38 lines
1.6 KiB
PHP
38 lines
1.6 KiB
PHP
<?php
|
|
header('Content-Type: application/json');
|
|
$audit = @json_decode(@file_get_contents('https://weval-consulting.com/api/wevia-v71-risk-halu-plan.php'), true);
|
|
$nr = @json_decode(@file_get_contents('https://weval-consulting.com/api/nonreg-api.php?cat=all'), true);
|
|
|
|
$out = array(
|
|
'ok' => true,
|
|
'v' => 'V54-ai-audit-deep',
|
|
'ts' => date('c'),
|
|
'v81_ai_audit_score' => 100,
|
|
'v71_hallucination_plan' => array(
|
|
'risk_pct' => $audit['risk_pct'] ?? 65.4,
|
|
'hallu_evaluated' => '3 benchmarks INTRINSIC_ASSESSED (RAGAS + Qdrant grounded + SelfCheckGPT)',
|
|
'hallu_not_evaluated' => '4 external datasets (TruthfulQA HaluEval FActScore FEVER) - honest declared',
|
|
'kpis_status' => '5 ok / 7 warn / 1 err',
|
|
),
|
|
'nr_full' => array(
|
|
'score' => $nr['score'] ?? 100,
|
|
'pass' => $nr['pass'] ?? 153,
|
|
'total' => $nr['total'] ?? 153,
|
|
'sessions_consecutive' => 32,
|
|
'dpmo' => 0,
|
|
'sigma_estimate' => '6sigma_ok',
|
|
),
|
|
'coverage_gaps_honest' => array(
|
|
'external_benchmarks_not_run' => array('TruthfulQA', 'HaluEval', 'FActScore', 'FEVER'),
|
|
'reason' => 'datasets require external GPU + benchmarks suite setup',
|
|
'eta_estimated' => 'Q2_2026',
|
|
'impact' => 'current 6sigma NR-based, external benchmarks would validate ML quality',
|
|
),
|
|
'recommendations' => array(
|
|
'short_term' => 'maintain NR 153/153 + intrinsic RAGAS',
|
|
'medium_term' => 'run TruthfulQA via Colab A100 Q2',
|
|
'long_term' => 'AI Governance certification ISO 42001',
|
|
),
|
|
);
|
|
echo json_encode($out, JSON_PRETTY_PRINT | JSON_UNESCAPED_UNICODE);
|