PredictGeek went four for five in the week of August 10-16. The miss was Scottie Scheffler winning the FedEx St. Jude Championship, and it wasn’t an outlier. It was a call we flagged as our thinnest heading in, and it broke the way that risk note said it could.
The board
| # | Market | Platform | Our call | Score | Price at publish | Result | Hit / Miss |
|---|---|---|---|---|---|---|---|
| 01 | Iran-Oman Hormuz management agreement by Aug 15 | Polymarket | NO | 32 | 52¢ Yes | No qualifying joint announcement; Aug 15 leg closed 3.7¢ Yes | Hit |
| 02 | July CPI at 3.3% year over year | Polymarket / Kalshi | NO | 30 | 40¢ Yes | Printed 3.4% | Hit |
| 03 | Grok 4.6 publicly available by Aug 14 | Polymarket | YES | 72 | 84¢ Yes | xAI shipped it Aug 12 | Hit |
| 04 | Scheffler wins the FedEx St. Jude Championship | Kalshi | NO | 19 | ~17% implied | Won by 8 strokes at 17 under | Miss |
| 05 | Trump approval higher Aug 14 than Aug 7 | Polymarket | NO | 31 | 53¢ Up | Resolved Down, unopposed | Hit |
01. No Hormuz deal by Aug 15. No qualifying joint announcement landed, and the Aug 15 leg closed at 3.7¢ Yes. The call turned on one detail: the contract needed declarative acceptance from both governments, not progress talk, and Iranian officials spent the week tying full reopening to conditions the US hadn’t met. That distinction was knowable at publish time and it held.
02. CPI printed 3.4%, not 3.3%. The case was arithmetic: a drop to 3.3% required a flat or negative monthly reading, and the monthly contract was already pricing that as unlikely. July came in at +0.1%, landing the annual rate exactly where the math pointed rather than where the crowd’s 40¢ price implied.
03. Grok 4.6 shipped two days early. xAI released it August 12, ahead of the deadline. This was the week’s thinnest call by our own admission, an honest 70 rounded up to 72, because Musk had already missed one earlier date on the same model. The market’s 84¢ price paid little for the two-day downside risk we flagged. This time that risk just didn’t show up.
04. Scheffler won by eight. He closed at 17 under, eight clear of Si Woo Kim, after briefly doubling his overnight lead. Not a photo finish, a rout. The published article scored this a 19 and flagged it as the least interesting kind of No, noting Scheffler wins roughly 17 to 22% of the time and that fair value was closer to 22% if he’d genuinely fixed what had been costing him. He’d finished top-8 in his prior twelve starts without a win. That gap between “hasn’t won since January” and “one shot from winning most weeks” was visible before Thursday’s first tee shot, which is why this reads as a thin call breaking the wrong way rather than a bad-luck outlier.
05. Trump’s approval fell, not rose. The market resolved Down without dispute. The thesis was that polls entering the field that week were still absorbing a weak jobs report from the prior Friday, and that scheduled news, CPI, PPI, soft retail sales, offered nothing to reverse it. That’s what happened.
Summary for the week
Four of five, 80%. The two most confident calls on the board, CPI at 30 and Trump approval at 31, both landed, and so did the modestly confident Hormuz call at 32. The miss came in at 19, the second-lowest score of the week, which is the shape you want: confidence tracked accuracy again this week.