443 Baltimore Vessel Visits Carry a "High Confidence" Cargo Estimate. 296 of Them Are Zero.

The Setup

Baltimore logged 900 vessel visits in the 90 days ending September 11. 443 of them carry a hydrostatic_confidence of "high" — the top tier Overwatch's cargo-estimation pipeline assigns to a call. 296 of those 443, 66.8%, have an estimated_cargo_hydrostatic of exactly zero. Across all 900 visits, 32.9% land in this high-confidence-zero bucket — the highest rate of any port logging 200+ calls in the window, ahead of Constanța (29.3%), Los Angeles (27.2%), Norfolk (26.9%), and Hamburg (25.8%). Singapore, the busiest port tracked, sits at 15.5%; Houston, the lowest among majors, at 7.9%.

The Chain

Two methods drive Baltimore's zero cluster. max_draft accounts for 278 of its high-confidence visits, all zero. no_draft_data accounts for another 18, also all zero — a method whose own name states that no draft reading was available to measure anything. Both are tagged "high" confidence. Meanwhile trim_saline_corrected, the trim- and salinity-aware hydrostatic path, produces 88 high-confidence Baltimore estimates averaging 28,973.6 tons, and fallback_tons_per_meter produces 59 averaging 50,406.8 tons. All four methods sit in the same confidence tier. The label spans a real range of zero to a 50,000-ton average.

The Implication

A downstream consumer filtering on hydrostatic_confidence = 'high' alone, without also checking hydrostatic_method, cannot tell "measured a large cargo exchange" from "had no draft data to measure anything." At Baltimore, that ambiguity covers two-thirds of the port's top-confidence estimates. A throughput or cargo-flow model built on confidence tier rather than method is treating a null result and a real 50,000-ton call as equally reliable — because the field meant to signal reliability doesn't distinguish them.

What to Watch

Whether no_draft_data keeps inheriting "high" confidence rather than a tier that reflects the absence of an underlying draft measurement. Whether Baltimore's rate — the highest among tracked ports — reflects something specific to its draft-reporting AIS feed, or whether other ports converge toward it as more calls resolve through max_draft instead of trim_saline_corrected.

Limitations

90-day window ending September 11; a longer window could shift the port ranking, and Baltimore's 900-visit sample is smaller than Singapore's 54,812. This does not evaluate whether max_draft's zero output is itself correct — a vessel that genuinely didn't exchange cargo would also read zero — only that it carries the same confidence label as methods producing large nonzero estimates. Port comparison limited to ports with 200+ visits in the window.

Data as of 2026-09-11. Source: Overwatch vessel_visits (hydrostatic_method, hydrostatic_confidence, estimated_cargo_hydrostatic), joined to ports on port_name='Baltimore', 90-day window.