Portal counts, talent totals, and broad reconstruction measures worsen held-season opening-gap accuracy.
The Ledger · Roster continuity
Portal and Roster Reconstruction
Does portal volume or broad roster reconstruction explain an additional signed season-opening market adjustment?
No separate adjustmentNo stable directional effect after existing team, quarterback, and game-flow information.Updated August 31, 2026
Decision
Do not add portal churn to the Factor sum. Retain the underlying counts for prospective validation and future interactions with quarterback, coach, or unit availability.
- Historical seasons
- 2022–25
- Primary sample
- 46 extreme-reconstruction openers
Start here
What the evidence says
The audited movement models do not show a stable week-of-game response to generic roster churn.
Result accuracy generally worsens when a broad reconstruction adjustment is added.
Much of the information is already in the team rating. Specific positions or units remain better research targets than total churn.
2026 prospective implementation
Extreme portal and returning-production states can be scored for 2026 from retained project files; the reviewed-condition detector is not yet connected.
Current inputs
- Portal entries and exits
- Transfer talent
- Returning PPA
- Coach and quarterback context
- Additional source need
- No new provider for the current preseason snapshot; future seasons require the existing roster/portal feeds to refresh.
- Refresh cadence
- Preseason, then after material roster corrections.
- Coverage and limitations
- Broad churn has no approved direction and should appear only as a zero-point evaluated condition.
Study results
| Test | Games | Result | Validation |
|---|---|---|---|
| Extreme FBS-FBS center | 26 | No stable direction | Mean driven by a minority of large cases |
| Extreme FBS-FCS center | 20 | 0.21 points toward reconstructed favorite | Effectively zero |
| Opening MAE | 127 | 3.027 baseline vs 3.221 reconstruction | Reconstruction worse |
| Kickoff MAE | 61 | 3.695 baseline vs 3.707 reconstruction | No useful gain |
| Opening-to-kickoff MAE | 42 | 1.773 baseline vs 1.841 reconstruction | Reconstruction worse |
| Actual vs OV MAE | 185 | 13.000 baseline vs 13.267 reconstruction | Reconstruction worse |
What we tested
Once the audited game-flow baseline is removed, portal volume, portal talent, broad reconstruction, and coach/QB interactions do not improve held-season opening, movement, or result prediction. A small kickoff change does not survive as a stable general rule.
- Incoming and outgoing portal counts and talent
- Returning PPA
- Season-relative reconstruction rankings
- Interactions with new coaches and low-experience quarterbacks
Leakage and validation safeguards
- Existing transfer talent and returning production in the team rating are accounted for
- Incomplete historical roster denominators are not converted into replacement percentages
- Leave-one-season-out validation
- Directional effects and variance effects tested separately
What we learned
- Broad portal activity is largely absorbed by existing annual team information.
- High reconstruction may increase uncertainty without giving a reliable direction.
- Specific position or unit interactions remain more promising than a generic churn score.
What this does not prove
- Historical roster endpoint coverage is inconsistent.
- The wider-result relationship is not stable enough for a calibrated range adjustment.
How uses this finding
Roster reconstruction is excluded from the displayed Factor sum. It may appear as an evaluated condition with a link to this negative result.
Historical data record
Revalidated August 31 under market-line contract 2026-08-31.2. The downstream audit uses the validated game-flow baseline and does not subtract the unsupported broad Week 1 QB model. All six endpoint families retain separate outputs.
Audited game-level table: being regenerated under the current market-line contract