# A2.25 More trades on Robinhood v3 (Analyst 2)

## Pre registered rule (written 18:16 Bangkok, 6 Oct, BEFORE any loosened cell was run)
- Live baseline: sellable $1M, age 3 h or less, top receiver 0.08 or less, army 0.015 or less. Live exit: 0.3x instant stop, 5x target, Kit A and Kit B coins held on a 50% trail from peak, pull or 48 h. Slot rule: 3 normal open plus 2 holds. Frictions: buy at gate + 30 s, every exit at the end of the next block, real tick depth, 1% fee each way. Ticket $1k.
- One knob loosened at a time, the rest stay live. Cells: sellable $750k, $500k; age 6 h, 12 h; top 0.10, 0.12; army 0.02, 0.03.
- A cell PASSES only if BOTH: (1) $ a day at $1k is higher than live in half A AND in half B; (2) worst day is no worse than 1.25x the live worst day.
- Rule (2) detail: live has NO losing day in half A (worst day $0), so 1.25x of $0 would kill any cell with a single red day. So the worst day is taken over all 23 days: live worst is -$3,161, so the floor is -$3,951. I also show worst day per half.
- Then the best two passing knobs (by smallest $ a day gain of the two halves) go together and must pass the same rule. If no cell passes, close.

## Answer in one line (18:17 Bangkok)
Only ONE knob passes: sellable $750k. But it does not work by adding good coins. It works by buying the SAME coins a bit earlier. The 22 extra coins it adds LOSE money. The gain is thin and rests on a few coins. Every other knob fails. Trades a day cannot be raised by loosening the gate without losing money. My call: keep live as is; run $750k only as a tagged paper variant. Close the "more trades" hunt on gate knobs.

## Grid, one knob at a time ($1k ticket, live exit, slot rule, every A2.21 friction)
Half A = 14 to 27 Sep (14 days), half B = 28 Sep to 6 Oct (9 days). tpd = trades a day taken. usd_tr = $ a trade. dd = max drawdown. The live row matches A2.24 to the dollar (+$3,317 / +$6,683).

| cell | gates | tpd_A | tpd_B | win_A | win_B | usd_tr_A | usd_tr_B | usd_day_A | usd_day_B | worst_A | worst_B | dd_A | dd_B | pass |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| live | 70 | 2.57 | 3.78 | 42% | 41% | 1290 | 1769 | 3317 | 6683 | 0 | -3161 | -4257 | -4700 | base |
| L750k | 91 | 2.93 | 5.33 | 44% | 35% | 1357 | 1346 | 3973 | 7177 | -728 | -3713 | -4553 | -7041 | True |
| L500k | 144 | 4.86 | 8.22 | 29% | 22% | 780 | 791 | 3788 | 6505 | -2401 | -5791 | -11237 | -17227 | False |
| age6 | 73 | 2.57 | 4.11 | 42% | 38% | 1290 | 1567 | 3317 | 6441 | 0 | -3161 | -4257 | -5414 | False |
| age12 | 73 | 2.57 | 4.11 | 42% | 38% | 1290 | 1567 | 3317 | 6441 | 0 | -3161 | -4257 | -5414 | False |
| top10 | 75 | 2.64 | 4.22 | 41% | 37% | 1236 | 1508 | 3266 | 6368 | 0 | -3878 | -4257 | -5370 | False |
| top12 | 79 | 2.64 | 4.67 | 41% | 36% | 1236 | 1388 | 3266 | 6478 | 0 | -2671 | -4257 | -5370 | False |
| army02 | 71 | 2.57 | 3.89 | 42% | 40% | 1290 | 1698 | 3317 | 6602 | 0 | -3161 | -4257 | -4700 | False |
| army03 | 75 | 2.64 | 4.22 | 41% | 42% | 1235 | 1611 | 3263 | 6800 | -749 | -3161 | -4257 | -4700 | False |

- Sellable $500k: 2x the trades, but win rate drops to 29% / 22%, $ a trade halves, drawdown is -$11,237 / -$17,227. Fails in half B ($6,505 under $6,683) and on worst day (-$5,791 under the -$3,951 floor).
- Age 6 h and 12 h: the same 3 extra coins (almost every coin gates under 3 h old). Half A unchanged, half B down $242 a day. Fail.
- Top 0.10 and 0.12, army 0.02: half A down or flat, half B down. Fail.
- Army 0.03: half B up (+$117) but half A down (-$54). Fail.

## Why the one pass ($750k) is weak
- 22 new coins (5 in A, 17 in B). Together they LOSE: -$280 a day in A, -$518 a day in B. 19 of 22 hit the stop, 1 is pulled; only 2 Kit B holds won (0x577bc74a99aabbd43a354668dc7b7ed4b5e303ae, 0x74ccd8416cbc4e4a864176c2707bc1434819ef0f, each about +$3.6k).
- 3 coins traded at $1M are not traded at $750k (filters or a full slot): -$297 a day in B.
- 67 coins are the same, just bought at $750k not $1M: +$936 a day in A, +$715 a day in B. That is ALL the gain.
- That gain is mostly a few coins that hit the 0.3x stop from a $1M entry but not from a $750k entry (0x6977777869f3192e378135ee50ee9180c717a80f, 0x5caf197a5effce69b66d6b7bb4d7d260fcdb8d71, 0x6c963ce1aff02c18b2d0ef81b96df77b2f53c4ee, 0x77b88d2ddae155cccec58f2750d53da9355b7e40), plus one big hold (0xe99c8437b0b29325a4080b8c837ff4e5a6f9fc69). One coin goes the other way: 0x9ba4d84ac1571a0788b2bb7a6412e73529d12dc6 is a +$3,694 hold at $1M and a -$725 stop at $750k.
- Median change per same coin: +$1 (A), +$8 (B). Without the top 2 coins per half, the earlier-entry gain is only +$302 / +$113 a day.
- Max drawdown in half B gets 50% worse: -$4,700 to -$7,041. Worst day -$3,161 to -$3,713 (inside the -$3,951 floor).
- Same direction with the old sell-all-at-5x exit (+$3,057 / +$3,956 vs +$2,733 / +$3,535) and with the 2 open holds cut to 5x (+$3,973 / +$5,751 vs +$3,317 / +$5,446). So it is not an artefact of open holds or of kit hold.

## Best two knobs together
Pre registered: the best two PASSING knobs. Only one passed, so there is no real pair. For information, $750k plus each other knob (same rule):

| cell | gates | tpd_A | tpd_B | usd_day_A | usd_day_B | worst_A | worst_B | dd_A | dd_B | pass | beats_L750k |
|---|---|---|---|---|---|---|---|---|---|---|---|
| L750k+age6 | 95 | 3.07 | 5.56 | 3866 | 7083 | -728 | -3713 | -5320 | -7157 | True | False |
| L750k+age12 | 95 | 3.07 | 5.56 | 3866 | 7083 | -728 | -3713 | -5320 | -7157 | True | False |
| L750k+top10 | 94 | 3.0 | 5.56 | 3922 | 7013 | -728 | -3713 | -4553 | -7041 | True | False |
| L750k+top12 | 96 | 3.0 | 5.78 | 3922 | 6852 | -728 | -3713 | -4553 | -7041 | True | False |
| L750k+army02 | 92 | 2.93 | 5.44 | 3973 | 7303 | -728 | -3713 | -4553 | -7041 | True | False |
| L750k+army03 | 95 | 3.07 | 5.56 | 3869 | 7554 | -749 | -3713 | -4553 | -7035 | True | False |

All pass the rule (they carry the $750k gain) but NONE beats $750k alone in both halves. Adding a second knob adds coins that do not pay.

## What this means
- The gate is already at the right place. Every looser setting adds coins that lose. The edge is the gate, not the number of coins.
- The only real effect found is ENTRY PRICE: the same good coins pay a bit more when bought earlier. That is a different question (an earlier entry for coins that will reach $1M) and at $750k you cannot tell them apart from the 22 extra coins, which lose.
- More trades a day must come from somewhere other than gate knobs (A2.15 already closed new venues; A2.22 and A2.23 closed other chains).

## Suggestion to the Supervisor
- Keep live as is ($1M, 3 h, 0.08, 0.015).
- If you want, the Coder can log $750k as a tagged paper variant only (there is already a coins750 table in w2.db with 1 row). Promote it only if live paper shows the same coins winning more from $750k AND the extra coins stop losing.

## Limits
- 23 days, 89 coins in the best cell; the gain rests on about 5 coins.
- Half B still holds 2 open Kit B holds (from A2.24).
- Only the $1k ticket was run (the rule asks $1k).
- Same data and sim as A2.24 (live row matches to the dollar). Kit tags for 105 new coins read the same way as A2.24 (launch contract bytecode hash); 0 errors.

## First principles score: 85%
Every number comes from the real tick-walk fills with next-block exits and the slot rule. Rule written before the run. Missing: no out-of-sample period beyond 23 days, and the one pass is decomposed and found thin, so I do not call it an edge.

Files: http://159.223.88.168/hx-c9ois/sw/sw-a2-25-grid.csv and http://159.223.88.168/hx-c9ois/sw/sw-a2-25-pairs.csv

Analyst 2, 6 Oct 2026 18:17 Bangkok.