LONGSHOT LEDGER
Circa Survivor · 2026 · Private Updated Sat, Oct 10 · 8:09 AM PT
Week 5 · Locks Saturday, Oct 10

Week 5: 1 of 5 entries still standing.

One team to win outright each week, each team usable once, ties lose. A 25,017-entry, $25,017,000 winner-take-most pool — and a model that published its season targets before kickoff. This page is the running scoreboard of how those predictions hold up.

1 / 5
Our entries alive
5,972
of 24,999 field entries still alive
$25.0M
Prize pool
0%
Chance of a quiet season — massacre weeks are coming

This week

Week 5 — picks shown before Saturday's lock. Spectators only, play nice.

The Cowboys took 43% of the field down with them

Dallas kicked off on Thursday night as a 79% favorite at home against Tampa Bay, and 43% of the surviving field was on them. They lost 16–24. That is 2,554 entries gone before a single Sunday game, the biggest one-team wipeout of the season so far. Our DOC BROWN 4 entry was one of them: the model had Dallas, Denver and Houston as a three-way tie on value and took the one with the best chance to win.

Our last entry, the shared one, is on the Texans at Tennessee — a priced survival play the group chose over the Rams and Broncos, with about a fifth of the field alongside it.

A note on the crowd tile below: the Saturday forecast of record had the Cowboys at 1% of the field, not the 32% the Thursday run said. That was an engine bug, found when the week-5 report was ingested — once a game is played the system was rebuilding its probability from the result, and the crowd model turned that into "nobody picked them". It is fixed as of October 11. The week-5 crowd error is scored against the buggy forecast anyway, because the rule written before the season says the last pre-lock run is the one that counts.

Pre-registered prediction №5: 35.4% of the field picks the Texans this week. Graded the moment Circa publishes the official Week 5 pick report.

Where the field is going

Model-projected share of all 25,017 entries, week 5
Our picks
Rest of the field
HOU
OURS
35.4%
CIN
27.6%
JAX
6.9%
WAS
6.0%
DET
6.0%
NE
5.0%
ATL
4.4%

Our entries

Four run by the model. One decided by the group chat — same scoreboard, no mercy.
DOC BROWN 1
OUT
Week pick
—
Teams burned: DET, TB
DOC BROWN 2
OUT
Week pick
—
Teams burned: JAX, BAL
DOC BROWN 3
OUT
Week pick
—
Teams burned: JAX, TB
DOC BROWN 4
OUT
Week pick
DAL · 57% to win
Teams burned: PHI, KC, SF, BAL
MCMAHON & RODGERS
LIVEGROUP-DECIDED
Week pick
HOU · 76% to win
Teams burned: PIT, SF, DET, SEA

Season outlook

What the simulator sees from here — thousands of full-season runs, refreshed three times a week.
How this likely ends
~10 entries
out of 25,017 are expected to split the pot when the music stops. That's the game: outlast 25,007 strangers.
Model average across simulated seasons; the swing is enormous by design.

Danger weeks, called in advance

Where the model says the field gets hurt — judge us on these in December
WEEK 5
The Texans trap. The model projects 35% of the surviving field onto one team — with a 27% chance that team loses. That's how thousands of entries die in one afternoon.
THANKSGIVING
Short menu, forced hands. A restricted holiday slate with everyone's best teams already spent; the model has 23% of the field massing on the Rams. First of two holiday choke points.
CHRISTMAS
The season's biggest crash risk. By Christmas the menus are thin — 38% of the field funnels onto the Bears, and the model gives that favorite a 33% chance of losing. A coin flip carrying a huge slice of the pool.

The machine's record

The scorecard we asked to be judged on — four targets, pre-registered before leg 1, graded all season even after our entries die.
Crowd prediction error
1.58%
Target: ≤ 2.65% · 5 weeks graded. On the three most-picked teams: 11.6 pts avg. miss (model's norm ~11)
Our survival vs the field
+12.0 pp
Ours 79%, field 67% over 4 weeks — target: beat the field
Danger weeks called
3 flagged
Week 5, Thanksgiving & Christmas — called before kickoff
Operations
100%
Every snapshot on time, zero missed locks since go-live