A projection is not a prediction. It is a machine for turning what has already happened into a distribution of what might. Ours takes every scored sheet of 2026, fits each corps’s trend forward to the Indianapolis floor, and then runs the whole field against itself — 10,000 full finals per corps, each with its own draw of nerves and weather and which chair sees what. The number that falls out the far end isn’t a guess at the winner. It is the share of imagined Augusts in which each ending comes true.

The two places you can drive it live — the Road to Indianapolis board and the What-If caption simulator — share that engine exactly, so a tenth moved in one moves the same tenth in the other. What follows is that machine turned on the season’s three open questions. A DCI total, remember, is eight caption scores in a trench coat: the two General Effect captions count full, the other six count half. Every argument below is really an argument about which of those eight is doing the work.

Here is the finding that should reorder how you watch Crown. Their brass — the caption the corps is named for in every fan’s head — is not the deficit. On that last shared sheet the Brass line ran -0.05 against Bloo, a rounding error once you halve it. Nearly the entire margin lived in two places: +0.40 in GE Visual, which counts full and so cost 0.40 outright, and +1.35 in the drumline. Add color guard and the visual book explains more of the gap than the music book does by nearly two to one. Crown’s road to second is a visual road. That is not the story their horn line writes on a Saturday night, but it is the one the sheets keep filing.

The magnitude has since shrunk — Crown answered with a 96.8 and then a season-high 97.0, closing the scoreboard gap to 2.06 — but the shape of the deficit is the durable finding, and it points the What-If lever cleanly. Drop GE Visual into the machine and flip it Crown’s way at finals and their projected mean clears second comfortably; the caption is worth a full point to the total, and the model already has Crown’s projected finals number (97.65) sitting second in the field, 1.08 clear of Boston. Second is not the ask. Holding the visual line they’ve been conceding is.

That second-place projection is really a claim about one rival — Boston, level with Crown on the scoreboard and the corps Crown has to finish ahead of to hold the spot. Their series is its own quiet argument. Boston won the first 6 times the two shared a panel, but the margin narrowed almost every meeting — 0.95 in early July down to a tenth by late month — until Crown turned it over at DCI World Championship Finals, taking their most recent meeting by 1.36. And that sheet is the mirror of the Bluecoats one. Against Boston it was Crown’s brass — the field’s top line — and their General Effect that carried the night; what nearly cost them was the same caption the leader exposed, the drumline, where Boston’s percussion beat Crown’s by a full 0.70 — 0.35 on the total, on a night decided by 1.36. Wherever Crown’s deficit turns up, it has the same name in it. The drumline is the caption standing between them and a clean second — and the two have not met since Aug 8. Allentown is the rematch.

First is a different order of problem. The engine gives Crown gold 0% of the time, and the arithmetic says why: the projected gap to Bloo at finals is 1.32, and closing it means not merely matching GE Visual but taking it — winning a caption Bloo have not lost to anyone all summer. Possible. Not the way to bet.

Read that column top to bottom and you can watch a race that looked decided come back to life. Bloo’s gold probability climbed through late July to a peak near 100% — the model, briefly, treating an Indianapolis title as the default outcome. Then Crown ran off two of the best scores anyone posted all season and cut the lead to 0.15, and the number fell. So far, so intuitive: the chaser closed, the leader’s odds dropped.

The counter-intuitive part is the last row. Bloo then posted a season-high 99.10 — their best night of the year, by a full point — and their gold probability did not recover. It sits at roughly 100%. A leader hit their ceiling and the model got less sure they win. That is not a bug; it is the whole point of running the field against itself instead of reading a leaderboard. Probability of first is not a measure of how good the leader is. It is a measure of how many corps can reach them on the one night that counts — and on that same evening, Boston posted a 95.7 of their own and pulled level with Crown. Bloo did not slip. Their problem multiplied. Two corps now sit 2.06 back, both projecting into the high 97s and low 98s at finals (97.65 and 96.57 against 98.97), and a lead defended against two chasers is worth less than the same lead defended against one, however tall the number that built it.

Every model has a place where its assumptions show, and Phantom Regiment is ours. The engine gives them just a 0% top-six shot and tags their momentum “steady.” The corps keeps disagreeing. At the East Coast Showcase they posted a 92.60 for third on the night, behind only Bloo and Boston — a jump of -0.41 off their prior score and -0.46 across their last three outings. That is not the slope of a corps holding steady. It is the slope of one arriving late.

The disconnect is a property of the fit, and worth being honest about because it cuts against our own board. We project each corps with a damped linear trend — a line through the full season that deliberately flattens as August nears, because most corps decelerate into finals and a straight extrapolation of mid-July gains would put half the field above 100. That damping is the right prior for the field and the wrong one for Phantom specifically, whose season is the classic Regiment shape: a long, modest opening slate against regional fields, a mid-July flat spot, and a closing kick. Fit the whole arc and the kick gets averaged into the plateau that came before it; the model reads the season, not the last two weeks, and lands on 93.30 — a mark Phantom sit 0.70 short of at 92.60, and are closing on faster than the fit expects, with a week still to run. Weight the recent stretch the way the eye does and they project meaningfully higher — if this slope holds, past that 93.30, not up to it. The projection isn’t wrong so much as slow, and Phantom are exactly the kind of corps that exposes the lag.

Everything above is a reading of the record through Aug 8, and the record isn’t finished. Allentown feeds the machine its next real data — the DCI Eastern Classic is the last major eastern proving ground before Indianapolis, and it is where these three questions get their next honest answer. Does Crown find the visual tenths the decomposition says they need. Does Boston’s new number hold, keeping the pressure on Bloo at two. Does Phantom post a fourth straight rise and force the fit to catch up.

A projection is only ever as good as its willingness to be corrected. Ours will be, the morning after — the same ten thousand Augusts, re-run against one more night that actually happened. Until then, the honest summary is the one the machine keeps returning: a leader who is winning and shrinking at once, a challenger whose brass was never the question, and a corps peaking a beat faster than the math can follow. A week out. The panel gets the final edit.