{"path":"research/progressive-disclosure.md","content":"# Progressive Disclosure of Complexity\n\n**Research compiled March 27, 2026** for the Deliberus project.\n\n**Core concept under investigation**: \"The same underlying graph, rendered at different cognitive resolutions. A casual user votes and reads summaries. A curious user examines premise chains. An expert drills into probabilistic weights and falsification history.\"\n\n---\n\n## 1. Progressive Disclosure in UX Design\n\n### The Principle\n\nProgressive disclosure, formalized by Jakob Nielsen in 1995, defers advanced or rarely used features to secondary screens, showing only the most important options initially. It resolves the fundamental tension between user demands for power and simplicity ([NN/G](https://www.nngroup.com/articles/progressive-disclosure/)).\n\nThree documented benefits:\n- **Learnability**: Novice users focus on core features without being overwhelmed\n- **Efficiency**: Advanced users avoid scanning irrelevant features\n- **Error reduction**: Hidden complexity prevents confusion and mistakes\n\n### Limits of the Standard Model\n\nNielsen warns that **beyond 2 disclosure levels, usability typically drops** — users get lost navigating between levels. Multiple secondary displays add primary-level complexity. This is a critical constraint for Deliberus: the vision of casual/curious/expert requires at least 3 levels, so the transitions between them must be exceptionally smooth — more like zooming than navigating to a different screen.\n\n### Beyond Form Design: Knowledge Systems\n\nThe standard examples (print dialogs, form fields) understate the principle's reach. Several knowledge platforms already implement progressive disclosure of *understanding*:\n\n- **Wikipedia**: Lead section (summary) -> detailed sections -> references -> talk page -> edit history. Research shows ~60% of user clicks go to lead section or infobox content, despite these containing only ~5-6% of total links. Users overwhelmingly engage at the summary level, but the deeper layers exist and are used by editors and researchers ([Dimitrov et al., 2016](https://pmc.ncbi.nlm.nih.gov/articles/PMC5468769/)).\n\n- **Stack Overflow**: Question -> accepted answer -> other answers -> comments -> edit history -> linked questions. Each layer adds context, dissent, and nuance.\n\n- **Metaculus**: Forecast probability -> community distribution -> individual rationales -> resolution criteria -> historical calibration. A casual user sees \"73% likely\"; a forecaster sees the full reasoning chain.\n\n- **Kialo**: Thesis -> pro/con tree -> individual claims -> voting weights. Over 1 million users engage with argument trees, with research showing argument-mapping tools are \"among the most effective methods for training critical thinking skills\" ([Kialo](https://www.kialo-edu.com/)).\n\n### The Inverted Pyramid Connection\n\nJournalism's inverted pyramid — most important information first, supporting details below — is progressive disclosure applied to text. Nielsen explicitly connects this to web reading behavior: users scan, don't read; attention is strongest at the top and decreases further down ([NN/G on Inverted Pyramid](https://www.nngroup.com/articles/inverted-pyramid/)). For web content, readers \"define their own reading paths\" rather than following the journalist's hierarchy — exactly the self-directed depth exploration Deliberus envisions ([Canavilhas](https://onlinejournalismblog.com/2007/10/19/from-the-inverted-pyramid-to-the-tumbled-pyramid-joao-canavilhas/)).\n\n### Implication for Deliberus\n\nThe vision of cognitive resolution levels is a *generalization* of progressive disclosure from UI features to *epistemic depth*. The same argument graph is not a set of hidden buttons — it is a rich structure that can be read at different levels of semantic granularity. This is closer to semantic zoom (see section 5) than traditional progressive disclosure.\n\n---\n\n## 2. Cognitive Load Theory (Sweller)\n\n### The Framework\n\nJohn Sweller's Cognitive Load Theory (CLT) distinguishes three types of load on working memory ([Sweller, 1988](https://www.sciencedirect.com/topics/psychology/cognitive-load-theory)):\n\n| Type | Source | Goal |\n|------|--------|------|\n| **Intrinsic** | Inherent complexity of the material | Cannot be reduced (but can be managed by sequencing) |\n| **Extraneous** | Poor instructional design | Minimize — this is where progressive disclosure helps |\n| **Germane** | Schema construction in long-term memory | Maximize — deep processing builds durable understanding |\n\n### Element Interactivity\n\nThe key construct is **element interactivity** — how many elements must be simultaneously held in working memory to understand a concept. High element interactivity = high intrinsic load. A single claim (\"inflation is rising\") is low interactivity. A multi-premise argument with probabilistic weights is high interactivity. Progressive disclosure manages this by presenting low-interactivity summaries first.\n\n### The Expertise Reversal Effect\n\nThis is crucial for Deliberus: **instructional techniques effective for novices become counterproductive for experts**, and vice versa ([Kalyuga et al., 2003](https://en.wikipedia.org/wiki/Expertise_reversal_effect); [Sweller, 2016](https://link.springer.com/article/10.1007/s10648-016-9359-1)).\n\n- **Novices** benefit from worked examples, structured guidance, and reduced element interactivity (i.e., summaries)\n- **Experts** find the same guidance redundant and even harmful — it creates extraneous load by forcing them to process information they already have as schemas in long-term memory\n- Information stored in long-term memory *reduces* element interactivity, so experts can handle (and prefer) the full-complexity view\n\n**This directly validates the cognitive resolution model**: showing a novice the full argument graph with probabilistic weights and falsification history would create counterproductive extraneous load. Showing an expert only summaries would waste their existing schemas and frustrate them. The same graph *must* be renderable at different cognitive resolutions.\n\n### Evidence\n\nThe expertise reversal effect has been replicated across dozens of studies in mathematics, science, engineering, and medical training. It is considered one of the most robust findings in CLT ([Kalyuga, 2007](https://www.uky.edu/~gmswan3/EDC608/Kalyuga2007_Article_ExpertiseReversalEffectAndItsI.pdf)). Adaptive interfaces that match information density to expertise level show significant performance improvements over fixed-complexity alternatives ([ScienceDirect](https://www.sciencedirect.com/science/article/abs/pii/S1071581984710044)).\n\n---\n\n## 3. Levels of Processing (Craik & Lockhart, 1972)\n\n### The Theory\n\nCraik and Lockhart's Levels of Processing framework proposes that memory retention depends on *depth* of cognitive processing, not duration of exposure ([Craik & Lockhart, 1972](https://www.simplypsychology.org/levelsofprocessing.html)):\n\n| Level | Processing | Retention |\n|-------|-----------|-----------|\n| **Structural** | Visual/surface features | Fragile |\n| **Phonemic** | Sound patterns | Moderate |\n| **Semantic** | Meaning, connections, implications | Durable |\n\nDeeper processing — engaging with meaning, making connections, evaluating implications — produces more durable memory traces.\n\n### Mapping to Deliberation\n\nIn a deliberation context, the cognitive resolution levels map naturally:\n\n| Deliberus Level | Processing Depth | User Action |\n|-----------------|-----------------|-------------|\n| **Casual** (vote + summary) | Structural/shallow | Reads conclusion, registers agreement/disagreement |\n| **Curious** (premise chains) | Intermediate | Traces reasoning, identifies connections between claims |\n| **Expert** (weights + falsification) | Deep/semantic | Evaluates evidence quality, probabilistic relationships, logical structure |\n\nThe theory predicts that **users who choose to go deeper will develop more durable and accurate understanding** of the issue — not just from spending more time, but from the *kind* of cognitive work involved in tracing argument chains vs. reading summaries.\n\n### The Self-Selection Dynamic\n\nDeliberus doesn't *force* depth (unlike a classroom). Users self-select their level. This creates an interesting dynamic: the system should make the *transition* to deeper processing attractive and low-friction. A user who encounters a surprising summary might think \"wait, why?\" — and the system should make it trivial to drill into the premise chain. Each depth transition is a moment of increased germane cognitive load that builds understanding.\n\n---\n\n## 4. Trust, Transparency, and the Inspectability Premium\n\n### The Non-Linear Trust-Transparency Relationship\n\nResearch on explainable AI (XAI) reveals a surprisingly complex relationship between transparency and trust. The naive assumption — \"more transparency = more trust\" — is empirically wrong ([Ehsan et al., 2022](https://pmc.ncbi.nlm.nih.gov/articles/PMC9023880/)):\n\n- Explanations can **backfire**, inducing automation complacency rather than appropriate vigilance\n- Feature importance explanations are \"particularly prone to misleading users\"\n- Too much transparency can cause people to **incorrectly follow** a system when it makes a mistake, due to information overload\n\n### What Works Better Than Explanations\n\nThe empirical evidence suggests alternatives to traditional explainability:\n\n- **Confidence/uncertainty data**: One study found confidence information \"more helpful to users than explanations\" — improving decision-making about task allocation\n- **Performance metrics**: Continuous (not cumulative) performance feedback calibrates trust better than explanations\n- **User control**: Simply allowing users to *modify* algorithmic outputs promotes appropriate reliance, sometimes better than explanations\n\n### The Inspectability Premium\n\nHowever — and this is the key finding for Deliberus — **the mere availability of transparency influences trust even before the system is used**. System transparency \"can significantly influence perceived effectiveness, and initial trust factors are formed by information revealed before the system is used\" ([Springer, 2022](https://link.springer.com/article/10.1007/s12525-022-00593-5)). Research on XAI found that \"perceived AI explainability shaped all dimensions of trust in AI engineers, even after controlling for trust in the AI system\" ([Nature, 2025](https://www.nature.com/articles/s41598-025-04189-9)).\n\nThis suggests an **inspectability premium**: the *knowledge* that you *could* drill into the reasoning — even if you rarely do — increases trust in the system's outputs. This is the \"trust but verify\" dynamic: most users never verify, but the option to do so changes the epistemic relationship.\n\n### Implication for Deliberus\n\nThe cognitive resolution model serves dual purposes:\n\n1. **Functional**: Different users genuinely need different levels of detail\n2. **Trust-building**: The visible *existence* of deeper layers (even when unused) signals that the system's conclusions are grounded in inspectable reasoning\n\nThe expert layer isn't just for experts. Its *existence* validates the casual layer. A summary that says \"Strong consensus (87% support, 12 premises analyzed)\" gains credibility from the fact that those 12 premises are *right there* if you want to check. This is fundamentally different from a system that just says \"87% support\" with no way to inspect why.\n\n### Design Caution\n\nTransparency should not be conceived solely in terms of explainability. For Deliberus, the right transparency model likely combines:\n- Confidence/uncertainty indicators at the summary level\n- Inspectable reasoning chains at the curious level\n- Full evidential structure at the expert level\n- User agency (ability to contribute, challenge, modify) at all levels\n\n---\n\n## 5. Semantic Zoom in Information Visualization\n\n### Shneiderman's Mantra\n\nBen Shneiderman's Visual Information-Seeking Mantra (1996) — \"**Overview first, zoom and filter, then details-on-demand**\" — is the canonical framework for navigating complex information spaces ([Shneiderman, 1996](https://www.cs.umd.edu/~ben/papers/Shneiderman1996eyes.pdf)):\n\n1. **Overview**: Show the entire dataset to establish context\n2. **Zoom and filter**: Let users focus on regions of interest, removing irrelevant items\n3. **Details-on-demand**: Provide full data about specific items when requested\n\n### Semantic vs. Geometric Zoom\n\nStandard zoom (geometric) changes magnification — things get bigger. **Semantic zoom** changes the *type and meaning* of information displayed at each level. A cluster of 50 nodes at overview level might show \"Climate Evidence (50 claims)\" — zooming in reveals individual claims, their relationships, and their evidence ([Wiens et al., 2017](https://www.researchgate.net/publication/321894105_Semantic_Zooming_for_Ontology_Graph_Visualizations)).\n\n### Evidence for Effectiveness\n\nUser studies on semantic zooming for ontology graph visualizations found \"an increase in readability, visual clarity, and information clarity\" compared to geometric zoom or static views. The approach uses **three discrete layers** of detail, abstracting and simplifying the graph structure at higher levels while preserving full detail at the lowest level.\n\nA performance analysis found that semantic zoom delivers \"improved legibility and compactness at all levels compared to uniform zooming or clustering-based approaches\" ([ScienceDirect](https://www.sciencedirect.com/science/article/pii/S0167739X17323610)).\n\n### Application to Knowledge Graphs\n\nSemantic zoom is directly applicable to argument graphs. At the broadest level, a contention might appear as a single node with a stance indicator. Zooming in reveals the argument tree. Zooming further reveals individual claims with evidence links, probabilistic weights, and falsification status.\n\nThe key insight from the visualization literature: **semantic zoom avoids the physical distortions of fisheye views** by using qualitative transitions between abstraction levels. In Deliberus terms: the casual view isn't a distorted version of the expert view — it's a genuinely different *rendering* of the same underlying structure, designed for a different cognitive purpose.\n\n### Deliberus Design Principle\n\nThe three-level cognitive resolution maps directly to Shneiderman + semantic zoom:\n\n| Shneiderman | Deliberus | What's Shown |\n|-------------|-----------|-------------|\n| Overview | Casual | Contention summary, stance distribution, key conclusion |\n| Zoom + Filter | Curious | Premise chains, argument tree, evidence links |\n| Details-on-demand | Expert | Probabilistic weights, falsification history, source reliability, logical structure |\n\n---\n\n## 6. Epistemic Trust and the Philosophy of Testimony\n\n### The Core Question\n\nWhen is it rational to believe what others tell you? The philosophy of testimony addresses this directly, and the debate between **reductionism** and **anti-reductionism** maps onto Deliberus's design decisions ([Stanford Encyclopedia of Philosophy](https://plato.stanford.edu/entries/testimony-episprob/)).\n\n### Key Positions\n\n**C.A.J. Coady** (anti-reductionist): We have \"a defeasible but presumptive right to believe what people tell us\" without requiring independent verification. Testimony is like perception — we don't need positive reasons to trust it, only an absence of reasons *not* to. We must *presuppose* the reliability of testifiers to even interpret their utterances ([Coady, 1992](https://philpapers.org/rec/FAUKOT)).\n\n**Paul Faulkner** (trust-based): When speakers tell us something, they \"invite their audience to trust\" them. Trust itself — not evidence — is epistemically foundational to testimonial justification. The speaker makes themselves accountable for the truth of what they assert ([Faulkner, 2007](https://philpapers.org/rec/FAUKOT)).\n\n**Richard Moran** (interpersonal): In telling someone something, the speaker \"is asking for a certain authority to be acknowledged — the authority to invest his utterance with a particular epistemic import.\" Testimony is fundamentally interpersonal, not merely informational — expressions like \"I assure you\" and \"Believe me\" are always implicit ([Moran, 2005](https://philpapers.org/rec/MORGTA)).\n\n### How Deliberus Changes the Epistemics\n\nIn ordinary testimony, the hearer must decide whether to trust the *speaker*. In a deliberation platform with progressive disclosure:\n\n1. **At the casual level**: The user trusts the *system's aggregation* — similar to trusting a reputation system or a prediction market. This is closer to Coady's anti-reductionism: presumptive trust unless defeaters appear.\n\n2. **At the curious level**: The user can inspect the *reasoning chain* — moving from trust in the system to evaluation of the *arguments*. This is closer to reductionism: the user now has positive evidence about the reliability of the conclusion.\n\n3. **At the expert level**: The user can inspect *evidence quality, source reliability, and logical structure* — reaching the deepest epistemic foundation.\n\n### The Transparency Shift\n\nMoran's insight is particularly relevant: testimony involves the speaker making themselves *accountable*. In Deliberus, progressive disclosure makes the *entire reasoning chain* accountable. A casual user trusts the summary; but the summary is backed by premises that are backed by evidence that is backed by sources — and all of it is inspectable. This transforms the epistemic relationship from \"trust the system\" to \"the system makes its reasoning transparent for your inspection.\"\n\nThis is a novel epistemic position — neither blind trust nor full verification, but *warranted trust grounded in inspectability*. The rational basis for trusting the summary is not that you've checked every premise, but that you *could*, and that others *have*, and that the system makes this visible.\n\n---\n\n## 7. Constructivism, ZPD, and Scaffolding\n\n### Vygotsky's Zone of Proximal Development\n\nThe ZPD is \"the difference between what a learner can do without help and what he or she can do with help\" ([Vygotsky, 1978](https://www.simplypsychology.org/zone-of-proximal-development.html)). **Scaffolding** is the temporary support structure that helps learners operate within their ZPD, gradually withdrawn as competence develops.\n\n### Self-Directed Scaffolding\n\nIn traditional education, the teacher provides scaffolding. In Deliberus, the *system's progressive disclosure structure* serves as scaffolding:\n\n- A casual user who reads summaries and votes is operating *within* their current ability\n- The visible existence of deeper layers creates the ZPD — \"I could understand this more deeply if I tried\"\n- Clicking into a premise chain is self-directed scaffolding — the system reveals structure that supports deeper understanding\n- The argument tree itself *is* the scaffold: it makes the logical structure visible, turning an opaque conclusion into a traceable chain\n\n### Constructivist Learning\n\nConstructivism holds that knowledge is actively constructed, not passively received. In a deliberation context:\n\n- A user who only reads summaries receives knowledge passively (low retention per levels of processing)\n- A user who traces premise chains is *constructing* understanding of the argument\n- A user who evaluates evidence quality and contributes counter-arguments is *co-constructing* the knowledge itself\n\nThe cognitive resolution model naturally supports constructivist engagement: each deeper level requires more active cognitive construction.\n\n### Social Constructivism\n\nVygotsky emphasized that \"knowledge begins as an external, social experience\" and is internalized through dialogue. Deliberus *is* a social knowledge construction system — the argument graph is built collectively, and individual understanding is constructed through engagement with others' reasoning. The progressive disclosure model lets users engage with this social process at their chosen depth.\n\n---\n\n## 8. Wikipedia as Case Study\n\n### Wikipedia's Progressive Disclosure Architecture\n\nWikipedia implements progressive disclosure of knowledge complexity, though it wasn't explicitly designed as such:\n\n| Layer | Audience | Content |\n|-------|----------|---------|\n| Lead section | Everyone | Summary, key facts, conclusion |\n| Article body | Interested readers | Detailed exposition, subsections, images |\n| References | Fact-checkers, researchers | Primary sources, academic citations |\n| Talk page | Editors, domain experts | Disputes, editorial decisions, quality assessments |\n| Edit history | Meta-analysts | Full provenance, contribution patterns, edit wars |\n\n### Empirical Usage Patterns\n\nResearch on Wikipedia clickstreams shows ([Dimitrov et al., 2016](https://pmc.ncbi.nlm.nih.gov/articles/PMC5468769/)):\n\n- **~60% of clicks go to lead section or infobox content** despite containing only ~5-6% of links\n- Users follow a **zoom-out-then-home-in** pattern in goal-directed navigation: start general, navigate through hub articles, then narrow to target\n- **Article structure is the strongest predictor of click behavior** in all except initial and final navigation phases\n- Links that appear in the lead section target \"significantly more general articles\" — the lead acts as a hub to broader context\n\n### Design Lessons for Deliberus\n\nWikipedia demonstrates that most users are casual consumers (lead-section readers) while a small minority engage deeply (talk pages, edit history). But the deeper layers are *load-bearing for the casual experience*: the quality of the lead section depends on the editorial process happening on the talk page. Similarly, in Deliberus:\n\n- Most users will engage at the summary level\n- The quality of summaries depends on the depth of analysis happening at the expert level\n- The availability of deeper layers validates the casual experience (inspectability premium)\n- The system must work well for the majority (casual) while remaining powerful for the minority (expert)\n\n---\n\n## 9. Flow Theory and Progressive Complexity in Games\n\n### Flow State\n\nCsikszentmihalyi's flow theory describes optimal experience as occurring when challenge and skill are balanced ([Csikszentmihalyi, 1990](https://medium.com/@icodewithben/mihaly-csikszentmihalyis-flow-theory-game-design-ideas-9a06306b0fb8)):\n\n- **Challenge > skill** = anxiety\n- **Skill > challenge** = boredom\n- **Challenge ≈ skill** = flow (deep engagement, loss of self-consciousness, intrinsic motivation)\n\n### Game Design Patterns\n\nGames master progressive complexity through several techniques relevant to Deliberus:\n\n1. **Tutorial levels**: Introduce core mechanics in isolation before combining them (cf. casual voting as \"tutorial\" for deliberation)\n2. **Stair-step difficulty**: Periods of high challenge followed by recovery plateaus — prevents exhaustion while maintaining engagement\n3. **Skill trees**: Players choose *which* complexity to unlock, creating personalized progression paths\n4. **Emergent complexity**: Simple rules combine to produce complex behavior (Go, Chess) — the player discovers depth rather than having it imposed\n\n### Application to Argument Exploration\n\nDeliberus could adopt flow-theory mechanics:\n\n- **Natural difficulty curve**: Voting (low challenge) -> reading summaries -> tracing premises -> evaluating evidence -> contributing arguments (high challenge)\n- **Achievement feedback**: \"You've examined 5 premise chains\" — flow requires clear goals and immediate feedback\n- **Adaptive complexity**: The system could suggest next-level engagement based on demonstrated interest (if a user keeps clicking into premises, surface the evidence quality layer)\n- **Emergent understanding**: Simple interactions (voting, agreeing) aggregate into complex collective intelligence\n\n### The \"Casual to Expert Pipeline\"\n\nGames solve the exact problem Deliberus faces: how to serve both casual players and hardcore enthusiasts with the same system. The key game design insight: **don't gate content behind mandatory progression**. Let casual players enjoy the summary experience indefinitely while making deeper engagement *discoverable* and *rewarding* but never *required*.\n\n---\n\n## 10. Empirical Evidence for Progressive Complexity Systems\n\n### Direct Evidence\n\n**Adaptive interfaces outperform fixed interfaces**: An empirical study found that \"the type of interface coupled to a particular user — novice, adaptive, or experienced — produces significant performance differences.\" Users matched to their appropriate complexity level performed better than those using a one-size-fits-all interface ([ScienceDirect, 1984](https://www.sciencedirect.com/science/article/abs/pii/S1071581984710044)).\n\n**Expertise reversal effect**: Dozens of studies confirm that information presentation optimal for novices is counterproductive for experts, and vice versa. This is \"one of the most robust findings\" in cognitive load theory ([Kalyuga, 2007](https://www.uky.edu/~gmswan3/EDC608/Kalyuga2007_Article_ExpertiseReversalEffectAndItsI.pdf)). Fixed-complexity systems necessarily fail one group or the other.\n\n**Semantic zoom improves comprehension**: User studies confirm \"an increase in readability, visual clarity, and information clarity\" for semantic zoom over static or uniform-zoom approaches in knowledge graph visualization.\n\n**Wikipedia's lead section dominance**: The empirical finding that 60% of clicks go to 5% of content validates the progressive disclosure model — most users want summaries, but the system's value depends on deeper layers.\n\n### Indirect Evidence\n\n**Metaculus calibration**: Prediction platforms that show forecast + rationale + community distribution achieve better calibration than those showing only point estimates. The layered structure lets users engage at their appropriate depth.\n\n**Kialo adoption**: Over 1 million users engaging with structured argument trees demonstrates that progressive complexity in argumentation is viable at scale, with research confirming argument-mapping as effective for critical thinking training.\n\n**DeliData study (2023)**: 64% of groups found better solutions than any individual — but this required structured deliberation with visible reasoning, not just voting. The depth of engagement correlated with quality of outcomes. (Referenced in Deliberus academic foundations.)\n\n### The Gap\n\nThere is no single study that directly tests \"progressive complexity vs. fixed complexity for deliberation systems.\" The evidence is assembled from adjacent domains: adaptive interfaces, cognitive load theory, information visualization, knowledge management, and game design. Each domain independently supports the same conclusion: matching information density to user expertise improves outcomes across learning, decision-making, trust, and engagement.\n\nThis gap represents a research opportunity for Deliberus — the platform could generate novel empirical evidence on progressive complexity in deliberation contexts.\n\n---\n\n## Synthesis: Why Cognitive Resolution Works\n\nThe ten research areas converge on a unified argument for the cognitive resolution model:\n\n### The Problem It Solves\n\nFixed-complexity knowledge systems face an irreconcilable tension:\n- **Too simple**: Experts find them shallow, untrustworthy, and frustrating (expertise reversal effect)\n- **Too complex**: Casual users experience information overload, increased error rates, and disengagement (cognitive load theory)\n- **No system can optimize for both simultaneously** with a single rendering\n\n### How It Works (Theoretical Foundation)\n\n| Mechanism | Research Base | Role in Deliberus |\n|-----------|-------------|-------------------|\n| Reduced extraneous load | CLT (Sweller) | Summaries hide unnecessary complexity from casual users |\n| Expertise reversal | CLT (Sweller, Kalyuga) | Experts need full detail; novices need structure |\n| Deeper processing → better retention | Levels of Processing (Craik & Lockhart) | Users who drill deeper develop more durable understanding |\n| Inspectability premium | XAI research | Knowledge that depth *exists* increases trust at all levels |\n| Overview-zoom-detail | Shneiderman's Mantra | Natural navigation pattern for complex information |\n| Semantic zoom | InfoVis research | Same graph, qualitatively different renderings per level |\n| Warranted trust via transparency | Epistemology of testimony | Inspectable reasoning chains change epistemic status |\n| Self-directed scaffolding | Constructivism (Vygotsky) | Users choose their depth, system supports their ZPD |\n| Challenge-skill balance | Flow theory (Csikszentmihalyi) | Each level matches a different skill/interest level |\n| Inverted pyramid | Journalism, Wikipedia | Most users want summaries; depth is for the few who need it |\n\n### The Novel Contribution\n\nWhat Deliberus adds to this body of research is the application to **collective deliberation and argumentation** specifically. The existing platforms (Kialo, Polis, ConsiderIt) have argument structures but don't implement true semantic zoom — they have a single rendering that users navigate. The cognitive resolution model proposes that the *same argument graph* should be renderable at fundamentally different levels of abstraction, each designed for a specific cognitive purpose and user type.\n\n### Design Principles Derived from Research\n\n1. **Three levels is the right number** — matches Nielsen's warning about >2 levels, but the transition mechanism (zoom, not navigate) may mitigate the usability concern\n2. **Transitions must be seamless** — clicking from summary to premise chain should feel like zooming in, not switching contexts\n3. **Each level must be self-sufficient** — a user at any level should get value without needing to go deeper\n4. **Deeper levels validate shallower ones** — the inspectability premium means expert-level detail increases casual-level trust\n5. **Let users self-select** — never force depth, always invite it\n6. **Show confidence/uncertainty at the summary level** — XAI research suggests this is more useful than explanations for calibrating trust\n7. **Make the existence of depth visible** — subtle indicators that deeper layers exist (e.g., \"12 premises analyzed\", \"3 counter-arguments\")\n8. **Adapt to demonstrated interest** — if a user keeps drilling deeper, surface more depth cues (flow-theory stair-stepping)\n\n---\n\n## Sources\n\n### Progressive Disclosure & UX\n- [Progressive Disclosure - NN/g](https://www.nngroup.com/articles/progressive-disclosure/)\n- [Inverted Pyramid: Writing for Comprehension - NN/g](https://www.nngroup.com/articles/inverted-pyramid/)\n- [What is Progressive Disclosure? - IxDF](https://ixdf.org/literature/topics/progressive-disclosure)\n- [Progressive Disclosure - The Decision Lab](https://thedecisionlab.com/reference-guide/design/progressive-disclosure)\n\n### Cognitive Load Theory\n- [Cognitive Load Theory - Sweller (ScienceDirect overview)](https://www.sciencedirect.com/topics/psychology/cognitive-load-theory)\n- [Expertise Reversal Effect - Kalyuga (2007)](https://www.uky.edu/~gmswan3/EDC608/Kalyuga2007_Article_ExpertiseReversalEffectAndItsI.pdf)\n- [Expertise Reversal Effect - Sweller (2016)](https://link.springer.com/article/10.1007/s10648-016-9359-1)\n- [Expertise Reversal Effect - Wikipedia](https://en.wikipedia.org/wiki/Expertise_reversal_effect)\n\n### Levels of Processing\n- [Levels of Processing - Craik & Lockhart (1972)](https://www.simplypsychology.org/levelsofprocessing.html)\n- [Original paper (PDF)](http://wixtedlab.ucsd.edu/publications/Psych%20218/Craik_Lockhart_1972.pdf)\n\n### Trust, Transparency & XAI\n- [How Transparency Modulates Trust in AI (PMC)](https://pmc.ncbi.nlm.nih.gov/articles/PMC9023880/)\n- [Effect of Transparency and Trust on System Acceptance (Springer)](https://link.springer.com/article/10.1007/s12525-022-00593-5)\n- [Explainability + Trust - Google PAIR](https://pair.withgoogle.com/chapter/explainability-trust/)\n- [Trust in Transparency: How XAI Shapes User Perceptions (arXiv)](https://arxiv.org/html/2510.04968v1)\n- [Effectiveness of XAI on Human Factors in Trust (Nature)](https://www.nature.com/articles/s41598-025-04189-9)\n\n### Semantic Zoom & Information Visualization\n- [Shneiderman - The Eyes Have It (1996 paper)](https://www.cs.umd.edu/~ben/papers/Shneiderman1996eyes.pdf)\n- [Semantic Zooming for Ontology Graph Visualizations](https://www.researchgate.net/publication/321894105_Semantic_Zooming_for_Ontology_Graph_Visualizations)\n- [Visualizing Large Knowledge Graphs (ScienceDirect)](https://www.sciencedirect.com/science/article/pii/S0167739X17323610)\n- [Semantic Zoom - EmergentMind](https://www.emergentmind.com/topics/semantic-zoom)\n\n### Epistemic Trust & Testimony\n- [Epistemological Problems of Testimony (Stanford Encyclopedia)](https://plato.stanford.edu/entries/testimony-episprob/)\n- [Paul Faulkner - Knowledge on Trust](https://philpapers.org/rec/FAUKOT)\n- [Richard Moran - Getting Told and Being Believed](https://philpapers.org/rec/MORGTA)\n- [Moran - Exchange of Words](https://philpapers.org/rec/MORTEO-81)\n\n### Constructivism & ZPD\n- [Zone of Proximal Development - Simply Psychology](https://www.simplypsychology.org/zone-of-proximal-development.html)\n- [Vygotsky's Sociocultural Theory](https://www.simplypsychology.org/vygotsky.html)\n- [Vygotsky Scaffolding - PrepScholar](https://blog.prepscholar.com/vygotsky-scaffolding-zone-of-proximal-development)\n\n### Wikipedia Navigation\n- [How Article Structure Influences User Navigation (PMC)](https://pmc.ncbi.nlm.nih.gov/articles/PMC5468769/)\n- [Information Architecture in Wikipedia (Wiley)](https://asistdl.onlinelibrary.wiley.com/doi/full/10.1002/bult.2015.1720410508)\n\n### Flow Theory & Game Design\n- [Flow Theory in Game Design](https://medium.com/@icodewithben/mihaly-csikszentmihalyis-flow-theory-game-design-ideas-9a06306b0fb8)\n- [Cognitive Flow: Psychology of Great Game Design](https://www.gamedeveloper.com/design/cognitive-flow-the-psychology-of-great-game-design)\n- [Flow in Games - Jenova Chen (MFA thesis)](https://www.jenovachen.com/flowingames/Flow_in_games_final.pdf)\n\n### Adaptive Interfaces & Deliberation\n- [Productivity Gains via Adaptive UI (ScienceDirect, 1984)](https://www.sciencedirect.com/science/article/abs/pii/S1071581984710044)\n- [User Experience with Adaptive UIs (ScienceDirect, 2025)](https://www.sciencedirect.com/science/article/pii/S0164121225002675)\n- [Gamification Strategies via Formal Argumentation (Springer)](https://link.springer.com/article/10.1007/s42979-022-01164-3)\n- [Kialo - Structured Debate Platform](https://www.kialo-edu.com/)\n- [Stanford Online Deliberation Platform](https://deliberation.stanford.edu/tools-and-resources/online-deliberation-platform)\n"}