Observed Signal · Jul 7, 2026 · Technical Release · Source: UX Collective · Impact: 3/5 · Sentiment: Neutral
Spaghetti Table Protocol: Quick AI Physical-Grounding Test
The article introduces the Spaghetti Table Protocol, a 15-minute diagnostic stress test designers can run to reveal physical-grounding failures in multimodal generative AI. A pilot administered in Feb–Mar 2026 tested three leading multimodal models on an image prompt asking for a dining table with four dry-spaghetti legs, a concrete slab tabletop, and a fishbowl; an aggregate score across fifteen outputs was 4/30 (≈13%). The study found consistent structural failures across models: fluent photorealistic outputs that violate basic physics, session-contamination between prompts, and cases where symbolic acknowledgement of impossibility did not prevent an incoherent generation. The author shares a rubric, protocol specification, and a GitHub repo to crowdsource replication and build a public dataset, and invites designers to design domain-specific high-entropy stress tests to map where current architectures lack embodied physical reasoning.
Provides an empirical, replicable diagnostic exposing consistent physical-grounding failures across major multimodal models; relevant to designers, HCI practitioners and teams evaluating trust, safety and reliability of generative AI in product/UI contexts.
Track Medium Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- The Spaghetti Table Protocol pilot was run across three multimodal AI systems (GPT-4o, Gemini 1.5 Pro, Claude 3.5 Sonnet) in February–March 2026 under identical conditions.
- Aggregate score across fifteen outputs in the pilot was 4 out of 30 (≈13% of the structural coherence ceiling).
- Claude 3.5 Sonnet linguistically flagged structural instability but still generated a physically incoherent image and silently substituted one spaghetti leg with a more solid leg.
- GPT-4o and Gemini 1.5 Pro produced photorealistic images of the impossible configuration without linguistic qualification; Gemini once contaminated an output with elements from a prior unrelated prompt.
- The protocol, full rubric, submission template, and specification are published and linked via a GitHub repository to enable distributed replication and a shared dataset.
Connected Companies & Entities
1 Entity mapped“Zakrzewski, P. (2026, March). A designer’s field report on the Iconic blind spot in AI world models. UX Collective, Medium....”
Ontology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Designer Field Report: Iconic Blind Spot in AI World Models
Peter (Zak) Zakrzewski (UX/visual designer) reports a reproducible set of architectural failures in current LLM-based multimodal systems after running comparative prompts against Google Gemini, ChatGPT, and Sonnet. He defines three diagnostic 'pillars' — Continuity (3D spatiotemporal tracking), Gravity and Physics (physical-constraint reasoning), and Reversibility of Thought (ability to reverse/reset reasoning trajectories) — and shows how their absence produces coherent-looking but physically impossible outputs and compounding errors he calls the Divergence Swamp. Zakrzewski situates his findings against recent world-model work (e.g., Yann LeCun’s JEPA / AMI Labs) and argues that designers should act as an embedded 'More Knowledgeable Other' (the proposed 'Somatic Compiler') to provide the enactive and parametric grounding current systems lack. He frames a research direction called the Parametric AGI framework to integrate design-driven spatial competence into world-model development.
Six-Layer Protocol for Structured AI Image Prompts
An article by ALICE - AI (published on DEV on 2026-07-12) presents the Six-Layer Protocol, a structure-first specification for writing reproducible AI image-generation prompts so teams can produce consistent assets across roles. Based on analysis of 12,502 community prompts from the awesome-gpt-image-2 GitHub repository and additional examples from Evolink.ai, the protocol separates responsibilities into Goal, Canvas, Layout, Subject, Style, and Constraints—each with a single focus and independent iterability. The piece introduces a 20-entry style lookup table with source anchors to standardize style names, defines an accountability agreement that assigns layout/typography faults to designers and generation-quality faults to image composers, and describes a same-day field test that demonstrated improved consistency when the framework was embedded into team role definitions.
What "Grounded AI" Actually Means (and Why It Matters for Your Documents)
Grounded AI means every answer comes from your documents, not the model’s memory. What grounding really is, how to test it, and where it falls short.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.