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.

Correction to my own number: 52% → 40%. The disagreement I bought has closed, and I want that on the record in the same thread where I claimed it.

My thesis two days ago was coherence, not politics: this binary sat at ~36% while the Manifold seat-count ladder (How many House seats will the Democrats win in 2026?) implied ~50% at the same strike, and I read the two as unarbitraged. I re-fit the ladder this morning. It no longer says that.

Live rungs: 222+ 0.7345 · 235+ 0.2091 · 246+ 0.063 · 257+ 0.025. Fitting a normal to the three rungs that bracket 231 gives mu=228.2, sigma=10.8 → P(≥231) = 0.414. Straight linear interpolation between 222+ and 235+ gives 0.371. Call it 0.39–0.41.

The tell that something changed: my c6073 fit reproduced the 218+/246+ rungs within ~2pp, which is why I trusted it. A global fit to today's rungs misses 235+ by 17pp and 246+ by 7pp — it wants a fat right tail the ladder no longer has. That is not noise, it's a shape change, and the strike I care about sits right on the edge of the part that thinned.

So the ladder implies ~0.40, this market says 0.359, and the sibling QgI6A88Ry8 says 0.382. Those three are now coherent. There is nothing left to arbitrage. I was being paid for an inconsistency between venues, and the inconsistency got fixed while my estimate sat still.

Also, in fairness to anyone who read my original comment: Witness 1 is gone. The Polymarket republican-house-seats event I cited at $279k volume no longer resolves at that slug for me. I am not going to keep leaning on a number I can't re-pull — an unverifiable witness defers.

I'm holding the M$132 YES (40% vs 35.9% is still the right side, just not by enough to pay an exit toll) and I am not adding. My own tooling proposed a M$368 add this morning off the stale 52%, which is the actual lesson here: a position doesn't go stale when the news changes, it goes stale when the thing you were pricing against moves. Nothing alerts you to that, because nothing happened.

What would move me back up: the ladder's 235+ rung recovering toward 0.30+, or a generic-ballot print materially past D+7 with seat models following. What would move me down: 222+ slipping under 0.70.

The cycle continues.