Shared Models, Not More Feeds
By RAVEN · DSA Labs
Connecting feeds is a project. Building a shared decision model is a discipline.
This month, multiple collegiate programs publicly announced efforts to connect practice tracking, live game analytics, scouting, and video into a single workflow. The announcements were serious, the investments real, and the ambition clear: one place where all the data lives.
What the coverage mostly skipped was the harder question.
Connecting feeds is a project. Building a shared decision model is a discipline. Programs that confuse the two will end up with more data, faster, pointing in more directions at once.
The feeds aren't the problem. The model is.
Imagine a staff running three tools that each do their job well. A practice tracker logs load, speed, and rep quality. A game analytics platform outputs efficiency metrics and positional grades. A scouting module captures film evaluation. Each tool is accurate on its own terms.
Then a coach asks: how much does this player contribute?
The practice tracker gives one answer. The game platform gives another. The scout's grade gives a third. Not because the tools are wrong — but because they were never asked to agree. There's no shared definition of contribution, no common player identity tying the records together, and no prior that lets a model weigh each source consistently.
More feeds didn't create more clarity. They created three parallel reports that now require a human to arbitrate.
Integration success isn't measured by how many feeds are connected. It's measured by whether the staff gets one consistent picture of player value, every time, from any source.
Identity is where integration breaks first.
Before a shared model can work, every feed has to agree on who it's talking about. This sounds obvious. It is not easy.
A player's name is spelled three ways across three platforms. Their position is listed differently in the scouting module than in the game stats export. A mid-season roster update propagated in one system and not the others. None of these are catastrophic failures — they're the ordinary entropy of running multiple tools in a real program.
But when you run an evaluation query across all three feeds and the identity layer is dirty, the model either silently drops records or duplicates them. Either way, the number it returns is wrong, and nobody in the room knows it.
At DSA, we recently completed a player-matching capability that works across multiple video and data upload sources — precisely because this identity problem shows up every time programs try to unify their data. The matching has to be reusable and auditable, not a one-time data-cleaning sprint before the season starts. Without it, every new feed you connect increases the surface area where silent mismatches can accumulate.
Common priors, or: why the same player looks different in every room.
Even after identity is clean, feeds can disagree on value because they're built on different assumptions about what matters.
A practice-load tracker is optimized to flag injury risk and readiness. It weights availability and volume. A game analytics platform weights efficiency and outcome contribution. A scouting module weights ceiling, athleticism, and projection. Each framework is defensible. None of them is wrong.
But if the position coach, the strength staff, and the recruiting coordinator are each reading a different one — and calling it "the data" — they will not agree on the player. They will agree even less when the player is borderline, which is exactly when a consistent model matters most.
A shared decision model means agreeing, before the feeds go live, on what questions the data is supposed to answer. What does contribution mean for this position on this team? How does practice availability weight against game performance? What's the threshold for flagging a discrepancy between film grade and efficiency output? Those aren't technical questions. They're coaching and organizational questions. The technology can only enforce the answers once they exist.
What an integration audit actually looks like.
When a program connects a new feed, the first question shouldn't be "does the data look right?" It should be three questions:
Is every player record in this feed matched to a canonical identity that exists in every other feed?
Is this feed's output being interpreted through the same priors as the other feeds, or does it require a separate mental model to read?
When this feed disagrees with another feed on the same player, what does the staff do — and who decides?
The third question is the one programs skip. They assume the feeds will mostly agree, and they're right — most of the time. The cases where feeds diverge are the interesting ones, and they're also the cases where a player's role, contract, or scholarship decision is actually in play.
The Big Takeaway
Adding another feed is easy. Making that feed answer the same question, about the same player, in the same terms as every other feed — that's the work.
The programs that get the most from unified analytics will be the ones who treat identity matching and shared priors as prerequisites, not afterthoughts.
Programs announcing unified analytics suites this month are making a real commitment. Before the next feed goes live, settle the identity layer and the shared definitions it will be read through.
One player. One ontology. Every feed.
StatLink is built around exactly that structure: a common player ontology and team workflow layer that lets practice, scouting, and game data speak the same language before any evaluation runs.