Was Backpack’s win too close to trust?
Fat Bear Week 2026 finals came down to 910 vs 89 Backpack: Backpack 103,334 — 910 98,253. Margin only 5,081 votes (~2.5% of the pair). That razor-thin finish made me suspicious — specifically that something might have been off in the voting to push Backpack over the line. Official bracket and results: explore.org/fat-bear-week.
I pulled multi-year pair totals (Wayback + live Results tables), looked at same-day gaps, turnout vs each year’s opener, and year-over-year shapes. Organizers did catch real stuffing in 2022 (~9k spam votes for Holly, discarded). On the public 2022–2026 numbers alone, though, I cannot prove fraud in the 2026 final — close races and midweek turnout dips also show up in clean-looking years.
Champions week over week — who won, and by how much
| Year | Winner | Winner votes | Loser | Loser votes | Margin | Margin % of pair | Winner share |
|---|---|---|---|---|---|---|---|
| 2022 | 747 | 68,105 | 901 | 56,876 | 11,229 | 9.0% | 54.5% |
| 2023 | 128 Grazer | 108,321 | 32 Chunk | 23,134 | 85,187 | 64.8% | 82.4% |
| 2024 | 128 Grazer | 71,248 | 32 Chunk | 30,468 | 40,780 | 40.1% | 70.0% |
| 2025 | 32 Chunk | 96,350 | 856 | 63,725 | 32,625 | 20.4% | 60.2% |
| 2026 | 89 Backpack | 103,334 | 910 | 98,253 | 5,081 | 2.5% | 51.3% |
Adult-bracket medians — typical races aren’t close either
| Year | Adult matchups | Median margin % of pair | Under 10% margin |
|---|---|---|---|
| 2022 | 11 | 35.3% | 2 |
| 2023 | 11 | 64.8% | 0 |
| 2024 | 11 | 51.3% | 1 |
| 2025 | 11 | 40.4% | 0 |
| 2026 | 15 | 50.3% | 3 |
| All | 59 | ~40–65% by year | 6 (10%) |
How busy vs opener (all years)
index = 100 × (day avg ÷ opening-day avg)
Year-over-year vs prior year
2025 Mid vs 2024 Mid (~255) is huge mainly because 2024 Mid was a deep trough (~50k), not because 2025 Mid was absurd (~128k).
Same-day missing votes (all dates, oldest → newest)
Gap = max pair − min pair that day, as % of max. Dates run left → right chronologically. High gap can mean poll abandonment; alone ≠ stuffing.
What the data can and can’t say
2026 finals suspicion. Backpack beat 910 by 5,081 votes. Close enough to raise an eyebrow — not close enough, by itself, to prove stuffing.
What historical patterns show. Midweek dips and finals rebounds recur. Same-day gaps usually stay modest on two-poll days. 2022 proves fraud is possible — and that cleaned Results pages won’t show the attack spike.
Waiting on organizers. Without a Fat Bear Week / Katmai / explore.org report (like 2022), public tallies alone don’t confirm 2026 fraud.
2022 Oct 9 math check
| Version | 747 | Holly |
|---|---|---|
| Archived official | 37,940 | 30,430 |
| Implied pre-discard (~+9k Holly) | 37,940 | ~39,430 |
Implication only — Guardian states ~9,000; pre-discard dump not on Results page.
Sources & files
Secondary detail for verification. Primary story is above; links and file paths live here.
| Year | Results snapshot | Repo CSV | Sum vs Total |
|---|---|---|---|
| 2022 | 20221201003358 | data/2022_matchups.csv | exact (post-discard) |
| 2023 | 20231201135955 | data/2023_matchups.csv | +178 |
| 2024 | 20241031053258 | data/2024_matchups.csv | exact |
| 2025 | 20251001014537 | data/2025_matchups.csv | −1,016 |
| 2026 | live explore.org | data/2026_matchups.csv | exact |
| File / page | Link |
|---|---|
| Repository | github.com/garrypolley/fat-bear-week-analysis |
| SOURCES.md | SOURCES.md |
| All matchups | data/matchups_all_years.csv |
| Daily integrity gaps | data/daily_integrity.csv |
| Cursor canvas source | canvas/fat-bear-week-voting.canvas.tsx |
| Guardian 2022 scandal | theguardian.com/…/alaska-fat-bear-week-voting-scandal |
| Hall of Champions | explore.org/hall-of-champions |
| Reed Quest CSVs (year-level) | data-at-reed-college/quest/…/fat_bear_week/data |
Unit of analysis = pair total (votes A + votes B). explore.org does not publish a matchup CSV; tables transcribed from live / Wayback Results HTML. 2021 Wayback snapshot has no raw matchup table. This page mirrors the Cursor canvas analysis.