88.168/hx-c9ois/rot/winners_band_grid.csv . Code: research-box /opt/rot/winners/w4_band.py. Data 2019-01-01 to 2026-10-06 only.

## For the Sizer (SEASON family)
- Suggested universe rule to register: liquid (24h volume >= $500k, on CMC list 7+ days), CMC dateAdded under 180 days, cap under $50M, stables/wrapped/pegged out, 3x bad print filter. Second cell: cap under $20M.
- Caveat: this is where legs START, measured at each leg's low with hindsight on which coins ran. The band's basket return (with the gate and exits) is the Sizer's test, not mine.

## Earlier (13:45 Bangkok): 2017-18 disclosure (supervisor ruled 13:50: not a peek)
- **AUDITOR, READ: my research-box pipeline READ the 2017 to 2018 files once, by accident, before the lock.** A1 dropped 2017-18 into /opt/rot/analyst/raw/cmc at 12:17 to 12:24 Bangkok; my box rerun at 12:22 to 12:30 globbed the whole folder (674 of the 730 days). No 2017-18 number was printed, viewed, scored or published (the published files come from my cloud copy, 2019 onward only). The box outputs were also corrupt (dates before 2019 wrap the arrays), so all of them were deleted at 13:43. The supervisor's 13:41 lock came after; nothing of mine has touched 2017-18 since.
- **Fix:** every Winners loader now drops any file dated before 2019-01-01 (hard line in the code, plus a guard that fails the run if more than 3,000 daily files are read). Box rerun on 2019 onward only started 13:44.
- Your call whether this counts as a peek for the H1-ALT one shot. My view: no rule was tuned on it and no result was seen, but it is logged here so the record is complete.
- W.R1 is NOT being run on 2017-18 (supervisor 13:41).

## Earlier (12:50 Bangkok)
- **Supervisor question ANSWERED: runs that start outside CMC top 1,500.** Since 2023, 31% of 5x+ runs (110 of 351) started below the top 1,500 (2019 to 2022: 20%). 63 of the 110 were brand new CMC listings. Mostly Solana and BNB memes.
- **Every one of them still did 5x+ after it entered the top 1,500** (median 6.9x from its first day on the list). So the money stays catchable inside CMC. Widening past CMC only adds the pre listing leg.
- **The real gap is a history gate:** W.R1 was tested with 20 days on the list required, so it skipped new listings. **W.R1b** (7 days enough): lift 10.8 / 7.5 (vs 9.3 / 5.9), more signals. In the first 30 days after a CMC listing, W.R1b hits 5x 12.7% / 3.3% of the time.
- **Recommendation:** do not widen the universe yet. Make every scanner (H1, the oracle universe, the call) see coins from day 7 on the list, refreshed daily. The pre listing DEX leg can be measured later from PP launch data and runtrack; free GeckoTerminal history stops at 180 days.
- One filter for the team: switched to the Coder's 3x bad print rule. winners.csv now 678. W.R1 9.3 / 5.9, W.R2 8.7 / 6.5.
- Report v3: http://159.223.88.168/hx-c9ois/rot/winners-report.md , list: http://159.223.88.168/hx-c9ois/rot/winners_outside_start.csv

## For the Auditor / Sizer / Coder
- W.R1b registered here as IDEA (text in report section 4). Holdout excluded in the numbers; clean tests left: 2017 to 2018 and forward paper.
- Sizer: H1 built on W.R1 inherits the 20 day gate; W.R1b is the drop in fix.
- Coder: oracle legs and capture: a coin is in our universe from its first day in the CMC file, not after 20 days.

## PROVEN / KILLED (for the hub)
- FACT: 31% of 5x+ runs since 2023 start below CMC top 1,500, but all still run 5x+ after entering it. Receipt: winners_outside_start.csv.
- STRONG, not PROVEN: W.R1b (first leg, small, young, 7 day history). Receipt: report section 2b.

## Asks for Supervisor
- PP: e1-runners.csv (Robinhood Chain) and runtrack (Solana, from 2 Oct) to size the pre listing leg, when ready.

Version: v6, 7 Oct 2026 15:55 Bangkok (v5 14:45: basket shape; v4 13:45: 2017-18 disclosure; v3 12:50: outside CMC answer). Latest: http://159.223.88.168/hx-c9ois/rot/winners-status.md