Methodology status and validation

Where CILM stands today: what has been observed in a pilot window, what remains to be validated, and what is deliberately not yet published.

Methodology status and validation

CILM is under active development. This page states what has been observed, which validation steps have been specified but not completed, which elements were developed by the author, and which parts of the quantitative model remain outside the public disclosure layer. It serves as the current public record of the methodology’s maturity and evidentiary limits. The pillar pages explain the reasoning and architecture; this page states the status of the supporting evidence.

This page is not a certification statement, an endorsement record, or an implementation guide. It is a status document describing the current maturity of the public methodology and the evidence available to support it. Throughout this page, “public layer” refers only to information intentionally disclosed on cilm.tech; internal working documents, draft calculations, and restricted technical documentation sit outside that boundary.

Current status

CIRS, the Composite Integrity Risk Score, is an expert-defined scoring model at an early validation stage. At its current stage of development, CIRS is intended as an engineering decision-support model built from identified evidence sources, explicit decision rules, and structured expert judgment. It has not been independently validated across industry settings, certified by an external body, or supported by a completed peer-reviewed validation study.

Its status is best described as initial calibration on a pilot corpus, not a validated predictive model.

What has been observed

The current calibration is anchored to evidence collected during a pilot observation window from 15 February to 30 March 2026, a period of 44 days.

The primary public empirical corpus associated with that window is the anonymized procurement-communications dataset published on IEEE DataPort under DOI 10.21227/34y3-zj88, containing 67 records. It does not constitute an independently labelled validation dataset or ground truth on its own.

Any inference drawn from this stage remains bounded by the pilot corpus and the observation window. Earlier procurement records supported reference-data construction and normalization only; they are not treated as validation evidence and do not enlarge the stated empirical base.

CILM does not claim a validated accuracy rate, a demonstrated reduction in error, or a proven improvement in procurement outcomes.

What has been specified but not completed

The validation design has been specified; the full procedure has not yet been executed. It includes independent, model-blind review of frozen test cases, measurement of inter-reviewer agreement, sensitivity and ablation analysis, checks of manual reproducibility, and comparison with simpler baseline approaches. Reproducibility is treated as a separate question from predictive performance.

Because these steps remain incomplete, no quantitative validation result is reported. Claims about accuracy, precision, recall, error reduction, or comparative performance require the procedure, sample, uncertainty, and result to be reported together — until that condition is met, such claims stay outside the public layer.

What is author-developed

The five-factor structure of CIRS, the separation of prior assessment from post-verification evidence, and the L0–L3 verification-depth framework were developed by Alexander L. Litvin as the core engineering structure of CILM. These elements are described at a public architectural level on the pillar pages. They have not yet been independently replicated or externally validated.

What is not published

The quantitative implementation parameters currently used for internal development — factor weights, tier boundaries, missing-data defaults, and quantitative limits on verification credit — are maintained in restricted technical documentation that is not part of the public methodology.

The boundary is deliberate. Publishing mutable draft parameters before a stable, citable version exists would create version ambiguity and could encourage use of values that have not completed validation. It could also narrow the available route to peer-reviewed publication before a stable, dated, citable version has been established.

The public layer discloses the architecture needed to understand the model: the five factors, the bounded score range, the distinction between prior and posterior assessments, the existence of risk tiers and evidence confidence, and the principle that verification credit is limited and auditable. It does not disclose the implementation parameters that realize those principles.

What happens next

As of 13 July 2026, the quantitative core had not been submitted to a journal. The next intended publication step is a DOI-bearing preprint that establishes a stable, dated, citable version before a later journal submission is considered.

This publication sequence is intentional. It avoids conflicting public versions of the quantitative core, preserves a stable citable record, and allows subsequent results to be interpreted against a defined model version.

Limitations and open questions

• The pilot corpus contains 67 records and covers a 44-day observation window. It supports initial calibration, not generalization.
• Inter-reviewer agreement has not yet been measured and reported publicly.
• Sensitivity, ablation, reproducibility, and baseline-comparison results have not yet been published.
• No third-party implementation or external pilot has yet been documented in a form that can be independently reviewed.
• The quantitative parameter set remains provisional and outside the public disclosure layer.
• Material changes to the maturity or evidentiary status described on this page will be recorded in accordance with the Editorial Policy.

Last reviewed: 13 July 2026