APEX ranking components and recovered designs

This is a reference for the ranking page and future model work. It preserves what was actually calculated, records the proposed subcategories, and keeps unresolved choices open. The percentages below are not final APEX weights unless explicitly described as part of an old, reproduced spreadsheet.

November 6 direction — model still in development

APEX should evaluate accomplishments in the season so far to support playoff selection and seeding. Future opponents do not contribute to completed-game SOS or the intended season APEX score. Moneyball's contribution to the new APEX model is capped at 5%; the final weights and missing-finance policy remain undecided. The historical board below still uses its original 15% Moneyball weight and is preserved as an archive, not relabeled as the new model.

SOS is a separate ranking view, initially displaying its top 75. The intended release schedule is weekly alongside APEX, Power and Résumé, with a shared games-through date. Automated weekly calculation/publication is not implemented in this preview.

Availability, replacement quality, actual crowd conditions, travel, time zones and kickoff timing are candidate game-context inputs. Opponent absences affect the opposition actually faced; a team's own absences do not directly lower its SOS. Missing reports are unknown, not evidence of healthy rosters. Fixed subcategory percentages below remain proposals. Fan votes do not feed independent APEX; whether stadium business efficiency enters through Moneyball is unresolved.

Current project structure

The creator confirmed that four separate rankings feed a fifth, combined APEX ranking:

Each input should have a page or expandable breakdown showing its component definitions, actual data, weights and missing-data status. A proposed component must be identified as proposed until its formula and inputs are implemented. Separate a score's inputs from descriptive team statistics displayed alongside it.

The creator recalled Power at approximately 50%, Résumé approximately 25%, and SOS/Moneyball sharing the remainder, but clarified these were where discussion left off, not final percentages. The split and every top-level coefficient remain undecided. The older 50/35/15 and proposed 50/40/10 formulas have only three top-level inputs and must not be silently presented as the confirmed four-input model. A future composite requires defined comparable score scales, weights that sum to 100%, a missing-component policy and an explicit version.

Historical calculation already recovered

The existing historical SOSv2 boards are a different, three-input prototype, not a calculated instance of the newly confirmed four-input structure. They use games through October 31, 2026, and reproduce these formulas:

The SOS values and Résumé/Power components use source proxies and assumptions; they are not finished opponent-adjusted game-level models. The payroll estimates are not audited player contracts. Missing finance values are unknown, not zero. The v2 Moneyball residual percentiles were carried forward from the earlier version rather than refitted after the Résumé change. Historical formulas gives exact equations and references; the later schedule audit reproduces the SOSv2 scenario from the recovered game set while retaining its source and timing limitations.

Recovered proposed subcategories

These tables describe distinct proposals from earlier conversations. The old assistant's wording, screenshot order and version labels do not establish that the user finalized or that software implemented them. Percentages are within the named parent model, unless explicitly marked as percentage points of the parent. They are candidates for display with source/status labels, not a single reconciled formula.

Résumé proposal

The proposal describes accounting for road/neutral difficulty, weather and margin within each game's result rather than automatically adding separate bonuses. Measured game control, adjustment methods, normalization and treatment of unavailable information still need definitions. A different résumé-plus-strength framework and its revised CFP Version 3.0 are preserved in the chat reconstruction; they must not be blended into this table without a new decision.

Predictive Power and matchup proposal

This historical proposal describes both neutral-field Power and a specific matchup forecast. Its last 25 percentage points are game-specific; the neutral-field rescaling and treatment of opponent-specific QB/trench factors remain undefined. The proposed percentages are a hypothesis, not calibrated effects. One separate idea proposed a 4% resource prior; the table already totals 100%, so that prior requires an explicit reallocation rather than being appended.

QB versus pass-defense sub-index

The proposal compares performance with the opposing defense and player quality; equal passing-yard totals are not necessarily equal performances. Its scoring scale, pressure/depth and turnover-worthy definitions, replacement policy and reliable national coverage are unfinished. NCAA passer efficiency, NFL passer rating and ESPN QBR are different measurements. The recovered 8% QB row and this six-part table are compatible structural evidence, not proof of an implemented integrated formula.

Venue and environment alternatives

One proposed 13-point allocation within the full matchup model lists: home/away baseline 2; crowd/noise/attendance 2; weather/climate adaptation 2.5; travel/geography 1.5; neutral-site imbalance 1.5; kickoff/body clock 1.5; altitude/surface familiarity 1; and rest/acclimatization 1. These add to 13 percentage points of that model.

Two other screenshots propose different 13-point allocations: 4/3/2.5/2/1.5 for venue, weather, attendance, travel/rest and altitude/familiarity; and 4/4/2/1.5/1.5 for venue, weather, travel/rest, altitude and familiarity. They are alternatives with unresolved chronology, not extra percentages to add together. Home field, crowd, travel, climate and acclimatization can interact; the 13% allocation is not evidence of a measured home-field effect.

Moneyball alternatives

One six-category proposal allocates results above financial expectations 30%, roster spending efficiency 20%, recruiting/player development 20%, coaching/staff efficiency 10%, fan/stadium efficiency 10%, and roster retention/sustained success 10% within Moneyball.

A separate five-category proposal allocates results above financial expectations 40%, opponent-adjusted quality of wins 20%, development/retention 15%, coaching/staff spending efficiency 15%, and sustained performance/consistency 10%. Neither proposal establishes finished formulas or changes the historical 60/25/15 proxy.

Possible inputs include estimated roster payroll, NIL and revenue sharing, coach and staff compensation, recruiting and transfer resources, development, retention, stadium capacity and actual attendance. Distinguish resources available from outcomes produced; enrollment and tuition are only institutional context. Avoid presuming a school's recruiting strength and spending are independent measures of the same resource.

SOS components

The intent is a separate ranking of schedule difficulty independent of whether the team won those games. Potential ingredients include opponent strength at the relevant time, FBS and FCS opponents, home/away/neutral venue and other documented game circumstances. The historical SOSv2 calculation is a specific provisional Elo scenario; its FCS 1,200 is a placeholder, not measured FCS strength. Opponent-rating provenance, full FCS data, venue effects and historical availability require further work. The imported Team Receipts SOS is a separate source/method whose scope is not fully reproduced.

How to handle overlapping inputs

SOS appeared inside the proposed and historical Résumé models, while the current design also makes SOS a separate APEX input. Likewise, historical Moneyball already includes Résumé and Power, and QB, efficiency, venue and game control can overlap. This does not decide that overlap is forbidden or automatically erroneous: it means direct and indirect influence must be made visible before selecting weights. A dependency graph and effective-contribution report should make any deliberate repeated influence reviewable. Do not silently remove SOS from Résumé, double it, append weights that already total 100%, or equate an internal 15% with 15% of APEX.

AP and official CFP rankings were also discussed as a separate optional poll-informed comparison. Their historical sandbox percentages are not adopted as part of independent APEX; poll availability, missing teams and potential overlap need policies before a comparison is calculated.

Data currently available and open work

• The frozen schedule covers 622 completed games involving FBS teams and 138 FBS programs; an additional November 5 scoreboard capture supplies five later finals for descriptive records but does not recalculate the historic boards.

• The national team sheet contains records and basic team/opponent box totals and rates for all 138 teams in the frozen game set. Coverage is not a guarantee of correctness; retained source-row flags and later official-source conflicts require review. These are descriptive box stats, not necessarily opponent-adjusted features.

• Player-game observations and play-by-play sources have been preserved or audited, but complete player identity/participation, game-control and opponent-adjustment definitions, EP/WP artifacts and source conflicts remain open. QB pressure, air depth and turnover-worthy information need demonstrated coverage.

• National game-specific injury and replacement data, kickoff weather, attendance/crowd composition, travel/acclimatization and comparable financial/recruiting/coaching/development data remain incomplete. The old payroll estimates cover 69 teams. Missing reports are not proof of healthy rosters or zero spending.

• Score reliability, original data-publication timing and historically available pregame information need further verification. Résumé's judgment of earned results must be evaluated for consistency; predictive performance can be tested against later game outcomes. Neither the presence of a weight table nor a replay alone establishes a validated model.

For full metric definitions and source routes, see the team-stat requirements, model map, screenshot evidence and chat reconstruction.