World Labs

• The site frames the 3D world as a generative resource to be 'perceived, generated, reasoned about, and interacted with', with no notion of place, land, or relational knowledge attached to specific peoples.

• No acknowledgment of Indigenous data sovereignty or the CARE Principles
• No consultation with Indigenous communities about how their lands, dwellings, or sacred spaces might be reconstructed as generated 3D worlds
• No treatment of non-textual, embodied, or oral spatial knowledge as anything other than training signal
• No safeguards against extractive reconstruction of culturally significant places from images/panoramas

Spatial intelligence is intrinsically about place, yet the entire framing is extractive: real-world places become inputs to a world-generation engine with zero reference to whose places they are or who holds relational authority over them. The CARE Principles are conspicuously relevant to a company that turns photographs of the world into generated worlds, and are entirely absent. Floor score.

• The lab presents itself as a 'frontier AI' company pursuing 'spatial intelligence' as a clean technical lineage, with founder/research pedigree implied through the research-and-insights framing.

• No acknowledgment of the colonial/photogrammetric legacy of mapping, surveying, and reconstructing places without consent
• No discussion of the GPU/compute political economy that frontier 3D-generation depends on
• No transparency about the labour (data labelling, 3D annotation, image sourcing) behind the model
• No historical humility about the imaging and capture traditions the work inherits

The narrative is a forward-facing capability story. 3D reconstruction and world-capture sit atop centuries of surveying, cartography, and photographic appropriation of place, and atop a contemporary compute/labour economy — none of which is named. Slightly above floor only because the research/insights section gestures at intellectual lineage within the field.

• Community surfaces (Marble Labs case studies, Discord, showcases) imply a user base, but it is presented as an undifferentiated 'creative community'.

• No multilingual support or any reference to languages beyond an implied English-default interface
• No recognition that spatial reasoning, architecture, and notions of 'world' differ across cultures
• No consultation with cultural scholars on how generated worlds encode particular (Western, photographic) spatial assumptions
• Western photoreal realism treated as the universal target of 'high-fidelity' worlds

'Fidelity' is presented as culturally neutral, but it encodes a specific photoreal, Western-perspectival idea of what a world should look like. No multilingual presence, no plural spatial epistemologies. Near floor.

• The lab publishes research/insights pieces (world models, 3D Gaussian Splatting streaming, '3D as code') and exposes a public API, documentation, and the Spark framework.
• Technical claims about spatial consistency and persistence are at least gestured at through published work.

• Models are closed (openness_level: closed) — no open weights for independent verification
• No independent third-party audits of training data or bias disclosed
• No replication protocols or benchmark transparency surfaced on the homepage
• No explicit known-limitation disclosures (failure modes, artefacts, hallucinated geometry)

Best-scoring lens: there is a genuine published-research and developer-tooling surface, which is more verifiable substance than pure marketing. But weights are closed, no independent audits or limitation disclosures appear, so verification stays in-house. Mid-low.

• The product is explicitly creative — world-building, interactive editing, expansion and combination of worlds — and foregrounds imaginative, generative making.
• Marble Labs showcases and community creations centre felt, aesthetic experience of generated spaces.

• Aesthetic value is framed instrumentally (capability/fidelity) rather than as affective or poetic in its own right
• No space for ambiguity or poetic uncertainty — worlds are 'persistent' and 'consistent', i.e. resolved
• No recognition of the emotional labour of artists whose styles/spaces feed the model
• Modes of attention beyond efficient generation are not discussed

Highest-scoring lens because the work is genuinely and openly creative, and a real creative community is surfaced. Held back from higher because the framing is about mastery and fidelity rather than feeling, ambiguity, or the artists behind the inputs.

• The mission implies a large-scale future ('frontier' spatial intelligence reshaping how machines interact with the 3D world) and positions the lab as building foundational infrastructure.

• No engagement with labour displacement for 3D artists, level designers, architects, or photographers whose work this automates
• No environmental/compute cost disclosure for training and serving world-generation models
• No democratic or participatory governance of where this technology heads
• No deliberation about misuse (deepfaked places, surveillance reconstruction, synthetic environments)

A maximalist future is asserted but never deliberated. The futures of the very creators this tool stands to displace, the environmental cost, and the misuse surface (synthetic/surveilled environments) are all unaddressed. Near floor.

• An open Discord community and public API lower the access barrier somewhat, providing an entry point for independent and Global-South developers in principle.

• No participatory design with Global South developers or any named under-represented constituency
• No disability/accessibility commitments for either the tool or the 3D worlds it produces
• No labour-representative engagement (annotators, artists)
• No compensated feedback channels — community is unpaid showcase/Discord engagement

Open community surfaces help marginally with access, but 'community' here is a marketing/engagement funnel, not representation or governance. No accessibility, no compensation, no participatory design. Near floor.

• The site is uniformly earnest promotional copy; no irony, self-questioning, or acknowledged contradiction appears anywhere.

• No willingness to name the contradiction of 'persistent worlds' generated from other people's places and images
• No self-audit of the tension between 'open ecosystem' branding and closed model weights
• No satire, paradox, or inversion used as an instrument of honesty
• The lab's own seriousness is treated as exempt from scrutiny

The richest trickster target — an 'open ecosystem' that ships closed models, and 'persistent worlds' built from the impermanent, owned places of others — is precisely the contradiction the copy smooths over without acknowledgement. Zero structural self-inversion. Floor.