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, wind fields from the ERA5 reanalysis, a harmonic tide model, and drift (leeway) coefficients from the published search-and-rescue taxonomy used by the United States Coast Guard.
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 engine'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 engine 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 engine's 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 engine fixes (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 pattern is consistent: alongshore transport at day scale matched in every scorable case; the nearshore strip and the offshore tails are where public data runs out. 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 Service — Hourly global ocean current analysis and forecast (1/12° grid), the drift engine's current field.
- 2.ERA5 (ECMWF) — Reanalysis wind fields used for windage forcing.
- 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.