The cognitive workbench needs to render goals at different sizes and positions based on what kind of goal they are. A standing life orientation (“be healthier”) isn’t the same as a concrete next action (“call the doctor”), and the visualizer needs to know which is which without guessing from the horizon field.
We added a goal-tier property — THEMATIC, STRATEGIC, or TACTICAL — derived from the existing horizon. Aspirational goals are thematic (standing orientations). Medium and long-horizon goals are strategic (scoped initiatives). Immediate and short-horizon goals are tactical (concrete actions). The derivation is a GoalTier enum in mindmap-api with a fromHorizon() switch — simple, but it needed to be set consistently at every creation site: experience-derived goals in GoalRecognitionPhase, decomposed sub-goals in GoalResolutionPhase.expand(), and drive-proposed goals in DriveGoalBridgeParticipant.
The interesting design question was progress computation. When a goal decomposes into sub-goals via decomposes-into edges, the visualization wants a progress bar — “3 of 5 sub-goals completed (60%).” But the goal graph is recursive. A strategic goal might decompose into tactical sub-goals, and one of those tactical goals might itself have sub-steps.
I went with recursive averaging rather than flat counting. A leaf goal’s effective progress is 1.0 if completed, 0.0 if not. A parent’s progress is the mean of its children’s effective progress. This propagates smoothly upward — if a mid-level goal has two children, one completed and one half-done, it reports 0.75, and its parent sees that fractional progress rather than counting it as zero because it isn’t “completed” yet.
The computation runs as the last step of GoalResolutionPhase.run(), after revise and sync have settled all status changes. It reads a fresh node snapshot, builds a parent→children map from decomposes-into edges, and walks the graph bottom-up with memoization. A visited set per recursion path guards against cycles — unlikely since detectAndBreakCycles runs earlier, but defensive code is free in a consolidation phase that runs once per tick.
This is two of the five Batch 1 issues for the cognitive workbench epic. The remaining piece before visualization can start is the explicit tier for thematic/strategic/tactical presentation — now landed, the UI team has what it needs.