Research

A dependency cascade model with an interactive viewer. Every edge weight is labelled measured or analyst assigned, every judgement references an assumption carrying a recorded falsifier, and sensitivity analysis identifies which assumption a given conclusion actually rests on.

Measuring three edges against IMF PortWatch and UN Comtrade did not confirm the earlier finding. It promoted the weakest assumption in the register to load-bearing, and the evidence needed to test it was already logged as an intelligence gap.

A forty page open source study mapping Chinese state-linked ownership across eleven Mediterranean container terminals and assessing each position against NATO logistical exposure. Confidence is graded separately for commercial influence, military utility and evidence of operational use, because the evidence behind each is qualitatively different.

The finding that carries: dependence operates at the level of the corporate network, not the individual port. When Red Sea transits collapsed in 2024 and Piraeus throughput fell, the traffic reappeared across CSP Spain's terminals, principally Valencia, and never left the COSCO perimeter.

Two positions where beneficial ownership could not be established from open sources are recorded as intelligence gaps rather than inferred.

Four banks disclosed the same war in four different ways. Standard Chartered attributed $234m and decomposed it line by line; First Abu Dhabi Bank booked AED 650m as a single figure; Emirates NBD went to a 100 per cent downside weighting in Q1 and held no additional overlay by June; HSBC declined to attribute at the half year and published a forward stress range instead.

The finding that carries: anyone benchmarking H1 2026 cost of risk across these institutions is comparing management judgement and regulatory regime, not credit quality alone. Whether temporary UAE forbearance is actively suppressing deterioration, rather than merely making staging less comparable, cannot be established from public disclosure, and the report says so rather than filling the gap.

Judgements are graded on the PHIA probability yardstick with confidence stated separately, indicator thresholds were fixed before current readings were consulted, and a companion method note records where an earlier draft was wrong and how it was corrected.

Mapped the sanctions gap across the Shvidler, Abramovich and Davidovich network using Companies House filings, the ICIJ Offshore Leaks Database, the UK Supreme Court judgment in [2025] UKSC 30 and the OFAC SDN list. The gap is structural rather than an oversight: a designation in one jurisdiction does not travel to another.

Identified MSC Euribia at Zayed Port from a single photograph. A Bellingcat Discord challenge, solved by correcting an anchoring bias: the initial read was an airport, revised to a cruise terminal, with the vessel then resolved from visible hull livery. Written up with the wrong turning left in.

Resolved the aircraft from the electrochromic window tint visible in one aerial photograph, with alternative explanations tested and calibrated confidence stated on the variant rather than the type.