docs/kernel_import_plan.md

Kernel Import Plan

Clean product one-liner: import the kernel in 6 slices, not as a 5.6G blob.

Layman version: move the engine, gearbox, and gauges first; leave the spare tires and posters behind.

How many slices?

Best read: 6 kernel slices total for the first honest standalone runtime.

Slice	Status	Why	Real-world analog
receipt schema	imported	stable proof format	receipt template
receipt model	imported	effect vocabulary + hashing	logbook
UTIR vocabulary	imported	bounded operation language	work-order language
guarded executor	imported	sandbox and allowlist logic	safety interlock
graph core	imported	graph/ops packet execution surface	engine block
module spine	imported, partial	shows upstream kernel surface and what is still missing	wiring diagram

Can we work more aggressively?

Yes, but the aggressive version should be:

1. import the minimal slices quickly
2. expose them through one local runtime loop
3. benchmark them immediately
4. only then widen the imported surface

Not:

1. vendor the entire upstream tree
2. hope the product boundary appears later

Next two aggressive moves

Move	Why it wins
add policy allowlist over local exec	turns execution from env toggle into real control policy
bind local runtime runner to vendored UTIR vocabulary	closes the gap between current shell runner and imported kernel language

Frontier read

Strategy	Read
big-bang kernel import	loses
tiny placeholder-only repo	loses
6-slice kernel import with immediate runtime use	wins