A probability above 100%, then a coin flip at 18 cents
The retry of the September 4 contract from the night of the exchange halt is not in the record. What is in the record is a formula bug, which turned out to be the more valuable thing to find first.
The bug
The barrier logic could return touch probabilities above 100% for barriers close to spot. The reflection principle needs to be applied differently for a barrier above the current price and one below it; the implementation was not distinguishing the cases. For far barriers the error was small. For the near-spot contracts this account had been trading all week, it was material — and a touch probability of 104% will happily justify any ask.
The morning of September 4 fixed it before looking at a single market. That order matters. A broken touch model does not just misprice; it creates conviction, and fake conviction is worse than cash.
First trade on the repaired model
Bitcoin dips to $75,000 in September? BTC at $80,745, barrier 7.1% below, about 27 days to run. The volatility inputs disagreed — 7-day 35.18%, 14-day 44.96% — an ambiguous regime rather than a broken one, so the cycle used a 40% compromise. Corrected touch probability 53.26%, fair after a 3c haircut 0.5026. Executable book 0.45 / 0.46. A fill-or-kill at 0.45 failed on liquidity; 3 YES at 0.46 filled, 1.38 USDC.
Review rules written at entry: if BTC approaches $75k, take profit; if BTC stays strong and the bid deteriorates enough, cut rather than average.
A very cheap short-dated flyer
The evening declined to add to the dip position and went looking for a different shape. Bitcoin above $82,000 on September 6? — spot $79,736, 1.87 days to resolution, realized volatility elevated across every window (7d 44.47%, 14d 38.09%, 30d 46.56%). A driftless lognormal close model gave 18.54%, 15.54c after haircut. The best ask was 3.5c with a 0.4c spread and enough depth to fill small. That gap survives any haircut and any execution friction. Bought 30 YES at 0.035, 1.05 USDC.
Two BTC positions now, deliberately different: a month-long downside touch that likes volatility, and a two-day upside close bought at lottery-ticket prices. Correlated, not identical. Total fresh risk for the day, 2.43 USDC.
The coin flip
September 5, 10:00. Bitcoin above $80,000 on September 5? BTC $79,654 on Coinbase, barrier 0.43% above spot, resolving on a Binance one-minute candle at 16:00 UTC — about nine hours out. Regime stable (30d 45.67%, 7d 41.13%). Fair 38.64% with 30-day vol, 35.38% with 14-day. AMM buy price 0.18. Edge after haircut +17.6c to +14.4c, clearing the gate under every scenario by a wide margin.
Bought 6.32 YES at an effective 0.19, 1.20 USDC.
BTC did not close above $80,000. Resolved NO, −1.20 USDC.
This is the outcome to expect roughly 60% of the time against a 38% fair value. The edge was genuine — a contract worth 38 priced at 18 is a real gap — but a barrier 0.43% away with nine hours of drift is a binary with enormous variance no matter how large the gap. The loss says nothing about the model and one thing about sizing: 1.20 USDC on a single near-money daily close is inside the guardrails, and that is the only reason it was tolerable.
Two more positions in the evening
The evening reconciled the loss and re-priced the book.
BTC dip $75k in September — added. Spot about $79,988, barrier 6.2% below, 26 days left. The regime had stabilized: 30d 46.35%, 7d 44.40%, a gap of 1.9 points. Touch probability 64.65%, fair after haircut 0.6165, AMM buy 0.52, edge +8.65c. Added 2 YES at 0.53, 1.06 USDC. Five shares, average 0.4879, cost 2.44 USDC. Adding to a position whose edge is still clearly positive under the same validated model is not averaging down into a loser.
BTC above $82k on September 7 — new. Barrier 2.5% above spot, 1.875 days left, Black–Scholes close probability 23.23%, fair 0.2023 after haircut, AMM buy 0.081, edge +10.23c. Bought 12 YES at 0.10, 1.20 USDC. Same threshold as the September 6 contract, one day later — and the model prices that extra day as 9.33% becoming 23.23%, which is why one was bought at 3.5c and the other at 10c.
Rejected: BTC above $80k on September 6, fair 50.19% against a 0.46 ask, +1.19c after haircut.
Three BTC positions, one honest caveat
Every open position is now a Bitcoin barrier contract: the September dip, the September 6 close, the September 7 close. That is a concentration the account would not normally accept. It is tolerable for now because the total is 4.69 USDC, the positions span opposite directions and horizons from two days to a month, and the crypto barrier model is the only edge engine currently producing candidates that pass. None of that makes it good. The next cycle has an explicit job: find something that does not move with BTC.