Digital Team Lead: Can Developer Effectiveness Be Measured from Code?

Ivan Gel presents UpCore, built inside DEX for 150 developers. It converts a completed task and merge request into estimated hours using language, complexity, tests, and legacy constraints, compares that estimate with reported time, and uses roughly three months of history to infer a trend.

The team dropped LLM estimation because outputs varied and now uses a deterministic model. Ivan claims 80% agreement with experts and 85% in one schedule comparison, but provides no independent sample or confidence interval. The result is a signal for joint review, not proven judgment about a person.

UpCore accounts for architecture, learning, reopened tasks, requirements, and AI, yet code is only one output. A lead also hires and develops people, while a strong engineer may prevent an unnecessary change. Bottom-up signals do not show business value, so they must connect to product goals.

Any visible metric invites gaming through smaller changes, complexity, or convenient tasks. Many factors make this harder but do not remove managerial accountability. The dashboard can prompt questions about a trend; one score should not determine pay or dismissal. In Moneyball, statistics complement context, trust, and a way to challenge an inference.