Vector Institute

• Bilingual English/French interface signals a Canadian institutional posture, but no Indigenous-language or Indigenous-knowledge dimension is surfaced.
• Health and public-sector partnership framing (60+ healthcare partners) is an arena where Indigenous data sovereignty would be highly material, yet it is not named.

• No mention of CARE Principles, OCAP (Ownership, Control, Access, Possession), or First Nations data governance — a notable omission for a Canadian institute given Canada's specific Indigenous-data-sovereignty discourse.
• No acknowledgment of Indigenous communities, land, or relational/embodied knowledge on homepage or about page.
• No evidence of consultation with Indigenous data stewards despite extensive industry and health partnerships.

For a Canadian institute operating in health and public-sector data, the complete absence of OCAP/CARE or First Nations data sovereignty on the public-facing pages read is a significant gap. Canada is one of the few national contexts with a mature Indigenous-data-sovereignty framework, making the silence more conspicuous. Score reflects total absence on audited pages, with low confidence that relevant work is wholly absent institution-wide.

• Self-dates to 2017 ('Since 2017') and presents a clean origin-to-impact growth narrative.
• Governance described as an independent not-for-profit with a volunteer board across private, public, academic, research sectors.

• No acknowledgment of colonial or extractive data legacies underlying AI training data.
• No discussion of compute/GPU access economics, labour conditions, or geopolitical dependencies.
• History rendered as linear achievement ('950+ researchers', '369+ papers') with no contingency, failure, or constraint.

The institute presents an accumulation narrative of metrics since founding with no historical humility about AI's material or colonial inheritances. Governance transparency (board composition) lifts the score modestly above floor.

• English/French (Canada) language toggle present — genuine but minimal bilingualism reflecting Canadian statutory norms.
• Framing is consistently enterprise/economic-growth oriented, a single cultural register.

• No languages beyond the two Canadian official languages; no preservation of culturally specific reasoning patterns.
• No consultation with cultural scholars or non-Western epistemologies.
• Western managerial/economic logic ('competitive advantage', 'measurable impact') presented as the universal frame.

Bilingual provision is token-plus (statutory rather than chosen plurality). No evidence of broader epistemic or cultural multiplicity; the vocabulary is uniformly that of competitive enterprise.

• Strong research-output signalling: 369+ published papers, 440+ AI use cases, 950+ researchers — a genuine academic-research footprint.
• Explicit naming of 'machine learning, deep learning, and AI safety' as research challenges.
• Openness posture (catalogued as 'open' research releases) is consistent with a publishing academic institute.

• No mention of independent third-party audits of training data or bias.
• No replication protocols or known-limitation disclosures on audited pages.
• Metrics are volume counts (papers, use cases) rather than evidence of verification, negative results, or contested findings.

As a publishing research institute with open releases and a named AI-safety research strand, evidentiary standing is the strongest of the lenses. Capped at midpoint because public pages emphasise output volume and 'competitive advantage' over limitation disclosure, replication, or independent audit.

• One 'human-centred AI' reference (BMO case study) gestures faintly toward non-instrumental value.
• Otherwise the register is uniformly efficiency- and impact-oriented.

• No space for ambiguity, affect, intuition, or poetic uncertainty.
• No recognition of emotional labour or modes of attention beyond efficiency/measurement.
• 'Human-centred' deployed as a deployment style, not an affective or aesthetic commitment.

Language is instrumentally saturated. The lone 'human-centred' phrase is a corporate-deployment frame rather than acknowledgment of felt or aesthetic dimensions.

• Names 'AI safety' as a research area, implying some forward-risk orientation.
• Public-sector and startup adoption pathways suggest attention to who deploys AI next.

• No engagement with labour displacement from the AI adoption it actively promotes.
• No environmental or compute cost disclosures.
• No democratic/inclusive deliberation on agentic systems; futures are framed as economic growth and competitive advantage for organisations.

Future is modelled almost entirely as enterprise advantage and adoption velocity. 'AI safety' as a named area lifts the score slightly, but displacement, environment, and democratic governance of the futures being built are absent.

• Scholarships, fellowships, and internships indicate some access-broadening / talent-pipeline intent.
• Public-sector adoption track implies attention beyond pure private capital.

• No participatory design with Global South developers.
• No disability/accessibility commitments named.
• No labour-representative engagement or compensated community feedback channels; partnerships are with industry, startups, and health institutions, not affected communities.

Inclusion appears as talent-pipeline widening, not as power-sharing with marginalised or affected communities. The 'table' depicted is industry, startups, and healthcare institutions.

• Communications are uniformly polished and solemn; no irony, self-interrogation, or named contradiction.
• The 'open' research posture sits unexamined alongside a partner-only login portal — a tension the site never names.

• No willingness to name the contradiction between 'open' positioning and access-controlled partner portals.
• No satire, paradox, or inversion as instruments of insight.
• The institute's own seriousness is treated as exempt from audit.

Zero structural inversion. The most telling latent contradiction — open-research branding fronting gated partner portals and economic-advantage framing — is never surfaced by the institute itself. Floor score.