ReVISit 2: Full-Lifecycle Visualization Framework

• The paper presents a full life cycle framework that integrates design, data collection, analysis, and dissemination, addressing stimulus complexity and reproducibility gaps.
• It employs a JSON-based DSL and Python bindings (reVISitPy) for declarative experiment design, enabling rigorous and reproducible study specifications.
• The framework leverages client-side instrumentation and pluggable storage to capture detailed interaction logs and support replay-oriented provenance for in-depth analysis.

Searching arXiv for the ReVISit 2 paper and closely related visualization user-study framework work. to=arxiv_search 天天中彩票网json_string {"^^^^^^^^2query2 OR title:\2^^^^^^^^"ReVISit 2: A Full Experiment Life Cycle User Study Framework\"","max_results":5,"sort_by":"relevance"} tool call നടത്തി to=arxiv_search 玩彩神争霸json_string {"^^^^^^^^2query2 user study framework browser-based experiments replay provenance DSL static-site arXiv","max_results":^^^^2id:(Cutler et al., 5 Aug 2025) OR title:\2query2^^^^,"sort_by":"relevance"} tool call നടത്തി ReVISit 2 is a software framework for browser-based visualization user studies that is designed to support the full experiment life cycle rather than isolated stages of experiment construction or deployment. It addresses stimulus and instrumentation complexity, experimental-design overhead, and reproducibility and dissemination gaps by combining a domain-specific configuration layer, client-side instrumentation, pluggable storage, replay-oriented provenance capture, and static-site deployment. In the formulation of "ReVISit 2: A Full Experiment Life Cycle User Study Framework" (&&&^^^^2query2^^^^&&&), the system is intended for publication-quality studies and is organized around five experiment phases: Design, Debug & Pilot, Data Collection, Analysis, and Dissemination.

^^^^2id:(Cutler et al., 5 Aug 2025) OR title:\2^^^^. Conceptual scope and design principles

ReVISit 2 was motivated by three interlocking challenges in contemporary browser-based visualization studies. The first is stimulus and instrumentation complexity: rich interactive web stimuli, including Vega or React visualizations, must be coupled with reliable logging of clicks, hovers, and keystrokes. The second is experimental-design overhead: factors, sequences, attention checks, and between-/within-subject allocations often require ad hoc code that is difficult to maintain or reproduce. The third is reproducibility and dissemination gaps: commercial survey tools are closed-source, domain-specific frameworks often cover only parts of the life cycle, and lab-built solutions tend to be brittle or unsupported.

The framework is organized around three core design principles. “Full-Life-Cycle Support” denotes coverage of Design → Debug & Pilot → Data Collection → Analysis → Dissemination under one unified framework. “Scientific Sovereignty” denotes purely static web deployments, pluggable storage, and forkable code on GitHub so that researchers remain in full control. “Lean On What’s Good” denotes deliberate avoidance of re-implementing mature tooling such as crowd-recruitment platforms and R/Python analysis, while supplying the “missing middle” specific to visualization studies.

This positioning defines ReVISit 2 less as a narrow questionnaire engine than as infrastructure for experiment specification, execution, inspection, and republication. A plausible implication is that the framework’s central contribution lies in integration across stages rather than in any single isolated feature.

2. Experiment specification and study design

At the design stage, ReVISit 2 provides a JSON-based domain-specific language for declaring components, sequences, skip and interruption logic, and randomization strategies including fixed, random, Latin-square, and dynamic sequencing. A representative DSL fragment specifies a sequence block with order: latinSquare, numSamples: 24, and components such as ["stimA","stimB"]. Client-side JSON Schema validation and a custom linter detect missing components or invalid sequences before deployment.

The same design space is exposed programmatically through reVISitPy, a set of high-level Python bindings described as Altair-style chaining for constructing the same DSL within Jupyter notebooks. ReVISitPy includes in-notebook preview of stimuli, sequence sampling, and synthetic pilot data, thereby tightening the design loop. This couples declarative study specification with notebook-based exploratory workflows familiar to many visualization researchers.

The design model explicitly treats experimental sequencing as first-class structure rather than peripheral control flow. Dynamic/staircase designs, Latin-square orderings, skip logic, and attention-check insertion are part of the shared configuration rather than embedded in opaque custom code. This is significant because the experiment definition itself becomes inspectable and reproducible.

3. Debugging, piloting, and runtime architecture

For debugging and pilot work, ReVISit 2 includes a Study Browser with two modes. Participant View steps through one simulated participant’s path, including randomization and skips, and provides “Next Participant” to sample another path. All-Trials View provides instant access to every component irrespective of sequence logic, enabling rapid navigation. Data Preview adds live dashboards of incoming form responses, attention-check pass rates, and simple summaries. The framework also includes built-in getUserMedia hooks and Whisper integration for think-aloud pilots.

The system architecture is described as a client-heavy, static-site React/TypeScript application. A Node.js build step compiles the JSON DSL into a TypeScript typed config and then into a React component tree. Custom plugins, including a Vega wrapper, are instrumented with trrack at build time. At runtime, the React app fetches studyConfig.json, initializes the sequence engine, and renders components in order. The event logger serializes each event and POSTs it to the configured StorageAdapter, for example through the Firebase REST API.

The codebase is modularized into core/, stimuli/, analysis/, and utils/. The core/ directory contains the DSL parser, sequence engine, and storage adapters. stimuli/ contains Markdown, Form, Image, Vega, and Custom React stimuli. analysis/ contains Timeline, Replay, and Data dashboards. utils/ contains trrack integration and the attention-check scheduler. This decomposition places sequence control, instrumentation, and analysis support within a single implementation surface.

4. Data collection, provenance, and replay

Data collection in ReVISit 2 centers on pluggable storage and automatic provenance capture. Supported storage modes include browser-only storage via IndexedDB, Firebase, Supabase, and custom back-ends exposed through a single StorageAdapter interface. Automatic logging covers DOM events, Vega signal changes, component-level responses, and timestamps. Attention-check insertion rules are first-class DSL constructs and may be deterministic every PRESERVED_PLACEHOLDER_^^^^2query2^^^^ trials or random.

The framework’s event model is specified in two equivalent representations. For replay, logs are represented as PRESERVED_PLACEHOLDER_^^^^2id:(Cutler et al., 5 Aug 2025) OR title:\2^^^^, where $e_i = (t_i, c_i, d_i)$ with $t_i$ a timestamp, $c_i$ a component ID, and $d_i$ event data. In the technical affordances section, the atomic event model is written as

$e_i = (\,t_i,\ c_i,\ \alpha_i\,)$

where $\alpha_i$ may wrap a DOM event or a Vega signal change, and the provenance stream is

$P = \langle e_1,\,e_2,\,\ldots,e_n\rangle.$

Replay operates by sorting events by timestamp, locating the DOM element for each component, and dispatching a synthetic event using the recorded event data at simulated time $t_i$. The replay scheduler is intended to reinstate the exact visual state by sequentially replaying PRESERVED_PLACEHOLDER_^^^^2id:(Cutler et al., 5 Aug 2025) OR title:\2query2^^^^ in time-relative fashion. This makes participant behavior auditable at interaction granularity rather than only at the level of terminal responses.

The same provenance substrate also supports analysis. The Timeline View aggregates task durations, correct/incorrect flags, hover versus click density, and embedded audio waveform. Export utilities provide tidy CSV and JSON for R/Python analysis while preserving event-level granularity. The combination of event capture, replay, and export means that study code, interaction traces, and analysis artifacts remain tightly coupled.

5. Dissemination and empirical evaluation through replications

ReVISit 2’s dissemination model follows its static-site orientation. Studies can be published through GitHub Pages, Netlify, or any static host. Deep-linking into trials uses a URL fragment carrying both sequence and participant ID so that a figure in a paper can link to a live replay of that trial. Demo Mode supports read-only navigation for reviewers while deactivating data collection, and versioned forks fix the framework version for each study so that later framework updates do not break earlier experiments.

The framework is evaluated through three full replications that each emphasize a distinct capability (&&&^^^^2query2^^^^&&&). In the replication of Harrison et al. ’^^^^2id:(Cutler et al., 5 Aug 2025) OR title:\2^^^^4 on just-noticeable difference in correlation, dynamic staircase sequencing was implemented through a DSL dynamic block in which, when a response is correct, PRESERVED_PLACEHOLDER_^^^^2id:(Cutler et al., 5 Aug 2025) OR title:\2id:(Cutler et al., 5 Aug 2025) OR title:\2^^^^, and otherwise PRESERVED_PLACEHOLDER_^^^^2id:(Cutler et al., 5 Aug 2025) OR title:\22^^^^. Attention checks were inserted every ^^^^2id:(Cutler et al., 5 Aug 2025) OR title:\2query2^^^^ trials by comparing PRESERVED_PLACEHOLDER_^^^^2id:(Cutler et al., 5 Aug 2025) OR title:\2^^^^3 versus ^^^^2id:(Cutler et al., 5 Aug 2025) OR title:\2^^^^, with reversed correct location from the previous check, and participants were excluded if failure rate exceeded 2^^^^2query2^^^^%. The study involved 24^^^^2query2^^^^ participants, allocated as 3^^^^2query2^^^^ per visualization × 2 directions × 4 chart types. The reported outcome was that scatterplots and hexbins yielded the lowest mean JNDs; for PRESERVED_PLACEHOLDER_^^^^2id:(Cutler et al., 5 Aug 2025) OR title:\2^^^^4 (positive), scatterplot had PRESERVED_PLACEHOLDER_^^^^2id:(Cutler et al., 5 Aug 2025) OR title:\2^^^^5 and parallel-coordinates had PRESERVED_PLACEHOLDER_^^^^2id:(Cutler et al., 5 Aug 2025) OR title:\2^^^^6. The replication also reports the Weber’s-law form

PRESERVED_PLACEHOLDER_^^^^2id:(Cutler et al., 5 Aug 2025) OR title:\2^^^^7

In the replication of He et al. ’24 on expert pattern-design workflow, the study used a mixed design with between-subjects variation on chart type and within-subjects variation on pattern order, geometric versus iconic. ReVISit 2 captured full trrack provenance on participant adjustments such as sliders and color toggles, together with think-aloud audio. The study involved 2^^^^2query2^^^^ expert designers. The reported analysis found no strong correlation between number of events and BeauVis aesthetic rating, while replay exposed new insights on “for-all” button usage.

In the replication of Feng et al. ’^^^^2id:(Cutler et al., 5 Aug 2025) OR title:\2^^^^8 on search in visualization, two conditions, search versus no-search, were crossed with two chart types, bubble and “255 charts.” The system logged timestamped hover-and-click traces and think-aloud audio. The participant counts were 99 for bubble and 93 for “255” charts. The replication reproduced the key finding that items highlighted by search had longer dwell times. For bubble, the Wilcoxon signed-rank result was

PRESERVED_PLACEHOLDER_^^^^2id:(Cutler et al., 5 Aug 2025) OR title:\2^^^^8

and for 255 charts it was

PRESERVED_PLACEHOLDER_^^^^2id:(Cutler et al., 5 Aug 2025) OR title:\2^^^^9

Qualitative traces also revealed “edge-then-center” exploration patterns.

These replications collectively demonstrate the framework’s support for dynamic sequencing, provenance-intensive interaction analysis, audio-assisted pilot and qualitative workflows, and deployment in publication-quality studies.

6. Relation to existing tools, limitations, and open directions

ReVISit 2 is explicitly contrasted with GUI-only tools such as Qualtrics and SurveyMonkey and with partial DSLs such as Sweetpea and Touchstone. In that comparison, the framework is characterized by coverage of all five life-cycle stages in one open-source package, fine-grained interaction provenance with replay, dynamic sequencing as a first-class DSL feature, and fully static-site deployment without vendor lock-in while remaining pluggable into recruitment and analysis ecosystems. The comparison is therefore not simply about feature count; it is about which parts of the experiment life cycle are treated as native concerns.

A common misconception would be to interpret ReVISit 2 as a no-code experiment builder. The stated limitations argue against that reading. The JSON DSL and React/Vega knowledge can present a steep learning curve for non-programmers. The framework is presently focused on desktop browsers, with native mobile, AR/VR, and head-mounted displays not yet first-class. It does not include built-in simulation of synthetic participants for power analysis or semantic checks for confounders. It also relies on third-party storage options such as Firebase, which introduces external dependency risks.

Planned enhancements and open research questions are correspondingly concrete. The roadmap includes a GUI “Experiment Builder,” participant simulators for power-and-error-checking frameworks, native mobile and WebXR support, automated experiment-graph analyses for order- or learning-confound detection before deployment, and archival packaging through snapshot screenshots, video recordings, and Web Archive bundles. This suggests that the framework’s future trajectory is toward broader modality coverage, stronger pre-deployment validation, and more durable preservation of executable study artifacts.

Within visualization methodology, ReVISit 2 occupies the niche of a full-life-cycle, provenance-aware, browser-native framework whose emphasis is not merely experiment deployment but the preservation, replay, and rehydration of experimental process as part of the scientific record.