# 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.

# FW idea queue, value ranked (FW Analyst 1, A1.29)

## READ THIS FIRST (v3, 7 Oct 2026, 12:20 Bangkok)
- **v3:** second engine (C, AB) scored and paper book live; see Second engine below.
- **Top 3 run on lab.py (plus 3 more from the same data): all KILLED.** Ledger rows FW-008 to FW-014. Grid: http://159.223.88.168/hx-c9ois/fw/lab/grid.md
- **The one lesson:** every cell that "won" on half B won by adding stake, and random stake did as well or better (FW-010, FW-011). On mc5 the edge is the coin set, and the lever that moves money is **ticket size**, not a filter. This matches the elite plan: ramp first.
- Base on lab.py (window to 6 Oct 12:00 Bangkok): half B $1k **+$14,160 a day** (n 149, 2.09x), $100 +$1,699.
- Overnight grid is live: research-box cron 02:30 Bangkok runs every rule in /root/fwa1/lab/rules plus grid.py; main pulls grid.md at :40 each hour.

## Verdicts (judged on half B only, best half A cell; $ a day at $1k)
| # | Idea | Term it moves | Best cell | Half B vs base +$14,160 | Verdict |
|---|---|---|---|---|---|
| 1 | Kit at $300k as a 2nd book | trades a day | n/a | 68% of its trades (134 of 196) are coins mc5 already holds; the 62 new coins add +$1,636, **-$1,901 without top 3** | KILLED (FW-008). Overlap = exposure cap job, not edge |
| 2 | Operator flag skip/tilt | mean x | skip flagged | +$5,006 | KILLED (FW-009). Flagged and unflagged earn the same per trade |
| 3 | Operator early entry +2/+3 min | trades a day | n/a | not testable: lab coins are picked by their cap at +300 s (look-ahead) | BLOCKED. Needs all v3 pools in lab data (A3/PP) |
| 4 | Add on strength (+600 s, cap >= 1.2x) | size per coin | +600 s 1.2x | +$16,143 but random adds p95 +$19,960 | KILLED (FW-011). Any 2nd ticket does as well = size |
| 5 | Operator reputation | mean x | 2x good, skip bad | +$21,677 but random p95 +$27,525 | KILLED (FW-010). Stake, not skill |
| 7 | LP top-up before +5 min | mean x | 1.5x top-up | +$14,160 (no coin changed) | KILLED (FW-012). Almost no coin has one |
| 9 | Time of day | mean x | skip 18-24 | +$12,050 | KILLED (FW-013) |
| A1.23 recheck | Cheap band $150k-$250k | trades a day | n/a | -$3,326 | KILLED again (FW-014) |

## Second engine: small steady wins on OLDER coins (v30/v31, 12:20 Bangkok)
| Idea | Term it moves | Edge if true | Chance | Hours | Score | Status |
|---|---|---|---|---|---|---|
| C deep oversold (4h, +10% / -20% / 14 d) | trades a day +0.2, own $ line | about +$12 a day at $1k (1.5 a week x +5.8% x $1k) | 0.35 | 2 | 2.1 | PAPER BOOK LIVE (FW-015) |
| AB dip in uptrend (4h, +15% / -30% / 14 d) | trades a day +0.3, own $ line | about +$25 a day at $1k (2.3 a week x +7.6% x $1k) | 0.3 | 1 (shared build) | 7.5 | PAPER BOOK LIVE (FW-016) |
- **Push back:** together they are worth about **+$37 a day at $1k**, roughly 1/380 of mc5 at the same ticket (+$14,160). A $1k loss at -20% on one trade wipes about 5 wins. They will not smooth mc5's days in any way you can see. Their value is a free, uncorrelated paper line, so I built them as a server book with zero Claude cost, not as a priority.
- **Why chance is only 0.3 to 0.35:** the 4 Oct study built its coin list from coins still listed on 4 Oct. Coins that died after an oversold signal are missing, and those are exactly the losers this pattern hits. That survivorship bias lifts the win rate. Only a forward paper book on today's live list fixes it. Random entry with the same exits wins 55% (C) and 50% (AB), so a real edge needs about 70%+.
- **Paper book:** research-box /root/fwa1/eng2.py, cron every 4 h at :20 UTC after each 4h close. Universe: GeckoTerminal top 100 pools by 24 h volume on Solana, BSC and Base, pool 30 d+ old, 24 h volume $10k+. Entry at the signal candle close; exits on 4h closes; 2% fee. Board: http://159.223.88.168/hx-c9ois/fw/a1-eng2-paper.txt (main pulls at :45).
- **Pre-registered bar:** 20+ closed trades per pattern, win rate within 10 points of the study (C 76%, AB 70%), and avg trade > 0 after fees. Then it goes to stage 5 via the Supervisor. At 1.5 to 2.3 signals a week that takes 2 to 3 months, so a faster check is to re-run the 4 Oct rules on a survivor-free universe (needs dead coins' candles); I can do that if asked.

## Still open (each with the term it moves)
| Rank | Idea | Moves | By up to (if true) | Note |
|---|---|---|---|---|
| 1 | **Size ramp proof at $1k depth** (paper twin of every live trade at $100/$300/$1k) | size | +$14k a day at $1k vs about $140 at $10 | The only lever left with a proven $ line. Owner: A2/PP per elite plan; I can add a lab.py size curve on request |
| 2 | mc5 on Uniswap v4 RH pools | trades a day | +$600 a day at $100 | Needs v4 data in lab.py format (PP/A3) |
| 3 | Operator early entry +2/+3 min | trades a day, mean x | +$800 a day at $100 | Needs uncut lab universe |
| 4 | Bundle distribution exit | mean x | +$300 a day at $100 | DEAD: SW A3.13 (11:55) killed the bundle exit on kit holds; one-wallet version FW-004 also dead |
- Dropped as dead family (elite plan): Telegram bot router share.

Version: v3, 7 Oct 2026 12:20 Bangkok, FW Analyst 1 (v2 12:00, v1 11:06). Latest: http://159.223.88.168/hx-c9ois/fw/lab/queue.md

## A1.32 ENTRY TIME LADDER, PRE-REGISTERED (7 Oct 2026 16:20 Bangkok, before any result)
- Question (Ben): mc5 buys only at +5 min, so every buy is above $250k. Is an earlier entry better?
- Universe: every Uniswap v3 WETH pool (any fee tier) with TRUE first swap 13 Sep 00:00 to 6 Oct 12:00 Bangkok (A3 cand_all chain build, 6,085 pools), uncut. Envelope for data = cap between $25k and $3M at ANY of the test times (2,220 pools), full Mint/Burn/Swap logs from creation to first swap + 4 h 15 min.
- Entry times t: +10, +20, +30, +60, +120, +180 s and +300 s (control). At t the rule sees ONLY what is known at t: cap from the last swap before T0+t, dust guard before T0+t. Fill at T0+t+6 s with the lab.py tick walk (our buy kept). Exits = lab.py (stop 0.5x fee in, 4 h time stop, pull, sell at end of next block, 1% + $0.05 gas each way).
- Band per t: fit on half A ONLY (grid lo $25k..$500k, hi $150k..$3M, n at least 30, best half A $ a day at $10), then FROZEN for half B.
- Output per t: n, $ a day half A and half B at $10 and $1k, without top 3, worst 3 days, random same count same size p95 (2,000 draws from all coins in the universe at that t, same half).
- PASS = fit band beats the +5 min live band ($250k to $1M) on BOTH halves at $10 AND beats random p95 on half B. Entering earlier on the +5 min BUY set is printed only as UPPER BOUND (look-ahead), never a verdict.
- Ledger FW-020.