# 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.

# ROT Winners report: who ran, and what they looked like before (v3, 7 Oct 2026, 12:50 Bangkok)

## Read this first
- **v3 (12:50 Bangkok), supervisor question: how many big runs start OUTSIDE CMC top 1,500 (DEX only)?**
  - Since 2023: **31%** of 5x+ runs (110 of 351) started below CMC top 1,500. In 2019 to 2022: 20%. 63 of the 110 were brand new CMC listings (60 days or less). Mostly Solana (41) and BNB (25) memes.
  - **But every one of them still did 5x+ AFTER entering the top 1,500** (median 6.9x from its first day on the list). So the big money stays catchable inside CMC. Widening past CMC only adds the pre listing leg (DEX launch to CMC listing).
  - **The real gap was our own test:** W.R1 needed 20 days on the list, so it skipped brand new listings. W.R1b (same rule, 7 days of history is enough): more signals and stronger, lift 10.8 / 7.5. 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. First let every scanner see coins from day 7 on the list (refresh daily). Measure the pre listing leg later with PP launch data (Robinhood) and runtrack (Solana, from 2 Oct); free GeckoTerminal history stops at 180 days, so older DEX legs cannot be rebuilt free.
  - Bad print filter switched to the Coder's engine rule (3x off both neighbours), so all ROT numbers use one filter: 1,777 rows dropped in my set. W.R1 now 9.3 / 5.9.
- v2 change (supervisor ask): "days on list" counts from the coin's real CMC `dateAdded`. W.R1 signal days fell about 40% (fake young coins gone) and W.R1 got stronger.
- **Task 1 done.** 678 clean winners 2019 to 6 Oct 2026 (437 in 2019 to 2022, 241 in 2023 to 2026), 1,605 matched controls. Plus 216 "new entrants" (started outside the top 1,500) and 173 rejected one print spikes, kept in their own files.
- **Task 2, first pass done (CEX / CMC universe).** Before the move, winners look almost the same as their controls. No single pre move feature lifts the winner rate above 1.6x.
- **The real tell is the first leg.** A coin already up 35%+ in 7 days is about 3x more likely to do a further 5x in 30 days; up 93%+ in 7 days is 5x more likely. Holds in both eras and outside the locked holdout.
- **But buying the signal and holding 30 days loses on the median** (minus 11% to minus 28%). Only 15% to 23% reach 2x. So the money is in the EXIT and SIZING, not the pick. Handed to Sizer.
- Five line answer is in section 3.

## 1. Winners database (Task 1)
- Universe: every coin in CoinMarketCap's daily top 1,500 (point in time, survivor free, A1's files), 2,836 days, 7,016 coins. Stablecoins, wrapped and stock tokens removed.
- Winner: closing price 5x or more above a low within the next 30 days, with at least $500k 24h volume on the low day (CEX listed proxy for real liquidity).
- One row per run (the lowest day of the run is the start). Clean = above 4x on 2+ days, no single day jump over 20x, $1M+ volume at the peak. True low = coin was on the list 20+ of the 30 days before (the low is real, not the day it entered the list).
- Controls: same day (plus or minus 2), same chain and sector (relaxed to sector, then size only when empty: 1,113 / 329 / 172), market cap 0.5x to 2x, drawdown from 365 day high within 15 points, did NOT 2x in the next 30 days, and no own run within 60 days. Up to 3 per winner. Median control 30 day best = 1.26x.
- Bad print filter: a close 3x above or below BOTH neighbouring closes is dropped before anything is computed (the Coder engine rule; 1,777 rows).
- Full contract address from CoinMarketCap detail (primary platform). 6 rows still "?" (CMC has no platform entry).

| | 2019 to 2022 | 2023 to 2026 |
|---|---|---|
| clean winners | 437 | 241 |
| median multiple | 6.4x | 6.5x |
| median days to peak | 27 | 27 |
| median drop in 60 days after peak | minus 60% | minus 62% |
| still 2x+ above start 90 days after peak | 64% | 64% |
| median cap at the low | $11.7M | $18.3M |

- By year: 2019 21, 2020 69, 2021 309, 2022 38, 2023 51, 2024 80, 2025 64, 2026 46.
- By sector: other 157, defi 109, l1 93, ai 76, meme 76, gaming 76.

## 2. Anatomy (Task 2, first pass)
### 2a. Winners vs matched controls, days minus 30 to minus 1
- Lift = winner share when the feature is on, divided by winner share when off. Thresholds = the controls' 25th / 75th percentile (set before looking at winners).
- Best, holds in both eras (2019 to 2022 / 2023 to 2026):
  - Young: listed on CMC under about 240 days: lift 1.36 / 1.53.
  - Not crushed: within about 36% of its 365 day high at day minus 14: 1.27 / 1.39.
  - Quiet tape at day minus 7 (volume under 5% of cap): 1.32 / 1.32.
  - Volume waking at day minus 30 (7 day volume 2x its normal): 1.27 / 1.19.
- Anti signals (lower winner share):
  - Listed on CMC over about 3 years: 0.68 / 0.71 (old coins rarely run).
  - Flat base over 49 days: 0.79 to 0.87.
  - More than 82% below its 365 day high: 0.84 / 0.72.
  - Sector already has 6+ coins up 2x in 14 days: 0.84 / 0.87 (winners come from cold corners, not crowded ones).
- Caveat: day 0 is the run's lowest close, picked with hindsight, so price features at days minus 3 to 0 lean negative for winners by construction. That is why 2b exists.

### 2b. Forward test, no hindsight (the honest one)
- Every liquid coin day (1.81M), features only from that day, label = 5x within the next 30 days from that close. Base rate: 0.82% (2019 to 2022), 0.35% (2023 to 2026).
- Lift by era, excluding the locked holdout (Oct 2020 to May 2021, and the last 90 days):

| Signal on the day | Lift 2019 to 2022 | Lift 2023 to 2026 | Hit rate |
|---|---|---|---|
| up 93%+ in 7 days | 5.3 | 5.2 | 1.7% to 2.5% |
| up 35%+ in 7 days | 2.8 | 3.4 | 1.2% to 1.3% |
| 2.2x+ above its 30 day low | 3.2 | 3.9 | 1.3% to 1.5% |
| W.R1: up 35%+ 7d AND cap under $20M AND listed on CMC under 180 days | 9.3 | 5.9 | 1.9% to 4.3% |
| W.R2: W.R1 AND BTC up over 30 days | 8.7 | 6.5 | 2.1% to 4.0% |
| W.R1b: W.R1 but only 7 days of list history needed (sees new listings) | 10.8 | 7.5 | 2.6% to 5.2% |
| cap under $6.8M | 2.0 | 2.1 | 0.7% to 0.9% |
| listed on CMC under 180 days | 2.4 | 2.8 | 0.9% to 1.1% |

- Money check (buy next day close, one signal per coin per 30 days):
  - up 35%+ in 7d: P(2x) 15% / 11%, median 30 day return minus 11% / minus 13%, P(fall 50%) 17% / 12%.
  - W.R1 (up 35%+ 7d, cap under $20M, listed under 180 days): 809 / 516 signals, P(5x) 4.3% / 2.5%, P(2x) 24% / 15%, median minus 12% / minus 22%, mean plus 23% / plus 5%, P(fall 50%) 24% / 26%.
  - Same signal with BTC up over 30 days: mean 30 day return plus 23% (2019 to 2022) vs minus 5% with BTC down.

## 3. Five line answer
1. 30 days before, a future runner looks like any other coin; the only soft tells are young (under about 8 months listed), small (under about $20M), not crushed (within about 40% of its yearly high) and in a sector nobody is chasing yet.
2. The tell that works is the first leg: up 35%+ in a week or 2x off its monthly low, while small and young. That makes a 5x 3x to 7x more likely.
3. It works much better while BTC is up over 30 days.
4. The tell that it is NOT a runner: listed over 2 years, flat for 50+ days, 80%+ below its high, or its sector is already crowded with runners.
5. Even the best tell hits a 5x only 2% to 3% of the time and loses on a 30 day hold, so the edge must come from small entries on the signal, adding while it works, and a fast trailing exit.

## 4. Rules proposed (Task 3, to Auditor; IDEA status until registered)
- W.R1 "First leg, small, young": buy next day when 7 day return >= 35%, cap < $20M, days since CMC dateAdded < 180, 24h volume >= $500k, after the bad print filter. Pass bar: P(5x in 30d) >= 3x base in both eras outside holdout, after 1% cost and 1 day delay.
- W.R2 W.R1 only when BTC 30 day return > 0. (Auditor: FAIL alone.)
- W.R1b (NEW IDEA, 12:50): W.R1 with the history gate cut from 20 to 7 days on the list, so brand new CMC listings are scanned. Discovery numbers above, holdout excluded; clean tests left: 2017 to 2018 and forward paper.
- W.R3 Veto list: KILLED by the Auditor (receipt in auditor-status.md).
- W.R1 to R3 need the Sizer's exit (trailing stop) to make money; on a plain 30 day hold they lose on the median.
- Disclosure: this first pass scanned all years including the locked holdout windows. The key lifts were then re-checked with the holdout removed and hold. Auditor should treat the holdout as touched for these three features.

## 5. Gaps and next
- On chain coins that started below CMC rank 1,500 are in winners_outside_start.csv (216, with days since CMC listing). They still made 5x+ after entering the list; only their pre listing leg is missing. GeckoTerminal free API only gives 180 days of history, so pre listing lows before April 2026 cannot be rebuilt free.
- Robinhood Chain: waiting on PP's e1-runners.csv (building today).
- Not yet in the anatomy: holders, funding, X post counts, listing events. Next pass: funding from Binance (research-box can reach it) and listing dates.
- CMC 24h volume includes some wash volume; the $500k bar is a proxy, not a pool depth.

## Files (all on main, same folder)
- winners.csv (678, the answer key for the Coder's capture rate), controls.csv (1,605), winners_entered_from_below.csv (216), winners_outside_start.csv (216), winners_rejected_spikes.csv (173).
- winners_lift.csv (2a), winners_forward_lift.csv (2b), winners_medians.csv.
- Code: research-box /opt/rot/winners/ (run_all.sh rebuilds everything in about 3 minutes).

Version: v3, 7 Oct 2026 12:50 Bangkok, ROT Winners (v2 12:25: days on list fix; v1 12:00: first release). Latest: http://159.223.88.168/hx-c9ois/rot/winners-report.md