nx_research_unified_gate.nx source
↩ module page · 146 lines · 9056 B
1// nx_research_unified_gate.nx -- liar-kill gate for the UNIFIED sovereign researcher (S1+S2+S3+S4 -> one verdict).
2// TIER-1 (deterministic, $0): nxr_assess grades fixture report bodies and shows the VERDICT:
3// PUBLISH (grounded + numbers confirmed), CONTINUE (a fabricated number / uncited claim -> investigate again),
4// STOP-INCOMPLETE (gaps remain but budget spent -> ship honestly). TIER-2 (marker knowledge/index/nxr_model.on):
5// the full nxr_research pipeline with real Qwen. license_tier: ORIGINAL
6import "nx_research_unified.nx"
7import "nx_syscalls.nx"
8
9func st_puts(s: *u8) -> i64 { var n: i64=0; while s[n]!=(0 as u8){n=n+1} sys_write(1,s,n); return 0 }
10func st_putn(v: i64) -> i64 { let bb: *u8=sys_mmap(28); var m: i64=v; if m<0{m=0-m;sys_write(1,"-" as *u8,1)} let t: *u8=sys_mmap(28); var k: i64=0; if m==0{t[0]=48 as u8;k=1} while m>0{t[k]=(48+(m%10)) as u8;m=m/10;k=k+1} var i: i64=0; while i<k{bb[i]=t[k-1-i];i=i+1} sys_write(1,bb,k); return 0 }
11func st_len(s: *u8) -> i64 { var n: i64=0; while s[n]!=(0 as u8){n=n+1} return n }
12func st_verdict(v: i64) -> i64 {
13 if v == 0 { st_puts("PUBLISH" as *u8) }
14 if v == 1 { st_puts("CONTINUE" as *u8) }
15 if v == 2 { st_puts("STOP-INCOMPLETE" as *u8) }
16 return 0
17}
18func st_row(tag: *u8, out: *i64) -> i64 {
19 st_puts(tag)
20 st_puts(" grounded_sents=" as *u8); st_putn(out[0] - out[1])
21 st_puts(" uncited=" as *u8); st_putn(out[1])
22 st_puts(" dangling=" as *u8); st_putn(out[2])
23 st_puts(" numeric=" as *u8); st_putn(out[3])
24 st_puts(" confirmed=" as *u8); st_putn(out[4])
25 st_puts(" discredited=" as *u8); st_putn(out[5])
26 st_puts(" gaps=" as *u8); st_putn(out[6])
27 st_puts(" -> VERDICT=" as *u8); st_verdict(out[8])
28 st_puts("\n" as *u8)
29 return 0
30}
31
32func main() -> i64 {
33 var pass: i64 = 0
34 var total: i64 = 0
35 let out: *i64 = sys_mmap(128) as *i64
36 let out2: *i64 = sys_mmap(128) as *i64
37
38 let srcv: *i64 = sys_mmap(8*8) as *i64
39 srcv[0] = ("The Eiffel Tower stands in Paris, the capital of France." as *u8) as i64
40 srcv[1] = ("Human performance reaches an F1 of 86.8 percent." as *u8) as i64
41
42 // T1 PUBLISH: fully grounded, the numeric claim (86.8) confirmed against source 2 -> 0 gaps -> PUBLISH
43 total = total + 1
44 let b_pub: *u8 = "The tower stands in Paris [1]. Human performance is 86.8 [2]." as *u8
45 nxr_assess(b_pub, st_len(b_pub), srcv, 2, 0, 3, 1000000, out)
46 st_row("T1 PUBLISH " as *u8, out)
47 if out[8] == 0 { if out[6] == 0 { pass = pass + 1 } else { st_puts("T1 FAIL gaps\n" as *u8) } } else { st_puts("T1 FAIL verdict\n" as *u8) }
48
49 // T2 CONTINUE: a FABRICATED number (50.0 not in the cited source) -> discredited -> gaps=1, budget left -> CONTINUE
50 total = total + 1
51 let b_bad: *u8 = "The tower stands in Paris [1]. Human performance is 50.0 [2]." as *u8
52 nxr_assess(b_bad, st_len(b_bad), srcv, 2, 0, 3, 1000000, out)
53 st_row("T2 CONTINUE " as *u8, out)
54 if out[8] == 1 { if out[5] == 1 { pass = pass + 1 } else { st_puts("T2 FAIL discredited\n" as *u8) } } else { st_puts("T2 FAIL verdict\n" as *u8) }
55
56 // T3 STOP-INCOMPLETE: same fabricated-number report but the budget is spent (iter 2 of 3) -> ship honestly
57 total = total + 1
58 nxr_assess(b_bad, st_len(b_bad), srcv, 2, 2, 3, 3, out)
59 st_row("T3 STOP-INC " as *u8, out)
60 if out[8] == 2 { pass = pass + 1 } else { st_puts("T3 FAIL verdict\n" as *u8) }
61
62 // T4 CONTINUE (uncited claim): an uncited sentence -> gaps>=1 -> CONTINUE
63 total = total + 1
64 let b_unc: *u8 = "The tower stands in Paris [1]. It is very old today." as *u8
65 nxr_assess(b_unc, st_len(b_unc), srcv, 2, 0, 3, 1000000, out)
66 st_row("T4 CONTINUE " as *u8, out)
67 if out[1] >= 1 { if out[8] == 1 { pass = pass + 1 } else { st_puts("T4 FAIL verdict\n" as *u8) } } else { st_puts("T4 FAIL uncited\n" as *u8) }
68
69 // T5 determinism
70 total = total + 1
71 nxr_assess(b_pub, st_len(b_pub), srcv, 2, 0, 3, 1000000, out)
72 nxr_assess(b_pub, st_len(b_pub), srcv, 2, 0, 3, 1000000, out2)
73 var t5: i64 = 1
74 if out[6] != out2[6] { t5 = 0 }
75 if out[8] != out2[8] { t5 = 0 }
76 if t5 == 1 { pass = pass + 1 } else { st_puts("T5 FAIL not-deterministic\n" as *u8) }
77
78 // T6 CORROBORATION AXIS (nx_research_crossval integration): the same value carried by >=2 distinct-origin
79 // sources -> out[9]=1; carried only by its cited source -> out[9]=0 (single-source is honest, not inflated).
80 total = total + 1
81 let srcv3: *i64 = sys_mmap(8*8) as *i64
82 srcv3[0] = ("The Eiffel Tower stands in Paris, the capital of France." as *u8) as i64
83 srcv3[1] = ("Human performance reaches an F1 of 86.8 percent." as *u8) as i64
84 srcv3[2] = ("An independent replication also measured 86.8 for humans." as *u8) as i64
85 let b_cor: *u8 = "Human performance is 86.8 [2]." as *u8
86 nxr_assess(b_cor, st_len(b_cor), srcv3, 3, 0, 3, 1000000, out) // origins 2+3 both carry 86.8
87 nxr_assess(b_cor, st_len(b_cor), srcv, 2, 0, 3, 1000000, out2) // only the cited source carries it
88 st_puts("T6 corrob: 3src confirmed=" as *u8); st_putn(out[4]); st_puts(" corroborated=" as *u8); st_putn(out[9]); st_puts(" | 2src corroborated=" as *u8); st_putn(out2[9]); st_puts("\n" as *u8)
89 var t6: i64 = 1
90 if out[4] != 1 { t6 = 0 }
91 if out[9] != 1 { t6 = 0 }
92 if out2[9] != 0 { t6 = 0 }
93 if t6 == 1 { pass = pass + 1 } else { st_puts("T6 FAIL corroboration\n" as *u8) }
94
95 // T7 LIAR-KILL (citation laundering): claim cites [1] (which carries NO number) while 2 OTHER sources carry
96 // the value -> confirmed=0 AND corroborated=0. Corroboration must never launder a bad citation.
97 total = total + 1
98 let b_mis: *u8 = "Human performance is 86.8 [1]." as *u8
99 nxr_assess(b_mis, st_len(b_mis), srcv3, 3, 0, 3, 1000000, out)
100 st_puts("T7 mis-cite: confirmed=" as *u8); st_putn(out[4]); st_puts(" corroborated=" as *u8); st_putn(out[9]); st_puts(" (2 non-cited carriers give NO credit)\n" as *u8)
101 var t7: i64 = 1
102 if out[4] != 0 { t7 = 0 }
103 if out[9] != 0 { t7 = 0 }
104 if t7 == 1 { pass = pass + 1 } else { st_puts("T7 FAIL cite-laundering\n" as *u8) }
105
106 // T8 REGRESSION-GUARD (why host ids matter): a same-origin pair -- 2 sources carrying the value from the SAME
107 // real host. nxr_assess (index-as-host default) OVER-CREDITS corroborated=1; nxr_assess_hosts (real host ids,
108 // both = same id) correctly says corroborated=0. Locks the exact caveat cvx_host_id fixes (proves the BUG,
109 // complementing crossval T13 which proves the FIX).
110 total = total + 1
111 let svp: *i64 = sys_mmap(8*8) as *i64
112 svp[0] = ("Human performance reaches an F1 of 86.8 percent." as *u8) as i64
113 svp[1] = ("A mirror of the same page also lists 86.8 for humans." as *u8) as i64
114 let hs: *i64 = sys_mmap(8*8) as *i64
115 hs[0] = 777; hs[1] = 777 // SAME real origin (two paths, one host)
116 let b_pair: *u8 = "Human performance is 86.8 [1]." as *u8
117 nxr_assess(b_pair, st_len(b_pair), svp, 2, 0, 3, 1000000, out) // index-as-host default: over-credits
118 nxr_assess_hosts(b_pair, st_len(b_pair), svp, 2, hs, 0, 3, 1000000, out2) // host-TRUE: echo-suppressed
119 st_puts("T8 host-guard: index corroborated=" as *u8); st_putn(out[9]); st_puts(" (over-credit) vs host-TRUE corroborated=" as *u8); st_putn(out2[9]); st_puts("\n" as *u8)
120 var t8: i64 = 1
121 if out[9] != 1 { t8 = 0 }
122 if out2[9] != 0 { t8 = 0 }
123 if t8 == 1 { pass = pass + 1 } else { st_puts("T8 FAIL host-guard\n" as *u8) }
124
125 // T9 (marker-gated): the FULL pipeline with real Qwen (synthesize 2 sections -> grounded report -> verdict)
126 let mk: i64 = sys_openat_rd("knowledge/index/nxr_model.on" as *u8)
127 if mk >= 0 {
128 sys_close(mk)
129 total = total + 1
130 let heads: *i64 = sys_mmap(8*8) as *i64
131 heads[0] = ("Location" as *u8) as i64
132 heads[1] = ("Human performance" as *u8) as i64
133 let qs: *i64 = sys_mmap(8*8) as *i64
134 qs[0] = ("In what city is the Eiffel Tower?" as *u8) as i64
135 qs[1] = ("What F1 does human performance reach?" as *u8) as i64
136 let buf: *u8 = sys_mmap(8192)
137 let rn: i64 = nxr_research("Research Brief" as *u8, heads, qs, 2, srcv, 2, buf, 8192, out)
138 st_puts("T9 MODEL pipeline bytes=" as *u8); st_putn(rn); st_puts(" -> VERDICT=" as *u8); st_verdict(out[8]); st_puts(" (gaps=" as *u8); st_putn(out[6]); st_puts(" corroborated=" as *u8); st_putn(out[9]); st_puts(")\n---REPORT---\n" as *u8); if rn > 0 { sys_write(1, buf, rn) } st_puts("\n---END---\n" as *u8)
139 if rn > 0 { pass = pass + 1 } else { st_puts("T9 FAIL empty\n" as *u8) }
140 } else { st_puts("T9 SKIP (no nxr_model.on; unified assess+verdict+corroboration+host-guard proven $0 in T1-T8)\n" as *u8) }
141
142 st_puts("NX-RESEARCH-UNIFIED " as *u8); st_putn(pass); st_puts("/" as *u8); st_putn(total); st_puts("\n" as *u8)
143 if pass == total { st_puts("NX-RESEARCH-UNIFIED ALL-PASS\n" as *u8); sys_exit(0) }
144 sys_exit(1)
145 return 1
146}