{"slug": "signal-a-draft-ux-framework-for-llm-based-systems", "title": "Signal: A draft UX framework for LLM-based systems", "summary": "A new open-source UX framework called SIGNAL aims to standardize communication design for LLM-based systems, focusing on semantics, intent, grounding, navigation, agency, and cognitive load. The draft framework consolidates research from human-AI interaction, conversation design, and cognitive accessibility to bridge the gap between messy human language and useful AI behavior.", "body_md": "**Semantic Interaction Guidelines for Natural AI Language**\n\nSIGNAL is a pattern framework for LLM UX, human-AI interaction, conversational AI, agent UX, semantic clarity, cognitive load, and, most importantly, user experience in LLM-based systems.\n\n| Status | Scope | Focus | Format | License |\n|---|---|---|---|---|\n| Draft / under active development | LLM-based systems | Communication UX | README-first framework | MIT |\n\nWarning\n\nSIGNAL is under active development.\n\nThis framework was created after researching adjacent work in Human-AI Interaction, conversation design, cognitive accessibility, plain language, generative AI UX, AI governance, and LLM user research.\n\nThese fields contain strong ideas, guidelines, and empirical studies, but I could not find a focused, open, README-first pattern framework dedicated mostly to user experience in LLM-based systems, especially the communication layer: semantics, intent, grounding, navigation, agency, and cognitive load.\n\nSIGNAL is not a claim of invention from zero.\n\nIt is a consolidation attempt: a practical pattern language for a space that appears fragmented across research papers, product guidelines, accessibility notes, governance frameworks, benchmark culture, prompt engineering, and chatbot design practices.\n\nA signal is a gesture, token, or action used to transmit information, communicate a command, or serve as a warning.\n\nWhen conversation is the interface, communication is the product experience.\n\nUsers do not want to write perfect prompts.\n\nThey communicate like people: short messages, vague references, metaphors, idioms, incomplete context, frustration, shortcuts, corrections, and indirect requests.\n\nLLMs are optimized to generate plausible continuations of language. Product experiences require more than plausible continuation: they require visible intent, grounding, progress, agency, cost, and value.\n\nThat gap is where UX matters.\n\nSIGNAL designs the communication layer between messy human language and useful AI behavior.\n\n| Letter | Dimension | Product question |\n|---|---|---|\nS |\nSemantics |\nDoes the system create clear meaning? |\nI |\nIntent |\nDoes it understand what the user means, not only what the user explicitly asked? |\nG |\nGrounding |\nDoes it show what the answer or action is based on? |\nN |\nNavigation |\nDoes it keep the user oriented through context, state, progress, and next steps? |\nA |\nAgency |\nDoes it ask before taking technical actions, using profile data, storing preferences, changing state, or creating consequences for the user? |\nL |\nLoad |\nDoes it reduce mental load by summarizing complex facts, keeping context clear, preserving semantic consistency, and avoiding unsupported user decision burden? |\n\nSIGNAL uses product-facing names, but each dimension is grounded in established concepts from Human-AI Interaction, linguistics, cognitive psychology, information retrieval, and agent research.\n\n| SIGNAL | Canonical concepts | Primary references |\n|---|---|---|\nSemantics |\n|\n\n[ISO 24495-1](https://www.iso.org/standard/78907.html);[W3C COGA](https://www.w3.org/TR/coga-usable/)**Intent**[Speech acts](https://en.wikipedia.org/wiki/Speech_act), indirect speech acts, intent recognition,[query rewriting](https://arxiv.org/abs/2408.17072)[LLM UX intent taxonomy](https://arxiv.org/abs/2401.08329);[MaFeRw](https://arxiv.org/abs/2408.17072)**Grounding**[retrieval-augmented generation](https://arxiv.org/abs/2005.11401), calibration, source attribution[Lewis et al. 2020](https://arxiv.org/abs/2005.11401);[HELM](https://arxiv.org/abs/2211.09110);[Microsoft HAX](https://www.microsoft.com/en-us/haxtoolkit/library/)**Navigation**[Nielsen](https://www.nngroup.com/articles/response-times-3-important-limits/); Clark and Brennan 1991;[Myers 1985](https://doi.org/10.1145/317456.317459)**Agency**[Amershi et al. 2019](https://www.microsoft.com/en-us/research/publication/guidelines-for-human-ai-interaction/);[Microsoft HAX](https://www.microsoft.com/en-us/haxtoolkit/library/);[NIST AI RMF](https://airc.nist.gov/airmf-resources/airmf/)**Load**[Cognitive load](https://doi.org/10.1016/0364-0213(88)90023-7), working memory, cognitive accessibility, progressive disclosure[Sweller 1988](https://doi.org/10.1016/0364-0213(88)90023-7);[Cowan 2001](https://doi.org/10.1017/S0140525X01003922);[W3C COGA](https://www.w3.org/TR/coga-usable/)SIGNAL is not inventing these concepts from scratch. It organizes them into a practical framework for teams building AI experiences through language.\n\nSIGNAL does not explain how an agent is implemented internally.\n\nIt explains how SIGNAL components can sit on top of an example agent loop and turn user/context information into proactive behavior, visible value, and precise communication about what the user expects.\n\nThe loop shown below is only an illustrative example, like using `example.com`\n\nin documentation. It is not a SIGNAL recommendation, pattern, or required architecture.\n\nThe implementation can use RAG, tools, memory, workflows, agents, MCP, databases, or only prompting. The UX responsibilities stay the same.\n\nThis is not an internal architecture diagram. It is a reusable allocation model.\n\nSIGNAL shows what an AI experience must preserve for the user, regardless of the actual implementation flow.\n\nAI conversation UX is not only about producing an answer with cosmetic quality, concision, or explanation.\n\nA good AI experience makes the AI context closer to what the user is trying to communicate and what the user actually needs.\n\nIt should divide understanding into reactive and proactive responsibilities.\n\nReactive understanding:\n\n- the current turn;\n- what the user is uncertain about;\n- what context helps answer the user;\n- what the AI needs from the user now.\n\nProactive understanding:\n\n- what the system is doing actively and passively;\n- what may become relevant from earlier context;\n- what may cost, risk, or create consequences for the user;\n- what AI can do for the user without being explicitly asked;\n- what follow-up or action suggestion the AI can offer.\n\nIt should communicate:\n\n- what changed;\n- what value was delivered;\n- what remains uncertain;\n- what the user controls;\n- what the AI can do next.\n\nIf the user says something that does not clearly connect to the last message, the system should not immediately ask \"what do you mean?\". It should first check whether the user is referring to something earlier in the conversation, something visible in the current environment, or something that just happened.\n\nSIGNAL can be applied to any prompt engineering, agent, bot, assistant, workflow, or AI product by mapping its communication UX responsibilities to the product's actual behavior.\n\nThe following loop is only an example used to explain the concept:\n\n``` php\nflowchart LR\n    U[\"User / Context\"] --> A[\"Understand\"]\n    A --> B[\"Reason\"]\n    B --> C[\"Act\"]\n    C --> D[\"Validate\"]\n    D --> E[\"Respond\"]\n    E --> U\n```\n\nIn this example, SIGNAL defines what each stage must preserve.\n\n| Example loop stage | SIGNAL responsibility | What belongs here |\n|---|---|---|\nUnderstand |\nSemantics + Intent |\nTranslate messy user language into clear meaning and likely goal. Handle vague references, metaphors, idioms, corrections, missing context, and indirect requests. |\nReason |\nGrounding + Navigation |\nDecide what the answer or action should be based on, what evidence is available, what state matters, what remains uncertain, and what path should be followed. |\nAct |\nAgency |\nUse tools, memory, workflows, profile data, databases, or external systems only inside clear user-control boundaries. Ask before actions that create consequences. |\nValidate |\nGrounding + Agency + Navigation |\nCheck whether the result is supported, whether the action stayed within scope, whether anything failed, and whether the user needs a recovery path or approval. |\nRespond |\nLoad |\nCommunicate the result with the lowest useful cognitive effort: clear summary, visible value, relevant evidence, next step, and no unnecessary reading or typing burden. |\n\nThis makes SIGNAL reusable across architectures without requiring this specific loop.\n\nIt does not matter whether the system is only a prompt, a RAG assistant, a tool-using agent, an MCP workflow, a database-backed bot, or a multi-agent system.\n\nThe implementation may change. The UX responsibilities remain the same:\n\n``` php\nUnderstand -> preserve meaning and intent.\nReason -> preserve evidence, uncertainty, and state.\nAct -> preserve user control.\nValidate -> preserve correctness, scope, and recovery.\nRespond -> preserve clarity, value, and low effort.\n```\n\nUse it in five steps:\n\n- Map the product's AI behavior to the loop.\n- Assign SIGNAL responsibilities to each stage.\n- Choose patterns for the stages that are failing.\n- Define checks for prompts, tools, retrieval, memory, workflows, and responses.\n- Convert failures into product changes: better wording, better retrieval, clearer state, safer approval, lower user effort, or a stronger action boundary.\n\nThe output of a SIGNAL review should be concrete: rewritten responses, clearer action boundaries, better tool behavior, improved retrieval overlap, evaluation checks, and visible value receipts.\n\nEach SIGNAL pattern is a reusable answer to a recurring AI UX problem.\n\n``` php\nflowchart TD\n    P[\"User communication problem\"] --> D[\"SIGNAL dimension\"]\n    D --> C[\"Canonical concept\"]\n    C --> R[\"Reference-backed principle\"]\n    R --> X[\"Reusable UX pattern\"]\n    X --> E[\"Evaluation criterion\"]\n```\n\nExample:\n\n| Problem | SIGNAL dimension | Pattern |\n|---|---|---|\n| User sends vague follow-up | Intent / Navigation | Context Recovery |\n| AI may act externally | Agency | Action Boundary |\n| Answer depends on uncertain evidence | Grounding | Confidence Split |\n| Long task leaves user waiting | Navigation / Load | Visible Work Trace |\n| User faces too many choices | Load | Few Useful Options |\n\n| File | Purpose |\n|---|---|\n`docs/FRAMEWORK.md` |\n\n`docs/RESEARCH_AND_BENCHMARKS.md`\n\n`docs/PATTERNS.md`\n\n`docs/WHY_SIGNAL.md`\n\n`docs/FOR_TEAMS.md`\n\n`templates/conversation_ux_review.md`\n\nSIGNAL is not:\n\n- a model benchmark;\n- a prompt engineering trick;\n- an internal agent architecture;\n- a replacement for product research;\n- a leaderboard;\n- a claim that one assistant is universally better than another.\n\nSIGNAL is a communication UX layer for AI product experiences.\n\nIt helps teams ask:\n\nDid the AI understand what the user meant, act within the right boundaries, reduce user effort, and make the delivered value visible?", "url": "https://wpnews.pro/news/signal-a-draft-ux-framework-for-llm-based-systems", "canonical_source": "https://github.com/hi-mundo/SIGNAL", "published_at": "2026-07-01 13:41:34+00:00", "updated_at": "2026-07-01 13:50:45.560879+00:00", "lang": "en", "topics": ["large-language-models", "ai-products", "ai-tools", "ai-ethics", "natural-language-processing"], "entities": ["SIGNAL", "ISO 24495-1", "W3C COGA", "Microsoft HAX", "NIST"], "alternates": {"html": "https://wpnews.pro/news/signal-a-draft-ux-framework-for-llm-based-systems", "markdown": "https://wpnews.pro/news/signal-a-draft-ux-framework-for-llm-based-systems.md", "text": "https://wpnews.pro/news/signal-a-draft-ux-framework-for-llm-based-systems.txt", "jsonld": "https://wpnews.pro/news/signal-a-draft-ux-framework-for-llm-based-systems.jsonld"}}