Published and citable; not automatically empirically validated.
Evidence before claim · Updated September 22, 2026
Evidence and maturity
This page separates scientific evidence, software implementation, testing, publication, and external review. A status in one layer never transfers automatically to another.
Peer-reviewed spectral-compression result: AIP Advances 16, 095208 (2026) · DOI 10.1063/5.0337045 ↗
01 / Canonical evidence layers
Seven categories, each making a different claim.
Follows within explicitly stated model assumptions.
Tested in simulations or methodological reference cases.
Tested with real data; the validity domain remains explicit.
A falsifiable claim whose confirmation remains open.
Planned research or software; not yet implemented.
Further theoretical exploration; not additional empirical evidence.
Separation rule: Software status such as IMPLEMENTED, AVAILABLE, IN DEVELOPMENT, or PLANNED describes technical availability—not scientific validity.
02 / Evidence matrix
What is defined, implemented, and actually tested?
| Element | Scientific status | Software status | Testing | External evidence |
|---|---|---|---|---|
| Frozen Legacy Kernel | Mathematically defined | Implemented in Studio 1.4.0 | Reference benchmark reproduced on Apple Silicon; one self-reported run | No independent replication |
| Spectral compression / Φ(Σ) | Peer-reviewed model-derived result in AIP Advances | Not a standalone Studio 1.4.0 module and not identical to the broader CRTI software indicator | Mathematical analysis documented in the AIP article; broader cross-domain empirical validation remains open | Peer-reviewed: AIP Advances 16, 095208 (2026), DOI 10.1063/5.0337045 |
| FCQ | Described in a versioned publication | Research and architecture component | Domain-specific validation remains open | Preprint / open-science contribution |
| Studio 1.4.0 | Released baseline | macOS package built, installed, and launched | Reference values confirmed in one real-hardware run | No independent replication yet |
| Studio 2.0 | RC2 · current release preparation | Four EWS registered as Experimental: variance, lag-1 autocorrelation, skewness, and CRTI; CRTI includes a runnability check and explicit explanation | RC2 tests 18/18 + 15/15 PASS; from-zero reinstall/launch PASS; public-release and cross-platform hardware validation remain separate gates | Not yet released as Stable 2.0 |
| Studio 3.0 / ECHO | Planned research architecture | Not implemented | No architecture-wide empirical validation | Planned for mid-2027; no availability guarantee |
| Domain transfer | Testable hypothesis | Requires separate operationalization for each domain | Permitted only through specified validation cases | No universal evidence of effectiveness |
| Ontology of Oscillation | Theoretical horizon | Not part of the mathematical core | No empirical claim for the method | Not a basis for validation status |
Reading rule: Peer-reviewed publication establishes a reviewed publication status for the mathematical work; it does not by itself establish universal validity, independent replication, or software validation.
02A / AIP boundary and falsifiability
What the peer-reviewed result shows — and what remains open
The AIP Advances result concerns spectral compression of the stationary covariance matrix in a defined class of linear stochastic systems. Under assumptions A1–A4, the dominant covariance eigenvalue diverges while the remaining eigenvalues stay bounded, implying Φ(Σ) → 1. This result is mechanism-specific; it is not a universal proof of CRTI as an early-warning detector for complex adaptive systems.
Open empirical question: Whether Φ or derived BenchEWS indicators provide robust lead time, calibration, discrimination, or incremental information beyond established early-warning signals must be tested separately for each system class. Defective eigenvalues, insufficient excitation of the critical mode, high-dimensional bulk effects, exogenous shocks, and strong non-normality are explicit limitations or failure modes.
No detection threshold, false-positive rate, or lead-time distribution is established by the mathematical result alone.
03 / Claim protocol
Every claim needs an explicit boundary.
What exactly is being investigated?
Which data, model, code, or source supports it?
Where does the claim stop?
What result would count against it?
How far has testing progressed?
Which version is authoritative?
04 / Non-claims
What BenchEWS explicitly does not claim
- No certaintyAn early-warning signal is not a deterministic tipping-point prediction.
- No universal scoreBenchEWS does not provide one universal stress score for arbitrary systems.
- No automatic transferEach domain requires its own data, measurement model, criteria, and validation cases.
- No diagnostic shortcutHomo sapiens is a research domain, not a medical or psychological diagnostic instrument.