nx_capability_triage_core.nx
buildroot/runtime/nx_capability_triage_core.nx
about
nx_capability_triage_core.nx -- the INTELLIGENT CATCH with ANTI-NAVEL-GAZING + OPERATOR-PROPOSAL
(07-15 operator: "make sure we have 3rd-party benchmarks for our research vs yours and all the
other things so this isn't navel gazing like it has been with our graders; and when we see a gap
the flag should be raised to PROPOSE INVESTMENT to me in what should be improved even if we are
doing another workstream so we can stop and eat our debt").
TWO load-bearing properties, both gate-proven:
1. ANTI-NAVEL-GAZING: a PARITY/EXCEEDS claim is credited (-> ROUTE) ONLY when backed by an
EXTERNAL 3rd-party benchmark. A self-graded (our-own-grader) or unbenchmarked SOTA claim is
NEVER trusted -> TR_SELFGRADED_SUSPECT -> go get a 3rd-party number. "Our research vs yours
(Claude) vs the external tool" is a valid external benchmark; "our grader says we're SOTA" is not.
2. OPERATOR-PROPOSAL: a real gap on a CORE capability, OR any suspected self-graded overclaim,
RAISES an interrupt-worthy investment proposal to the operator (stop current work, eat the
debt) rather than silently work-listing it.
Pure decision funcs here (gate-locked); a scan CLI reads the registry + emits proposals.
license_tier: ORIGINAL
dependencies 2 imports · 3 importers
imports: nx_syscalls.nxnx_crashresume_census_core.nx
imported by: nx_capability_triage.nxnx_capability_triage_gate.nxnx_verify.nx
structs
| none |
consts
| 21 | const SOTA_UNMEASURED: i64 = 0 |
| 22 | const SOTA_BEHIND: i64 = 1 |
| 23 | const SOTA_PARITY: i64 = 2 |
| 24 | const SOTA_EXCEEDS: i64 = 3 |
| 27 | const BENCH_NONE: i64 = 0 // no benchmark at all |
| 28 | const BENCH_SELF: i64 = 1 // OUR OWN grader said so (navel-gazing -- NOT trusted for a SOTA claim) |
| 29 | const BENCH_EXTERNAL: i64 = 2 // 3rd-party suite / tool / Claude-as-baseline (the ONLY trusted SOTA proof) |
| 32 | const WEIGHT_PERIPHERAL: i64 = 0 |
| 33 | const WEIGHT_STD: i64 = 1 |
| 34 | const WEIGHT_CORE: i64 = 2 |
| 37 | const TR_BUILD: i64 = 1 |
| 38 | const TR_WIRE: i64 = 2 |
| 39 | const TR_MEASURE: i64 = 3 |
| 40 | const TR_USE_IMPROVE: i64 = 4 |
| 41 | const TR_ROUTE: i64 = 5 |
| 42 | const TR_SELFGRADED_SUSPECT: i64 = 6 // claimed SOTA on self/none grading -> get a 3rd-party benchmark |
| 45 | const PROP_QUEUE: i64 = 0 // work-list it, keep going |
| 46 | const PROP_RAISE: i64 = 1 // interrupt-worthy -> propose investment to the operator NOW |
functions
| 49 | func tr_decide(has_organ: i64, wired: i64, sota: i64, bench: i64) -> i64 |
| 61 | func tr_proposal(verdict: i64, weight: i64) -> i64 |
| 70 | func tr_routes_to_nishi(v: i64) -> i64 called by 1: main |
| 77 | func tr_is_gap(v: i64) -> i64 called by 1: tr_proposal |
| 83 | func tr_verdict_name(v: i64) -> *u8 |
| 93 | func tr_streq(a: *u8, b: *u8) -> i64 |
| 101 | func tr_trim_copy(buf: *u8, s: i64, e: i64, sc: *u8) -> i64 |
| 117 | func tr_sota_match(buf: *u8, s: i64, e: i64) -> i64 |
| 127 | func tr_bench_match(buf: *u8, s: i64, e: i64) -> i64 |
| 136 | func tr_weight_match(buf: *u8, s: i64, e: i64) -> i64 |
| 149 | func tr_parse_row(buf: *u8, ls: i64, le: i64, out: *i64) -> i64 |