# A2.20: what a slow fill costs (6 Oct 2026, 16:04 Bangkok)

## Read this first
- **A slow fill costs almost nothing up to 45 s.** Every delay from 2 s to 300 s stays at or above $2,000 a day in both halves at $1k.
- **Suggested check 3: fill within 30 s of the gate.** That covers the 20.6 s real fill already seen, and it leaves a safety margin before money starts to leak (60 s and later).
- **Up to 45 s, the differences come from ONE coin.** 0x6c963ce1aff02c18b2d0ef81b96df77b2f53c4ee ends at 0.14x or 4.65x depending on the exact fill second (2 s: 0.14x, 5 s: 4.65x, 10 s: 0.14x, 20 s: 4.65x). It is a coin-flip on the 0.3x stop level, not a delay cost.
- **Real cost starts at 60 s.** Without the knife-edge coin, the second half drops from 2.01x to 1.88x at 60 s (0x0d974113699e786f4fc56d760ec7f5270b147a88 misses its 5x), then to 1.83x at 90 s and 1.69x at 120 s. At 120 s that is about -$1,500 a day.

## Table (wide gate, 0.3x stop, block-end exits, $1k real depth; first half 14 days, second half 9 days)

| Fill after gate | Mean x first / second | $ a day first / second | Worst day first / second | Median price move gate to fill | 5x hits (of 70) |
|---|---|---|---|---|---|
| 2 s | 2.07 / 1.96 | 2,756 / 3,618 | 0 / -3,053 | +0.6% / +0.7% | 28 |
| 5 s | 2.06 / 2.09 | 2,730 / 4,099 | 0 / -2,000 | +0.9% / -1.0% | 29 |
| **6 s (check 3 as written)** | 2.06 / 2.08 | 2,721 / 4,091 | 0 / -2,000 | +1.0% / -0.1% | 29 |
| 10 s | 2.06 / 1.95 | 2,736 / 3,599 | 0 / -3,049 | -2.3% / -0.3% | 28 |
| 15 s | 2.08 / 1.95 | 2,780 / 3,590 | 0 / -3,047 | -1.6% / -2.8% | 28 |
| 20 s | 2.08 / 2.08 | 2,779 / 4,084 | 0 / -2,000 | -2.9% / -2.4% | 29 |
| **30 s (suggested)** | 2.07 / 1.96 | 2,743 / 3,616 | 0 / -3,075 | -1.9% / -0.9% | 28 |
| 45 s | 2.06 / 1.96 | 2,721 / 3,608 | 0 / -3,033 | -3.8% / -1.4% | 28 |
| 60 s | 2.07 / 1.96 | 2,753 / 3,617 | 0 / -2,000 | -4.1% / -1.4% | 28 |
| 90 s | 2.06 / 1.83 | 2,720 / 3,125 | 0 / -3,026 | -3.4% / -0.3% | 27 |
| 120 s | 2.04 / 1.69 | 2,686 / 2,617 | 0 / -3,072 | -6.8% / -6.4% | 25 |
| 180 s | 2.06 / 1.77 | 2,721 / 2,904 | 0 / -3,080 | +0.2% / +0.3% | 26 |
| 300 s | 1.94 / 1.79 | 2,409 / 2,973 | 0 / -3,142 | +0.9% / -1.2% | 24 |

- The second-half swing between 1.96x and 2.09x up to 45 s is entirely 0x6c963ce1aff02c18b2d0ef81b96df77b2f53c4ee. Without it, every delay from 2 s to 45 s gives 2.00x to 2.01x in the second half and 2.06x to 2.08x in the first.
- The worst day of about -$3,050 at some delays is 2 Oct, the day of that same coin's stop. Otherwise the worst day is 28 Sep at -$2,000 (two pulls to 0).
- Table: http://159.223.88.168/hx-c9ois/sw/sw-a2-20-delay-table.csv

## Why (first principles)
- Target, stop and pull are measured from OUR fill. So a late fill only costs money if the coin's path after the fill is worse.
- At the $1M gate the price moves only a few percent in the first minute (median -4% to +1%). The 5x runs take minutes to hours, so 30 to 45 s does not change which coins reach 5x.
- From 60 s on, a few runners have already left. A late buy then pays more for the same top and misses 5x on 2 to 4 coins.
- Earlier A3.1 found +60 s cost 0.17x. That was on the old 0.6x stop. With the 0.3x stop, a dip right after the fill no longer stops us out, so a late fill matters less.

## Caveat
- This prices only the delay itself. A slow feed can also mean we act on an old price; the CLAUS fill fell 22% in 15 s. That is covered because the exits are measured from the fill price, not the gate price (Coder fix v1.9).

## How
- The same 70 wide trades. The entry is the first event at or after gate + delay. Exits: 0.3x instant stop, 5x, in-range pull, 12 h; each exit is priced at the end of its block (A2.19). Fills by tick walk at $1k (A2.14).
- Script: container /tmp/w2/a220/delay.py

## First principles score
8/10. Real depth, both halves, 13 delays. The one knife-edge coin is named. Weak: 70 trades; the second half is 9 days.

Version 1, 6 Oct 2026 16:04 Bangkok. Latest: http://159.223.88.168/hx-c9ois/sw/sw-a2-20-fill-delay.md