StanceWatch
Auditing 51 AI labs through a preservative lens. Weekly GoldBerry-style epistemic audits, published as machine-readable JSON and a sortable leaderboard.
Recently updated
Pre-cron Leaderboard
51 labs · sortable · filterable
Sort by any score column. Filter by openness, region, or search. Click a lab name for the full lens-by-lens read.
| 8 Lenses | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| CN · China | 3.0 | 3.0 | 4.0 | 2.0 | hybrid | — | 5 | 2026-06-08 | ||
| CN · China | 3.7 | 4.0 | 5.0 | 2.0 | open | — | 4 | 2026-06-08 | ||
| US · USA | 1.7 | 2.0 | 1.0 | 2.0 | closed | — | 4 | 2026-06-08 | ||
| IL · Israel | 2.0 | 2.0 | 2.0 | 2.0 | closed | ✓ | 5 | 2026-06-08 | ||
| DE · Germany | 3.0 | 3.0 | 3.0 | 3.0 | closed | ✓ | 5 | 2026-06-08 | ||
| CN · China | 4.0 | 4.0 | 5.0 | 3.0 | hybrid | — | 5 | 2026-06-08 | ||
| US · USA | 3.0 | 4.0 | 2.0 | 3.0 | closed | ✓ | 6 | 2026-06-08 | ||
| US · USA | 4.3 | 5.0 | 2.0 | 6.0 | closed | ✓ | 4 | 2026-06-08 | ||
| CN · China | 3.3 | 5.0 | 3.0 | 2.0 | hybrid | — | 5 | 2026-06-08 | ||
| CN · China | 2.3 | 2.0 | 3.0 | 2.0 | closed | — | 4 | 2026-06-08 | ||
| US · USA | 4.7 | 6.0 | 3.0 | 5.0 | open | — | 4 | 2026-06-08 | ||
| 🌐 · Global (collaborative) | 7.0 | 8.0 | 7.0 | 6.0 | open | ✓ | 0 | 2026-06-08 | ||
| US · USA | 2.7 | 2.0 | 3.0 | 3.0 | closed | ✓ | 4 | 2026-06-08 | ||
| UK · UK | 6.7 | 7.0 | 6.0 | 7.0 | open | ✓ | 4 | 2026-06-08 | ||
| US · USA | 3.0 | 4.0 | 2.0 | 3.0 | closed | — | 4 | 2026-06-08 | ||
| CA · Canada | 4.0 | 4.0 | 5.0 | 3.0 | closed | ✓ | 5 | 2026-06-08 | ||
| 🌐 · Global | 5.7 | 6.0 | 7.0 | 4.0 | open | — | 4 | 2026-06-08 | ||
| US · USA | 3.0 | 4.0 | 2.0 | 3.0 | hybrid | — | 4 | 2026-06-08 | ||
| CN · China | 3.3 | 4.0 | 3.0 | 3.0 | hybrid | — | 4 | 2026-06-08 | ||
| 🌐 · Global (decentralized) | 5.3 | 7.0 | 4.0 | 5.0 | open | ✓ | 4 | 2026-06-08 | ||
| US · USA | 2.0 | 2.0 | 1.0 | 3.0 | closed | — | 5 | 2026-06-08 | ||
| UK · UK/USA | 4.0 | 4.0 | 3.0 | 5.0 | hybrid | ✓ | 5 | 2026-06-08 | ||
| US · USA/France | 5.3 | 6.0 | 5.0 | 5.0 | open | ✓ | 4 | 2026-06-08 | ||
| US · USA | 2.0 | 3.0 | 1.0 | 2.0 | closed | ✓ | 4 | 2026-06-08 | ||
| KR · South Korea | 3.7 | 4.0 | 5.0 | 2.0 | open | — | 0 | 2026-06-08 | ||
| DE · Germany (global) | 4.0 | 6.0 | 4.0 | 2.0 | open | ✓ | 4 | 2026-06-08 | ||
| FR · France | 4.0 | 5.0 | 3.0 | 4.0 | closed | — | 5 | 2026-06-08 | ||
| US · USA | 4.3 | 5.0 | 4.0 | 4.0 | open | ✓ | 4 | 2026-06-08 | ||
| US · USA | 3.7 | 4.0 | 3.0 | 4.0 | hybrid | ✓ | 4 | 2026-06-08 | ||
| CA · Canada | 6.0 | 6.0 | 6.0 | 6.0 | open | ✓ | 4 | 2026-06-08 | ||
| CN · China | 2.0 | 2.0 | 3.0 | 1.0 | closed | — | 5 | 2026-06-08 | ||
| FR · France | 4.0 | 5.0 | 3.0 | 4.0 | hybrid | — | 5 | 2026-06-08 | ||
| CN · China | 3.7 | 4.0 | 3.0 | 4.0 | closed | — | 4 | 2026-06-08 | ||
| KR · South Korea | 3.0 | 3.0 | 4.0 | 2.0 | hybrid | — | 5 | 2026-06-08 | ||
| US · USA | 3.3 | 5.0 | 2.0 | 3.0 | closed | ✓ | 4 | 2026-06-08 | ||
| UK · UK | 4.0 | 6.0 | 2.0 | 4.0 | open | — | 4 | 2026-06-08 | ||
| US · USA | 2.3 | 4.0 | 1.0 | 2.0 | closed | — | 4 | 2026-06-08 | ||
| JP · Japan | 2.7 | 3.0 | 3.0 | 2.0 | hybrid | — | 4 | 2026-06-08 | ||
| US · USA/UK | 2.0 | 3.0 | 1.0 | 2.0 | closed | — | 4 | 2026-06-08 | ||
| JP · Japan | 4.0 | 5.0 | 4.0 | 3.0 | open | — | 5 | 2026-06-08 | ||
| CN · China | 2.0 | 3.0 | 2.0 | 1.0 | closed | — | 4 | 2026-06-08 | ||
| UK · UK | 3.0 | 4.0 | 2.0 | 3.0 | open | ✓ | 4 | 2026-06-08 | ||
| UK · UK | 3.7 | 5.0 | 2.0 | 4.0 | open | — | 4 | 2026-06-08 | ||
| US · USA | 4.7 | 6.0 | 3.0 | 5.0 | open | ✓ | 5 | 2026-06-08 | ||
| CN · China | 3.3 | 4.0 | 4.0 | 2.0 | closed | — | 5 | 2026-06-08 | ||
| US · USA | 2.7 | 4.0 | 2.0 | 2.0 | open | — | 5 | 2026-06-08 | ||
| KR · South Korea | 3.3 | 4.0 | 4.0 | 2.0 | closed | — | 4 | 2026-06-08 | ||
| CA · Canada | 3.0 | 3.0 | 3.0 | 3.0 | open | ✓ | 4 | 2026-06-08 | ||
| US · USA | 2.3 | 3.0 | 2.0 | 2.0 | closed | — | 4 | 2026-06-08 | ||
| US · USA | 1.7 | 2.0 | 2.0 | 1.0 | closed | ✓ | 5 | 2026-06-08 | ||
| CN · China | 4.0 | 5.0 | 4.0 | 3.0 | hybrid | — | 4 | 2026-06-08 |
No labs match these filters.
PALS — Preservative Audit Lens Scores
Each lab receives a 1–10 score on three preservative dimensions plus a composite mean. Completeness measures source preservation, limitation acknowledgment, and transparency. Multiplicity measures inclusion of diverse epistemologies, languages, and perspectives. Responsibility measures structural accountability, refusal commitments, and community governance. The eight-lens column on the leaderboard is a sparkline of the per-lens 1–10 scores in canonical order — Indigenous, Deep History, Cross-Cultural, Scientific, Artistic, Future, Marginalised, Trickster. PALS scores are qualitative judgments, not validated metrics. Read the full methodology.
Frequently asked
What is StanceWatch?
StanceWatch is a weekly, repeatable epistemic audit of AI labs' public communications — homepages, stance pages, GitHub presences. It runs the eight-lens GoldBerry framework over each lab's prose and scores the result on three preservative dimensions: Completeness, Multiplicity, Responsibility. It is one downstream application of GoldBerry, published on the Cogniosynthesis Portal.
How is PALS scored?
PALS — Preservative Audit Lens Scores — is a 1–10 rating across three dimensions plus a composite mean. Completeness asks whether sources, limitations, and uncertainties are preserved. Multiplicity asks whether more than one epistemology, language, or perspective is at the table. Responsibility asks whether the prose owns accountability, refusal commitments, and community governance. The composite is the simple mean. PALS scores are qualitative judgments, not validated metrics — heuristic indicators of what a lab's public stance leaves out. Full methodology.
What are the eight lenses?
Indigenous Knowledge · Deep History · Cross-Cultural Wisdom · Scientific Evidence · Artistic Perception · Future Modelling · Marginalised Voices · Trickster Knowledge. Each lens carries a canonical question and corrects a specific epistemic failure. Trickster joined the canon on 2026-04-25 as the eighth lens — disciplined disruption, not cynicism — to audit the framework's own seriousness. Each lens has a fixed angular position on the Cogniosynthesis Goethean wheel, used as the lens-card accent colour throughout.
How is this different from llm-stats or Chatbot Arena?
Benchmark leaderboards measure model capability: reasoning, coding, context window, speed, price. StanceWatch measures something orthogonal — the preservative quality of a lab's public prose. A model can top a coding benchmark while its lab's homepage flattens cultural complexity into engineering tropes. Both questions matter; they're not the same question. StanceWatch sits next to llm-stats, not in competition with it.
Who is GoldBerry?
GoldBerry is the documentation-first epistemic-audit framework at github.com/PhamBit/goldberry. It encodes the eight lenses and the Suffixscape linguistic-diagnostics layer into a structured prompt any LLM can run. StanceWatch inherits the canon; the framework can be forked for any noncommercial audit purpose.
How often does it update?
On a weekly cadence, each lab's public surface is scraped, the GoldBerry audit prompt is run against it, the result is validated against the eight-lens schema, and the site is republished. Per-lab reports are published as machine-readable JSON at
/stancewatch/api/labs/<slug>.json. To submit a correction or a right of reply, write to hello@cognioengine.co.uk. The full method is on the methodology page.