What somebody can buy on day one.

Canadian data residency is not sovereignty — the Government of Canada says so in its own guidance. The U.S. CLOUD Act lets American authorities compel a U.S.-headquartered provider to produce data held in a Canadian region, without notice. So "it is in a Canadian region" is not an answer to an OCAP possession question. The commitment has to be architectural: the community holds the corpus; derived artifacts — indexes, embeddings, extracted structure — are community property under the same terms as the source documents, because that is the loophole every AI vendor currently leaves open; and there is a repatriation and destruction clause that has been rehearsed, not just written.

Fair, and asked in good faith. The differentiator is not ownership percentage — Dalian met 51% on paper. It is who does the work. Publish Indigenous labour content as a measured number, report it per contract, and invite the pre-award audit rather than waiting for it. R8dius set the visible benchmark at 80% Indigenous staff. If the honest answer today is "we are small and building toward that," say that, with a date.

Do not contest the relationship — it is real. Contest the artifact. MNP and Deloitte’s Indigenous practices are advisory: the deliverable is a report, a rights study, a governance framework. Nobody operates it afterward. The gap is between advice about data and a system that runs on the data — the Indian Lands Registry has no API, First Nation Profiles has no export, and none of the Big Four is going to write that software. With MNP the right posture is usually partnership, not displacement.

Name the envelope on the first call. Budget 2025 cut Indigenous Services and Crown-Indigenous Relations by roughly $2.3B by 2030 while adding capital — so band operating budgets are the wrong pocket. Specific Claims funds digitization and database creation directly. Listen, Hear Our Voices funds up to $100,000 per project for digitizing textual records. RLEMP funds land management capacity with continuous intake. If a pitch cannot be mapped to a named programme with an intake window, the answer should be no.

Concede the premise, because it is true. The First Nations Technology Council named it digital colonialism in April 2025. The extractive default — ingest community records, improve a vendor’s model, sell it back — is a real pattern, not a hypothetical. What is defensible is narrow: retrieval over records the community already owns, answering questions the community already asks, with citations back to the source page. Not a model trained on sacred or governance knowledge. Not generation that speaks for the Nation. And the scoping decision belongs to the community before the build, not in a consultation after it.