FROZEN PREDICTION

Iowa vs Michigan

My CFB Baseline · 2026 Week 4 · revision 3 · Independent

Open this immutable run

Michigan by 2.1

Why this pick?

Game controlSmall effect

Avoiding bad lossesSmall effect

Close-game gritNeutral effect

Defensive toughnessModerately leans away

Scoring punchSmall effect

Transfer portal impactModerately leans away

Previous-season performanceSmall effect

Recruiting strengthSmall effect

Returning productionSmall effect

Team talentModerately leans home

Big-game winsModerately leans home

Recent formModerately leans away

Wins away from homeNeutral effect

Tough schedulesModerately leans home

Statement winsModerately leans away

WinningNeutral effect

Show the technical details

Frozen 2026-09-26T13:29:59.255436-05:00 · Snapshot 193 · Home win probability 0.5647 · 80% margin interval -14.5 to 18.8.

Provisional calibration: real-season accuracy is not established.

Model configuration used

Configuration: Revision 3
Engine version: 3.8.0-kickoff-chronology
Evidence cutoff: Sep 21, 2026 · 4:13 PM CDT

Factors unavailable across the whole snapshot are excluded. Their weight is redistributed among available factors, so the effective model can differ from the configured sliders. Missing evidence for just one team remains neutral. The tables show saved weights; omitted factors are not reconstructed from current model settings.

Factor key	Normalized share
`adj_margin`	0.108108
`avoid_bad_losses`	0.040541
`close_game_wins`	0.027027
`defense_ppg`	0.067568
`offense_ppg`	0.067568
`preseason_portal`	0.040541
`preseason_previous_performance`	0.054054
`preseason_recruiting`	0.067568
`preseason_returning`	0.054054
`preseason_talent`	0.067568
`quality_wins`	0.081081
`recent_form`	0.054054
`road_neutral_wins`	0.040541
`sos`	0.081081
`statement_wins`	0.040541
`win_pct`	0.108108

Retained calibration/context	Frozen value
`version`	provisional-all-slates-policy-v1
`calibration_kind`	documented_policy
`validation_status`	not_historically_validated
`applied_home_field_points`	2.5
`applied_fcs_baseline_z`	-1.5
`residual_scale`	13.0
`residual_scale_source`	documented_policy_not_validated
`intercept`	0.0
`difference_coefficient`	7.0
`home_field_coefficient`	2.5

These values came from this immutable run, not the model’s current settings.

Factor	Home evidence	Away evidence	Normalized share	Home effective share	Away effective share	Weighted difference
Game control	Measured	Measured	0.108108	0.12417134115247323	0.12417134115247323	-0.034527654192065424
Avoiding bad losses	Measured	Measured	0.040541	0.04656425293217746	0.04656425293217746	-0.005731670133838643
Close-game grit	Measured	Measured	0.027027	0.031042835288118308	0.031042835288118308	0.0
Defensive toughness	Measured	Measured	0.067568	0.07760708822029577	0.07760708822029577	-0.06566351891117488
Scoring punch	Measured	Measured	0.067568	0.07760708822029577	0.07760708822029577	-0.04423480880506697
Transfer portal impact	Measured	Measured	0.040541	0.02533783783783784	0.02533783783783784	-0.07192656244136243
Previous-season performance	Measured	Measured	0.054054	0.033783783783783786	0.033783783783783786	-0.02244061318156216
Recruiting strength	Measured	Measured	0.067568	0.04222972972972973	0.04222972972972973	0.049709892490937116
Returning production	Measured	Measured	0.054054	0.033783783783783786	0.033783783783783786	0.04543113162485053
Team talent	Measured	Measured	0.067568	0.04222972972972973	0.04222972972972973	0.062201390289948554
Big-game wins	Measured	Measured	0.081081	0.09312850586435492	0.09312850586435492	0.07407456007887106
Recent form	Measured	Measured	0.054054	0.062085670576236617	0.062085670576236617	-0.05015294294892749
Wins away from home	Measured	Measured	0.040541	0.04656425293217746	0.04656425293217746	0.0
Tough schedules	Measured	Measured	0.081081	0.09312850586435492	0.09312850586435492	0.06560102859806773
Statement wins	Measured	Measured	0.040541	0.04656425293217746	0.04656425293217746	-0.056751966338688405
Winning	Measured	Measured	0.108108	0.12417134115247323	0.12417134115247323	0.0

Missing team data was represented neutrally in the calculation; it was not a measured neutral result. Measured equal or offsetting inputs can also have zero effect.

This is the exact saved prediction. Current model settings and current prediction selection are not used here.