About
Vaethra exists because the information needed to see a disruption coming is almost all public, almost all scattered, and almost never assembled by anyone without a reason to shade it.
The problem is assembly, not access
Almost everything you would need to see a crisis forming is already published. Seismic networks publish. Port authorities publish. Statistical agencies publish trade flows. Grid operators publish load. Coded conflict datasets publish incidents. Nearly none of it is secret.
What does not exist is the assembly: one place where a strait's traffic falling can be read next to who depends on that strait, next to what those parties in turn supply, with every step carrying the source it rests on. That work is unglamorous, it is mostly plumbing, and it is the difference between a headline and an early warning.
The organisations that do assemble it sell access for tens of thousands a year, which means the people most exposed to disruption — journalists, researchers, small operators, anyone in a country the market considers too small to sell to — are looking at the same headlines as everybody else.
Four commitments, in the software
These are not values on a wall. Each one is enforced somewhere in the code, and each one has cost something to keep.
Provenance travels with the record
Every figure carries the publisher it came from and the moment it arrived. Not in a footnote — attached, everywhere it appears, so any claim can be walked back to something you can check without asking us.
The accuracy record is published, including the misses
The backtest reports how often signals actually preceded confirmed events, scored against two null baselines. Where the sample is too small to support a claim it says so and prints no rate. There is no parameter anywhere that suppresses a bad result.
Estimates are labelled as estimates
Observation and model output are separated everywhere in the interface, because the fastest way to destroy an intelligence product is to let the second quietly acquire the authority of the first.
The observation layer stays free
No account, no cookies, no tracking beyond a cookieless page count, no paywall on the evidence. Anyone can open the Terminal, read any figure back to the record behind it, and export what they are looking at. It stays free because it is also the proof: an accuracy record behind a paywall is worth nothing to anyone deciding whether to believe it.
Calling the API from your own code is licensed, and that is what pays for the above. We would rather charge the machines than the reader.
Ziad Soliman
Vaethra is built and run by Ziad Soliman, who designed and wrote the whole of it — the collection layer, the dependency graph, the cascade model, the scoring harness and the interface.
The shape of the product follows from a single conviction that runs through all of it: a system that cannot be checked should not be believed, including this one. That is why the evidence tiers exist, why the backtest publishes its failures, why model output is fenced off from observation, and why the free layer is the evidence rather than a sample of it.
It is also why the work is unusually public. The methodology, the glossary, the coverage gaps and the limits are all documented here rather than described in a sales call, on the view that a claim nobody can inspect is a claim nobody should act on.
Enquiries — press, partnership, data licensing, or anything about the method — contact@vaethra.com.
The order of construction
Four layers, built deliberately in sequence rather than in parallel. Observation first, because everything above it inherits its errors. Consequence second, because knowing what rests on what is the part that turns an event into a forecast worth making. Prediction third, once there is a harness that can tell whether a forecast was any good. And a fourth layer studying how populations interpret all of it, because belief moves people as reliably as scarcity does.
| Layer | Question | State |
|---|---|---|
| Terminal | What is happening? | Live |
| Atropos | What does it break? | Live |
| Oracle | What happens next? | Pre-release 0.19.0 |
| Eschaton | What do people believe it means? | Not started |
Where this is, honestly
Two of the four systems are live and free to use with no account. Oracle is a pre-release whose production run's forward record held 3,019 published forecasts, 0 of them resolved, when this page was built, on 11 October 2026 (an earlier log holds one day of forecasts, 25 September 2026), and Eschaton is not started; neither is sold, and the commercial page describes only what exists. The API is licensed and in production. There is no outside funding and no sales team.
The company is small — one engineer, who wrote all of it. That is a real risk to weigh and the honest mitigations are written down rather than implied: the data is reproducible from public sources, source-code escrow can be arranged on enterprise agreements, and recovery codes exist for every platform with a second holder named in the access review. It is also the reason a customer gets a call with the person who built the system rather than an account manager relaying questions.
What follows from that: expect fast answers and unusual depth on how the numbers are made, and expect us to say no to things that need a team we do not have. If your procurement requires SSO, a SOC 2 report in hand, or twelve months of uptime history, we do not have those yet and the trust centre says so in each case.
The company
Vaethra is built and operated by Vaethra Technologies LLC, a limited liability company registered in the State of Wyoming, United States — filed 20 July 2026, Wyoming filing ID 2026-002035574, registered agent Northwest Registered Agent — with its mailing address at 30 N Gould St Ste N, Sheridan, Wyoming 82801. The engineering is done from Vienna. That is the entity behind every agreement on this site — the Terms, the Data Processing Agreement and the Privacy Notice.
Two things a funded competitor cannot buy this quarter
This section exists because it was missing. A reader could see the assets on this site and still have to construct the argument for themselves, and most will not bother. So here it is, with its weaknesses named, because an argument that only lists its strengths is a pitch.
1. The graph is time, not money
The record carries 78,000 recorded dependencies between countries, ports, commodities, companies and facilities, of which 9,980 carry a published measurement rather than a structural or inferred link. Each one has a provenance: which method produced it, which source, which evidence class. That is visible in /api/v1/graph/stats, not asserted here.
The links themselves are not secret — most derive from public trade,
shipping and infrastructure data. What is expensive is the reconciliation: one
entity per real-world thing, across sources that spell it differently, with the
duplicates collapsed and the merges auditable. Three different ports are called
Georgetown. commodity:alumina is not
commodity:aluminium. Getting that wrong is invisible until a
cascade is wrong, and a competitor starting today spends the same months on it
that we did, regardless of budget.
The weakness. A well-resourced team with the same discipline gets there — this buys quarters, not years, and it buys nothing at all against someone who already has a resolved entity graph and chooses to extend it.
2. Publishing the misses is a position, not a feature
The track record publishes every scored run, including the 24 across 6 corridors, of which the harness counts 0 as demonstrating skill, and the harness's own verdict that the flagship test was uninformative. The API that proves it needs no key.
An established vendor cannot match that by deciding to. Publishing a null result is cheap for a company with no customers and expensive for one with a renewal book — every existing contract was sold against a number that a published accuracy record would put in context. The asymmetry is structural: it costs us almost nothing and costs them a quarter of revenue, so the longer it runs the more it is worth. It also compounds; a record is only credible in proportion to how long it has been kept in public.
The weakness. It is a moat only while the numbers eventually get good. A published record of never beating the null is an honest record of a system that does not work, and we would have to publish that too. That is the deal, and it is the reason to believe the number when it does move.
The alternatives
Commercial disruption and risk intelligence exists — Windward, Everstream, Resilinc, Kpler and Recorded Future among others sell it to enterprises, and the free public datasets Vaethra reads (ACLED, GDELT, PortWatch) are open to anyone. The difference is not the feeds. It is that the observation layer here is free and account-less, every record carries its publisher and both timestamps, the dependency graph states the method behind every link, and the accuracy record is public with its misses. We have not found a published, scored accuracy record with the misses shown from any of the vendors named; if one exists, tell us and this sentence changes.
What is not a moat
The feeds are public and anyone can read them. The interface can be copied from a screenshot. The cascade method — every rule, every channel, what a share of dependence is and how the ordering is tested — is described on the methodology page on purpose; only the calibration constants are withheld, and the page says which. None of those is the defensible part, and claiming otherwise would be the first thing a technical reviewer disproved.
Research that is not a product
Eschaton is a research note, not software. It is listed here rather than beside the products because there is nothing to use and no date.