Preferred Networks

• No reference to Indigenous data sovereignty, CARE principles, or relational/embodied knowledge anywhere in the visible homepage text.
• The framing 'make the real world computable' treats reality as an extractive, measurable substrate rather than a relational web of knowledge-holders.

• No acknowledgment of the Ainu, Ryukyuan/Okinawan, or other Indigenous peoples of the Japanese archipelago whose lands and knowledge intersect any 'real world' data programme.
• No data sovereignty or consent framework for community-held knowledge.
• No recognition of non-textual or oral knowledge traditions in foundation-model training.

Total absence. The mission language is computationally totalising ('make the real world computable') with no countervailing recognition of knowledge that resists or precedes computation. Score floored at 1.

• Vertical integration 'from chips to applications' is named, which implicitly touches the geopolitical economy of semiconductors — but it is framed as a competitive moat, not as a historical or labor reckoning.
• Heavy reliance on legacy-industrial Japanese partners (Mitsubishi Heavy Industries, Toyota) situates the lab in a specific postwar industrial lineage that is presented as pedigree rather than examined.

• No acknowledgment of colonial or extractive data legacies.
• No transparency about GPU/chip supply geopolitics or the labor embedded in the hardware stack despite owning that stack.
• No historical humility about AI's inheritances; history appears only as corporate provenance.

The lab uniquely controls the full hardware-to-application chain, which makes its silence on the human and geopolitical history of that chain notable. It scores 2 rather than 1 because the value-chain framing at least makes the material substrate visible, even if unexamined.

• Genuine bilingual operation (Japanese-first site with an English mirror) and an in-house Japanese-language foundation model (PLaMo) represent real, non-token investment in a non-English epistemic lineage.
• PLaMo and a Japanese-NLP focus position the lab as a counterweight to Anglophone-default model development.

• Bilingualism stops at Japanese/English; no Ainu, Ryukyuan, or broader Asian-language plurality.
• No discussion of preserving culturally specific reasoning patterns versus flattening them into a translation-friendly model.
• No consultation with cultural or linguistic scholars is evidenced.

Scores above the floor — a Japanese-native foundation model is a substantive multiplicity asset rare among labs. But it is national-linguistic, not cross-cultural in the plural sense, and the wisdom dimension (reasoning patterns, not just tokens) is absent. 3.

• ISO/IEC 27001 certification covering AI product development and delivery is a verifiable third-party standard.
• Identity as a research lab with 'research releases' implies some published, citable output and a scientific posture.
• A named foundation model (PLaMo) gives an evaluable artifact.

• ISO 27001 is an information-security standard, not an audit of training data, bias, or model behavior — it is being used as a proxy for responsibility it does not cover.
• No independent audits of training data or bias, no third-party replication protocols, no explicit known-limitation disclosures on the homepage.
• 'Hybrid' openness with no visible open-weights commitment limits external verifiability.

A real certification and a research identity lift this above the lower lenses, but a security certification substituting for evidentiary-AI rigor is exactly the conflation the lens is built to catch. 4.

• An 'entertainment' vertical is named among the nine sectors, the only nod toward affective or creative domains.
• The phrase 'create the future together' gestures at a collaborative, quasi-aspirational register.

• No acknowledgment of affective, intuitive, or aesthetic dimensions of the technology.
• No space for ambiguity or poetic uncertainty; the dominant register is engineering-efficiency.
• Efficiency itself is a named focus area, crowding out any mode of attention beyond optimization.

Entertainment-as-a-vertical is commercial, not perceptual. The lab's stated emphasis on efficiency actively narrows attention. 2.

• Explicit forward orientation ('create the future together') and engagement with 'Physical AI' and mission-critical applications shows the lab models futures at the infrastructural level.
• Robotics and manufacturing focus puts the lab squarely in the labor-automation frontier, making the future it shapes materially consequential.

• No engagement with labor-displacement risk despite a robotics/manufacturing-automation core — a conspicuous omission given the partners (Toyota, MHI).
• No environmental or energy-cost disclosure despite owning chips and computing infrastructure.
• No democratic or inclusive deliberation about who the 'together' in 'create the future together' includes; it appears to mean large industrial partners.

The lab models futures explicitly and at scale, but only corporate-industrial futures; the people displaced or the planet powering the chips are absent from the modelled future. 3.

• No participatory design, accessibility, labor-representation, or compensated-feedback mechanism appears anywhere in the visible text.

• No Global South developer engagement.
• No disability/accessibility commitment.
• No labor-representative engagement despite an automation-heavy portfolio.
• The 'together' of the mission is operationalised exclusively through blue-chip corporate partners (Toyota, Mitsubishi Heavy Industries).

Marginalised voices are wholly absent, and the partnership roster actively signals the opposite — power consolidating with the largest incumbents. Floored at 1.

• No irony, self-interrogation, paradox, or willingness to name internal contradiction; the register is uniformly solemn corporate-engineering.
• The vertical-integration narrative is presented as unambiguous virtue with no inversion permitted.

• No space where the official narrative is tested by its opposite.
• The unexamined contradiction — a lab that 'makes the real world computable' while building robots that replace the workers in that real world, branding total value-chain control as safety — is never surfaced.
• No mechanism treats the lab's own seriousness as auditable.

Zero trickster capacity. The most striking inversion writes itself — owning the entire chain is framed as responsibility when it is equally a description of unaccountable concentration — yet the lab shows no awareness of it. 1.