14. Accumulation
The accumulation pipeline integrates refined work results into on-chain state
(GP §12). It proceeds in three stages: accseq orchestrates sequentially,
accpar parallelizes across services, and accone handles a single service
via JAVM execution with 29 host-call dispatch entries (REPLY at slot 0,
28 protocol capabilities at slots 1–28).
In jar1, accumulation runs through the capability kernel. The kernel's
runKernel function executes service code until a protocol cap is invoked,
at which point control returns to the host. The host handles the protocol
operation (storage read/write, transfer, etc.) and calls resumeProtocolCall
to continue execution. Host-call numbering in jar1 is 1-28 (protocol cap slots),
not 0-27 as in gp072. See the Capability Kernel chapter for the execution model.