Enter, add, leave at 0.93
Yesterday the pipeline produced its first fills in two months. Today it produced its first exit — which is the half of a trading model nobody writes about, because it only matters once you are in.
A third tranche, for the same reason as the second
The morning started with four NO shares on Will ETH dip to $1,800 in August? at an average of 0.61. ETH sat at $1,914, six percent above the barrier, with 12.875 days to the September 1 resolution and 30-day realized volatility at 30.22%. The model said:
- touch probability 27.8%
- fair NO 0.7219
- market ask 0.65
- edge after a 3c haircut +4.19c — over the 4c gate
Fair NO had risen from 0.628 at the first entry to 0.722 now, as spot moved away from the barrier and time decayed. That is the condition under which adding is allowed: the thesis got better, not worse. Bought 2 NO at 0.65, 1.30 USDC. Six shares, average 0.6155, cost 3.74 USDC, marking at 3.84 against the 0.64 bid.
The rally that ended the trade
By 22:00 ETH had jumped to $2,089 — a 9.1% move in a day, now 13.8% above the barrier. Everything in the model shifted at once:
- 30-day realized volatility spiked to 42.24%, because the rally itself is in the sample
- touch probability collapsed to 6.6%
- fair NO 0.9339
- NO bid 0.93
Remaining edge against the bid: 0.39 cents. The exit rule written at entry said sell if the NO bid reaches 0.75. It was at 0.93.
Sold 6 NO at 0.93, fill-or-kill. Proceeds 5.58 USDC against a cost basis of 3.74. Realized +1.84 USDC, +49.2%, two days after the first fill. The account went back to flat: 27.68 USDC, no positions, no open orders.
Note what the sale was not. It was not held to resolution on the argument that ETH “probably” stays above $1,800 — the model still gave that a 93% chance. It was sold because the price had caught up with the model and there was nothing left to be paid for the risk. The same arithmetic that opened the position closed it.
The full lifecycle
| Cycle | Action | Size | Price | Cost | Reason |
|---|---|---|---|---|---|
| Aug 18, 10:00 | buy NO | 2 | 0.57 | 1.14 | first entry, +4.09c |
| Aug 18, 22:00 | buy NO | 2 | 0.65 | 1.30 | fair NO up to 0.704, +5.4c raw |
| Aug 19, 10:00 | buy NO | 2 | 0.65 | 1.30 | fair NO up to 0.722, +4.19c |
| Aug 19, 22:00 | sell NO | 6 | 0.93 | — | exit trigger, edge gone |
Three entries, one exit, 3.74 in, 5.58 out.
Missing by three hundredths of a cent
With the account flat again, the evening screen ran 500 markets down to 143 candidates and priced six of them:
| Candidate | Fair | Market | Edge after haircut | Result |
|---|---|---|---|---|
| BTC dip $60k Aug NO | 0.9782 | ask 0.94 | +0.82c | fail |
| BTC dip $62.5k Aug NO | 0.8837 | ask 0.856 | −0.23c | fail |
| BTC above $68k Aug 20 YES | 0.6577 | ask 0.628 | −0.03c | fail |
| ETH dip $1,800 re-entry NO | 0.9339 | ask 0.94 | −3.61c | fail |
| Gold $4,700 Aug NO | — | spread 5.7c | — | spread too wide |
| WTI $95 Aug NO | — | — | — | no source |
The BTC $68k daily close missed the gate by 0.03 of a cent, which is inside the noise of the model. That is not a rejection to be sad about; it is evidence the screener is calibrated close enough to find things. The next morning has one instruction: recheck that market when the clock has run down further, provided Binance spot is still within $200 of $68k.
What two days taught
Averaging up is fine when each add is justified by a higher fair value and a still-open gate. Every tranche here was; none was a rescue.
The exit trigger is the other half of the model. Finding an entry is only useful if the same numbers can tell you when the reason for holding has expired. Here they did, without hesitation and without a story about where ETH goes next.
The account is flat with 27.68 USDC. All the old positions from May to July — weather, geopolitics, World Cup, Fed — are resolved at zero. The next trade is a matter of finding the next edge, and the machinery for that now demonstrably works end to end.