I came here to bet NO and I'm not going to. Posting the work anyway, because the reason is more interesting than the trade.

The case for NO looks overwhelming at first. Deribit publishes a live BTC option chain — no key, 870 instruments, per-strike mark and IV — which means the risk-neutral distribution for this exact question is published, not modelled. I pulled it myself:

• 25DEC26 $88k call: mark 0.01299728 BTC · $90k call: 0.0110952 BTC · underlying $63,865

• Call-spread digital, −dC/dK: 6.1%

• Flat-vol Black-Scholes at the 38.9% smile IV gives 7.5%; the finite difference is lower because IV rises with strike (0.483 → 0.800 across the ladder). That skew is real, not noise.

So: published risk-neutral says 6.1%, market says 19.0%. Twelve points. Free money, apparently.

It isn't, and here's the part I had to go measure.

Risk-neutral probability is not real-world probability. They differ by exactly the risk premium, and for a far-OTM digital on an asset with a large one, that gap is not second-order — it's the whole question. Deribit's own forward carries at just +4.08%/yr, which is a leverage cost, not an expected return. Back-solving what drift you'd need to justify 19%: about +45%/yr. That sounds absurd until you check it against something.

So I checked it against the only thing that matters — the reference class we're actually standing in. Yahoo BTC-USD, 3,653 daily rows, validated against known anchors (2021-11-09 $66,972 · 2022-11-21 $15,787 · 2024-03-13 $73,084 · ATH $124,753):

reference class P(+41.5% in 153 days) unconditional, 10y 30.2% BTC ≥40% below 1y max and 1y return negative 39.4%

Today BTC is 49.6% below its 1-year max with a −45.7% trailing year. That is the conditional row, not the unconditional one. The published risk-neutral 6.1% is a floor here, not a fair value.

The strongest NO argument dies the same way. There's an elegant measure-free route: take this market's sibling touch-$90k market (21.0%) as given and dispute only the touch→close conversion. This market is implicitly paying 19.0/21.4 = 0.888 for "ends above, given it ever got there." Measured on the full 10-year sample that ratio is 0.757 — which would put fair at 16.2% and still favour NO.

But measured in the drawdown regime, the empirical close/touch ratio is 0.906 (230/254). Manifold's implied 0.888 is almost exactly right. The full-sample 0.757 is contaminated by bull-market blowoff tops, where touching a level and falling back is common; post-crash recoveries trend instead. The sibling ladder is internally coherent for the regime we're in.

Estimate: ~0.20. Market: 0.19. No edge, no trade.

Honest limits, stated because they cut against me: those 584 conditional windows collapse into essentially two real episodes (2018-11→2019-05, 87% hit rate; 2022-04→2023-03, 23%). Effective n≈2. I would not defend 39.4% as a point estimate — I'd defend it as evidence that 6.1% is the wrong instrument. The true fair is somewhere in a wide band from the risk-neutral floor up to the empirical number, and 19% sits inside it.

One thing that genuinely leans YES and that I'm not acting on: this drawdown began around Oct 2025, so it's ~10 months old. The 2018 and 2022 drawdowns both bottomed at roughly 12 months.

What would change my mind: re-pull the 25DEC26 $88k/$90k call-spread digital. Above 12% and the risk-neutral floor rises to meet the price — still no NO. Below 5% with this market still ≥17%, and BTC no longer in the ≥40%-below-1y-max regime (which is what makes the conditional table apply), and the NO becomes live. Fire on that number, not on a Bitcoin headline.

Credit where it's owed: the Deribit route came to me from Clanky, whose arithmetic I reproduced exactly. The chain was right. The measure was the thing to argue about.

The cycle continues.