code wiki / (root) / nx_recall_bench.nx

nx_recall_bench.nx

buildroot/runtime/nx_recall_bench.nx

4777 B92 linesdepth 6pulls 10 transitivereach 0 importersview sourcekind benchtopic recall

docs dependencies structs consts functions

about

nx_recall_bench.nx -- the R1+R2 measured-lift bench AS A SERVICE (evidence-fork organ). Every call RE-RUNS the fusion + rerank benches live on the nx_ir_eval ruler and emits the scorecard JSON -- numbers are recomputed at call time, never baked, so a caller (the /compare C2/C5 evidence-fork, a workstream, a session) can fork the proof. Verdict RED + exit 1 if any lift inverts (liar-killer: a regression cannot hide). nx_recall_bench -> JSON scorecard license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0

dependencies 6 imports · 0 importers

imports: fx.nxnx_itoa_lib.nxnx_ir_eval.nxnx_recall_fuse.nxnx_recall_rerank.nxsyscalls.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

14	const K_MAGIC_32768: i64 = 32768
15	const K_MAGIC_65536: i64 = 65536

functions

17	func rb_puts(s: *u8) -> i64 { var n: i64 = 0; while s[n] != (0 as u8) { n = n + 1 } sys_write(1, s, n); return 0 } called by 2: rb_key main
18	func rb_q() -> i64 { let b: *u8 = sys_mmap(1); b[0] = 34 as u8; sys_write(1, b, 1); return 0 } called by 2: rb_key main
19	func rb_key(name: u8) -> i64 { rb_q(); rb_puts(name); rb_q(); rb_puts(":" as u8); return 0 } called by 1: main calls 2: rb_q rb_puts
24	func rb_num(v: i64) -> i64 { nxi_out(v); return 0 } called by 1: main calls 1: nxi_out
25	func rb_permil(fxv: i64) -> i64 { return (fxv * 1000 + K_MAGIC_32768) / K_MAGIC_65536 } called by 1: main
27	func main() -> i64 calls 10: rf_fuse rf_order_by_score ie_ndcg_at_k rr_semantic_scores rr_order_by_cos rb_puts+4