PortWatch · Global Port Intelligence
An internal decision view connecting live activity, throughput nowcasts, report calibration, and provider evaluation
Industry Decision-Support Case / 12
PortWatch puts live port activity, throughput nowcasting, official-report calibration, and provider evaluation into one traceable data chain.
The product supports internal situational awareness across the top one hundred container ports. It distinguishes observed, estimated, and reported data and never presents minute-level TEU nowcasts as actual terminal handling. DEMO data currently validates the engineering chain; production gates require licensed commercial data.
The local MVP includes one hundred ports, a map, congestion ranking, live table, and port-level trends.
Data objects are explicitly labeled observed, estimated, or reported.
Nowcasts include confidence, versioning, and a protection that hides TEU when MAPE is too high.
The current provider is DEMO; commercial-provider benchmarking and production readiness are not complete.
Start by Naming What Cannot Be Observed
Vessel position, arrivals, departures, and waiting can be observed in near real time; actual global terminal crane moves cannot. The product puts this fact into interface, schema, and acceptance rules rather than hiding it behind polished charts.
When estimation error exceeds the threshold, TEU is hidden while an activity index remains. Monthly revisions create new records instead of overwriting history, preserving decision provenance.
Productize Data-Provider Procurement
Production data requires licensed providers. A normalized gateway contract and a fourteen-day blind test cover completeness, freshness, event accuracy, licensing, and total cost instead of binding the app to one raw AIS source.
Demo delivery and production readiness use separate machine-readable gates. Passing DEMO proves code and safeguards—not commercial data quality.
PROCESS
Key Decisions
Never mix three data classes
Observed, estimated, and reported values keep consistent labels across storage, API, and UI.
Hide output when error is high
The product would rather show less than expose precise-looking numbers it cannot defend.
Separate demo and production acceptance
DEMO validates the chain; licensed data, benchmarking, and backtesting are distinct production conditions.
Role & Collaboration
I led problem framing, evidence classification, dashboard architecture, nowcast boundaries, provider evaluation, and acceptance design.
AI supported engineering and documentation; I governed industry facts, boundary language, and production-readiness claims.
Validation, Outcome & Reflection
The local demo chain loads one hundred ports, generates DEMO observations, and supports machine-readable demo acceptance.
No licensed commercial live data is in use, so the project does not claim real global operations coverage or production accuracy.
A decision product earns trust not through more numbers, but by knowing where every number came from, what it proves, and when it should disappear.
DEMO / 12
Product demo · 10 sec
A 720p local demo of the global port overview and primary situation views; displayed data is for product demonstration only.
