This document sets out the Compile-Time Inference (CTI) marine corrosion, electrical, and cathodic protection ontology, its 293 attributes across 28 classes, and the live PostgreSQL 16 compiled assessments over the 6 sample vessels.
Quote the class and field name with each comment, for example half_cell_reading.cp_margin_mv or lab_fluid_sample.coolant_diagnosis.
| Phase | What fills the fields | Fields |
|---|---|---|
| Alpha (proven on 6 sample reports) | Deterministic PDF/table/lab ingest, three-clock separation, and core ontology schema | |
| Beta (5-wave compile pipeline) | Vision and text models with SQL engineering expressions, longitudinal rollup, P-Space, and Argmax decider | |
| Production | Confirmation and overrides by senior surveyors in the field, plus outside manufacturer datasheets |
| Beta wave | What it does | Fields |
|---|---|---|
| Wave 1: ingest and registry | Values captured from PDF/tables/ALS lab certificates, or looked up from vessel and product registries | |
| Wave 2: VLM per photo | The vision model reads inspection photos (crevice, pitting, thread tape, unpassivated welds, anode consumption) | |
| Wave 3: LLM per observation | The text model extracts typed findings, corrosion mechanisms, electrical faults, and standard breaches with exact spans | |
| Wave 4: SQL calculation | Calculated deterministically in PostgreSQL 16 (CP margin, continuity verdict, cable duty cycle, glycol %, coolant diagnosis) — zero model calls | |
| Wave 5: rollup and relationships | Longitudinal repeat-survey links (2022 → 2026), causal chains, P-Space log-likelihood ratios, and Argmax recommendation ranking |
Three clocks stay separate throughout: when the vessel was built or suffered an incident (incident_or_build_at), when the surveyor inspected and measured it (survey_observed_at), and when the report was signed and issued (report_issued_at). Every value cites an immutable evidence_span.
ClientOrganisation insurer, ferry operator, defence maintainer, shipyard
└─ Fleet / Contract sister vessels (e.g. Rottnest Island & Norfolk Island)
└─ Vessel IMO / registration, hull material (Al, Steel, Carbon, GRP), particulars
├─ Compartment Port/Stbd Engine Room, Void 1..4, Cofferdam, Bilge, Tiller Flat, Hull External
│ └─ VesselSystem DC Power, AC Power, Shore Power, Propulsion, Cathodic Protection, SW Cooling, HVAC Chiller
│ └─ Component Battery Bank, Solar Regulator, Cable Run, MEN Link, Anode, Rudder, Shaft, Seachest, Flange
└─ Survey OOW Corrosion, In-Water CP & Stray Current, Electrical Fire Forensic, Follow-up
├─ SurveyDocument PDF Report (sha256), Drawing Reference, ALS Lab Certificate, Disclaimer
├─ EvidenceSpan exact page, char_start..char_end, verbatim substring, or photo sha256 + bbox
├─ VisualObservation Photo, surface morphology, contaminant/salt/swarf, PTFE thread tape / crimp issue
├─ InstrumentTest
│ ├─ HalfCellReading Ag/AgCl mV vs ABYC/ISO hull window (-900..-1050 mV Al; ≤-850 mV Steel)
│ ├─ AnodeReferenceReading Green Amp R1..R4 reference mV vs Zinc -950 mV delta
│ ├─ ContinuityTest Fitting A to Fitting B Ohms / Open Circuit (OC) vs bonding rule
│ ├─ ElectricalLeakageTest DC/AC Volts & mA/A across crank, idle, high-idle, solar, and shore states
│ ├─ FaultMonitorTest Intentional +/-24V hard ground fault vs MarineLink alarm response (7A..7K)
│ └─ LabFluidSample ALS Work Order (pH, uS/cm, TDS, Turbidity, Cl, Na, Cu, Zn, Fe, PG/EG glycol %, HPC)
├─ Finding Corrosion mechanism, electrical fault kind, severity, recurrence state
├─ StandardCompliance AS/NZS 3000, AS/NZS 3004.1/.2, ISO/DIS 12473, ABYC E-2/E-11 clause & verdict
└─ Recommendation Argmax-prioritised remediation action, product spec, and verification step
Each value and relationship carries a level on the 11-point Juster scale (0 to 10), declared from deterministic evidence rules in Beta and calibrated against resolved follow-up survey outcomes in Production.
| Field | Applies to | Route | Phase | Draft meaning |
|---|---|---|---|---|
| juster_level | Every value and link | SQL | Beta | The ordinal grade on the Juster scale, from 0 to 10 |
| juster_basis | Every value and link | SQL | Beta | The deterministic rule and evidence span that set the grade |
| juster_scale_version | Every value and link | SQL | Beta | Which version of the scale was used (juster11_v1) |
| juster_resolved_outcome | Every value and link | Use | Production | Whether surveyor confirmation or follow-up survey confirmed or corrected it |
| juster_calibrated_level | Every value and link | SQL | Production | The grade once resolved outcomes support calibration (n ≥ 30) |
| edge_subject, edge_predicate, edge_object | Link | Wave 5 | Beta | What is linked to what, and how (causes, persisted_since, remediated_since, mitigates) |
| edge_evidence | Link | Wave 5 | Beta | Exact evidence_span_101 IDs backing the link |
| edge_support_count | Link | Wave 5 | Beta | Independent measurement modalities supporting the link |
| edge_valid_during | Link | Wave 5 | Beta | Bitemporal period [valid_from, valid_to) the link holds |
| edge_origin | Link | Wave 5 | Beta | Workflow or SQL expression that asserted the link |
Executed on stock PostgreSQL 16.15 (EXPLAIN ANALYZE compile time: 17.98 ms, 0 loops, 0 model calls in Waves 4 & 5, 0 derived fields assigned to model tier).
| Survey / Vessel | Hazard Evaluated (P-Space) | Sum LLR (L) | Threshold (c) | Posterior Prob | Triggered | Contributing Signals (Log-Likelihood Ratios) |
|---|
| Finding Key | 2026 Follow-Up Observation | Longitudinal State (Wave 5 SQL) | Linked 2022 Finding ID |
|---|
Every field has one route and one execution tier enforced by PostgreSQL check constraints. Zero derived SQL fields are sent to a model.
| Class | Field | Draft meaning | Type | Route | Phase | Wave | State | Canonical home |
|---|
Each list is stored as versioned rows in logix.ont_vocab_101, so terms can evolve without schema rewrites.