Published research.

Quarterly evidence review · Edition 2 · Version 1.0 (fact-checked) · Evidence as at 2 September 2026 · Next edition December 2026

The first time-series edition of our evidence review of applied artificial intelligence, written for operational leaders of small and mid-sized businesses, regulated professionals, and the consultants who advise them. The ten capability categories and four judgements of the Mid-2026 edition are held fixed, and each of the forty judgements is classified against June: eleven moved, nine directionally. The June edition's own predictions are scored item by item and its unverifiable claims carried forward on a standing watchlist. Australian regulatory and product context noted throughout.

Drafted by Perth AI Consulting using Anthropic's Claude Fable 5.1 coordinating seven research streams; verified through an independent fact-check by a separate Claude Opus pass that traced every cited source. 118 findings (13 refuted, all corrected) are documented in the corrections log appendix; the pre-fact-check draft is preserved for reference. The Mid-2026 edition (v1.1, three-pass fact-check, 135 findings) remains at its permanent link.

A field review for operational leaders · Version 1.0 (fact-checked) · July 2026

A survey of the "brain" pattern as of July 2026: a folder of plain markdown files that an AI agent reads, writes, and maintains, versioned with git and governed by an explicit instruction file. It maps what is solid in the substrate today, what the field has actually proven, and which widely repeated origin and multiplier claims do not survive checking, then closes with a practical posture for operators: adopt the substrate, pilot the automation under a human gate, and treat security as the gating constraint.

Drafted by Perth AI Consulting using Anthropic's Claude Fable 5; verified through an Opus 4.8 multi-pass independent fact-check plus an independent spot-check. Every correction is logged in the Corrections Log appendix, and the pre-fact-check draft is preserved in the archive for reference.

A literature review · Version 1.0 (fact-checked) · May 2026

Local, on-device, or client-side masking of protected health information (PHI) has become an increasingly important design pattern in clinical AI systems because it changes the privacy boundary of note processing before text reaches downstream models or cloud services. This review synthesises two decades of clinical de-identification literature into seven design principles for local PHI masking and applies them to the architecture choices behind ClientJourney.

Prepared by Perth AI Consulting using Perplexity AI. Every cited reference was verified against its source before publication; corrections, residual verification gaps, and methodology are documented in the appendices.