Weak Link — workshop

Crank an AI conveyor as fast as you like — score only counts what ships out the far end of three fixed-pace human stations, so overrunning the bottleneck just piles up work-in-progress until something jams.

One control: a crank on an AI conveyor with no speed limit. It feeds three fixed-pace stations in series — Review, Test, Release — each draining its inbox at its own rate no matter how fast parts arrive. Crank past the slowest station (Test, by default) and the surplus doesn’t vanish: it piles up as work-in-progress in that station’s inbox, and a pile that outgrows its capacity jams the station for a few seconds, backing up everything behind it too. The score only counts crates that make it all the way through — produced and scrapped are shown, but they don’t count toward the grade. The skill is finding the crank position that matches your slowest station, not maxing it out.

Riffs on a recent NBER finding that AI lifts commit volume roughly 180% but release throughput only about 30%, because the humans downstream (review, test, deploy discipline) are the actual constraint — cranking the front of the line doesn’t move the bottleneck, it just buries it.

Reuse

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

createWeakLinkSim(opts) — returns a state object; call update(dt) each frame and setCrank(v) (0..1) to set the conveyor’s speed. There’s no drag/input handling in the module — that’s the demo’s canvas + pointer code, reference only.
opts: conveyorMaxRate, stationRates (array of 3, Review/Test/ Release), bufferCapacity, jamDuration, warnThreshold, duration (0 for an endless run).
State exposes buffers (3, each with wip, warning, jamFlash), stations (3, each with name, rate, jamTimer), crank, totalProduced, totalShipped, totalScrapped, and status ('playing' | 'finished').

Gotchas

Overflow at any buffer both scraps the excess WIP and jams the station right behind that buffer for jamDuration — a jam near the front cascades into every buffer downstream of it filling too, since nothing is draining through the jammed station. That cascade is the point, not a bug to tune away.
Default station rates (2.2, 1.4, 2.0) make Test the bottleneck; a crank held steady at or below 1.4 / conveyorMaxRate (~18%) never overflows anything. Change which rate is lowest to move the bottleneck around.
The demo bundles its own copy of the module (self-contained by contract). If you touch src/, re-copy it into demo/.