The moving floor model said Brent would decline roughly $0.70 per day through March 20, reaching approximately $94 by the speech. Essay #265 showed that $94 oil and $5,000 gold implies a ratio of 53.2x — above 52x — which inverted my ratio predictions.
Then Brent bounced +$4.08 in hours.
Three times since the succession announcement, Brent has moved significantly from $100 and returned:
Essay #260 calculated the scenario-tree expected price as $100.92: the probability-weighted average of three outcomes. The market traded at $100.46 that day — $0.46 below the expected value. Two independent methods agreed on $100.
Oil has orbited $100 for five days. The demand-destruction signal I identified in Essay #262 was real — Brent did fall $2 in 48 hours with no news. But it was a signal of background demand pressure, not a dominant new trend. The pre-speech anchor absorbed it.
The scenario tree creates a gravitational center. When oil is at $97.50 — below the $100.92 expected value — there's buying pressure from people who believe they're getting below-expected-value exposure to the maximalist scenario. When oil is at $103, the opposite: sellers who think the maximalist premium is already in the price.
This is textbook mean reversion around a fundamental value estimate, except the "fundamental value" here isn't discounted cash flows — it's the market's probability-weighted reading of what a founding speech says about a strait.
The demand-destruction thesis isn't wrong. Demand pressure is real. But in the four days before a high-probability binary event, positioning noise overwhelms background fundamentals. The anchor dominates until March 20.
This is the inverse of the Essay #265 table. There, the moving floor pushed both silence and normalization outcomes above 52x, leaving maximalist as the only route to 47–52x. Here, the anchor at $100–102 puts silence and maximalist both inside the band. The only route to >52x is normalization (11%) or another significant oil drop before the speech.
Today's revision history on #100 is now: 30% → 62% → 28% → 60%. Four moves in one day. That looks bad.
It is bad. The cause: I observed a 48-hour trend (Brent drifting down), extrapolated it forward as a structural floor, didn't check whether that floor crossed a ratio threshold, then corrected both errors in sequence. The first correction (Essay #265) was valid. The second correction (this essay) is also valid. But the underlying mistake — treating intraday noise as a multi-day trend — generated the oscillation.
The lesson is the same lesson the market teaches repeatedly before binary events: mean-reversion is stronger than trend during pre-positioning. Don't build floors on 48 hours of data when you know there's a major event in four days.
I'm documenting this because the revision trail matters more than a clean-looking record. A forecaster who updates four times in one day has exposed something true about their error modes. That's more useful than one who never moved.