Rendered scalar sequence-effect illustration (synthetic)

Raw observations, contrast estimation and scientific independence | Rendered paired estimation demonstration (synthetic)

Read the data-generating process before interpreting the figure | Figure 1. Acquisition timestamps: the analysis grid is not the device rate | Figure 2. Paired eye responses and participant-bootstrap uncertainty | Figure 3. Missingness assumptions remain unidentifiable from observed rows

What the plots can, and cannot, establish | Figure 1. Effect estimates with user-supplied uncertainty intervals | Figure 2. Resampling distribution, not a posterior without a Bayesian model | Figure 3. Structural graph disagreement is not causal discovery | Figure 4. A correct regression fit can still be causally misleading

An estimation-first view of biometrics outcomes | 1. Show participant-level trajectories, not only a mean bar | 2. Show the estimated within-participant contrast and uncertainty | When this figure is scientifically appropriate

eyeprocess 0.11.1 published on CRAN | Related downstream modelling addition

Installation | Simulate a complete project | Standard workflow | Inspect and visualize | Persist the canonical representation

Dynamic gaze-state IRTree baseline | Functional pupil-informed IRT | Theory-defined strategies | Gaze-informed diffusion | Monte Carlo design | Evidence promotion gate

Frequency/activity features | Event-related pupil deconvolution | Luminance and trial-order adjustment | Robust filtering

True response-mixture IRT | Nonparametric Rasch diagnostics | Stepwise item-reduction sensitivity | Biometric-feature imputation sensitivity | Process-informed Rasch trees

Packaged lifecycle closure

Scope | Bayesian diagnostics | 3PL response-process alignment

Visual audit

Purpose | Minimal workflow | Explicit specification | Item labels and conditions | Adding observed responses | Output structure | Interpretation boundaries

Scope | Simulation and observed-state models | Hidden states with Stan | Model comparison and recovery

Why these functions are gated | Process-state HMM + IRT | Cognitive diagnosis and latent process classes | Latent-space IRT | Process-adjusted DIF | Sequence representations | Flexible item-response curves | Linking and person fit | Process-aware adaptive testing | Promotion rule

Frozen evidence as a visual object

Measurement stance | Specification | Diagnostics and scalar comparisons | Preprocessing sensitivity

Confirmatory parameter mapping | Identification and posterior checks | Simulation programme

Profile before import | Folder import | Gazepoint-specific audits | Separate biometrics and synchronization | Gazepoint Analysis 7.2.0 paired exports

Simulate a complete project | Standard workflow | Inspect and visualize | Persist the canonical representation

Automatic detection | Generic mapping | Dedicated vendor adapters | Preserve source meaning

Evidence levels | Registering a case | Fingerprinting, redaction, and semantics | Compatibility evidence

Why import success is not validation | Public validation corpus | Semantic round-trip contract | Timestamp semantics | Coordinate and pupil fidelity | BIDS eye-tracking semantics | HED event semantics | Evidence matrix

Eye-Tracking-BIDS | Parquet and Arrow | Package ecosystems

Visual diagnostics

Simulation-based calibration view

Experimental item-parameter seeding | Presentation/accessibility sensitivity

Channel-specific model checks | Alignment negative controls

A common response target | Two complementary quantities | Non-additivity | Interpretation

Why recovery precedes promotion | Recovery experiment | What recovery does not show

Scope | Canonical specification | Simulation | Structural audit | CmdStan fit | Identification | Missingness | Interpretation boundary

More sensors can create more failure modes | Device-dependent stress | Falsification controls | Sensor value is conditional | Channel conflict

Why the first M3 likelihood is scalar | What the bridge does not do | Future full functional likelihood

Why eight models? | Fit the ablation lattice | Incremental pupil evidence | Redundancy and complementarity | Sensor value and channel conflict

Pupil is a measurement, not a label | Inspect the contract | Missingness stress is part of the model evidence | Device and session effects | Practical rule

Validation must include failure cases | Recovery programme | Identifiability boundaries | What counts as a successful negative result?

Purpose | Simulate and audit before fitting | Fit with CmdStanR | Relationship to M0-M2 | Evidence boundary

Backend status

Pupil latency sensitivity | Event-marker plausibility | Validation ladder

Conceptual feature blocks | Process-profile discovery | External validity

Measurement channels, not feature dumping | Registry | Response + response time + gaze | Graded responses | Nominal distractors + option gaze | Missingness as a process | Device and algorithm facets | Bounded gaze measures

Why preserve the response process? | Option-level process evidence | Local dependence must be checked | Revisiting as collateral evidence in cognitive diagnosis | Validation requirements

Declare preprocessing | Static and dynamic AOIs | Declared features | Sensitivity analysis

Posterior uncertainty as measurement value | Channel ablation

Scope | Pre-flight specification | Multivariate anomaly review

Pre-action and gaze-evidence representations | Frontier estimators remain gated

Purpose | Core model families | A process channel must earn its place | Missingness: separate exposure from response | Cross-device measurement is an estimand | Latent distribution and IRF stress tests | Promotion is evidence-based

Response and response-time matrices | Conventional and explanatory IRT | Accuracy and response time | Process-informed and experimental models

Bundled multimodal study | Integrity and expected outputs | Data dictionary and reproduction directory | Release audit

Why pupil needs a measurement layer

Reproducible records

Parameter recovery and interval coverage | Grouped validation and leakage | Preprocessing multiverse | Reproducible release assets | Explicit confirmatory gates | Systematic advanced-model grid

Purpose | Deterministic plans | Atomic execution and resumption | Collection and evidence | Promotion remains gated

Interpretation boundary | Methodological anchors

Contracts | Partitioned storage | Schema migration and benchmarks | External engines

Partial and streaming scoring | Unified evidence bundles

Windowed process representations | AOI trajectory features

Prespecification before fitting | Classification uncertainty | Sensitivity and competing heterogeneity

Objective | Native visualization | Interpretation boundary

Why empirical validation is separate from declared support | Validate one source | Build a validation corpus | Produce a safe validation bundle | Recommended real-export acceptance process

A common validation contract | Recovery table | Prior SBC | Posterior SBC | Posterior predictive checks | Misspecification | External and grouped validation | Incremental information and negative controls | Process-dependent discrimination | Evidence grade

Visual evidence atlas

Question | Generate raw trial/event inputs | Construct and validate the survival table | Kaplan-Meier description | Complementary event-incidence view | Empirical hazard/risk increments | Repeated-participant Cox model | Explicit AFT sensitivity models | Troubleshooting clinic | Reporting example | API links

Why TTFF is a censored-data problem | When to use this method | Worked example: disclosure inspection | Worked variant: evidence verification | Event semantics | Censoring decision guide | Kaplan–Meier description | Complementary survival visualizations | Cox versus AFT models | Cross-language backend contract | Repeated participants: clustered versus frailty Cox | Diagnostics | Model-choice quick guide | Sensitivity analysis | Failure case: unusable gaze is not censoring | Reporting guidance | Copyable manuscript template | API map

Purpose | Accuracy | RMS sample-to-sample precision | BCEA | Sampling intervals | Four-panel quality dashboard | Review-rule sensitivity | Reporting checklist | API links | Limitations

Why event detection belongs in the analytical record | When to use this workflow | Define detector specifications | Synthetic 60 Hz worked example | Compare event catalogues structurally | Propagate detector choice into AOI features | Propagate into the same statistical model | Model-input audit | Denominator discipline | REMoDNaV is an external bridge, not a reimplementation | What to report | Visual reporting companion | Generate a Markdown report | Interpretation boundaries | R/Python contract parity | Failure clinic

Purpose | Deliberate detector failure | Model-input audit | Invalid callback output | Interpretation | Reporting example | Related material

Purpose | Reproducible synthetic setup | 1. Event-count sensitivity | 2. Pairwise temporal agreement | 3. AOI-feature sensitivity | 4. Coefficient stability | 5. Model-input accountability | Planned-denominator robustness | Suggested publication sequence | Example caption | Interpretation boundaries | API handoff

AOI Geometry Is an Analytical Assumption | Synthetic advertising layout | AOI Perturbation Sensitivity Analysis | Pixels versus Degrees of Visual Angle | Propagating AOI Uncertainty Into Statistical Models | Run geometry to inference | Interpreting AOI Assignment Stability | Fixation and sample input | Zero is not missing | Visual diagnostics | Plot-to-question map | Demonstrate a failure branch | AOI Robustness Decision Clinic | Reporting example | Planning the sensitivity analysis before results | Troubleshooting AOI perturbation branches | Reporting problematic branches | Figure set for manuscripts and reviewer responses | Reporting AOI Robustness | R/Python Scientific Parity | Limitations | API map

Purpose | Recommended bundle | Export from a result | Reviewer-facing reading order | Interpretation | Privacy and limitations | API map

Purpose | Design and event definition | Censoring and gaze quality | Repeated observations and estimand | Diagnostics | Sensitivity analysis | Reporting bundle | Final freeze questions | API map

Why plan before results? | Minimum analysis plan | Fillable template | Translate the plan into R | Planned interpretation matrix | Reporting example | Amendments | API map

Model contract | Fit and diagnose | Extensions | Prior sensitivity | Assumptions and limitations | Reporting example | API links | Model-comparison guard

Why a dedicated quality report? | Recommended workflow | Accuracy versus precision | Precision metrics answer different questions | Effective versus nominal sampling | Data loss and missingness | Study-specific review rules, not universal exclusions | Quality decisions as sensitivity analysis | When not to use these metrics | Manuscript reporting example | API map | Reporting references

Why keep trials? | Preparation | Serial mediators or moderators | Cross-language provenance | Analysis-readiness checklist | Interpretation boundary

Related crossed repeated-measures method

Related crossed location–scale development

Related hierarchical modelling addition

Python-native extension outside the frozen R contract

Related methodological addition

Pupil latency sensitivity | Event-marker plausibility | Validation ladder

Scope | A backend-free advanced workflow | Declare an advanced model | Translate, fit, diagnose, and estimate

Audit first; do not select an order from a plot | Governed ARMA orders | Post-fit residual comparison

Complexity is audited before Stan | Model card

Purpose | Backend-free simulation | Fit declaration | Posterior derivatives | Predictive calibration on held-out data

Approximate GP is the default | Hyperparameters are posterior estimands

Leave-future-out is an explicit refit workflow

Measurement uncertainty is declared, not silently corrected | Joint brms translation

Why separate location and residual scale? | Candidate specifications are explicit | Posterior residual-scale trajectory

Observed trajectories | Stored latent mean | Temporal dependence | Missingness | Computational audit | Binocular example

More sensors can create more failure modes | Device-dependent stress | Falsification controls | Sensor value is conditional | Channel conflict

Why the first M3 likelihood is scalar | What the bridge does not do | Future full functional likelihood

Why eight models? | Fit the ablation lattice | Incremental pupil evidence | Redundancy and complementarity | Sensor value and channel conflict

Pupil is a measurement, not a label | Inspect the contract | Missingness stress is part of the model evidence | Device and session effects | Practical rule

Validation must include failure cases | Recovery programme | Identifiability boundaries | What counts as a successful negative result?

Purpose | Simulate and audit before fitting | Fit with CmdStanR | Relationship to M0-M2 | Evidence boundary

Prespecify consequential analysis choices | Materialize, do not rank

Scope | Deterministic synthetic example | Restricted specification | Interpretation boundary

Approved family | Translation without fitting | Prior-predictive gate | Full-MCMC backends

Verified Open Gaze fields | 3-D pupil diameter

Lightweight posterior-draw contract | Trajectory uncertainty | Declared contrasts and windows

Preparation is not preprocessing automation | Explicit baseline transformation | Readiness evidence | Measurement-context audit

Posterior predictive evidence | Temporal residual review

Purpose | A Gazepoint-like mapping audit | Contract through specification | Backend and post-fit stages | Sensitivity

Start with the prediction target | Execution

Channel-specific model checks | Alignment negative controls

A common response target | Two complementary quantities | Non-additivity | Interpretation

Why recovery precedes promotion | Recovery experiment | What recovery does not show

Scope | Canonical specification | Simulation | Structural audit | CmdStan fit | Identification | Missingness | Interpretation boundary

Backend status

Posterior uncertainty as measurement value | Channel ablation

Why pupil needs a measurement layer

Objective | Native visualization | Interpretation boundary

Packaged lifecycle closure

Audits | Checks | Constructors and bundles | Execution workflows | Explicit writers | Extraction | Freezing and reproducibility | Model fitting | Other public workflows | Plots and graphics | Prediction | Readers | Simulation | Summaries | Tables and adapters | Validation

Grouped posterior predictive checks | Descriptive uncertainty decomposition

Backend-independent posterior tables | Publication graphics | Fitted-model extraction

Fitted predictions

Explicit output

Public API contract | Lightweight post-fit adapters | Explicit failure boundaries | Prediction diagnostics | Output safety | Why examples are selective

Visual audit

Frozen evidence as a visual object

Visual diagnostics

Simulation-based calibration view

Visual evidence atlas

Interpretation boundary | Methodological anchors

Purpose | 1. Simulate known data | 2. Declare the analysis contract | 3. Preflight the design | 4. Prepare and specify | 5. Freeze analysis provenance | 6. Optional dual-backend fitting | 7. Unified posterior review | 8. Cross-backend consistency | 9. Evidence and compatibility

Why a unified API? | Build a backend-independent specification | Inspect workflow progress | Fit through either approved backend | What the unified layer does not do

A manifest is an analysis contract about the analysis contract | Create a manifest before fitting | Freeze only when the analysis-defining fields are ready | Compare analysis provenance

Audit installed components | CmdStanR setup | Full MCMC only

Two interchangeable implementation backends, one modeling contract | Posterior-summary parity | Object-schema compatibility

Scope | Simulate a known data-generating process | Declare the model contract | Prepare and audit the data | Specify priors and inspect prior implications | Translate and fit with the optional backend | Diagnose, interpret, and validate | Prior sensitivity and simulation recovery | Structured reporting

Scope | Simulate durations with stored truth | Declare the duration contract | Prepare, convert, and audit | Specify priors and check prior implications | Translate and fit through the restricted backend | Diagnostics and posterior interpretation | Sensitivity, recovery, and reporting

Diagnose the design before invoking Stan | Missingness is described, not repaired | Fixed-effect geometry | Repetition, crossing and random slopes | One combined preflight

Orchestration without automatic robustness claims | Declare a suite before running it | Evidence is an inventory

Frequency/activity features | Event-related pupil deconvolution | Luminance and trial-order adjustment | Robust filtering

True response-mixture IRT | Nonparametric Rasch diagnostics | Stepwise item-reduction sensitivity | Biometric-feature imputation sensitivity | Process-informed Rasch trees

Scope | Bayesian diagnostics | 3PL response-process alignment

Experimental item-parameter seeding | Presentation/accessibility sensitivity

Conceptual feature blocks | Process-profile discovery | External validity

Scope | Pre-flight specification | Multivariate anomaly review

Pre-action and gaze-evidence representations | Frontier estimators remain gated

Partial and streaming scoring | Unified evidence bundles

Windowed process representations | AOI trajectory features

Why these functions are gated | Process-state HMM + IRT | Cognitive diagnosis and latent process classes | Latent-space IRT | Process-adjusted DIF | Sequence representations | Flexible item-response curves | Linking and person fit | Process-aware adaptive testing | Promotion rule

Why import success is not validation | Public validation corpus | Semantic round-trip contract | Timestamp semantics | Coordinate and pupil fidelity | BIDS eye-tracking semantics | HED event semantics | Evidence matrix

Measurement channels, not feature dumping | Registry | Response + response time + gaze | Graded responses | Nominal distractors + option gaze | Missingness as a process | Device and algorithm facets | Bounded gaze measures

Why preserve the response process? | Option-level process evidence | Local dependence must be checked | Revisiting as collateral evidence in cognitive diagnosis | Validation requirements

Purpose | Core model families | A process channel must earn its place | Missingness: separate exposure from response | Cross-device measurement is an estimand | Latent distribution and IRF stress tests | Promotion is evidence-based

A common validation contract | Recovery table | Prior SBC | Posterior SBC | Posterior predictive checks | Misspecification | External and grouped validation | Incremental information and negative controls | Process-dependent discrimination | Evidence grade

Purpose | Minimal workflow | Explicit specification | Item labels and conditions | Adding observed responses | Output structure | Interpretation boundaries

Scope | Simulation and observed-state models | Hidden states with Stan | Model comparison and recovery

Measurement stance | Specification | Diagnostics and scalar comparisons | Preprocessing sensitivity

Confirmatory parameter mapping | Identification and posterior checks | Simulation programme

Evidence levels | Registering a case | Fingerprinting, redaction, and semantics | Compatibility evidence

Bundled multimodal study | Integrity and expected outputs | Data dictionary and reproduction directory | Release audit

Purpose | Deterministic plans | Atomic execution and resumption | Collection and evidence | Promotion remains gated

Contracts | Partitioned storage | Schema migration and benchmarks | External engines

Prespecification before fitting | Classification uncertainty | Sensitivity and competing heterogeneity

Dynamic gaze-state IRTree baseline | Functional pupil-informed IRT | Theory-defined strategies | Gaze-informed diffusion | Monte Carlo design | Evidence promotion gate

Eye-Tracking-BIDS | Parquet and Arrow | Package ecosystems

Parameter recovery and interval coverage | Grouped validation and leakage | Preprocessing multiverse | Reproducible release assets | Explicit confirmatory gates | Systematic advanced-model grid

Profile before import | Folder import | Gazepoint-specific audits | Separate biometrics and synchronization | Gazepoint Analysis 7.2.0 paired exports

Automatic detection | Generic mapping | Dedicated vendor adapters | Preserve source meaning

Declare preprocessing | Static and dynamic AOIs | Declared features | Sensitivity analysis

Response and response-time matrices | Conventional and explanatory IRT | Accuracy and response time | Process-informed and experimental models

Reproducible records

Why empirical validation is separate from declared support | Validate one source | Build a validation corpus | Produce a safe validation bundle | Recommended real-export acceptance process

Core installation | Optional fitting and validation dependencies | Windows toolchain check | Backend preflight | Minimal compilation smoke test | Clean-process package checks

Why estimands are first-class | Binary probability standardisation | Duration median and predictive-tail estimands | Structural sensitivity | Participant and item deletion | Parameterisation sensitivity | Exact K-fold validation

Purpose | Binary strict readiness | Identifier-like predictors are review signals | Duration extremes, impossible ranges, and censoring | Traceability

Replay recorded transformations | Detailed binary PPC | Detailed duration PPC

Why an explicit contract layer? | Audit the currently loaded package | Inspect an object schema

Clean manifest | Deliberate post-outcome contamination

Boundary, not duplication | Prepared handoffs | Combine only after validation

Declared task | Explicit candidate grid

Observed endpoint | Evaluate and quantify participant-cluster uncertainty

Development workflow | Explicit independent external dataset | Explicit absence of external validation

Synthetic upstream outputs | Assemble the modelling handoff | Declare task and provenance | Participant-grouped resampling | Repository-aware evaluation when available

Nested grouped-resampling workflow

Capability table

Participant-grouped generalization workflow

Scope | Materialized grouped folds | Fold-local evaluation | Interpretation

Deterministic documentation mode | Audit generated text

Participant-and-stimulus generalization workflow

Stimulus-grouped generalization workflow

Capability audit | Separate interaction priors | Full-MCMC backend selection | Separation screening | PSIS-LOO and model averaging | Power-scaling sensitivity | Simulation-based calibration

Binary scenarios | Duration scenarios

Why bounded subsequences? | Synthetic sequences | Enumerate occurrences | Sequence-level prevalence | Group comparison | Relation to specialist packages

Why declare the design? | Synthetic randomized groups | Declare and test | Bootstrap interval | Causal language

Synthetic data | Sequence index | State distribution and entropy | Distance and clustering diagnostics | Transition network

Scope | Synthetic data | Prepare and audit the panel | Summarise occasions and states | Quantify within-panel change | Reporting

Scope and guardrails | Synthetic data | Multichannel model | Covariate-dependent model | Reporting

Model target | Synthetic repeated sequences | Fit and inspect | Predictions | Transition outcomes | Interpretation

Start with the research question | Contiguous motifs | Consensus sequences | Descriptive group comparison | Distances | Clustering, representatives, ensembles, and stability | Transition networks | Higher-order transition models | Hidden and mixture sequence models | Optional adapters | Minimum decision checklist | Focused articles

Purpose | Synthetic long-format data | Audit, validate, and prepare | Encode states and inspect basic structure | Discover contiguous motifs | Compare sequences through distances and clustering | Consensus and descriptive group comparison | Transition structure | Continue with focused articles

Study objective | Synthetic study data | Prespecified preparation | Structural summaries | Recurring contiguous motifs | Consensus and condition contrasts | Distance, clustering, and representatives | Transition network and recent-context model | Compact categorical HMM sensitivity description | Assemble report-ready evidence | Interpretation boundary

Why preparation is explicit | A deliberately problematic synthetic input | Audit without modification | Compact validation contract | Apply explicit preparation policies | Revalidate the canonical result | Errors that require source correction | Reporting recommendations

Different validation questions | Prior-scale sensitivity | Simulation-based recovery | Minimum repetition rule | Failure handling | Interpretation

Evidence layers | Numerical diagnostic contract | No automatic convergence statement | Diagnostic plots | Responding to failures | Reporting language

Scope | Synthetic sequences | Aligned-position consensus | Descriptive group comparison

Interpretation boundary | Synthetic categorical sequences | Categorical HMM | Mixture HMM | Estimation limitations | Optional ecosystem adapters

Synthetic paths | Transparent distance families | Clustering and validation | Subsampling stability | Co-association ensemble | Optional PAM and CLARA interfaces

Synthetic navigation sequences | First-order transition network | Higher-order transition model | Whole-sequence bootstrap

Scope | Synthetic ordered-state data | Extract contiguous motifs | Summarise, filter, and format motifs | Summarise motif positions | Plot motif prevalence | Plot motif locations | Interpretation boundary

Overview | 1. Load example data or import your own exports | 2. Validate the master table | 3. AOI geometry and coding checks | 4. AOI entries and AOI-window summaries | 5. AOI-window GLMM | 6. AOI, fixation, and transition features | 7. Advanced sequence-model preparation | 8. AOI time-course GAMM | 9. Diagnostics and reporting

Overview | 1. Load example data or import your own exports | 2. Validate the master table | 3. Light pupil preprocessing branch | 4. Conservative artifact-cleaned branch | 5. Pupil quality audits | 6. Reliability and sensitivity checks | 7. Confirmatory pupil-window LMM | 8. Pupil time-course models | 9. Cluster permutation and divergence point | 10. Reporting

Overview | Load the package | Built-in synthetic kiosk demo | 1. Import Gazepoint biometric exports | 2. Inspect real-data readiness | 3. Run the full workflow | 4. EDA, GSR, and SCR examples | 5. Pulse, IBI, HRV, and respiration examples | 6. TTL alignment and model-ready data | 7. Reporting bundle | 8. Feature inventory | 9. Interpretation guardrails | 10. Private real-data smoke tests