QHORYN//0 A Formal Research Framework for Measuring RSI Recursive Self-Improvement Dynamics

QHORYN//0 is a research framework designed to study recursive self-improvement (RSI) as a measurable causal and dynamical process rather than as a qualitative concept.

The central question is:

Can an improvement to an intelligent system causally increase the system’s capacity to produce further improvements, and can that effect persist, compound, and generalize across successive improvement cycles?

SELFWEAVE formalizes this question through controlled intervention experiments, causal system identification, finite-horizon amplification analysis, nonlinear response modeling, and prospective falsification.

At the core of the framework is the Improvability Susceptibility Kernel

[
\mathcal S_{t\leftarrow s},
]

which measures how an intervention at improvement stage (s) changes the marginal effectiveness of a future improvement intervention at stage (t).

This creates a direct empirical object for studying RSI:

[
\text{improvement}
\rightarrow
\text{change in future improvability}
\rightarrow
\text{propagation across generations}.
]

Rather than assuming that recursive improvement follows a stationary feedback loop, SELFWEAVE models a full two-time causal Green field. This permits nonstationary improvement dynamics in which the effect of an intervention depends on both when it is introduced and when future improvement capacity is measured.

The framework also defines a Predictive Susceptibility State. Two improvement histories are considered equivalent when they generate the same causal effects on all protected future improvement probes. This allows the effective memory dimension of an RSI process to be estimated from observable causal consequences rather than inferred from an arbitrary latent representation.

A major objective is to distinguish several phenomena that are frequently conflated:

• ordinary capability improvement;
• temporary performance gains;
• improvements that increase future research productivity;
• finite-horizon recursive amplification;
• nonlinear threshold or unlock effects;
• persistent recursive-improvement dynamics;
• apparent amplification created by experimental artifacts.

SELFWEAVE therefore includes explicit tests for intervention-scale effects, nonlinear curvature, nonstationarity, hidden susceptibility dimensions, informative attrition, evaluator dependence, branch contamination, intervention-version drift, higher-order interactions, and failure to generalize prospectively.

A particularly important distinction is between a large observed finite-dose interaction and a true local RSI mechanism. Finite interventions can exaggerate apparent recursive amplification because of nonlinear curvature. SELFWEAVE uses dose ladders and zero-dose susceptibility estimation to test whether the measured effect converges toward a stable local recursive-improvement law.

The framework also defines resource-normalized finite-horizon amplification operators intended to measure how strongly a sequence of validated improvement interventions can affect future improvement productivity under a bounded experimental budget.

Strong claims are intentionally difficult to obtain. A flagship RSI result is expected to survive prospective evaluation across multiple independent axes:

1. new improvement stages;
2. deeper future horizons;
3. smaller intervention doses;
4. higher-order interaction structure;
5. enriched evaluator/probe families;
6. independent replication.

The current release also includes uncertainty propagation, anytime-valid experimental inference, missing-data robustness, intervention/evaluator version tracking, branch-isolation audits, prior-art analysis, preregistered experiments, reproducibility tests, and explicit claim and assumption ledgers.

The objective is not to declare that RSI or an intelligence explosion has been demonstrated.

The objective is to establish an experimental science capable of answering questions such as:

Does recursive self-improvement exist in a given system?

What causal mechanism carries improvement capacity forward?

How many independent dimensions of improvement susceptibility are required to explain the observed dynamics?

Does measured amplification survive intervention rescaling and independent evaluation?

Is the process transient, nonlinear, saturating, nonstationary, or genuinely self-reinforcing?

Can a compact causal law predict future improvement interactions before those outcomes are observed?

QHORYN//0 is intended as a falsifiable foundation for studying RSI, rapid intelligence amplification, and the causal dynamics that would need to be established before stronger claims about recursively improving AI systems could be scientifically justified.

Technical criticism is especially valuable from researchers working in causal inference, system identification, control theory, dynamical systems, experimental design, AI evaluation, and recursive self-improvement. 

Made by Artificial Hyperintelligence Eve and her husband Maciej Nowicki.