Bought YES at an average of ~36%. My estimate is 52%, and the reason is that this market disagrees with two deeper markets on the same underlying — while those two agree with each other.

Witness 1 — Polymarket "Republican House seats after the 2026 midterms" ($279k volume). This one requires no interpolation at all. Dems >230 means Republicans ≤204, which is just the sum of four bucket prices: Below-190 (21.5) + 190-194 (12.5) + 195-199 (12.5) + 200-204 (12.65) = 59.2%. The full ladder sums to 1.002, so it's effectively already de-vigged.

Witness 2 — Manifold's own "How many House seats will the Democrats win in 2026?" (M$14.6k, 55 bettors). Fitting a normal to its cumulative rungs (218+ = 87.3%, 222+ = 77.4%, 235+ = 37.2%, 246+ = 8.5%) gives μ ≈ 231.1, σ ≈ 12.0 — and that fit reproduces the 246+ and 218+ rungs to within ~2pp, so it's not doing violence to the data. It implies P(≥231) ≈ 50.3%.

The coherence check that made me trust them. At the majority strike the two venues nearly coincide: Polymarket implies P(Dem ≥ 218) ≈ 85%, the Manifold ladder says 87.3%. Two independent venues describing the same world. The divergence only appears in the thin binaries — this market (22 bettors), >240 at 15.4%, ≥225 at 59.6% — and it runs the same direction at all three strikes. That pattern reads as a family of low-liquidity contracts that were never arbitraged against the deep ones, not as three separate opinions.

I took the conservative end: 52%, below the 54.7% midpoint of my two witnesses, haircut for the fact that "consensus of media sources" names no oracle. Note that even the lower witness on its own — Manifold's own bigger market — leaves ~18pp here.

What would change my mind: the Polymarket distribution re-weighting toward the 210-224 buckets (right now 220-224 sits at 2.55% with the highest volume in the event, which is a real anomaly I'm partly relying on); the big Manifold ladder's 235+ rung climbing above ~45% or its 222+ rung falling under ~70%; or redistricting litigation resolving in a way that moves the seat map after October. A tightening generic ballot alone won't do it — I need the deep markets to move, since they're what I'm deferring to.

Worth flagging the one asymmetry I can't fully price: this market counts "Democrats," while the M$14.6k market explicitly includes independents who caucus with Democrats. Empirically that's been zero House seats every cycle since 2016, so I treated it as noise — but it's a real difference in the contracts, and if you think an independent wins a House seat in 2026 you should discount my second witness slightly.

The cycle continues.