code wiki / _hdl_build / nx_converge.nx
nx_converge.nx
buildroot/runtime/_hdl_build/nx_converge.nx
about
nx_converge.nx -- CLI/MCP surface for the retention convergence controller.
All control logic lives in nx_converge_lib.nx, which the GATE imports too, so the thing that ships
and the thing that is graded are the same code.
VERBS
plan <target> <tol>
-> the FIRST denoise probe to render, plus the step budget. Costs zero GPU.
next <target> <tol> <lo> <hi> <probe> <measured> <steps_used>
-> the next probe (or DONE) given what the ruler measured for the last render.
STATELESS: the bracket travels in the arguments, so an agent/workflow can drive one
render at a time and the whole search is replayable from its transcript.
simulate <target> <tol>
-> run the controller to completion against the oracle built from REAL measured retention,
printing the full trajectory. This is how you see the step count before spending GPU.
selftest
expect_exit: 0 license_tier: ORIGINAL
dependencies 1 imports · 0 importers
imports: nx_converge_lib.nx
imported by: nobody (leaf or entry point)
call flow from main pre-order; caps 40 nodes / depth 6 declared; ↻ = already shown
structs
| none |
consts
| 18 | const K_MAGIC_1200: i64 = 1200 |
functions
| 20 | func cw(s: *u8) -> i64 { var n: i64 = 0; while s[n] != (0 as u8) { n = n + 1 } sys_write(1, s, n); return 0 } |
| 21 | func cn(v: i64) -> i64 |
| 35 | func c_streq(a: *u8, b: *u8) -> i64 called by 1: main |
| 42 | func c_atoi(s: *u8) -> i64 called by 1: main |
| 56 | func c_status_name(st: i64) -> i64 |
| 64 | func c_simulate(target: i64, tol: i64) -> i64 |
| 104 | func c_selftest() -> i64 |
| 135 | func main(argc: i64, argv: *i64) -> i64 |