Keeping the Core Together Is Overrated: Adaptive Capacity in Sports and Business

In an offseason assessment of a successful sports team, “they kept the core together” is considered good news. The players know one another. They have established roles and experience solving problems together. The expectation is that those relationships will help them perform again next season.

That expectation contains a question worth examining: how much of a team's strength depends on keeping the same people, and how much can be sustained through new members?

(Alongside my writing here at The Autogenic Realist, I founded DataBaller, a sports analytics company, in 2026. Both endeavors reflect my interest in viability and vitality: how systems sustain their ability to function, adapt, and develop under real conditions. Our sports research offers a concrete starting point for thinking about what a team carries through change.)

Testing the Continuity Bonus

In basketball, building forecasts from estimated player contributions reduced probability forecast error by about 2.4% across four development seasons, compared with carrying forward the previous season's team rating. The player-based approach won three of those four season comparisons. We estimated contributions from players' game histories and weighted them by expected playing time, using each team's previous game to estimate its next lineup.

That result followed an earlier investigation into roster retention. Across five seasons each of basketball, baseball, and football, we tested whether returning participation could tell us how much of last season's team rating to carry forward. We measured minutes in basketball, batting and pitching opportunities in baseball, and attempts in football.

Using the returning share to adjust the forecasts increased prediction error by 0.7% to 5.4%. Predictions were more accurate without that adjustment. Our research summary describes the investigations.

Those results initially left a gap. A better player-based forecast could still benefit from information about continuity. We therefore ran a follow-up that added four measures to the player-based model: a discount on contributions earned with another team, the share of expected minutes supplied by returning players, players' tenure with the team, and the shared playing history of pairs of teammates.

Before running the tests, we set the same requirement for each addition: it had to improve early-season forecasts in a majority of the four development seasons and in the final season evaluated. Across 1,161 early-season games, none of the four passed. Adding all four together also made forecasts worse in both the development seasons combined and the final season.

The proposed discount on contributions earned elsewhere illustrates why we required both comparisons. It improved forecasts in three development seasons, but worsened them in the final season. The apparent benefit did not hold up. Our evidence supported keeping the player's estimated contribution without that additional discount.

We can now make a more direct claim: once estimated player contributions were accounted for, none of these four continuity measures improved early-season forecasts consistently enough to meet the requirement we had set. That is evidence against assigning an automatic bonus to keeping a core together in this setting.

Sustaining Viability Through Change

In Autogenic Realism, viability is a system's capacity to maintain function, adapt to constraints, and develop coherence across changing conditions. One dimension of viability is adaptive capacity: the ability to reorganize while preserving core functions. This is the sense in which a system persists through the capabilities it sustains. Sustaining those capabilities may require changing its parts.

A team can sustain a capability through someone who has just arrived. In our player-based model, replacing someone with an equally contributing player leaves the estimated team strength unchanged. The follow-up tests examined whether a player's move, the lineup's retention, or its accumulated experience justified adjusting that estimate. None earned a place in the early-season forecast.

The strongest case for continuity concerns the ability to work together. A player may know when a teammate will move or how they will respond when a play breaks down. Our pairs measure addressed that argument by estimating how much members of the expected lineup had played together. It also failed the early-season requirement. Shared history did show a small improvement in a separate in-season analysis, so the finding belongs to the early-season setting we tested. The measure approximates shared experience; it does not observe every act of coordination.

For Autogenic Realism, the implication is that continuity of capability deserves its own assessment. Membership can change while the contributions needed for effective functioning remain available. Adaptive capacity concerns how the system makes those contributions work together through the change. The sports evidence supports taking that possibility seriously without assuming that the familiar lineup deserves extra credit.

The position and the capability

For a business reader, consider a fully staffed team last year and a fully staffed team this year, after replacements have completed onboarding. This is the comparison between complete rosters after training camp. The question is whether the team has sustained its capabilities across that transition.

Comparable individual skills make that possible, but completing onboarding does not establish that it has happened. New members need to learn the plays: how decisions are made, how the systems work, and how responsibilities fit together. Existing members adjust their roles. The organization has to assess whether those changes have produced an effective working team.

Suppose the team maintains a critical service. It can examine whether it still diagnoses failures reliably, makes sound technical decisions, and delivers changes without creating new problems. If those capabilities hold across comparable demands, the organization has evidence of functional continuity despite changes in membership. If they deteriorate, filled positions and completed training do not explain away the loss. The organization needs to identify which knowledge, coordination, or support remains missing.

This is the business application to draw from the framework. Adaptive capacity includes the work that makes new members effective together. Its costs belong in the assessment: time spent teaching, attention diverted from other work, and the demands placed on existing staff. Viability concerns whether the organization can sustain that process as well as the performance that follows it.

When a team has kept its core together, understand which capabilities that continuity protects. When membership has changed, understand how those capabilities have been sustained or developed. The evidence to look for in either case is the team's ability to do its work under the conditions it now faces.

Research note

The basketball comparison covered 1,161 early-season games across five seasons. In the four development seasons, the player-based model won three comparisons and reduced pooled Brier error from 0.22603 to 0.22058, a 2.4% reduction. Brier score measures the squared difference between forecast probabilities and outcomes; lower is better. The improvement was uneven: one season's win was very small and one season was a loss. These outputs do not establish statistical significance.

In the final season, 2025–26, error fell from 0.21815 to 0.19485 across 232 games, a 10.7% reduction. The underlying model was selected after that season's results had been consulted, so it was not an untouched test throughout development. The larger result should be read with that limitation. Neither percentage describes an increase in the number of games predicted correctly.

The follow-up continuity tests were registered before their run on 13 September 2026. Each addition had to improve Brier error in at least three of four development seasons and in 2025–26. The mover discount won three development seasons but lost on the final season; returning share won none despite a small final-season gain; tenure won two and lost on the final season; shared history won one and lost on the final season. All four together worsened pooled development and final-season scores. Failure to meet this requirement establishes no reliable improvement from these tested additions, rather than an exactly zero effect of continuity. The same historical seasons had already informed model development; registering this follow-up did not make them new, untouched data.

The original historical test estimated lineups from each team's previous game, including the previous season's final game at opening night. The follow-up approximated opening-night rosters using players who appeared in each team's first three games. That deliberately uses later information and is a diagnostic approximation, not a historically available opening-night forecast. It did not improve aggregate early-season scores. Against this revised base, Brier error was 0.22061 in development and 0.19537 in the final season before adding continuity terms. Tenure was reconstructed from consecutive prior seasons in game logs, capped at five. Shared history was approximated from pairs' appearances and minutes in the same games, without observing their exact time together on court.

In a separate in-season analysis, shared history reduced error from 0.21561 to 0.21507 across development seasons and from 0.20486 to 0.20407 in the final season, improvements of about 0.25% and 0.39%, respectively. That analysis had no predeclared acceptance requirement and remains exploratory. A small descriptive sample of 31 low-continuity opening games also showed a larger model-error gap to the betting market than the corresponding high-continuity sample; the same pattern was absent in the following one-to-five-game grouping. This does not establish a causal adjustment period or show that coordination is complete by game five.

The follow-up is recorded in the registered plan and run transcript. These are working source references for the draft. The organizational applications remain my interpretation; the analysis did not directly measure onboarding or business viability.