A keeper is made over a decade. The engine sees who made him.
It isolates the goalkeeping-development residual across the three keeper clusters and reads it patiently across a career.
The manager does not develop these clusters.
The goalkeeper sits inside the same rating scale as everyone else, but he is scored on three clusters no outfield player has, and developing them is a specialist craft. So the goalkeeping-development residual is pulled entirely out of the manager and given to the named coach who does the work.
Everyone else in the market rates the club's goalkeeping setup or hands the credit vaguely to the manager. But a keeper's shot-stopping, command of the box, and footwork are developed by one person doing a distinct job, and that is where they are attributed. The manager keeps a bounded share for hiring the coach; the coach keeps the development he produced. A goalkeeper is its own position, so his development has its own author.
Illustrative engine read on the real goalkeeping structure (the keeper scored on the three goalkeeping clusters inside the same machinery, the goalkeeping-development residual pulled from the manager to the named coach). Composite reads, demonstration figures.
Shot-stopping, commanding the box, playing out are different jobs.
A modern goalkeeper is not one skill. He stops shots, dominates his area, and plays out from the back, genuinely different crafts, so the coach's residual is read on all three separately. A coach who builds a shot-stopper who cannot pass and one who builds a sweeper who flaps at crosses are different coaches, and one number would hide it.
This composite coach reliably builds shot-stoppers who can play out from the back but adds almost nothing to command of the box. That is a precise, useful read: a keeper who needs to grow into crosses should go elsewhere, and a club chasing a sweeper-keeper has found the right developer. A single number would call all three the same. The keeper's three jobs are read as three, because a coach is rarely equally good at all of them.
Illustrative engine read on the real goalkeeping-development residual (scoped to the three goalkeeping clusters Shot-Stopping, Command & Aerial Control, and Sweeping & Distribution, read separately). Composite coach, demonstration figures.
Keepers age slowest, so their coaches show up slowest.
Goalkeepers are the longest-career profile in the whole model: they peak late and decline gently, and a good one plays into his late thirties. That changes how the coach's fingerprint is read. Where a set-piece coach's signal is noisy, a goalkeeping coach's is slow and deep, accruing over many seasons of a single keeper.
The long career is a gift to the read: a coach who takes a raw teenager and is still adding to him a decade later leaves a deep, unmistakable fingerprint, allowed to accrue rather than judged on one season. Like every portable actor, it must reproduce at another club to be his. The most patient craft on the pitch gets the most patient read, and its author gets the credit.
Illustrative engine read on the real goalkeeping confidence shape (the longest-career, lowest-physical-dependence profile, a fingerprint accruing over many seasons), portability validation, the manager credit-split, and the no-KR rule. Composite reads, demonstration figures.
Goalkeeping shows up over a career, and the engine is built to wait for it.
Give the keeper his own coach, and the credit its author.
The Goalkeeping Coach read isolates the three goalkeeping clusters, reads the modern keeper's three jobs apart, and lets a long-career fingerprint accrue patiently, never touching a rating.