Uncooked effectivity numbers can inform us what occurred. They’re much less dependable at telling us what that efficiency means.
A five-yard acquire on third-and-4 is completely different from a five-yard acquire on third-and-12. A powerful effectivity quantity in opposition to a troublesome schedule is completely different from the identical quantity in opposition to a set of struggling opponents. We all know these distinctions are essential, however accounting for each of them in a single constant score isn’t simple.
That’s the drawback CORE is designed to deal with.
CORE stands for Context & Opponent-Relative Effectivity. It’s a new CollegeFootballData workforce score derived from play-level Predicted Factors Added (PPA). It measures how a lot worth a workforce creates on offense and permits on protection after accounting for the conditions through which its performs occurred and the opponents it confronted.
CORE scores are already out there via the CFBD API, the general public methodology web page, and the official Python shopper. This submit explains what the numbers imply, how the 2 levels of the score match collectively, and the place CORE might be helpful.
Studying the 2025 CORE chart
The characteristic chart exhibits the ultimate CORE scores for the 2025 season. Offensive score runs from left to proper, with stronger offenses farther proper. Defensive score runs from high to backside, however the route deserves consideration: decrease defensive values are higher as a result of protection measures factors allowed above common. That places the strongest defenses close to the highest of the chart.
The dashed traces mark the rated-team common on either side of the ball. Groups within the upper-right mixed above-average offense with above-average protection. Groups farther proper however nearer to the underside have been extra offense-driven. Groups close to the highest however nearer to the middle leaned extra closely on protection.
That profile is among the predominant causes I wished CORE to publish separate offensive and defensive elements. A single total quantity is helpful for sorting, however it could actually cover how two groups arrived at related outcomes.
What CORE is measuring
CORE asks two associated questions:
How a lot worth did a play create relative to what we’d usually anticipate from that scenario?How a lot of a workforce’s accrued efficiency stays after accounting for the offenses and defenses it performed?
These questions are dealt with in separate levels. The primary stage adjusts for context whereas the second adjusts for opponent energy.
The separation is essential as a result of recreation scenario tells us what a play was being requested to perform whereas opponent adjustment tells us how troublesome the schedule made these accomplishments. Treating them as one imprecise adjustment would make the ultimate quantity tougher to examine and clarify.
Stage one: evaluate every play with its context
The start line is PPA, which estimates how a lot a play modified the offense’s anticipated factors. PPA is extra informative than uncooked yards as a result of the worth of a outcome relies upon closely on the place and when it occurs.
CORE goes one step additional by estimating the PPA anticipated from the scenario earlier than evaluating that expectation with the play’s precise worth. The context consists of:
interval and time remaining;the rating margin earlier than the play;down and distance;discipline place;residence, away, or neutral-site context; andwhether the play was a cross or rush.
Workforce and opponent identification should not a part of this situational expectation. The mannequin doesn’t start by assuming {that a} specific college ought to produce extra worth. It first asks what usually occurs in that soccer scenario.
The distinction between precise and anticipated worth turns into the play’s context-relative outcome:
context-relative worth = precise PPA – anticipated PPA from context
Suppose a qualifying play produces 0.8 PPA in a scenario the place the mannequin anticipated 0.2. Its context-relative worth can be +0.6 PPA. If a play produces -0.3 when the expectation was +0.1, its worth relative to context can be -0.4.
These are illustrative numbers, however the interpretation is the essential half. Optimistic values imply the offense created extra worth than the scenario sometimes produces. Damaging values imply it created much less.
That is nonetheless an outline of the play, not a causal declare a few quarterback, coordinator, or particular person resolution. Context adjustment makes the comparability extra helpful; it doesn’t clarify each purpose the play succeeded or failed.
Which performs rely
A score is partly outlined by the info it leaves out. CORE makes use of qualifying cross and rush performs from accomplished FBS-versus-FBS video games. It excludes extra time, rubbish time, FBS-versus-FCS video games, particular groups, non-scrimmage performs, and performs that do not need the context required for a legitimate comparability.
These are intentional boundaries. Particular groups and extra time introduce completely different conditions than the regulation offense-versus-defense query CORE is designed to reply. FBS-versus-FCS video games are excluded so the opponent adjustment operates throughout the meant workforce inhabitants. Rubbish-time performs are excluded as a result of rating and recreation incentives could make them behave in a different way from aggressive conditions.
The identical eligibility guidelines are utilized constantly inside a score snapshot. Consistency is extra essential than squeezing each doable play into the pattern.
Stage two: alter offense and protection throughout the schedule
After calculating context-relative worth, CORE seems to be on the linked community of offenses and defenses throughout the season. Each qualifying play hyperlinks one offense with one opposing protection.
The adjustment estimates offensive and defensive results collectively. In plain language, it asks how a lot worth an offense tends to create whereas accounting for the defenses it confronted, and the way a lot worth a protection tends to permit whereas accounting for the offenses it confronted.
That is the place a uncooked workforce common can change meaningfully. An offense could have a powerful unadjusted quantity after dealing with a number of weak defenses. One other could look merely good regardless of taking part in a way more troublesome group. The opponent-relative stage makes use of the total schedule community to differentiate these circumstances.
The adjustment additionally makes use of regularization, which pulls excessive estimates towards the center when the out there schedule doesn’t strongly help them. That’s particularly helpful early in a season, when groups have performed fewer opponents and the schedule community isn’t nicely linked.
Regularization improves stability, nevertheless it doesn’t remove early-season uncertainty. A four-game pattern remains to be a four-game pattern. CORE ought to grow to be extra informative because the season develops and extra qualifying performs join the groups.
Find out how to learn offense, protection, and total
CORE publishes three predominant values:
Offense is factors created above the rated-team common per 100 qualifying offensive performs. Increased is best.Protection is factors allowed above the rated-team common per 100 qualifying defensive performs. Decrease is best.Total is offense minus protection. Increased is best.
If a workforce has an offensive score of +8, its offense created eight factors above the rated-team common per 100 qualifying offensive performs. If its defensive score is -5, its protection allowed 5 factors fewer than common per 100 qualifying defensive performs. Its total score can be:
total = offense – protection
total = 8 – (-5) = 13
The defensive signal is the best half to misinterpret. Damaging protection is nice as a result of fewer factors allowed is nice. CORE doesn’t flip the defensive quantity merely to make each part level in the identical route. This retains issues in line with different metrics discovered on the CFBD platform.
Total is a composite of the 2 sides. It represents the anticipated web benefit throughout 100 qualifying offensive performs and 100 qualifying defensive performs, not a proportion and never one noticed 100-play charge.
The API additionally returns offensive and defensive play counts, the inclusive season-type and week boundary, and the mannequin model.
What CORE is nice for
I feel CORE is most helpful as an analytical start line moderately than a last reply.
Evaluating groups throughout uneven schedules
School soccer schedules should not balanced. CORE gives a approach to evaluate workforce effectivity after contemplating each play context and the standard of opposing models. That makes it a helpful various to a uncooked PPA common or scoring margin when the schedules are materially completely different.
Separating offense-led and defense-led groups
Two groups can have related total scores with very completely different profiles. The offensive and defensive elements make these variations seen, which is helpful for workforce pages, previews, charts, and exploratory evaluation.
Discovering disagreements value investigating
A big hole between CORE and one other measure isn’t mechanically proof that certainly one of them is incorrect. It’s a helpful immediate. The distinction could come from opponent combine, recreation context, excluded performs, pattern measurement, or the truth that the 2 scores reply completely different questions.
These disagreements usually make higher evaluation topics than the highest of a sorted desk. If a workforce seems to be a lot stronger in CORE than it does in scoring margin or an unadjusted effectivity metric, the following step is to examine its schedule and offensive-defensive cut up.
Constructing downstream instruments
As a result of CORE is on the market via the API, it may be utilized in notebooks, dashboards, workforce comparisons, and different purposes. I’d nonetheless evaluate CORE with a minimum of one impartial measure earlier than drawing a powerful conclusion. No single score captures each a part of workforce high quality.
What CORE doesn’t do
CORE is an effectivity score. It’s not a forecast, level unfold, win chance, résumé rating, or betting system. It doesn’t embrace particular groups, and it doesn’t present participant scores, matchup projections, or pass-rush splits.
It additionally doesn’t set up why a workforce carried out nicely. A constructive offensive score could mirror execution, expertise, scheme, teaching, well being, or a number of different elements. CORE measures the team-level outcome over qualifying performs; it doesn’t determine the causal contribution of every issue.
Public retrospective scores start with the 2016 season. These historic values are recalculations utilizing the launched CORE methodology.
Lastly, CORE doesn’t publish a built-in rank, percentile, grade, or confidence interval. You’ll be able to kind total values to create a rating, however that presentation selection shouldn’t be confused with a further output from the mannequin.
Getting CORE from the CFBD API
The endpoint is GET /scores/core. At the very least certainly one of 12 months or workforce is required, and an non-compulsory convention filter can also be out there.
Utilizing the official Python shopper, a minimal request for the ultimate 2025 scores seems to be like this:
import os
import cfbd
api_key = os.environ.get(“CFBD_API_KEY”)
if not api_key:
increase RuntimeError(“Set CFBD_API_KEY earlier than operating this instance.”)
configuration = cfbd.Configuration(
access_token=api_key,
)
with cfbd.ApiClient(configuration) as api_client:
ratings_api = cfbd.RatingsApi(api_client)
scores = ratings_api.get_core(12 months=2025)
for row in sorted(scores, key=lambda worth: worth.total, reverse=True)[:10]:
print(
row.workforce,
row.total,
row.offense,
row.protection,
row.model_version,
)
The instance checks for the API key earlier than making a request.
You’ll be able to assessment the whole discipline definitions within the API reference or use the generated instance within the Python shopper documentation. CORE can also be out there on the official TypeScript and .NET shoppers. In the event you want a key, begin at CollegeFootballData.com.
The place CORE goes from right here
The ultimate 2025 scores present a helpful completed-season instance, however the extra attention-grabbing take a look at begins with the 2026 season. As soon as the primary public snapshot is on the market, I plan to make use of CORE to look at which groups look meaningfully completely different after accounting for scenario and opponent energy, and which groups are being carried extra closely by one facet of the ball.
For now, I’d encourage you to retrieve the 2025 scores and evaluate one workforce with a uncooked effectivity measure, scoring margin, or one other score you already belief. If CORE adjustments the way in which you see that workforce, examine the schedule and the offensive-defensive cut up earlier than deciding why.
