Qualified Execution

01

In most organizations that rely on strategic decision-making, execution follows generation with minimal intermediary infrastructure. A strategy is created, perhaps validated against historical data, and then deployed. The gap between “this idea passed a test” and “this idea deserves real resources” is either nonexistent or managed through informal human judgment.

This default assumption — that generation implies readiness for execution — is inherited from an era when generation was expensive. If producing a credible candidate required months of analysis and years of domain expertise, the generation process itself served as an implicit quality filter. The difficulty of producing an idea was a reasonable proxy for the idea’s readiness.

That proxy has collapsed. AI-assisted generation can now produce hundreds of plausible candidates in minutes. The cost of generation no longer correlates with the quality or deployment-readiness of the output. Yet most systems continue to treat generation as implicit permission to execute — a structural vulnerability that becomes more dangerous as generation scales.

02

Qualification is implemented as 14 concrete, measurable criteria — 10 global and 4 regime-specific. Each criterion produces a machine-readable result with required and actual values. The evaluator is a pure function: it takes assembled data and returns a verdict. No database reads, no side effects, fully deterministic.

The criteria are deliberately heterogeneous. Some measure outcomes (win rate, ROI), but as binary thresholds rather than continuous scores — the system asks “did this strategy demonstrate positive returns?” not “how positive were the returns?” Others measure process quality: observation duration, evidence sufficiency, behavioral stability, fee viability. This heterogeneity prevents any single dimension from dominating qualification.

03

The most important architectural decision in qualified execution is the default behavior when evidence is uncertain. In most systems, uncertain evidence is treated as permission — if the system cannot determine that something should be blocked, it allows it. Qualified Execution inverts this: if the system cannot confirm that a candidate has earned execution, it blocks it.

This fail-closed principle runs through every layer of the architecture. The promotion gate blocks when qualification data is missing. The execution engine pauses when regime proof is absent. The lifecycle manager auto-pauses when degradation signals accumulate. Each layer independently enforces the same principle: absence of evidence is not permission to proceed.

04

Qualification is not a single gate that a strategy either passes or fails. It is a staged progression through seven distinct states, reflecting the continuous accumulation and potential degradation of evidence over time.

A strategy begins as unproven. As evidence accumulates through paper trading, it may achieve global qualification but remain unproven in the current market regime. When regime-specific evidence is sufficient, it reaches conditional qualification — the state that enables live deployment. If performance degrades, qualification is revoked and the strategy enters a degraded state, where it continues paper trading to potentially recover.

This staging reflects a fundamental insight: qualification is not permanent. The same strategy may be qualified in ranging markets but unproven in trending markets. It may be qualified today and degraded next week. The system continuously re-evaluates — qualification is earned and maintained, not achieved and assumed.

05

The strongest test of any qualification system is whether it separates outcomes. If qualified strategies perform similarly to never-qualified strategies, the gate is adding cost without value. If they diverge, the gate is making substantively different decisions.

The Orqis qualification system has been tested through trade-level attribution on production paper trading data. The results show statistically significant separation across multiple dimensions — win rate, account-level returns, regime-adjusted performance, and false-negative analysis.

06

Defining a principle requires defining its boundaries. Qualification is frequently confused with prediction, optimization, and guaranteed outcomes. These confusions are understandable because qualification interacts with all of them. But it is architecturally distinct from each.