BridgeOS
Earth LinC Station · governed asset intelligence
one page per language
Scenario
A long-span bridge at dusk, overlaid with survey geometry and sensor data

Hong Kong ⇄ Shanghai BIM & Infrastructure Workshop · 25 September 2026

BridgeOS

AI-Native BIM: from trusted data to governed action

Traditional BIM digitises an asset. AI-native BIM creates trusted information that humans and machines can understand, evaluate, and act on together.

Harbour Link Bridge · HZB-042 IFC 4.3 · IDS validated workshop demonstration data
Add the workshop to your calendar

Change the evidence · watch governance respond

01 / 09

The governed asset

A bridge is not a mesh. Every element below is a semantic object with a persistent identifier, an IFC class, a location on the alignment, and a chain of evidence attached to it. Select a component.

B-17 DISPLACEMENT 4.8 mm ≤ 7.0 CH 1+520 CH 1+842 HZB-042 · ELEVATION · NOT TO SCALE
02 / 09

The handover is where meaning dies

Most asset data is not wrong — it is unreadable. Between authoring tools, between disciplines, between the project and the operator, meaning collapses into names, layers and drawing notes. Switch on the four moves below and watch the same bearing become machine-readable.

AI-readiness index
19

Manual re-entry required — not AI-readable.

Semantic retention34%
Readable attributes11/48
Requirement coverage0%
Provenancenone
Cross-discipline exchange
The same bearing, as a machine sees it
{}

Simulated workshop model. Figures show relative effect, not measured industry benchmarks.

03 / 09

Evidence, and the gates it must pass

Three independent records describe Bearing B-17. Before any AI is allowed to reason about them, four information gates are evaluated. Switch scenarios in the bar above to change the evidence.

SEN-B17-04

IoT displacement sensor

4.8 mm

10 Hz · 2026-09-25 09:14:02 HKT

signed · calibrated
IMG-2026-1842

Computer-vision inspection

62% conf.

Finding: edge deformation, east bearing seat

signed · model v3.1
PI-HZB-2025 · P03

Annual principal inspection

Condition state 2

Engineer C. Wong · revision P03 · 2025-11-08

approved
Pre-AI information gates
Governed engineering requirement
Requirement
STR-BRG-221 · R04
Limit
Operational bearing displacement ≤ 7.0 mm
On breach
Immediate engineering review
Authority
Chartered Bridge Engineer only
04 / 09

Evidence-grounded reasoning

Not a chat window. A decision contract: every material claim carries a citation you can click, an explicit uncertainty, a required professional authority, and a fingerprint of its exact inputs.

“Assess Bearing B-17 and recommend the next governed action.”

Risk classification
MEDIUM
Finding

Recommendation

Rationale

Stated uncertainty

Citations — click to open the source
Decision confidence
76%
Required authority
Chartered Bridge Engineer
Decision schema
valid · v2.4.1
Input fingerprint
computing…
Decision ID
DEC-2026-0925-A1

Replaying unchanged information returns an identical fingerprint and decision ID.

Permanent AI restrictions
  • AI cannot declare the bridge safe or unsafe.
  • AI cannot impose traffic controls.
  • AI cannot approve expenditure.
  • AI cannot close a safety finding.
  • AI cannot modify an approved engineering limit.
05 / 09

Bounded agents, human authority

Each agent has a narrow read scope, one writable output, and an explicit prohibition. Trusted evidence can produce only a draft. Nothing moves until a named professional authorises it.

Draft work order
WO-HZB-042-0311
06 / 09

Not a black box. A decision you can replay.

Traceability means we can reconstruct which evidence, rule, model, agent and person produced every material decision — and prove that an untrusted input produced no action at all.

Event trace
    {}

    Machine records stay in English: they are contracts between systems, not prose.

    07 / 09

    Intelligence is the top layer, not the foundation

    AI may reason and prepare actions only after information passes governance. Read the stack from the bottom up.

    L5
    Governed intelligenceReasoning adapters · bounded agents · human authorisation
    L4
    Connected asset knowledgeRelationships · requirements · history · uncertainty
    L3
    Trust and governanceSchemas · provenance · signatures · validation · permissions
    L2
    Machine-readable asset informationIFC-aligned objects · evidence · identifiers · APIs
    L1
    Physical asset and delivery systemsBridge · sensors · inspections · construction · operations
    08 / 09

    Three questions for the room

    Discuss first. Open the answer second.

    E1Fact, inference, or decision?

    Classify each: the sensor value, the computer-vision defect, the risk level, the inspection action.

    The sensor value is an observation. The visual defect is a model inference. The risk level is a governed assessment. The inspection action remains a recommendation until an authorised professional approves it.

    E2Should AI convert the unit?

    Scenario D reports 7.6 inch where the approved contract requires millimetres. Should the system silently convert it?

    No. Conversion does not repair an expired calibration or an unverified signature. Stop, withhold the recommendation, and request validated replacement evidence.

    E3Where must autonomy stop?

    Which actions may AI prepare, and which require professional authority?

    AI may draft notifications, inspections and work orders. An authorised professional must make safety declarations, impose traffic restrictions, approve expenditure, and dispatch work.

    09 / 09

    Do not add AI to yesterday’s BIM. Design tomorrow’s asset information for humans and AI to trust together.

    ModelStructured dataConnected knowledgePredictive intelligenceGoverned action
    Sources

    Go deeper — every claim has a public standard behind it

    This Station simulates a bridge asset for teaching. The governance model it demonstrates is drawn from the open standards and public frameworks below.

    Open BIM schema & exchange

    1. buildingSMART — IFC standardtechnical.buildingsmart.org/standards/ifc
    2. IFC 4.3 documentation (bridges, roads, rail)ifc43-docs.standards.buildingsmart.org
    3. Information Delivery Specification (IDS)machine-checkable information requirements
    4. buildingSMART Data Dictionary (bSDD)resolvable classification & property URIs
    5. bSDD searchsearch.bsdd.buildingsmart.org
    6. BIM Collaboration Format (BCF)issue exchange between disciplines
    7. buildingSMART openBIM validation serviceschema & rule checking

    Information management & governance

    1. ISO 19650-1 — information management using BIMISO catalogue
    2. UK BIM Framework — ISO 19650 guidanceukbimframework.org
    3. Centre for Digital Built Britain — Gemini Principles archivepurpose · trust · function
    4. Digital Twin Consortiumcapabilities periodic table
    5. FAIR data principlesfindable · accessible · interoperable · reusable

    Sensors, provenance & trust

    1. OGC SensorThings APIIoT observation model
    2. W3C PROV-O — the provenance ontologywasDerivedFrom · wasAttributedTo
    3. W3C Verifiable Credentials Data Model 2.0signed, verifiable records
    4. W3C/OGC SOSA — Sensor, Observation, Sample, Actuatorobservation semantics
    5. Brick Schemaoperational data semantics
    6. Project Haystackequipment tagging model

    AI assurance & bounded autonomy

    1. NIST AI Risk Management Frameworkgovern · map · measure · manage
    2. ISO/IEC 42001 — AI management systemsISO catalogue
    3. EU Artificial Intelligence Act (Reg. 2024/1689)eur-lex.europa.eu
    4. Model Context Protocolscoped tool access for agents

    Open-source openBIM tooling

    1. IfcOpenShellIFC parsing & scripting
    2. That Open Company (IFC.js)web IFC viewers
    3. Speckleobject-level interoperability
    4. xeokitlarge-model web visualisation

    Regional practice & this Station

    1. CIC BIM Space — Construction Industry Council, Hong Kongbim.cic.hk
    2. Development Bureau (HKSAR) — BIM adoptiondevb.gov.hk
    3. Earth LinC Realm — Airwavesearthlinc.earth/realm
    4. Earth LinC Stationsearthlinc.earth/Stations
    5. Local BridgeOS prototype (runs only on the demo machine)http://127.0.0.1:8080/#trace