Package: mfrmr 0.2.4

Choose the result you need | Put the network at the stage its edges describe | Quick start | Inspect individual estimates | Check what needs review | Use your own CSV | 1. Read and inspect the file | 2. Review the rows and the rubric | 3. Fit and read the results | If a check stops you | If criteria occupy separate columns | Continue with diagnostics and reporting | 1. Check the data and estimation record | 2. Read precision and model fit before interpreting differences | 3. Identify what needs closer inspection | 4. Assemble the evidence for a report | 5. Display a table and save the analysis | Further analyses should answer a stated question | An observed writing-assessment example

Follow the result through the existing API | Does an unavailable interval mean interval estimation is impossible? | An explicit corrected-JML workflow | Score a new cohort using the corrected calibration | Describe the observed ratings at the corrected estimates | A provisional two-family workflow | What the paired automatic-integration study supports | Saved-input replay after aligning optimizer and interval checks | Planning a practical two-family analysis | Estimate the same response model in Stan | Inspect observed ratings with posterior residuals | Scoring new Persons with the fitted two-family calibration | Model, estimation method and algorithm are different choices | What happened to the name "bounded GPCM"? | What changes if I use JML? | What can I report under MML? | Which uncertainty claims are supported? | How uncertain is a relative slope? | Choosing the interval for the question | Examine one relative discrimination with a profile likelihood | A fitted-model bootstrap alternative | Why an information warning needs a parameter-specific review | Uncertainty in probability and information curves | What the additional interval checks found | Checking the updated estimator on the affected design | Carry the selected inference into reports and figures | Check that the integration grid does not change the decision | Why intervals, information criteria and tests need separate checks | Relationship to generalized many-facet models | When would two slope families help in practice? | Start from an actual rating table | What if the analysis could affect a championship? | Can EM estimate two slope families? | What does changing the ability SD test? | Do two slopes handle testlets or local dependence? | How does this relate to multivariate G theory? | What can the related facets literature add? | Which facet receives different slopes? | Criterion discrimination with rater-specific category use | An explanation for a high-school reader | Before fitting: model-choice triage | Comparing PCM and GPCM | Save the comparison and carry its limitations into reporting | Response-family boundary | GPCM is not one estimator | Why total-score conditioning does not remove ability | Report wording templates | Checking the support boundary | Slope values, boundaries, and uncertainty | What can and cannot be compared across programs | Model structure: which effect does discrimination change? | Exact overlap and what has been checked | Inference: same estimates do not imply the same intervals | Source-grounded recovery interpretation | Supported routes and verification evidence | What works with caveats | What is intentionally restricted | Recommended substitutes | A worked example | Estimator references | Current boundary

First example: reuse selected criterion locations | Limits that matter before using real data | Annual speech contests with common recordings | Identify the performance, judge and rating occasion | Choose what is shared and what is allowed to change | Read the existing helper results for their stated purpose | Which uncertainty is available with anchors? | Reusing an analysis saved before the centering change | Comparing an anchor with TAM or ConQuest | Fit and diagnose the observed ratings | 1. Check connectedness first | 2. Export anchor candidates | Design the link, not only the percentage | 3. Compare group residuals | 4. Refit DFF when subgroup comparisons are defensible | Review a particular school comparison before fitting | Preserve that question alongside a residual screen | Estimate the declared contrast in one joint calibration | Estimate an identified comparison when the full model is deficient | Obtain CML uncertainty for the identified comparison | Standardize an MML interaction contrast by that fit's population SD | Comparing with MH or ordinal logistic DIF | Compare an explicitly shared response roster | 5. Cell-level follow-up | 6. Model-estimated facet interactions | 7. Multi-wave anchor review | Recommended sequence | Related help | References | Optional: explore rating assignments

Start with the decision you need to make | Choose what the response model needs to explain | Choose how Person ability enters estimation | Which combinations are currently implemented? | Check the population scale before comparing results | Why not start with the most complex model? | Compare conclusions, not just fit statistics | Continue to the analysis and reporting

Inspect the design before estimating a model | Compare score networks with MFRM | Model network configurations with ERGM | Connect to a manuscript, not a coefficient inventory | Export values and choose your own graphics | References

Declare what shares an effect | Fit and check the calibration | Review fixed-rater feedback | Score persons and interpret their intervals | Missing assigned scores and sparse designs | Reuse and save the result | Connect to the common graphics and table workflow | Compare fixed effects with an ordinary RSM | Inspect posterior predictive residuals | Compare predictions on the same definition | Compare Person scores and inspect model-aware maps | Build a report from saved results

Choose attributes and their meaning | Distinguish unrecorded and undefined values | Fit an explicit imputation model | Describe groups and compare choices | Compare feature selections | Interpret the retained profiles | Read separation, feature profiles, and imputation sensitivity | Custom distribution plots | Compare PAM with a hierarchy | How this relates to MFRM rater-type research | Link separately grouped persons, raters, and tasks | Distinguish sparse assignment from missing attributes | Many features, k-means, and PCA | Reopen an analysis without repeating imputation or clustering | Plan for the size of the analysis

Choose report sections by question | Start with an editable GPCM report | 1. Fit once and retain the analysis record | Show what was planned, observed, and excluded | 2. Check uncertainty and fit before writing the answer | 3. Show estimates, uncertainty, and the answer | Keep separation, agreement, and category functioning distinct | If the question concerns observed versus adjusted scores | If the score will determine pass or fail | A short Methods and Results example | 4. Review gaps and generate draft text | Manuscript coverage map | 5. Save the analysis and complete the references | References

Find a figure by purpose | Latent regression: coefficients and specified comparisons | Connect a figure to a claim | Minimal setup | 1. Targeting and scale structure | Establish what higher scores mean | Wright map | Fit pathway | Category probabilities | 2. Local response and level issues | Strict marginal follow-up | 3. Linking and coverage | 4. Residual structure and interaction screens | Checking one substantive dimension without fitting a multidimensional model | Shared Persons, sparse assignment and the choice of PCA columns | A residual block is not automatically another substantive ability | Residual networks, EGA and the assignment graph answer different questions | 5. Custom figures without losing the evidence boundary | Save a figure with its values and notes | Set an appearance once for several plots | Use a common title argument | Fair Scores and annotations outside the figure | 6. Secondary visual layer | Save the agreement result and its review rules | Response-time QC context | Small-N shrinkage with uncertainty | 7. Attributes and assessment planning | External-feature hierarchy | D-study planning | Portable score-batch review | Recommended sequence | Related help

Mental model | Visualize an existing FACETS analysis | Why results and decisions can differ | When a native definition has a useful purpose | Decisions near a cutoff | Ordinary JML, corrected JML and FACETS | Using TAM JML as a third comparison | A one-shot legacy-compatible call | Translating the specification file | FACETS and labels | Anchoring | Bias and interaction | Wright map / variable map | Fit df and ZSTD review | Group anchoring and DFF | Reviewing output contracts and fit tables | Producing FACETS-style output files | Recommended next steps

Keep assignment separate from missing scores | Use a joint score model | Inspect sampling and preserve provenance | Return completions to the original roster | Check the missing-score predictions | Analyze every completion on the same scale | Pool the rater contrast, including its covariance | Compare with inference from the observed scores | What repeated sampling showed | Challenge the missingness assumption | Save enough to review and reproduce the result | References

Why the MML calculations are shared | Mathematical Core | Current MML scope | When estimates change with quadrature order | Numerical readiness and bounded polishing | Choosing the iteration ceiling | Strict Marginal Diagnostic Target | Diagnostic Basis In The Package | Literature Positioning | Interpretation Boundaries | Grouped-count variance is a screening approximation | Documented simulation-check scope | GPCM model and inference limits | Equal weighting as a model-choice principle | Practical Reading Order | Analyses not provided by mfrmr | Recommended Expert Reading Of Package Output | Key References

Define the score and the population | Is a criterion a facet or a score? | Load ratings and inspect their design | Inspect the sources of variation | Compare feasible assessment plans | Explain which error sources change | Read a planning curve | Ask about a change, not just two separate intervals | Save the result and know which operation to use | Report the assessment-planning answer | Adapt the question to another setting | Examine a construct claim | Other designs and uncertainty questions | Common tasks without a rater facet | Different rater teams for each task | Compare interval methods on one shared axis | Incomplete data or unavailable results | Where to find calculations, assumptions and references

Calibration session | Fresh scoring session | Conditional ability-level probabilities | Reuse a GPCM calibration for a later cohort | If scoring stops or asks for review | When the calibration was estimated with JML | Using GPCM JML slopes in the same workflow | What supports these scores?

Choose the question before the coefficient | 1. Identify the rated unit | 2. Compare pairs on their shared responses | 3. Separate agreement, correlation and mean differences | Add a second reference only when the design supplies one | 4. Describe panel agreement under a declared definition | 5. See the disagreement, then compare the coefficients | Matched-score counts | Pairwise QWK with its sample size | 6. Save the evidence and state the answer | When a result needs investigation

Fit the shared-rater model | How accurate is the calibration likelihood near the estimate? | Give feedback about the observed raters | Choose a view and adapt its presentation | Compare the same observed panel with fixed raters | What the bounded validation shows | Earlier fixed-population evidence | Estimate population variation, including zero | Compare intervals using complete model refits | What the bootstrap comparison shows | Score Persons without treating raters as known | How close are these scores to the joint posterior? | Distinguish an observed rater from a replacement rater | Keep assignment and missing scores explicit | Inspect posterior predictive residuals | Compare predictions on the same definition | Compare Person scores and inspect model-aware maps | Save the complete analysis | Compare with a fixed-rater analysis

Fit the model and identify the comparison | Estimate a rater difference without an ability population model | Standardize by the same fit's estimated ability SD | Read fixed values and group constraints correctly | Prepare a sheet for one rater | Save the analysis and keep a copy of the sheet | Keep interpretation with exported tables | Carry the same intervals into figures and reports | Declare what can be treated as independent | Keep unavailable results visible | Understand what the sandwich can and cannot address | What the bounded simulation showed | Save the analysis and use the plotted data | References

Keep failed and incomplete trials in the planned roster | Connect simulated ratings to a declared screen | Compare thresholds without refitting the models | Distinguish low mean squares from ZSTD-only warnings | Check the mathematical target before interpreting a threshold | Design an accuracy study around the feedback decision | Evidence and limits | Stress-test thresholds, sample size and low-side warnings

A reproducible assessment with parallel criteria | Use a fixed rating budget to compare task allocation | Unequal blocks: five criteria versus two | Does a large local variance diagnose halo? | Can each task have its own dependence variance? | Save the actual inputs, assumptions and scores | Appendix: complete four-rating calculation | References