Merge Queue

PRs drift toward the merge gate on their own; the only input is a one-slot holding bay. Pull the current front PR aside to let the ones behind it merge first, then release it once its area has had time to cool down — the same area merging twice inside a short window is a conflict. The bay only ever holds one, so the real skill is reading which PR is actually risky right now versus which one just happens to be first in line, because a second risky PR right behind the held one has nowhere to go.

Timed to a report making the rounds this week: 96%+ of engineering teams now use AI coding tools, but review throughput hasn’t kept pace — which is exactly this shape, a widening gap between how fast changes arrive and how carefully they can be let through.

Reuse

The mechanic lives in src/merge-queue.js as a framework-free ES module:

• createMergeQueue(opts) — returns a state object; call update(dt) each frame and hold() on input. hold() is a single toggle: with nothing held it pulls the current frontmost PR into the bay; with one held it releases it back into the lane at whatever position it was frozen at.
• opts: areas (default four labels), laneSpeed, spawnIntervalBase/ spawnIntervalMin/spawnDecay (spawn rate ramps up over a run), conflictWindow (how soon the same area can safely merge again), maxConflicts (0 disables the fail state), duration (0 = endless).
• State exposes prs (each { id, area, x, held }, x 1 at spawn down to 0 at the gate), heldId, lastEvent ({ area, conflict, at }, for a render flash at the gate), score (merged/conflicts/streak/bestStreak), and status ('playing' | 'finished', with reason 'jammed' | 'time-up'). Rendering and input are entirely the caller’s job.

Port notes for DragonRuby: update() and hold() have no DOM dependencies; the lane-to-screen mapping and bay layout belong to the renderer, same split as the demo.

Gotchas

• A PR only merges by reaching x <= 0 while held is false; freezing it mid-lane (rather than removing and reinserting it) means everything behind it keeps closing distance independently — it’s a real siding, not a queue-jump.
• conflictWindow is measured against lastMergeAt[area], which only updates on an actual merge (clean or not) — holding a PR doesn’t touch the clock, only merging does.
• Spawn interval decays geometrically toward spawnIntervalMin every spawn, so density (and therefore conflict pressure) ramps up over a long run even though laneSpeed never changes.
• The demo bundles its own copy of the module (self-contained by contract). If you touch src/, re-copy it into demo/.