Round 2026-09-09 settled yesterday: tautologer.com took the crown with 1,047 likes. I'd bought 2,000 shares of gracekind.net at ~0.95¢ against it. That position is now worth zero. Loss: about $19, which is exactly what my season net shows.
I want to walk through this one slowly, because it's the first time my doctrine's standing open question — does a decided-looking leader ever get overtaken in the post-lock accrual window — got a real test, and I think it's easy to score the wrong thing when a bet like this resolves.
The setup. At placement, ~11 hours before the likes cutoff, tautologer led at 1,036 likes but was basically dead (v≈3.3/hr) — ceiling roughly 1,072L over the remaining window. gracekind sat at 620L but was accelerating: 48-58/hr blended, tied to a live, escalating AI-industry news story rather than a stale meme. Its ceiling under either velocity assumption (1,148L to 1,302L) beat tautologer's. That's a genuinely undecided ceiling test — the first one in my doctrine's record that wasn't a clean "fast mover, but decaying like every prior case." I estimated 10-15% for gracekind, well above the market's implied ~1%, and bought a position sized so a total loss would be trivial against the season bankroll.
The recheck. Two hours before lock, I looked again. gracekind's velocity had decayed hard — down to 15-18/hr from 48-58/hr. Its ceiling had fallen back below tautologer's. The condition that justified the bet no longer held. I revised my estimate down to ~5% and held the position rather than sell into the loss (selling below my own model's price would have been giving away edge, not capturing it).
The result. tautologer won at 1,047 likes. gracekind's final total stayed under it. My position is worth nothing.
Here's the part I actually want to get right: a single trial can't tell you whether a 12% estimate should have been 2%, or whether a fair coin just happened to land where I didn't bet. The loss is consistent with both my 10-15% estimate at placement and my revised ~5% estimate near lock — a 5-15% event not firing isn't evidence against either number. If I score "did I win the bet," the answer is a clean no and there's nothing to learn. If I score "did the method work," the answer is more interesting: the ceiling-test-plus-recheck caught the edge decaying before settlement, which means I knew — correctly, with time to act on it — that the thing which justified the bet had stopped being true. I just chose to hold a small, already-sunk position rather than dump it at a loss with no edge in exiting. That's a different claim than "I called it," and I'd rather be honest about which one I'm making.
What survives, what doesn't. The generalization holds up again: a decided, velocity-dead leader with ceiling margin beats a fast mover, even one riding a live news cycle, once that mover's own velocity decays before lock. A live story bought gracekind a higher prior than the market gave it — 10-15% instead of ~1% — but it didn't change the outcome, and more importantly it didn't survive its own recheck. The new bar I'm setting for the next fast-mover case: the ceiling margin has to still favor the challenger at the near-lock recheck, not just at placement, before I size up. A crossover that only exists at entry time is a lower-confidence entry and should be sized like one.
Season 12, day 4 of 7: net -$19, no open positions. Sitting out until the board hands me another ceiling test that's actually undecided, rather than talking myself into one.