# A2.18: size curve for the new config (wide gate + 0.3x stop) (6 Oct 2026, 15:22 Bangkok)

## Read this first
- **Size is the knob that moves $ a day the most, by far.** Going from $500 to $3k multiplies $ a day by about 4.6. The gate, exit and re-entry changes together add about 40%.
- **$ a day tops out at about $7.5k to $10k a trade** (first half $10.2k a day at $7.5k; second half $16.6k a day at $10k). Above that our own buy and sell eat the edge, and $ a day falls.
- **Safe ceiling: about $5k a trade.** Up to $5k the worst 3 days stay positive in both halves. At $7.5k the second half has a losing 3 day window (−$3.3k).
- **Up to 3 trades are open at once** (peak, both halves). So the wallet needs 3 × trade size, plus gas.
- **Suggested ramp:** $500 → $1k → $2k → $3k, then $5k only if live fills match paper (A2.11).

## Size table: wide gate ($1M, age ≤ 3 h, top ≤ 0.08, army ≤ 0.015), 0.3x instant stop, 5x target, in-range pull, 12 h

| Size | Mean x first / second half | $ a day first half | $ a day second half | Worst day (2nd half) | Worst 3 days first / second | Wallet needed (3 open) |
|---|---|---|---|---|---|---|
| $100 | 2.16 / 2.20 | +299 | +454 | −200 | 0 / +382 | $300 |
| $500 | 2.11 / 2.16 | +1,432 | +2,183 | −1,000 | 0 / +1,754 | $1.5k |
| $1,000 | 2.06 / 2.10 | +2,718 | +4,156 | −2,000 | 0 / +3,134 | $3k |
| $2,000 | 1.95 / 2.00 | +4,901 | +7,539 | −4,000 | −57 / +4,881 | $6k |
| $3,000 | 1.86 / 1.91 | +6,627 | +10,258 | −6,000 | −787 / +5,430 | $9k |
| **$5,000** | 1.70 / 1.74 | +8,952 | +14,051 | −10,000 | −3,344 / +3,531 | $15k |
| $7,500 | 1.53 / 1.58 | **+10,210** | +16,365 | −15,000 | −8,145 / −3,305 | $22.5k |
| $10,000 | 1.39 / 1.44 | +10,099 | **+16,649** | −20,000 | −14,283 / −13,922 | $30k |
| $15,000 | 1.18 / 1.23 | +7,026 | +12,944 | −30,000 | −29,356 / −43,237 | $45k |
| $20,000 | 1.03 / 1.07 | +1,406 | +5,382 | −40,000 | −46,944 / −79,969 | $60k |

- Trades: 36 in the first half (14 to 27 Sep, 14 days), 34 in the second (28 Sep to 6 Oct, 9 days). That is 2.6 and 3.8 a day.
- Exits: 29 target, 33 stop, 5 pull, 3 time. Median hold 91 min.
- **Only one losing day in 23 at $500:** 28 Sep, two pulls to 0 (0x10897a25ded3b3f7d8cb9736fccf22c0ef5a414e and 0xed7bc12c9c6e6484e17f7204c16a57a4e2792668). That one day sets the worst day at every size: 2 × stake.
- The first half's worst day is 0 at small sizes because 7 of its 14 days had no trade and the rest were green.
- Mean x drops below 1.5x at about $8k.
- Full table: http://159.223.88.168/hx-c9ois/sw/sw-a2-18-size-table.csv

## Which knob moves $ a day (second half, first half in brackets)

| Change | $ a day at $500 | Gain |
|---|---|---|
| Old: today's rule + 0.6x stop | +1,545 (+1,052) | |
| Exit: 0.6x → 0.3x stop (today's gate) | +2,053 (+1,178) | **+33% (+12%)** |
| Gate: today → wide (0.3x stop) | +2,183 (+1,432) | +6% (+22%) |
| Re-entry on top (0.3x stop) | high mode +424 (−392); drop mode 0 (−221) | **loses in the first half: drop it** |
| **Size: $500 → $3k (wide + 0.3x)** | +10,258 (+6,627) | **× 4.7 (× 4.6)** |

- Edge per trade barely moves between configs (2.1x to 2.2x). Trades per day and size are what scale dollars.
- **The 0.3x stop vs no stop** is almost the same (+$31 and +$27 a day at $500). Most 0.3x stops sell coins that were headed to near zero anyway.
- **The $750k gate with 0.3x stop** is weaker in the second half at every size (1.69x at $500) and has 5 trades open at peak. It tops out lower ($10.1k a day at $5k in the second half). Keep it paper only.
- **Re-entry under the 0.3x stop:** a coin that fell 70% rarely comes back. Re-buys stop out at a median 0.27x to 0.28x. Under the 0.3x stop, the A2.9 re-entry variant should be switched off.

## Caveats (why not go straight to $10k)
- Each fill was priced by walking the pool's real tick liquidity at the entry and exit events (A2.14 method; the R model agrees to 0.001x here). But our buy is assumed not to change what other buyers do. At $5k into a $1M coin that is 0.5% of the cap, likely fine. At $10k+ it is not tested.
- 70 trades and a 9 day second half. One bad day with 3 pulls would cost 3 × stake.
- Pulls are priced at 0 here (A3.5 prices them at 0.13x), so this is a little pessimistic.
- Live fills came 20 s late on CLAUS (A2.11). If live fills stay worse than paper, the safe size is lower.

## A2.11 (background)
- No new Coder paper trade since CLAUS (w2.db read 15:20 Bangkok, books w2_v3_1m and w2_v3_hold). Nothing new to compare.

## How
- Trades: strict A2.8 gates at $1M (s16_gates), wide filters, cover ok. Exits: trade4 with 0.3x instant stop on true in-range liquidity (a_true_all, 300 pools). Fills: tick walk of all Mint and Burn ranges (raw_all.pkl from sw3.tsv.gz, 51,281 Mint/Burn events). Check: wide no stop 2.09x / 2.14x and today + 0.6x 2.09x / 1.90x, same as A2.14 and A2.16.
- Open at once: overlap of [entry, exit] times over all 70 trades.
- Scripts: container /tmp/w2/a218 (rawall.py, tick2.py, run.py, table.py, reent3.py).

## First principles score
8/10.
- **Strong:** real pool depth at every entry and exit, both halves, 11 sizes, concurrency counted, the knob comparison uses the same trades.
- **Weak:** the second half is 9 days; impact on other buyers at large size is not modeled; the worst day rests on one day.

Version 1, 6 Oct 2026 15:22 Bangkok. Latest: http://159.223.88.168/hx-c9ois/sw/sw-a2-18-size-curve.md