`nx_examiner.nx` source

1// nx_examiner.nx -- the EXAMINER organ (operator: the team must PERFORM its own evaluation; put the
2// grading capabilities with the team member that owns this, or create that member -- created). The
3// Examiner is the team's assessor: it OWNS and RUNS the grading the team needs, so the team grades
4// ITSELF instead of Claude doing it by hand. Three jobs, one organ (single responsibility):
5//   1. grade a CAPABILITY for S-class exceed (the S-class grader, skeptical + triangulated).
6//   2. grade a NEW measured RESULT -- did it beat the naive baseline, and did it beat SOTA?
7//   3. grade the TEAMMATES (the scorecard) -> the next growth target + whether the team passes.
8// RACI partnership: the EXAMINER benchmarks/grades performance; the CRITIC judges belief-truth
9// (counter-theory, provisionality); the BUILDER builds what the Examiner finds weakest/below-S; the
10// growth loop consumes the Examiner's verdicts. The Examiner never builds or fixes -- it measures.
11// license_tier: ORIGINAL  Refs: separation of judging (Critic=truth) from grading (Examiner=performance).
12
13import "nx_sclass_grader.nx"
14import "nx_team_scorecard.nx"
15
16// JOB 1 -- grade a capability for S-class exceed (delegates to the skeptical grader the Examiner owns).
17func exam_grade_capability(has_ext: i64, verified: i64, beats: i64, triangulated: i64, matches: i64) -> i64 {
18    return sg_grade(has_ext, verified, beats, triangulated, matches)
19}
20
21// JOB 2 -- grade a NEW measured result by distortion (lower is better): beating SOTA = S_EXCEED;
22// beating only the naive baseline = A_PARITY (real progress, not exceed); else B_BEHIND. This is how
23// the Examiner judged the vector-quant run honestly (beats scalar, not SOTA -> A_PARITY).
24func exam_grade_result(metric_new: i64, metric_baseline: i64, metric_sota: i64) -> i64 {
25    if metric_new < metric_sota { return SG_S_EXCEED }      // genuinely beats the best (still owes triangulation)
26    if metric_new < metric_baseline { return SG_A_PARITY }  // improves on naive, not on SOTA
27    return SG_B_BEHIND
28}
29
30// JOB 3 -- run the teammate scorecard: the next organ to build (weakest x leverage), and whether the
31// whole team clears the bar. The Examiner performs the team grading the operator no longer hand-drives.
32func exam_next_growth(n: i64, scores: *i64, leverage: *i64) -> i64 { return tsc_growth_target(n, scores, leverage, TSC_BENCHMARK) }
33func exam_team_passes(n: i64, scores: *i64) -> i64 { return tsc_all_pass(n, scores, TSC_BENCHMARK) }
34
35// the Examiner's HEADLINE for the team: how many capabilities are truly S-class (the real benchmark).
36func exam_sclass_count(n: i64, grades: *i64) -> i64 { return sg_count(n, grades, SG_S_EXCEED) }
nx_examiner.nx source

`nx_examiner.nx` source
