NEWS

---

magmaan 0.2.2

• Restored model handles now use the lab's native pointer-address check, so serialized models rebuild correctly across platform-specific pointer displays.

• Portable package checks now run on Windows, macOS and Linux with current dependencies. Live lavaan comparisons accommodate 0.7-3; the pinned lavaan-0.7.2 preset retains its own oracle.

• Complete-data endpoint audits now support simple equality-constrained std.lv charts, including loading and latent-mean ties across factors or groups. Released marker ties preserve the constrained model dimension. Automatic constrained starts retain the existing normalized-unit fallback.

• The embedded lab fit now retains read-only deletion provenance by group and reason. The summary continues to show rows supplied, used and deleted.

• Complete-data GLS fits now use PORT by default, retaining the layered start and information-refined coordinates. Estimates at the same optimum are unchanged; iteration counts and runaway paths may differ. ML keeps L-BFGS.

• Complete-data ML and GLS fits with standardized extent above 10 are no longer reported converged when their supported std.lv endpoint Newton audit fails. Diagnostics record the trigger and unsupported transports; estimates are unchanged.

magmaan 0.2.1

• Supplied start tables and previous fits now appear as requested$starts = "table" and effective$starts = "<resolved convention>+table" in fitting reports, identifying first-attempt values and the unmatched-parameter fallback.

• Nested tests now refuse boundary_nesting only when a required tangent loses rank, rather than because the fitted factor covariance is improper or singular. PSD-boundary comparisons compute under the same interior-population assumption as global tests and retain psd_boundary; anova() also flags computed table rows. A 500-replicate weak-factor null check at each of N=100/200 gave score rejection 5.6%/6.0% and LR 6.4%/7.0% at nominal 5%, including 10.0%/1.8% improper null fits (SB and PEBA4, one restriction).

• The lavaan-0.7.2 marker component now also covers FIML and ordinal/mixed DWLS, using each route's saturated H1 covariance. Switched fits replay the selected fitting attempts and revert when rejected; marker = "default" keeps the requested markers.

• Fully fixed models with latent regressions or higher-order factors and fixed means now return their implied moments without crashing in the stationarity audit.

• Explicit quadratic references (including summary/anova SB and PEBA4) now clip roundoff-negative spectrum entries using the projected-score validator; materially negative spectra still raise a numeric error.

• Ordinary magmaan(estimator = "ML") on complete all-ordinal data now uses association ML, exact Stage-1 IJ covariance with observed sensitivity, All global/nested references, and robust modification indices/equality releases. DWLS remains the ordinal default. Global tests can be liberal in finite samples; see the association-ML help caveat and decisions/08 evidence.

• Identification refusals now suggest specific scale, location and rotation restrictions. Information deficits name affected parameters without suggesting an automatic fix; no identifying constraints are added.

• The lavaan-0.7.2 FIML preset now uses lavaan's scale-dependent H1 covariance ridge and retains stalled EM moments. fit$fitting$h1 records H1 convergence and update/repair counts. Converged H1 moments retain strict oracle agreement; stalled H1 endpoints are outside the compatibility contract. Native FIML H1 defaults are unchanged.

• Lab optimizer controls now accept opt-in objective and objective_downward scaling for continuous fixed-weight LS and ML2S Stage-2 LS, using the fitting GN diagonal. Defaults and audit rules are unchanged. ML/direct FIML and residual backends reject these modes as Unsupported. Ordinal and mixed routes currently ignore all coordinate scaling modes pending TASK-33.10.5.6.

• magmaan_model() now caches and prints structural identification. magmaan() refuses unidentified models before fitting with magmaan_identification_error, naming the free parameter directions. Unchecked models fit normally; deliberate ridge fits remain in magmaanlab.

• The lavaan-0.7.2 preset now adapts weak markers for complete-data ML. marker = "default" keeps the requested identification. Fits retain the actual fitted specification and a switch/revert table; nested inference explicitly refuses switched marker coordinates. FIML and DWLS adaptation follows separately; their preset currently reports marker = "default".

• Fits using convergence = "lavaan-0.7.2" (including the preset) report lavaan's post-estimation check separately from convergence. The six flags are available in fitting$post_check on the lab fit; ordinary fits expose them through as_lab_fit(). This adds no warning or inference restriction.

• ULS fits and the continuous part of mixed DWLS/WLS fits now warn when within-group observed variances differ by more than a factor of 1000, with advice to rescale. ML, GLS, FIML and ordinal-only fits do not emit this advice. The diagnostic is computed by the shared C++ core.

magmaan 0.2.0

Ordinary API and migration

Calls written for 0.1.0 need changes. Removed arguments raise errors naming their replacements.

• magmaan_model() constructs a reusable model with a frozen data schema. prototype declares groups and ordinal categories; factor levels allow a zero-row prototype. Schema violations raise magmaan_schema_error.
• magmaan(model, data, estimator, covariance, inference, options) fits that model. Syntax strings and lab specifications construct it from data; undeclared ordered factors are rejected. inference = FALSE estimates only, and infer(fit) adds inference later.
• Every model has a mean structure. Saturated intercepts add entries to coef() and vcov() without changing other estimates or inference.
• Observed covariates are random: fixed.x is removed. ML structural estimates stay unchanged, while GLS/ULS estimates can change in overidentified models. Lab specifications with observed covariates and fixed_x = TRUE are rejected.
• covariance = "unrestricted", "psd", "barrier" or barrier(lambda) replaces psd = TRUE. Barrier fitting is experimental and has no inference (penalized); barrier(0) gives the unrestricted fit.
• options$start replaces top-level start and options$starts. It accepts "default", "fabin3", "lavaan-0.7.2", a previous fit or a parameter table, and overrides preset starts.
• missing, cluster and meanstructure are removed. Estimators other than FIML and ML2S delete incomplete rows listwise; pairwise deletion and two-level fitting remain lab features.
• Group order follows prototype factor levels, setting the latent-mean reference group.
• Mplus specifications from magmaanlab::mplus_model() retain source, groups and category schema through fitting and serialization. Unsupported inputs raise magmaan_mplus_error with all required edits; model printing shows fittability. Plain strings continue to mean lavaan syntax.

Inference policy and evidence

• ML and FIML provide observed-bread, empirical-score sandwich covariance and global/nested score and LR tests. Score tests use observed sensitivity with an expected-information metric; nested LR calibration uses empirical scores at the larger fit, including restricted, misspecified means. Components retain separate typed unavailable reasons.
• All-ordinal and complete mixed DWLS use exact empirical first-stage influence and estimated-weight sandwich covariance. Global tests use the fit-function statistic; nested tests use its difference with observed-Hessian geometry. OPG fitting weights and estimates are unchanged. DWLS has no LR test and no nested score test; ordinal ULS/WLS have no ordinary policy inference.
• DWLS nested comparisons include moment-nested Wu–Estabrook threshold invariance. Three-category threshold steps report equivalent_models; released-scale thresholds-plus-loadings versus loadings-only is not_nested.
• DWLS global and nested tests use the exact weighted chi-square All reference. Exact-first-stage and reference-law studies reconfirmed the policy across ordinal, mixed and threshold-invariance designs and eight textbook models. These studies support the default without guaranteeing nominal size in every finite-sample setting.
• anova() retries a worse-fitting larger model from the embedded restricted estimate. A better endpoint is used only with a passing native verdict; input fits stay unchanged. Recovery is recorded in refit; internal warnings are retained in attr(result, "reseed")$warnings and printed as notes.

Test tables and references

• summary() and anova() use columns test, statistic, df, reference, pvalue, recommended and reason. Migrate p.sb/p.peba4 to pvalue selected by reference == "sb"/"peba4"; select tests by their stable codes score, lr, fit_function and fit_function_difference.
• ML/FIML default to PEBA4 only; DWLS global and nested tests default to All. references = c("sb", "peba4") requests the former comparator p-values. Other supported reference laws reuse the same statistic and spectrum. recommended marks only the primary policy test with its default reference. LR rows remain visible with a small-sample over-rejection note.
• Unavailable tests retain typed rows. Compatibility tables add unscaled statistics, scales and shifts; references cannot accompany lavaan_compat.

Modification indices

• modindices() combines robust one-df score indices and affine equality releases for ML, FIML and all-ordinal/mixed DWLS. Candidates include absent structural regressions among variables already in equations and reverse paths. Candidate order and typed unavailable rows are preserved.
• ML/FIML use observed nuisance sensitivity with expected-information metrics and EPCs, so negative observed release curvature alone no longer removes a candidate. DWLS uses exact first-stage rows and estimated-weight influence.
• test = "lr" refits candidates from embedded null starts and applies the nested policy (LR for ML/FIML, fit-function differences for DWLS).

Standardized estimates

• summary(fit, standardized = TRUE) adds std.lv/std.all estimates and delta-method SEs using the selected covariance, plus endogenous-variable R-squared and SEs. Fixed markers, groups and ordinal parameters are included. Without covariance, estimates remain and SEs have typed missing reasons; defined parameters have no declared standardized scale.

Fit measures

• fit_measures(fit) provides misspecification-robust policy point indices with a matching independence baseline and typed unavailable reasons. summary(fit, fit_measures = TRUE) attaches and prints them on request. Policy intervals and close-fit p-values await evaluation.
• fit_measures(fit, lavaan_compat = ...) provides lavaan's standard, scaled and robust families, intervals and close-fit p-values for supported bundles. FIML ML/MLR measures are validated against lavaan and use tighter saturated H1 EM precision.

Lavaan compatibility

• options$preset = "lavaan-0.7.2" selects lavaan starts, search and acceptance rules for continuous ML/FIML and all-ordinal or complete mixed DWLS, with unrestricted covariance and supported linear equalities. Explicit options override the preset. Prepared DWLS fits accept these options; ML2S retains its fresh-fit fallback.
• vcov(), confint(), summary() and anova() accept named bundles through lavaan_compat; infer() can cache an additional bundle. Complete-data ML supports ML/MLM/MLR; FIML supports ML/MLR. All-ordinal reporting supports DWLS/WLSMV, ULS/ULSMV and WLS, including default nested difference tests.
• Complete mixed DWLS supports WLSMV covariance, Wald intervals, scaled-shifted global tests and Satorra-2000 nested tests, including grouped delta/theta fitting. Missing mixed observations and other mixed bundles remain unsupported.

Fixes

• Grouped all-ordinal WLSMV covariance now preserves group-weighted sandwich geometry with the reporting denominator, correcting unequal-group SEs.
• Exact ordinal/mixed first-stage influence uses affected score blocks, pair counts and sparse Gamma-diagonal derivatives, reducing post-fit cost.
• Mixed modification indices now share the equality-release criterion scale, removing an extra factor two while preserving EPCs and fitting weights.

magmaan 0.1.0

First simulation prerelease, paired with magmaanlab 0.1.0. The ordinary API accepts lavaan syntax and lavaan-backed model specifications. EQS is lab-only.

Development is unfinished: remaining bugs and incomplete coverage of the main estimators and inferential procedures remain to be resolved and validated. Version 0.1.0 provides a versioned simulation snapshot, not a finished release.

• Saved specifications supply ordered variables, parameterization and grouping; conflicting call options error before fitting. Row accounting uses the resolved grouping.
• Defined parameter estimates are retained independently of inference.
• Confidence levels and parameter selections are validated. Unchecked convergence is distinguished from a failed fit.
• Help pages and the README document extraction for simulation consumers.

Automatic policy inference covers single-level complete-data ML. Other estimators retain their estimates and report unsupported inference components. This prerelease does not expand inference coverage or change its defaults.