Skip to content
AYON Ecosystem

Research

What we study, measure and get wrong, in public. Every article names its sources, carries its evidence class, and publishes misses with the same prominence as matches. Pick a subject.

AYON Coastal Intelligence

Forecast calls locked before the day and graded against instruments, drift reconstructions on public data, and the nearshore gap. Estimates are models, not guarantees.

Now running

Now running

Forecasts locked before the event and still being graded. Locks are never edited; each day is scored as it comes.

Forecast verification — AYON Wave

Calls locked before the day and graded afterwards against buoys and the model cell. Hits and misses carry the same weight.

Field note

Same swell, 1 km apart: why the Huntington Beach pier peels and 17th Street closes out

A research field note from Huntington Beach: on 17 October 2026 the same 18-second south swell closed out at 17th Street and peeled in A-frames off the pier's north side, 1 km away. On a lower tide on 20 October, the peak sat 50–90 m north of the pilings. Two records on two tides make a hypothesis, not a conclusion — and the hypothesis is about the sand.

Read the article
Verification

Hurricane Rachel: the swell call, locked before it arrived and graded PARTIAL

A research verification: Hurricane Rachel's swell call for Southern California was locked here on 1 October, before the first wave existed at any buoy, and graded once against instruments on 13 October. The result is PARTIAL: the swell arrived a day early, peaked 9.5 hours late and a little above the called band, and faded on schedule. The lock, the misses and the hit are all below.

Read the article
Verification

Trestles 2026: forecasts locked before the call, scored after

A closed research verification at Lower Trestles, California: a forecast lock hashed and committed before the day it calls — 10 locks in all, with the Sep 11–13 runs missed and recorded as missed — then scored against NDBC buoys and the public event record. The engine read the swell well and the wind badly, and the day it ranked first was the one shut down by wind. Over 7 scored days the error proved systematic on all three axes: height overestimated, period underestimated on 6 of 7, direction the strong axis. The event finished on 2026-09-18 and V10 closed the mission on 2026-09-19.

Read the article
Verification

Macaronis: a reef 8.3 km off, and a Sunday call graded

A research verification from the Mentawai Islands: the operator at the break moved AYON Ecosystem's point 8.3 km onto the reef, their local rules were written into the record, and a falsifiable call for Sunday 13 September 2026 was sent two days ahead. Graded at the reef cell on 13 September: four hits, two misses. Graded again on 14 September against the operator's report: the direction miss was AYON Ecosystem's own grading error, the size was never scorable, and the real miss was wind, 15–20 kn against 7–11 called.

Read the article
Finding

San Blas: a swell window that could never fire

A research finding from the AYON Wave spot catalogue: the two San Blas records carried swell windows that a year of archive hours shows firing 0.4% and 3.3% of the time, so the dominant swell never sat inside either window and carried a penalty in nearly every hour. The correction — 210–225°, where the model archive places the swell — is committed and reaches the app with the next update, and it exposes a second problem that direction alone cannot fix.

Read the article

Drift and the nearshore gap — AYON Coast

Real drift cases reconstructed on public data, and the last two kilometres where the public models stop.

Finding

Sixty kilometres of INMET silence: no INMET station within 60 km of Cumbuco

A research finding worth checking yourself: INMET's public archive of automatic weather stations shows no INMET station within 60 km of Cumbuco, one of Ceará's busiest kite coasts — the nearest, Guaramiranga, is 72.6 km away — and 124 of 639 stations nationwide recorded not one valid reading across the first eight months of 2026.

Read the article
VerificationIn progress

A Kelvin wave is coming for Newport: the call, locked before it arrives

A research verification: a claim made on television, checked against the instrument before the event it describes has even happened. On 12 September 2026, ABC7 Los Angeles reported a Kelvin wave expected to raise Southern California sea levels 6 to 12 inches within a month; AYON Ecosystem's own tide-gauge reconnaissance, run the same week, found the nearest working gauge to Newport Beach already running about 8 inches above the same week last year.

Read the article
Field note

One number for eight beaches: field-verifying a hurricane swell

A research field note from Newport Beach, California, September 2026: standing on the sand during Hurricane Marie's swell, comparing what the wave model said with what the ocean did — and correcting the data that ships in AYON Wave.

Read the article
Case series

Reconstructing real drift cases on the Ceará coast

A research series: four real incidents from the Ceará coast, reconstructed with public ocean-current and wind data under frozen parameters — every result published, including the one the reconstruction got structurally wrong.

Read the article
Finding

The last two kilometres: where global ocean data stops resolving the coast

A research finding that repeated itself until it became the program's central argument: the best public ocean-current data on Earth does not resolve the strip of water where coastal sports actually happen.

Read the article
Open problem

Kite equipment has no row in the drift tables

A research program's most honest sentence: the number that dominates every drift estimate for kite equipment has no entry in the standard search-and-rescue drift taxonomy, and we have found no published measurement of it. This article defines the gap and the experiment that closes it.

Read the article

Events and climate

Season-scale analysis: what the public climate products say, and what they do not.

Analysis

San Clemente, 21 October 2026: the survey points to straight-line wind

A research analysis of one storm night in Southern California: on 21 October 2026, trees were snapped and one uprooted 2 miles south of San Clemente, and early talk turned to a tornado. The National Weather Service surveyed the damage and found every limb pushed to the northeast, with the storm's motion — likely straight-line wind, possibly above 50 mph. This page lays that report beside the radar frames we froze that evening.

Read the article
Analysis

It will happen again: reading the August–September 2026 window

A research analysis of six extraordinary weeks: a Himalayan glacier collapse that killed more than 1,300 people, a hurricane swell that filled Californian lineups and rescue logs on the same day, and the historic El Niño that forecasters say is now more likely than not.

Read the article
How this subject works

Reconstructions run on public data with parameters frozen and hashed before scoring — one run, no tuning toward a known outcome. People in case studies are anonymised, and only facts already public in the press are used. Four words keep their exact meaning everywhere: a model is a computed estimate, never an observation; a field observation is a timestamped human report; verified means reviewed against source and repeat observation; confidence measures evidence strength — never a promise of safety.

AI and sustainability

What AI costs in electricity and water, what multiplies that cost, and how to use it well. Analyses of published measurements, every figure dated and sourced.

Analysis

The hidden cost of a prompt: why sustainable AI starts with how we use it

A research analysis of what AI costs in electricity and water, and where that cost comes from. A single text prompt is small — Google measured its median Gemini prompt at 0.24 Wh, about 9 seconds of television — but how people use AI multiplies it: reasoning modes about 30 times on average, agents up to 136 times, and every retry repeats the whole cost. The figures, their sources and their limits are below, with six habits that cut the cost without cutting quality.

Read the article
How this subject works

Articles in this subject analyse published measurements, company disclosures and studies; AYON Ecosystem does not meter these systems itself. Three labels keep their meaning: a measured figure comes with a published method; a stated figure is a company's own claim without one; a projection is an estimate of the future, never a reading.