About the Frost Log
The Frost Log is a daily strategy analysis of the Big Brother 28 live feeds. A new edition posts every morning at 6am PT, built from the previous day's publicly available feed coverage. It is not episode recaps and it is not spoiler-safe: the feeds run ahead of the broadcast, and so does the log.
It is also not neutral. Every edition commits to reads — who holds power, who's leaking, who blundered — and then grades those reads in public when the outcomes arrive. The log's first rule is that it scores itself before it scores anyone else.
How to read an edition
The Cold Read is the editorial: what actually shifted, stated plainly enough to be wrong. The Counter-Read closes each edition with the best honest case that the lead read is mistaken — ambiguous feeds turn into elegant, self-sealing stories otherwise, and the counter-read is the cheapest protection.
Alliance Standings are drawn as a transmission map: not just who is aligned, but who talks to whom, which relationships leak information, and which conspicuous silences matter. When the log says a player is an "edge" or a "leak line," it's describing how information moves through them.
The Prediction Ledger is the accountability engine. Every forward claim carries a class (vote, nomination, flip, comp-branch, structure, jury), a probability, and an evidence grade:
- A — said privately and backed by behavior
- B — repeated behavior or multiple independent conversations
- C — a single statement, possibly performed for an audience
- D — inference from silence, body language, or the edit
Houseguests lie, rehearse, and misunderstand their own motives; the grade is the reader's protection. Predictions are scored against naive baselines — "the stated target goes home," "the nominations stand" — because a forecaster only deserves credit where it beats the obvious guess. Resolved predictions are graded on both process and outcome: a lucky hit doesn't strengthen the method, and a well-calibrated miss doesn't weaken it.
The Retrospective scores yesterday's edition against what actually happened — the log grades itself before it grades anyone else.
One piece of shorthand the log leans on constantly: Big Brother is not one game, it's two games stapled together. Game 1 is survival — the weekly elimination gauntlet of nominations, vetoes, and votes; visible, eventful, and the source of nearly all the discourse. Game 2 is jury construction — the winner is chosen by the people the finalists helped evict, so every eviction is also an act of electorate design: you are building, one bitter voter at a time, the body that will judge you. When the log talks about a player's equity, it means their standing in both games at once — and a striking share of analytical error in this fandom comes from optimizing Game 1 while Game 2 decides the winner.
Some recurring vocabulary: a signature file is the running record of a houseguest's observed behavioral patterns, always tagged with how many scenes support it. The threat ledger tracks how the house prices each player's dangerousness — which is often very different from how dangerous they actually are. A regime call states how chaotic the game currently is; in chaos weeks, confidence ceilings drop and prediction horizons shorten on purpose.
A footnote marker carries the log's sourcing so the prose doesn't have to. Solid sourcing — multiple trusted feed sources, or something plainly seen on the feeds — is the default, and gets no marker at all. A superscript ¹ marks a claim as thin: single-source, an ambiguous name, an unclear timestamp. Every ¹ points at the same one-line footnote in the closing fine print, so it repeats rather than counting upward. It attaches only to factual claims where provenance actually changes how much weight you should give them; the log's own reads, opinions, and prediction probabilities never carry one, because editorial confidence is a different axis from source quality. An edition with no ¹ in it is simply an edition where nothing was shaky.
(Editions before 2026-08-06 used a pair of colored triangles instead — a green ▴ for solid and a yellow ▾ for thin, marking every graded claim. They interrupted reading more than they informed. Archived editions keep them.)
Timeline claims trace to public feed-update sources, which can be wrong; see the terms for the fine print. Corrections happen in the next edition, in the open.
The log posts daily. It is watching so you don't have to — though you should anyway.