About

SPEC is a synthesis paper, not an empirical one. It does not propose a new estimator. It consolidates four streams of work already drifting toward a shared diagnostic vocabulary — manager-side model validation, allocator-side AI-claim detection, firm-side governance, and trading-system-side backtest hygiene — and renders them as a per-strategy, falsifiable protocol that allocators can paste into IC memos and DDQs.

The protocol's structural moat is allocator-side per-strategy framing. Every prior published framework is either firm-facing (Big-Four governance, MAS Veritas, BoE-PRA SS1/23) or per-claim (Simonian 2025 AI Washing). The per-strategy, allocator-facing, falsifiable layer is the unoccupied position in the landscape of published artifacts; SPEC is designed to occupy it.

Four citation streams are designed into the protocol from §1.4: academic literature reviews; allocator IC memos; model-risk officers mapping AI claims to SR 26-2, EU AI Act, IOSCO PD684 / PD788 vocabulary; and AI-finance founders forced to demonstrate compliance against the standard, becoming unwitting citation drivers.

The v1 calibrations are anchored against the empirical distributions documented in the exemplar literature and the audit-practice distribution the author has observed across 2024–2026 retainer engagements. They are not derived from a population-level study of AI-manager failure rates — no such study yet exists in the published literature. As that literature develops over 2026–2028, revised calibrations will be released as SPEC v2 and successor versions, each commit-pinned to a dated rubric specification.

Milos Maricic, Maricic Advisory. AI strategy advisor working with sovereign wealth funds, central banks, and pension funds. The SPEC v1 calibrations are anchored against the audit-practice distribution observed across 2024–2026 retainer engagements with institutional allocators evaluating AI-driven investment managers.

More at milosmaricic.com/about · work-with-me

The v1 draft is circulated to:

• Luigi Pistaferri — Stanford University.
• Joseph Simonian — CFA Institute Research Foundation. Author of Investment Model Validation (2024) and AI Washing (2025), the closest published precedents to SPEC's manager-side and allocator-side framings respectively.
• Markus Leippold — University of Zurich. (Fallback: Derek Snow, PhD finance, AI-investment focus.)

Reviewer feedback informs the v2 rubric. Reviewers and adopters who disagree with a specific calibration threshold are encouraged to publish the alternative calibration and the empirical basis for it; the protocol is designed to absorb that work, not to resist it.

Diagnostic question library contributions. Additional diagnostic questions, axis-specific failure-mode taxonomies, and calibration evidence are welcomed under the same CC-BY 4.0 attribution discipline. The library transitions from "drafting" to "freeze-and-circulate" at v1; subsequent additions follow the §2.0 successor-version convention.

Supervisory and self-regulatory adoption. Supervisory authorities operating under the IOSCO, US Interagency, BoE-PRA, ESMA, and MAS frameworks, and self-regulatory bodies including CFA Institute, are invited to consider the v1 protocol as a candidate operationalization of the per-strategy testing-and-validation vocabulary their frameworks already specify at the firm-level perimeter. The v1 commitment is to a falsifiable rubric the supervisory community can challenge, replicate, and refine — not to a binding regulatory framework.