We study how complex systems lose stability — and learn to see it coming from their dynamics alone.
A small, independent lab working at the intersection of topological data analysis, matrix time-series methods, and the philosophy of how contradiction behaves inside a measurement. We build detectors for the moment a system begins to spiral — before the spiral closes.
The method is the identity.
We started from a stubborn conviction: that the contradictions and edge cases most pipelines discard as noise are usually the signal. Everything else followed from refusing to look away from them.
Edge cases as signal
Outliers, contradictions, and the points where a model breaks are treated as information about the system, not failures to be smoothed over. The anomaly is where the structure shows itself.
Falsify everything first
Before any claim, we tried to break it. We ruled out every statistical configuration for stability-under-constraint that did not require time — discarding each one that survived on coincidence rather than mechanism.
Mechanism, only then
What remained was a time-aware approach built on TDA and matrix time-series variance. The mechanism was named last, after it was the only thing left standing.
The same mathematical signature of impending instability appears across systems that share nothing physically — [public-safe one-liner: e.g. "from neural recordings to industrial process streams"].
That substrate-independence is the actual claim — not any single product. A pattern that holds across unrelated domains is harder to explain by accident, which is exactly why we trust it more than a result tuned to one dataset.
The full record — preregistrations, kill-criteria, and published validations — lives under Research. Operational details are held back by design; see the disclosure note there.
Published work & record
We publish what can be falsified and withhold what is operational. The list below routes to the work itself rather than summarizing it.
[Year] [Paper title] [One sober line on what it established — the domain and the result, no adjectives.] [status]
[Year] [Paper title] [One sober line — domain + result.] [status]
[Year] [Paper title] [One sober line — domain + result.] [status]
[Year] [Paper title] [One sober line — domain + result.] [status]
Program committee
Program committee, [AIES 2026 — full conference name]. Reviewing for the [2026] cycle.
What we hold back, and why
The published components are designed to be independently checkable. The operational payload that turns them into a working system is withheld — a deliberate choice given its dual-use potential, not an oversight. We'd rather under-claim in public than over-expose.
A note on what we do not claim. Where work is still under review or unconfirmed, it is marked as such. We do not assert results the field has not yet accepted. If a claim isn't here with a falsification trail behind it, treat it as something we are not yet making.
Working with the lab
The mechanism is available for licensing, and we take a small number of selected consulting engagements. This page is the door for those conversations — the research stands on its own elsewhere.
The detection mechanism
The core method can be licensed for integration into monitoring and early-warning settings where loss of stability carries real cost. We scope each license to the domain and validate against it before anything ships.
Selected engagements
We take on a limited number of engagements where the problem is genuinely a stability-detection problem and the data has the dynamics to support it. We will say so plainly when it isn't a fit.
- [Product line 1 — system / domain, one phrase]
- [Product line 2]
- [Product line 3]
We are open to conversations with partners and early funders aligned with a research-first pace. Seed and partnership discussions begin the same way as licensing — by email below.
[confirm or replace contact address]
The lab & the name
Anaskoros Labs is an independent research laboratory — small by design, with a publication trail rather than a pitch. The work has appeared under the Anaskoros name since it began, which is the simplest reason it has stayed.
The name began as a near-miss — a translation that didn't land where it was aimed. We kept it because the accident turned out to describe the work better than the intended word would have.
Not refutation for its own sake — but pulling a system out of a spiral before it closes. A moth follows a light into a tightening turn it cannot survive. The reversal is catching the spiral early, and opening it back out.
That image is the closest honest description of what the lab is for. The geometry on the overview page is the same idea drawn once: a coil that tightens, turns, and releases. We treated a small mistake as a signal worth keeping — which, as it happens, is also the method.