The Setup
Over the 30 days ending October 6, 1,099 vessel calls at Norfolk, Virginia produced a hydrostatic cargo estimate. 509 of them cleared the high-confidence threshold — 46.3%. Across the full vessel_visits table in the same window, 51,892 calls produced an estimate and 27.4% cleared that threshold. Norfolk is running at 1.69 times the network's high-confidence rate, on a base large enough that it isn't a fluke of small numbers.
The average estimated cargo load at Norfolk also runs above the network mean: 2,031.7 against 1,502.9 — 35.2% higher. Average confidence score across all methods: 0.783 at Norfolk versus 0.755 network-wide. Every comparison points the same direction.
The Chain
The hydrostatic pipeline assigns each call one of five methods, and Norfolk's method mix explains part of the gap. In the 30-day window: max_draft accounts for 423 calls (38.5% of Norfolk's volume) at an average confidence of 0.800. no_draft_data accounts for 348 calls (31.7%) at 0.755. fallback_tons_per_meter: 161 calls (14.6%) at 0.727. trim_saline_corrected — the method that applies port-specific water density to correct draft readings for salinity and tide — accounts for 145 calls (13.2%) at 0.868, the highest average confidence of any method running at this port.
That correction depends on port_water_density_cells. Norfolk has three of them covering the entire Hampton Roads estuary — a brackish, tidally mixed body of water where salinity in practice varies by location and by tide stage. Three cells is the entire spatial resolution available to the method producing the port's best numbers.
The Implication
Read as a single aggregate, Norfolk looks like one of the more trustworthy ports in the corpus for draft-based tonnage inference right now — a reasonable conclusion for a trade-flow or port-congestion read to draw from the 46.3% figure alone. That conclusion holds, but it's a blend. The 13.2% of calls running through trim_saline_corrected pull the port's average up because they're genuinely well-supported by a real physical correction. The 31.7% running through no_draft_data are contributing an average confidence of 0.755 with no underlying draft reading at all — meaning a meaningful share of Norfolk's above-baseline confidence number isn't coming from better underlying data, it's coming from how the no-draft fallback happens to be calibrated for this specific port.
What to Watch
Two numbers worth tracking into Q4: whether trim_saline_corrected's 13.2% share of Norfolk calls grows as more of the port's traffic gets draft readings, and whether the three-cell density coverage expands or stays fixed. If Hampton Roads traffic grows faster than density-cell coverage, the share of calls falling back to no_draft_data or fallback_tons_per_meter should rise, which would pull Norfolk's blended confidence number down even as the trim-corrected subset keeps performing well on its own. A falling blended number in that scenario wouldn't mean the underlying correction got worse — it would mean coverage didn't keep pace with volume.
Limitations
This is a 30-day read on one port. It doesn't establish whether Norfolk's confidence advantage is seasonal — Chesapeake Bay salinity and tidal range shift meaningfully across the year, which could inflate or depress the trim-correction's apparent performance depending on when the window falls — or structural, meaning the correction model is simply better calibrated for this estuary than for others in the corpus. Nothing here is checked against an independent cargo manifest or bill of lading; "high confidence" describes internal model agreement across input signals, not validated accuracy against ground truth. Three density cells for an estuary this size is, on its face, a thin base — this piece doesn't attempt to say what a sufficient number would be, only that it's small relative to Hampton Roads' area.
Data as of 2026-10-06. Source: vessel_visits hydrostatic cargo estimation pipeline and port_water_density_cells (Norfolk), 30-day trailing window. Axiom Overwatch.