Sequences

Each sequence lives in its own folder under specs/, written in our own words. Every spec has the same parts:

File	What it is
spec.md	The sequence in prose: interface, parameters, behavior, edge instants, validation
task.json	The task Muse Spark receives
reference.py	An independent model of the behavior (stdlib Python, no engine code) that self-checks against hand-worked checkpoints and writes the vectors
vectors.json	The test scenarios a draft must pass. Generated, never hand-edited; CI regenerates them byte for byte
graph.json	A hand-built reference block graph

Validated before any model sees it

A spec is ready only when its hand-built graph passes every scenario on the pinned engine and each of a list of deliberate mistakes fails at least one scenario. That second check is what makes the vectors strong enough to catch wrong drafts.

Sequence	Runs on	Scenarios	Reference graph	Mistakes caught
Chilled-water plant reset	BOPTEST office, plant	10	33 blocks	12
Air handler chilled-water requests	BOPTEST office, one per air handler	15	18 blocks	22
Duct static pressure reset	BOPTEST office, one per floor	6	20 blocks	11
Zone static pressure requests	BOPTEST office, one per zone	9	23 blocks	10
Home price response	BOPTEST home heat pump	22	33 blocks	21

The four office sequences chain together: zone requests feed each floor’s static pressure reset, and air handler requests feed the plant reset.

Your own sequence can go through the same gate from the workbench, with scenarios you accept instead of a reference model.