Method before results
This is research, not an operational rescue tool, and the discipline is the point. Each reconstruction uses only public data: hourly ocean-current analysis from the Copernicus Marine Service, hourly 10 m wind at one point per case from the Open-Meteo historical archive, built on reanalysis such as ERA5, and drift (leeway) coefficients from the published search-and-rescue taxonomy used by the United States Coast Guard. The reconstructions run on a short particle-tracking script written for this series, not on an operational drift model: 800 to 2,000 simulated objects per case, stepped every 10 minutes through the hourly current field, with a wind-drift term taken from the object's drift class.
Before a case is scored, the full configuration is frozen and hashed — one run, no iterating until it looks right. People are anonymised; only facts already public in the press are used. Misses are reported with the same prominence as matches, because a validation series that only publishes its hits is marketing, not science.
Case B — six days, sixty kilometres, direction exact
Two fishermen left a beach east of Fortaleza in March 2022 and were found six days later, still adrift, near the port entrance more than 60 km to the northwest — a fact pattern reported publicly at the time. Seeded from the declared fishing grounds with a swamped-hull drift class, the reconstruction's net drift direction came out at 309°; the real outcome axis is roughly 300–310°.
The subset of simulated particles that stayed afloat for the whole window — the population that corresponds to a raft that in fact floated for six days — passed within 10 km of the real rescue point in 100% of cases, with a median closest approach of 2.6 km. Declared limitations: the true capsize point is unknown (a distribution was used), and the outcome was public before the run — what the freeze guarantees is that parameters could not be tuned toward it.
Case A — 24 hours, direction right, magnitude at the tail
A father and son spent roughly 24 hours clinging to a capsized raft in November 2025 and were found by other fishermen 32 km offshore. The reconstruction again matched the alongshore transport axis (297°, northwest) — but its offshore magnitude put the median at 20 km, with the real outcome beyond the 90th percentile.
Two readings are possible and both are declared: the capsize may have happened farther out than the assumed start distribution, or offshore transport is under-represented in the coastal cells of the global current product. Without a field experiment the two cannot be separated. Verdict recorded: partial — direction yes, offshore magnitude at the limit.
Case D — the honest structural miss
A fishing vessel out of Itarema lost contact in August 2026 and its crew of six was found alive nine days later, roughly 400 km offshore — almost due north of the departure coast. The free-drift reconstruction achieved 0% containment: no particle came within 60 km of the target, under any windage hypothesis. West-northwest was the net direction in every configuration. The target to the north is unreachable by free drift in that current field.
That failure is the most informative result in the series. The leading hypothesis is that the vessel was never drifting — an intact boat and a well crew after 19 days at sea reads as a loss of communication, not a loss of propulsion, and leeway physics simply does not apply to a boat under way. The case forced three fixes to the script (domain-boundary handling, advective box sizing, survivor-bias reporting) and one product lesson that outranks all of them: a missing-vessel case needs triage — drifting, operating, or anchored — before any drift map is drawn. What protects a crew like that one is not a better map. It is a registered sail plan and a check-in that turns ten days of silence into a day-one alert.
Where the series stands
Across the series the record is mixed, and it is published that way: one match (Case B), one partial (Case A) and one outright structural miss (Case D, 0% containment). Where a raft truly drifted, the alongshore direction at day scale came out right; the nearshore strip and the offshore tails are where public data runs out. This is a research prototype that supports continued study; it does not validate operational use. Those two gaps are measurable, and closing them is the program's next field objective.
This is research. Estimates are models, not guarantees; a model is never an observation, and a machine estimate is never “confirmed” — only a human reviewing it in the field makes it verified. Nothing on this page is an operational-capability claim, a safety promise, or a prediction of a specific event at a specific place.
References & data
- 1.Copernicus Marine — Global Ocean Physics Analysis and Forecast (GLOBAL_ANALYSISFORECAST_PHY_001_024) — The exact product: 0.083° × 0.083° (1/12°) grid with hourly mean surface currents — the current field the script reads.
- 2.ERA5 (ECMWF) — The reanalysis behind the Open-Meteo historical archive from which each case's hourly wind was taken, one point per case.
- 3.Allen & Plourde (1999), U.S. Coast Guard — Review of Leeway — the drift-class taxonomy used for object windage coefficients.
- 4.Diário do Nordeste — Public reporting of the 2022 six-day drift case (ref. 1.3210654) and the 2026 kite incident (ref. 1.3772763).
- 5.CNN Brasil / G1 / O POVO — Public reporting of the 2025 24-hour case and the 2026 Itarema vessel case.