Spread: Celtics (-9.5)

Research grade

⚠️ AvoidMagic at 64%

Matches a pattern that lost clearly more than the plain consensus in our tests.

• − BasketballFull year, out of sample: basketball lost −4.8 % a trade (n 1,747, 95 % CI −9.2 to +0.3); Oct 22 backtest: −12.0 % and −9.2 % in its two windows.
• ≤ Spread or totalOct 22 backtest: spreads and totals −14.5 % a trade in spring (95 % CI −27 to −1), +1.5 % in October: no edge.
• ≤ Consensus under 60 %Every backtest started at 60 %+ of the holders; below that nothing was measured.
• · Favourite (60–95¢)Full year, out of sample: 0.70–0.90 lost −0.9 % a trade (n 26,050, 95 % CI −1.5 to −0.3): the least bad band, still a loss. The Oct 22 backtest's +0.4 % (n 4,327) was not confirmed.

The rules come from our Oct 22 2026 backtest (one entry per market, held to resolution, after a 1¢ half-spread and Polymarket's taker fees). Our full-year test without hindsight (Oct 26 2026) found none of them profitable out of sample, and the plain consensus lost −3.0 % a trade. The grade describes the market; it is not a forecast, not an edge and not advice.