Physical Intelligence

• No reference whatsoever to Indigenous knowledge, data sovereignty, or community consultation appears in any public-facing text.

• No acknowledgment of CARE Principles or Indigenous data sovereignty.
• Robotic 'open-world generalization' is asserted as universal without naming whose physical worlds, environments, or embodied practices are modelled or excluded.
• Human-video transfer learning ('transfer from human videos to robotic tasks') raises provenance and consent questions about whose bodies and labour are captured, none of which are addressed.

Total absence. The lab's entire frame treats 'the physical world' as a single neutral substrate to be generalized over, which is itself an erasure of plural embodied and relational ways of knowing. Human-video training data is mentioned with zero provenance or consent framing. Floor score of 1.

• The lab names its investor and capital lineage explicitly (OpenAI, Sequoia Capital and others), giving a thin trace of its geopolitical-economy position.
• Open weights/code release for π0 gestures at a reproducibility lineage in ML research culture.

• No acknowledgment of colonial or extractive data legacies behind 'human videos' or web-scale pretraining.
• No discussion of the GPU/compute supply chain, mineral or labour economy underpinning robotics hardware.
• No regulatory or historical context; the narrative is a clean technical progression (π0 to π0.7) with contingencies erased.

A small credit above floor for naming the capital lineage, which is more historical self-location than many labs offer. But the dominant register is teleological inevitability ('a Cambrian explosion of robotics applications'), with the material and historical conditions of production fully absent.

• No multilingual content, cultural scholarship, or perspective plurality is present; the site is English-only and technically framed.

• No multilingual support beyond English token presence.
• No engagement with culturally specific embodiment, gesture, or task ontologies — 'any robot' and 'diverse tasks' are presented as culture-free.
• Western efficiency/scaling logic is treated as the universal grammar of physical intelligence.

The promise of a single model that controls 'any robot' for 'diverse tasks' is the cross-cultural flattening lens made literal: heterogeneous physical, cultural, and embodied practice collapsed into one optimization target with no acknowledgment that this is a choice. Floor score.

• Open weights and code release for π0 and associated models genuinely supports third-party verification and replication — the strongest single signal in the audit.
• Specific, falsifiable technical claims are made (memory for tasks 'exceeding ten minutes', training from 'few hours of data', efficient online RL).
• Model versioning (π0 through π0.7) provides a public, trackable research progression.

• No independent or third-party audit of training data, bias, or safety is referenced.
• No disclosure of known limitations or failure modes in the public-facing text.
• Performance framings ('open-world generalization', 'Cambrian explosion') outrun the evidence presented and lack quantified caveats.

Highest lens score by a wide margin. Open weights for a closed-rated lab is a real verifiability commitment and the claims are concrete. Held below 7 because limitation disclosure, independent audit, and quantified caveats are absent on the public surface — open weights enable verification but the lab does not itself perform or surface it.

• The 'Cambrian explosion' metaphor and the π glyph branding show some aesthetic and rhetorical intent.
• An interest in embodiment and the physical world brushes adjacent to affective/intuitive dimensions without engaging them.

• No acknowledgment of affective, intuitive, or emotional dimensions of human-robot interaction.
• No space for ambiguity or poetic uncertainty; register is uniformly confident and technical.
• No recognition of emotional labour or modes of attention beyond efficiency and capability.

A flicker of metaphor (Cambrian explosion) lifts this just off the floor, but the metaphor serves hype, not reflection. Physical embodiment is treated purely as a control problem; the felt, tacit, attentional dimension of moving in the world is entirely absent.

• The lab is explicit and concrete about the future it intends to bring about: general-purpose robots deployed at scale via a 'Physical Intelligence Layer' with partners 'already solving real-world problems'.
• Naming deployment partners and a productization thesis makes its future-shaping ambitions legible rather than hidden.

• Zero engagement with labour displacement risks despite building general-purpose robotic automation — the single most conspicuous omission.
• No environmental or compute cost disclosure.
• No democratic or participatory governance of agentic embodied systems; the future is shaped by the lab and its investors alone.

Scored above floor only because the lab is unusually concrete about whose future it is building (a deployed robotic-automation layer). But a lab explicitly automating physical labour at scale that says nothing about labour displacement, environmental cost, or democratic governance is shaping powerful futures unilaterally. The clarity of ambition makes the governance silence more, not less, damning.

• References 'community development' and 'community engagement via social platforms', but only in the sense of developer/social-media reach.

• No participatory design with Global South developers or affected workers.
• No disability-community accessibility framing, despite embodied/assistive robotics being a natural site for it.
• No labour representative engagement or compensated feedback channels for those whose work robots would perform — the 'real-world problems' being solved are unnamed and the affected parties absent.

The only 'community' present is the open-source developer/follower community. Workers whose physical labour is the explicit target of general-purpose robots, disabled users, and Global South stakeholders are entirely outside the table. Floor score.

• No irony, self-interrogation, paradox, or willingness to test the official narrative against its opposite appears anywhere in the public text.

• No naming of the central contradiction: a 'closed' company that open-sources weights while saying nothing about the deployment risks of what it open-sources.
• No acknowledgment that 'bringing general-purpose AI into the physical world' is also bringing general-purpose automation into physical labour markets.
• The lab's seriousness is treated as exempt from audit; the absurd edge of 'any robot, any task' is never followed to where it implicates jobs, consent, or control.

Pure solemn promotional register. The richest available inversion — that a lab named for 'physical intelligence' is silent on the physical-world consequences (bodies, labour, consent) of its own technology — is never touched. Floor score.