The Demiton Classification Framework.

The Demiton Classification Framework turns rows from an Australian government contract-disclosure register into a stable, machine-comparable set of classification dimensions. It is intended for analysts, journalists, government policy researchers, and platform builders who need a defensible basis for comparing public procurement disclosures across firms, projects, and time. The framework's source-of-truth schema is data.qld.gov.au's Contract Disclosure Reports; other Australian disclosure registers will be supported as parsers are added.

data.qld.gov.au Contract Disclosure Reports are the framework's primary source. Each row carries supplier name, supplier postcode, contract value, procurement method, number of offers sought, variation flag, and contract title. The framework operates on these fields directly. The Demiton platform layers additional data — AusTender, ABR ownership chains, BOM weather, CKAN planning approvals — on top of the framework when applied to project-level analysis. Those additional sources are out of scope for the framework itself.

Each contract is classified along eight dimensions, each independent and rule-based. Value tier (under $50k, $50k–$1M, $1M–$10M, $10M–$100M, over $100M). Supplier locality (metro, regional, interstate, international — derived from supplier postcode and address). Procurement competition tier (limited / sole-source vs panel vs open tender — derived from procurement method and offers sought). Variation intensity (whether contracts have variations and at what scale relative to baseline). Engagement category (construction, professional services, ICT, goods, advisory — derived from title and description). Repeat-supplier flag (whether the supplier appears multiple times for the same delivery authority). Panel-vs-direct (whether the contract is panel-arrangement or direct award). Confidentiality flag (whether commercial-in-confidence redactions apply).

Each dimension uses explicit, testable decision rules. Value tier banding is fixed by AUD thresholds. Locality is determined by postcode lookup (state and metro/regional classification from Australia Post's locality data). Procurement competition is derived from the explicit procurement method field plus the number of offers sought. Variation intensity uses the disclosure register's variation flag and any quantified variation amount. Engagement category uses keyword and title pattern matching against an explicit category dictionary, falling back to UNKNOWN where no rule fires. Panel-vs-direct uses the explicit procurement method field. Confidentiality uses the explicit confidentiality flag. All rules are explicit — same input always produces the same output.

The framework reports a confidence level alongside each dimension. High confidence: the source field is present, unambiguous, and the rule fires cleanly. Medium confidence: the source field is present but ambiguous (e.g. partial postcode, generic title), and a fallback rule produces a best-effort classification. Low confidence: the source field is missing or contradictory, and the dimension is reported as UNKNOWN with the reason.

The framework's outputs are only as good as the disclosure register inputs. Suppliers that misreport their postcode produce wrong locality. Procurement methods recorded inconsistently across agencies produce uneven competition-tier comparisons. The framework does not resolve ownership chains — a Queensland-postcoded supplier wholly owned by an interstate parent classifies as local. Indigenous certification is not in this framework's scope; that data integration is platform-side and uses Supply Nation certification (research in progress). The framework does not classify subcontractor lineage; only first-party contract awards are in scope.

The classification framework is reviewed quarterly. Methodology revisions are versioned (v1, v2, ...) and pinned at /methodology/v1 etc. for permanent academic citation. The classifier package on PyPI follows semantic versioning; methodology version aligns with the package's major version. Changes are tracked in the GitHub repository's CHANGELOG.md.

Methodology proposals, dimension additions, and rule refinements are welcome via pull request at github.com/demitonapp/classifier. Contributions that change classification outputs require test coverage demonstrating the new behaviour and an explanatory note in the PR. Rule changes that affect existing classifications are deferred to the next major version to preserve reproducibility of prior analyses.

For academic, policy, and journalistic citation, use: Demiton (2026). The Demiton Classification Framework: A procurement classification framework for Australian government contract disclosure data. v1. https://demiton.io/methodology/v1. The pinned URL guarantees that future revisions don't change the cited methodology.