Engineering AI Benchmarks
Named Customers, Measured Outcomes

99.5% of extracted attributes (tags, lines, equipment IDs, instrument loops, IBR-scoped pressure parts, hazardous-area zones) are correct against the source document. Measured at field level, not document level. Sample-based audit per ISO 2859-1 statistical sampling. Independent QA engineer (not extraction team). Contractual remedy clause: re-process + credit if accuracy slips. McDermott engagement: 99.51% measured per most recent audit cycle.

Engagement scope (10,247 tags / 600 P&IDs / 99.5% accuracy) is publishable as named-outcome reference. Detailed engagement narrative (specific project, exact timeline, downstream asset management platform) is under NDA pending case study publication (target Q2 2026 release). Reference calls available under NDA on request via Pathnovo sales channel.

GCKC and Zetwerk are in active production builds. Reference availability is gated on (1) pilot completion, (2) commercial outcome confirmation, (3) client publication approval. Target Q3 2026 release per Pathnovo case study publication pipeline. Until release, engagement scope is publishable (P&ID revision delta + MR package for GCKC; BOM + datasheet compliance for Zetwerk) but detailed customer narrative is held back per pilot-stage NDA.

Three differences. (1) Field-level vs document-level: Pathnovo measures per extracted attribute; many vendors measure per document, which hides per-attribute performance. (2) Independent audit: Pathnovo's audit process uses an independent QA engineer with no extraction involvement; many vendors audit internally. (3) Contractual remedy: Pathnovo's 99.5% is contractual with remedy clause; most vendor accuracy claims are marketing-grade with no contractual binding.

PE exam (Process Engineering Principles & Practice) tests theoretical process engineering knowledge: thermodynamics, fluid mechanics, reaction kinetics, distillation, separations. It does not test extraction accuracy on production engineering documents (P&IDs, isometrics, datasheets, HAZOP studies). Pathnovo's benchmark is the engineering reality: actual extraction outcome on McDermott's 10,247 tags / 600 P&IDs at 99.5% measured accuracy. PE exam is a credibility proxy; McDermott is the production proof.

McDermott is the deepest production engagement: 10,247 tags is large enough to be statistically robust (not a marketing pilot of 50 drawings), and the engagement is live and continuous (not a finished-and-archived project). For credibility purposes, one large continuous engagement at 99.5% measured accuracy is more verifiable than ten small pilot engagements at 'up to' marketing claims. As GCKC + Zetwerk + future engagements complete, Pathnovo will publish each as a named-outcome reference following the same depth standard.

Yes. Every Pathnovo engagement provides per-attribute audit trail with bounding-box source evidence. Client QA engineers can re-audit any sample; results are reproducible from the audit trail. For tier-1 owner-operators with strict procurement standards, on-demand third-party audit (TÜV SÜD, Bureau Veritas, SGS, Lloyd's Register, DNV) is supported with audit fee absorbed by Pathnovo if accuracy validates.

P&ID extraction: 99.5% on tags / lines / instruments / equipment / valves; measured per ISO 2859-1 sample-based audit. Isometric MTO: 99.5% on pipe lengths, fitting counts, flange ratings; sample-based audit on randomly selected isometrics. Mill certificate traceability: 99.5% on heat number / chemistry / mechanical properties; per-cert audit with chemistry value verification. HAZOP register extraction: 99.5% on action capture (no missed recommendations) + 98% on field-level extraction. Each document type has its own methodology; full methodology documentation available on request.