Trust & Transparency

    Research Methodology

    How every piece of research published by Cassandra Research is produced, validated, and held accountable.

    Our standard of proof

    Cassandra Research operates in YMYL (Your Money or Your Life) domains — Australian tax, law, and medical reasoning. Every output must be defensible against the same evidentiary standard as a signed partner file note. We do not publish AI output that has not been traced to primary source and reviewed by a qualified human expert.

    01

    Primary-source retrieval

    Every research question is grounded in primary Australian sources: legislation.gov.au, ato.gov.au, austlii.edu.au, Federal Court and High Court judgments, and Treasury releases. We do not retrieve from secondary blog content or paraphrased summaries. Source URIs are captured at retrieval time and persisted with the response.

    02

    AI synthesis with cited inference

    Retrieved passages are passed to a tightly-scoped reasoning prompt that produces a structured answer with inline citations. The model is constrained to refuse questions where retrieval confidence is below threshold rather than hallucinate. Model version, prompt version, and retrieval set are logged for every published article.

    03

    Expert human review

    Tax content is reviewed by a qualified accounting professional (CA/CPA) before publication. Legal content is reviewed by an Australian-admitted lawyer. The reviewer's role is to (a) verify each citation resolves to the claim made, (b) confirm the reasoning is current as at the review date, and (c) flag any drafting that overstates certainty.

    04

    Citation & freshness audit

    Before publication, an automated audit confirms every citation link returns HTTP 200, every legislative section reference exists in the current consolidated Act, and every case citation matches the AustLII record. Articles carry a visible 'Last reviewed' date and are re-audited at least every 6 months, or sooner after a relevant legislative amendment.

    05

    Correction & accountability

    If an error is identified post-publication — by a reader, an internal review, or a third party — we correct the article, log the correction in the public Corrections Register, and update the dateModified. We do not silently edit. Material corrections are linked from the affected article.

    Model governance

    • Aligned to the NIST AI Risk Management Framework (AI RMF 1.0) across Govern, Map, Measure and Manage.
    • Tracking ISO/IEC 42001:2023 AI Management System controls.
    • Compliant with the Australian Privacy Act 1988 and the 13 Australian Privacy Principles.
    • Zero-log architecture for client research sessions. TLS 1.3 in transit, AES-256 at rest.
    • Model versions and prompt revisions are version-controlled. Material changes are disclosed.

    What this is not

    Cassandra Research publishes research, not advice. Our articles are general in nature, current as at the published review date, and do not take into account a specific taxpayer or client circumstances. They are intended for use by qualified professionals as a research input — not as a substitute for professional judgment, formal advice, or a private ruling request to the ATO or a written legal opinion.

    Found an error? Email service@cassandraresearch.com. We acknowledge all reports within one business day and publish corrections in the Corrections Register.

    Last reviewed: 13 September 2026 · Owned by Imran Hussain, Founder & CEO