Our research north star
The principles, layers, and boundaries that guide how DevPlay turns research into evidence.
Building a longitudinal understanding of human development through sport
Children develop continuously, but most of that development disappears — captured only in fragments across training sessions, notebooks, and memory. DevPlay Lab exists to turn repeated sessions into a trustworthy longitudinal history of a child in sport: who the same physical child is, what they are doing, and how they are developing over time.
That history must never be built by confusing children, by allowing weak evidence to become fact, or by pretending a tracker label is the same as a verified identity. Every claim must survive explicit review.
Identity & Physical Continuity
Who is the same physical child?
Establish safe continuity of the same player through training, including dense and ambiguous situations. Later layers depend on this foundation.
Football Semantics
What is that child doing?
Recognise meaningful football actions and training context around each identified player. Layer 2 does not replace Layer 1; it builds on it.
Development Intelligence
How is that child developing over time?
Transform longitudinal player history into meaningful development signals for children, parents, and coaches. Only as strong as the identity and semantics beneath it.
Each layer becomes a foundation for the next while continuing to mature as new evidence arrives.
UNKNOWN > wrong child
A missing or uncertain identity is preferable to an incorrect one. Conservative rejection protects every downstream conclusion.
Tracker ID ≠ physical identity
A tracking label is a hypothesis, not a fact. Physical continuity must be established by evidence, not assumed from a number.
Small diagnostics teach mechanism
Focused experiments reveal why something works or fails. Broad claims without mechanism do not advance the research.
Proposal ≠ acceptance
A method may propose a possible identity association without being trusted to accept that association as true. Candidate discovery and identity acceptance are separate responsibilities.
Correct ownership > complete coverage
Assigning less data to the correct child is better than assigning more data to the wrong child. Missing observations are acceptable; identity contamination is not.
Evidence before claims
Interpretation and strategic direction must be grounded in recorded observations, not inferred from convenience or expectation.
Research Question
A precise, bounded question about identity, football semantics, or development intelligence.
Diagnostic
A small, repeatable test designed to expose mechanism and failure boundaries.
Qualification
Evidence that a method performs safely under defined conditions; not a proof of universal fitness.
Population Benchmark
Controlled comparison across enough sessions and conditions to understand variability.
Decision
An explicit research decision — incorporated, rejected, parked, or marked for follow-up — recorded with provenance.
Methods progress only when evidence supports the next confidence tier. A diagnostic result is not a qualification; a qualification is not a population benchmark; a benchmark is not a decision.
What this programme does not do
- It does not infer the identity of a child from a single camera frame, tracker ID, or appearance cue alone.
- It does not promote a method, model, or signal to operational use without qualification evidence and reviewed decision.
- It does not treat an experiment proposal, GitHub push, or preliminary result as accepted scientific fact.
- It does not replace human research-lead review with automated classification, scoring, or strategic direction.
- It does not publish, share, or retain individual child data beyond the explicit research purpose and consent boundaries.