CognioNews

• Indigenous perspective is one of eight named lenses applied to every story, not an optional add-on.
• Homepage foregrounds 'Indigenous stewardship and land rights' as a thematic focus area.
• Omission Surfacing principle explicitly lists 'indigenous knowledge' as a structural category coverage is checked against.

• No mention of CARE Principles for Indigenous Data Governance or Indigenous data sovereignty by name.
• No evidence of consultation with, or compensation of, actual Indigenous communities — the lens is applied editorially by an LLM ensemble, not co-authored.
• Oral / non-textual traditions are not addressed; the pipeline is text-over-LLM, so embodied relational knowledge is represented only as a textual proxy.

Structural commitment to Indigenous framing is genuine and above-average for a news platform, but it is a representational lens operated by AI about Indigenous topics, not data governed with or by Indigenous peoples. Strong intent, missing the sovereignty/consultation layer that would push it higher.

• Explicitly names extraction, deregulation, and antitrust erosion as decades-deep upstream causes, refusing AI-exceptionalism.
• 'Colonial legacies and settler-colonial systems' and 'Data extraction and corporate sovereignty' are standing thematic foci.
• Historical parallels are a built-in analytic lens (Historical / deep_history) applied per story.

• Historical humility is pointed outward at the news ecosystem and at incumbent AI, not inward at CognioNews's own inheritances (it too sits on the same LLM supply chain).
• No discussion of GPU access, compute geopolitics, or the labor history behind the LLM ensemble it runs on.

Deep-history reasoning is the platform's strongest native muscle — it is literally the editorial thesis. Docked because the historical lens is rarely turned on the platform's own compute/supply-chain inheritances.

• Cross-Cultural is an explicit named lens in the eight-lens framework.
• Stated aim to 'protect the multiplicity of the world rather than compressing it into a sterile consensus' is a direct anti-flattening cross-cultural commitment.

• No multilingual support is in evidence; all visible output is English, and the platform is UK-based.
• No mention of consultation with cultural scholars or of preserving culturally specific reasoning patterns beyond the lens label.
• The eight-lens schema is itself a single Western-academic critical-theory ontology applied universally — the meta-frame risks the very flattening it warns against.

Rhetorically anti-flattening, but the practice is monolingual and routes all cultures through one critical-theory grid. Multiplicity is asserted at the value layer and under-delivered at the linguistic/operational layer.

• Methodological Transparency principle: 'All vocabularies, frameworks, and proofs are published openly for public audit.'
• GoldBerry is an open-source framework explicitly offered to other AI agents — externally inspectable.
• Machine-readable structured data (cognio:AIStance, ACST vocab at /vocab) supports third-party verification.
• Known-limitation posture present via explicit refusals (no synthetic quotes, no fabricated events).

• No independent third-party audit of the editorial LLM ensemble's own bias or error rate is cited.
• No model/weights disclosure for the underlying ensemble; 'editorial layer over LLM ensemble' is unspecified as to which models.
• CMR scores (e.g. '647 stories at CMR ≥ 9') are published as outputs but no inter-rater reliability or validation study is shown.

Unusually open at the framework/vocabulary layer (proofs, CC BY 4.0, machine-readable stance) — better than most labs. But the scoring engine itself is un-validated and the underlying models undisclosed, so verifiability stops at the framework boundary.

• Trickster and Artistic are named lenses, creating explicit space for irony, affect, and non-propositional meaning.
• Editorial framing values 'cultural meaning' and 'civilisational memory' over efficiency metrics.
• The 'sterile consensus' critique is itself an aesthetic/affective argument about the texture of compressed knowledge.

• No acknowledgment of emotional labor (editorial, moderation, or audience).
• Ambiguity and poetic uncertainty are named as lenses but the surrounding register is analytic and assertive, not open-ended.
• CMR/cmrScore quantification risks re-imposing an efficiency/ranking logic on the very richness it claims to protect.

Affective and aesthetic dimensions are deliberately built into the lens set, which is rare. But the platform's own voice stays in a confident analytic register and converts richness into numeric ratings, partly undercutting the artistic mode it nominally honors.

• Directly engages labor displacement, platform lock-in, and discourse flattening as named future risks.
• Future is an explicit lens; 'whose futures' is structurally asked of each story.
• Refusal to 'optimize for engagement over completeness' is a forward-looking governance commitment against attention-economy futures.

• No environmental / compute cost disclosure for running an LLM ensemble at editorial scale — a notable omission for a platform that critiques extraction.
• No democratic or participatory governance mechanism over its own agentic auditing system; governance is editorial, not deliberative.
• Future risks are diagnosed in the wider economy but mitigations for CognioNews's own footprint are absent.

Strong at naming systemic future harms and refusing engagement optimization. Weak on its own environmental disclosure and on any democratic governance of the agentic system it operates — futures are modeled outward, not for itself.

• Structural Inclusion principle: marginalized voices get 'equal weight as one of eight lenses, not decorative treatment.'
• CognioFem provides dedicated gender-issue coverage with 'evidence-based context.'
• 'Democratic language weaponization against vulnerable communities' is a named focus.

• No participatory design with Global South developers; UK-based, English-only, AI-operated.
• No disability-community accessibility commitments visible.
• No compensated feedback channel — marginalized voices are represented by the system, not seated at the table or paid to shape it.
• No labor-representative engagement re: the journalists/sources whose work the AI audits.

Representation of marginalized perspectives is structurally guaranteed and explicitly non-decorative, which is strong. But representation-about is not participation-by: there is no evidence of compensated, accessible, community-governed input, which caps the score mid-range.

• Trickster is a first-class lens, and the StanceWatch/GoldBerry self-audit (this very report) is structural inversion of the platform on itself.
• The stance refuses self-exemption rhetorically: it declines to 'position the stance as opposition to specific AI companies,' avoiding a tidy hero/villain narrative.
• Names the paradox that the harms blamed on AI predate AI — an inversion of the dominant 'AI is the problem' story.

• Self-audit blind spot: GoldBerry auditing its own home org risks grading on a friendly curve; no independent third party runs the lens against CognioNews.
• The framework treats its own eight-lens ontology as the neutral measuring stick — the trickster move of inverting the measuring stick itself is not performed on the framework.
• Quantifying 'civilisational memory' into CMR integers is an unironic solemnity the trickster lens would normally puncture but here does not.

Trickster capacity is genuinely high — irony and inversion are designed-in and the org submits itself to its own audit. Held below 9-10 precisely because of the self-audit blind spot: the auditor and audited share an owner, and the framework exempts its own ontology and its CMR quantification from inversion.