Frost Log — October 2, 2026

Executive Summary

Rick Devens won Big Brother 28, 6–1 over Taylor. Every juror but Lala voted for him, including Drew, the friend he evicted to get there.

The endgame resolved along the branch we had moved toward but not committed to:

Drew called the cut "the right move I think man, it just hurts." Devens also won America's Favorite Houseguest, for $810,000 in total winnings. Taylor takes $75,000 as runner-up.

The feeds ended for the season at about 9:06am PT. The houseguests said individual goodbyes. The "talk in the morning" Devens had promised Taylor never appeared on camera.

Part 3 itself broke the format expectation. The questions came as juror-delivered statements to be judged true or false, not the duration-and-trivia drill Taylor had been running.

Our scorecard is mixed:

The Cold Read

Read 1: We called taking Taylor a blunder candidate. It became a landslide, and that is the most important thing to learn from this finale.

Yesterday we had Devens beating Drew about 75% of the time and Taylor about 62%. We said taking Taylor would be "the clearest case this season of a player picking the opponent who scares him less over the one he beats more." He took her and won 6–1.

Our juror-by-juror model for Devens–Taylor is worth setting out in full:

Juror Our lean Actual vote
Barrett Devens Devens
Dee Devens Devens
Angela Devens-lean Devens
Drew Devens, at 60 Devens
Yash Toss-up Devens
Lala Devens-lean Taylor
Melody Taylor Devens

The two misses cancel on the count but not in meaning. Melody voted for the man whose tiebreak sent her to the jury. We had her warm on Taylor and cold on Devens. Lala was the lone Taylor vote. She was the juror with Devens pledges on record, and the one Taylor told Devens she could move.

Here is what we can and can't claim from this.

Proven: Devens–Taylor was not a close pairing, and our 62 underpriced Devens in it badly. Calling the take a blunder candidate was wrong on the outcome. It was also probably wrong on the process: we priced Taylor as a stronger jury opponent than anything in the jurors' own experience justified.

Not proven: that taking Taylor added equity versus taking Drew. One plausible story is that cutting your closest ally to sit next to the Part 1 winner reads as agency, while dragging the goat reads as fear. The other is that Devens beats either of them comfortably and the choice barely mattered.

We can't tell those apart from one season. Exit interviews and the finalists' jury answers would be the evidence. We have nothing solid on either beyond the vote itself.

What does stand: Devens won on the signature we logged more often than any other for him. He comforted whoever he might need as a voter, at least fourteen times on record. Every juror walked out of this house having been handled warmly by him, including the ones he evicted. Juries vote on what they personally experienced, and what they experienced of Devens was kindness on the way out the door.

Read 2: The Part 3 lean we gave Taylor rested on a format assumption, and the format changed.

Our 52/48 Taylor edge was a prep edge. She was drilling the season-events format the show has run for years, and we treated Drew's insistence on a juror-statement game as a relay of a retired design.

Part 3, "Back in My Day," had jurors in old-age costumes deliver statements, and the finalists picked out the false ones. Its shape was closer to what Drew kept describing than to the "how long did the Pressure Cooker last" trivia Taylor was rehearsing at midnight. Whether the statements' content was about season events isn't documented. Either way, her odd-minute duration strategy never came into play. Devens won, and Taylor missed questions 2 and 4.¹

The lesson is not that Drew read the show better. It's that a format expectation is a prior, not a fact, and a prep edge only counts if the prep matches the comp. We let a strong expectation stand in for knowledge and then traded on it for two editions.

In fairness, the trim we made, from 54 to 52, was driven by Devens studying while Taylor spent days on hair. That part of the read held: the player who reviewed broadly beat the player who drilled narrowly.

Read 3: Taylor's pitch worked, and its own arithmetic predicted her loss.

Taylor's Wednesday pitch got her the seat. In the branch where she lost Part 3, that is the best result she could have bought, and it was worth $75,000 she wouldn't otherwise have had.

But look at what she argued. If Devens took her, he'd have Drew's vote "for sure" and wouldn't have to fight Drew for Barrett. Both came true. She sold herself as the opponent Devens could beat, and that is the opponent she turned out to be.

The threat half of her pitch was that, if Devens took Drew, she'd vote against him and bring Lala. Lala did end up voting Taylor, from the other branch entirely. We don't know whether the two spoke before the vote.

In post-show interviews, Taylor said she had been lying about honoring the deal if she won Part 3.¹ That is consistent with her stated target of Devens all season. The mutual offer was insurance, never intent.

Read 4: Drew was a passenger to the last stop, and he voted for the driver.

Drew never learned of the pitch on any feed we have, and he never got an answer to "were you actually plotting to get me out?" He was evicted by the vote of the man he'd spent the week calling destined to win.

He then voted for that man. His signature, warmth toward whoever is kind to him, survived the cut, which is exactly what the 60 on P438 was betting. "The right move I think man, it just hurts" is the line of a juror who has already forgiven.

Detailed Timeline

(All times PT. Thursday, October 1 unless noted.)

The last feeds

Live finale, about 5:00–6:30pm PT. Minute-level stamps are approximate.¹

Broadcast documentation of earlier events. This is new detail about old events, dated to when they happened.

Post-show (color only):

Alliance Standings

The game is over. These are the final positions of the information network, and what it cost the people at either end of it.

1. Devens–Taylor: the alliance that was never declared, and the one that decided the season.

2. Devens–Drew: broke at the last moment it could, cleanly.

3. Taylor–Drew: warm to the end, and never told the truth.

The jury, as a network:

Conspicuous non-conversations, final ledger:

Player Status

Prediction Ledger

Evidence grades: A = said privately and backed by behavior; B = repeated behavior or multiple independent conversations; C = a single statement, possibly performed; D = inference from silence, body language, or the edit.

Resolved this window. The scores below are Brier-style: lower is better, and 0.25 is a coin flip.

# Class Prediction P Outcome Brier Grid
P439 comp-branch Part 3: Taylor 52 / Devens 48 48 (Devens) Devens 0.27 Process-wrong / outcome-wrong. The Taylor lean rested on a format assumption that failed. We lost slightly to a coin flip.
P402 comp-branch Final HOH overall: Devens 40 / Drew 35 / Taylor 25 40 Devens — Top pick right on thin margins. Part 1 went to our low name.
P434 finalist-choice Devens wins Part 3 → takes Drew 55 Took Taylor 0.30 Process-right in direction (cut from 80) / outcome-wrong. The original 80 would have scored 0.64.
P435 flip Devens tells Drew of the pitch before Part 3 35 No (not on record) 0.12 Process-right / outcome-right.
P438 jury In Devens–Taylor, Drew as seventh juror votes Devens 60 Yes 0.16 Process-right / outcome-right. Drew's sympathy signature carried into the jury.
P378 jury Devens beats Taylor in the final 75 filed (live ~62) Yes, 6–1 0.06 filed / 0.14 live Outcome-right, but our live number underpriced a landslide.
P440 jury Season winner: Devens 40 / Taylor 38 / Drew 22 40 Devens — Top pick right, low conviction.
P413 finalist-choice Final two is Drew–Taylor 75 filed (live ~42) No 0.56 filed / 0.18 live The filed number was process-wrong; the repricing was right.
P416 finalist-choice Drew sits in the final two 80 filed (live ~68) No 0.64 filed / 0.46 live Process-wrong. We priced the pledge's repetition, not its exposure.
P426 flip No third Devens pitch to Taylor before Part 3 75 Yes 0.06 Clean.
P429 flip Drew never concedes the season-events format before Part 3 75 Yes 0.06 Outcome-right, process-wrong. We read the stubbornness as a misread, and the format then moved toward his version.

Closed without scoring, because their branches didn't occur or their line was never captured:

The jury model's verdict. Of the six jurors we actually called in Devens–Taylor, we got four right; Yash was a toss-up and went Devens. The two misses: Melody, our only firm Taylor call, and Lala, a Devens lean.

Baselines.

New predictions: none. The season is closed.

Booked: 1. A prep edge is only as good as the format assumption under it. When the comp's shape is unconfirmed, a drilling edge is a branch, not a bonus. 2. The pairing you beat less on paper can be the landslide. Before calling a seat choice a blunder, price what the jurors personally experienced of each finalist, not who looks scarier on the feeds. (Hypothesis: choosing a comp-winning rival may itself count as a résumé item.) 3. A critique voiced at the jury table is not a critique adopted. Barrett named the case against Devens, then voted for him.

The Counter-Read

The lead read is that the choice of Taylor was fine, and our jury model, not Devens, got it wrong. Here is the best honest case against that.

Devens may have gotten lucky in his opponent, not shrewd in his choice. If Taylor's jury performance was simply weak on the night, a 6–1 margin says more about one Thursday than about the pairing. Drew might have lost 6–1 to him too, or 7–0.

In that reading, our 75-versus-62 ranking was directionally right. The Taylor pairing won for Devens, the Drew pairing was never tested, Drew may still have been the safer opponent, and Devens chose the riskier seat and got away with it. The landslide doesn't prove the choice was optimal; it proves it was survivable. That is the same trap we warned about with the loyalty keep a season ago.

Melody's vote may not mean what we want it to. We can't say whether she voted for Devens or against Taylor, her co-nominee the night she left. If it was against Taylor, Melody tells us about Taylor's relationships, not about Devens's goodbye game.

Lala's vote cuts against the "warmth converts" story. She had pledges with Devens and voted for Taylor. If his comfort-the-voter style was the engine of the win, it failed with exactly the juror it was most aimed at.

On the format, the counter is simpler. Devens may have won Part 3 on any format. He out-studied Taylor all week, and a broader review may beat narrow drilling regardless of shape. If so, our process error on the format cost us accuracy we would never have earned anyway.

What would settle it: the jurors' own explanations of their votes, which, as of this writing, we don't have on solid sourcing.

Retrospective

Grading the October 1 edition against outcomes, with its ex ante branches preserved.

Branch table as filed:

What happened: the 32% final two came true, and inside it the 62% favorite won 6–1. We had the eventual winner on top, but within two points of the runner-up.

Read 1 (Taylor insured her losing branch, and it was right): process-right, outcome-right. The pitch got her the seat, and post-show color is consistent with it having been insurance.¹ The best read of the edition.

Read 2 (Devens torn; taking Taylor is a "blunder candidate"): split.

Read 3 (Drew is a hub with no inbound information): process-right, outcome-right. He never learned of the pitch on any feed, and was cut.

Read 4 (Taylor's prep slipped; trim to 52/48): process-wrong at the foundation. The trim was in the right direction, but the edge we were trimming assumed a format that didn't appear.

Counter-read discriminators:

Net for the finale:

The process miss worth carrying into next season: we argued the jury from feed-level "who's scarier" for one day too many, instead of from what each juror had personally been through.


Coverage: October 1, 05:54 PT → October 2, 05:54 PT (Day 87 and post-finale). The live finale aired in this window (about 5:00–6:30pm PT). Its live content, which was Part 3, the final eviction, the jury vote, America's Favorite Houseguest and the renewal, is dated to October 1. All other broadcast material documents earlier days. Regime call: SEASON COMPLETE. Devens wins Big Brother 28, 6–1 over Taylor; Drew finishes third. Confidence ceilings, retired with the season: comp outcomes near base rate and written as branches; finalist-choice calls capped near 85 and cut when stated intent conflicts; jury forecasts held at a hard ceiling and argued from what each juror personally experienced. Evidence grades: A = said privately and backed by behavior; B = repeated behavior or multiple independent conversations; C = a single statement, possibly performed; D = inference from silence, body language, or the edit. ¹ Thin sourcing: single-source, an ambiguous name, or an unclear timestamp. Unmarked claims are solidly sourced.