PsychoMatic 1.0.1

Fix report-path test assertions on macOS, including CRAN’s M1 R-devel environment, by comparing canonical filesystem destinations instead of literal /var and /private/var aliases. The same checks support Windows short paths and repeated path separators.
Add regression coverage for output-path identity before and after creating a report, while retaining file-existence and overwrite-protection checks.
Include the corrected tests in the source distribution. Statistical functions, numerical results, public APIs and dependency requirements are unchanged from 1.0.0.

PsychoMatic 1.0.0

Release scope

Change the package license from MIT to GPL version 3 or later (GPL (>= 3)), copyright 2026 José Gamarra-Moncayo.
Consolidates the unreleased 0.4.0 development series into the first stable 1.0.0 API. The preceding published release is 0.3.0.
Complete English and Spanish reference PDFs are generated reproducibly from every documented topic; editorial sources stay outside the source tarball.
See docs/MIGRATION_1.0.0.md for warning-table compatibility and workflows.
esem_auto() now defaults to seed = NULL, using the session RNG unless the investigator supplies a reproducible seed. Its console omits preparation, internal maps and versions while preserving them in objects and exports. References precede the final Interpretation section in both languages; the generic closing manuscript sentence was removed. Negative degrees of freedom independently block interpretation, even without a native warning; zero degrees of freedom remain a separate saturation case.
Add esem_auto() for a single measurement block with target or geomin oblique rotation, native rotated inference, input mappings and admissibility gates. Raise minimum lavaan to 0.7-3. Provide descriptive CFA comparison, bilingual console/Word/Excel reports and a final numerical manuscript draft. Documentation explains rotation choice, soft targets, epsilon and inference limitations.
invariance_partial() now defaults to the same native lavaan settings, item preparation and metric/scalar/strict constraints as factorial_invariance_auto(). Explicit WLSMV/ULSMV default to ordered indicators, marker identification and delta parameterization. Existing positional arguments are preserved.
Added fit_variant (scaled by default, robust on request), missing and min_group_n. Variant selection propagates to tables, comparisons, admissibility gates, interpretations and bilingual console/Word/Excel reports; unavailable robust indices have no fallback.
Identified threshold and latent-response partial invariance remains available through explicit workflow = "ordinal". semTools is required only for this workflow. Researcher rationale, retained anchors, effective release and exact matched-prerequisite nesting checks remain in place.
Simplify traditional invariance console and reports: display observed response options instead of the measurement-level label and omit the item coding table. Raw coding metadata and all model estimation settings remain available.
Reproduce native lavaan calls in the traditional four-stage workflow. factorial_invariance_auto() now always fits only configural, metric, scalar and strict models, constraining loadings, intercepts and residuals. Explicit WLSMV/ULSMV default to ordered = TRUE, marker identification and delta parameterization, matching native CFA defaults. Automatic selection uses WLSMV for detected ordinal responses and ML/MLR otherwise. Explicit ML/MLR/ULS analyze numeric scores. Ordinal comparisons without positive additional df remain unevaluable. Use ordinal_invariance_auto() for identified ordinal stages.
Add fit_variant = c("scaled", "robust") to both invariance APIs. Scaled is the default; explicit robust selection governs CFI/RMSEA, changes, decisions and reports. Corrected chi-square/df/p remain scaled; SRMR is base. Requested unavailable robust indices are not silently substituted.
Add ordinal_invariance_auto() with exactly configural, equal thresholds, and equal thresholds and loadings stages, using Wu-Estabrook identification. Restrict estimators to WLSMV/ULSMV, mark untestable threshold stages explicitly, and provide bilingual reader tables and Word/Excel reports with APA references to Wu and Estabrook (2016), Svetina et al. (2020), and Chen (2007).
Preserve signed fit changes and use the loading-specific SRMR guideline when adding loading equality to the ordinal sequence. Chen cutoffs remain descriptive extrapolations for ordinal outcomes.

This is a local development candidate until the release checks and publication process are completed.

Default report destination

Save reports in the current R working directory (getwd()) when no output path is supplied, including per-analysis reports, integrated reports and EFA/CFA exports. Generate unique timestamped names and share one base name for Word/Excel pairs. Preserve explicit paths and overwrite protection.

Network manuscript presentation and APA typography

Enlarge fixed node labels and stability values; add label_cex for plots and exports. Add explicit model, algorithm, and bootstrap_type choices to the network workflow, with accurate methods and TMFG edge captions. Defaults remain GLASSO, consensus Louvain and resampling. BGGM availability depends on EGAnet (disabled in 2.4.1); never silently substitute an estimator.
Harmonize community colors with distinct blue, coral, green, purple and gold fills; use opaque node/point backgrounds and matching legends. Keep positive edges neutral and negative edges dashed. Loading maps add a signed color key and contrasting black/white cell labels without changing estimates.
Include signed cross-loadings beneath primary loadings and a complete item stability table beneath community structural consistency.
Add combined EGA/item-stability panels, colored replication dot plots and soft community fills. Two-group networks mark complete BH loading tests, retaining unavailable tests and exploratory configural interpretation.
Classify loading magnitude using unrounded values and explain rounded boundary values in reader output.
Default network output emphasizes community membership, bootstrap structural consistency, centrality stability and separate NCT/loading interpretations.
Retain full technical conditions in the object and print(..., detail="audit"); translate unavailable assessments and inadmissible CFA into scientific caveats.
Display only BH-significant loading differences in the compact item table; missing tests and unsupported configural structure remain explicit.
Share partial APA reference italics across analyses in compatible consoles, Word and Excel, while preserving plain reference strings and numeric results.
Select relevant publication figures by default; retain all diagnostic types on request. Add community legends, consistent colors, signed edges and BH-aware loading difference plots. Separate TEFI comparison sets into panels.

Network runtime and correlation corrections

Add ordinal corr="polychoric" to network estimation and resampling.
Show stage/substage progress, elapsed seconds and optional-stage interruptions.
Avoid unused consensus Louvain in fixed-partition loading permutations and respect requested consensus iterations in hierarchical bootstrap.
Serialize a self-contained bootnet estimator for Windows workers, independent of other installed PsychoMatic versions.
Distinguish network hierarchy/TEFI from supplementary CFA and explain negative variances, WLSMV AIC/BIC and second-order components with three factors.

Psychometric networks (2026-10-03)

Make UVA explicitly optional: NULL (default) skips it, "diagnostic" reports candidates and reasons without removal, and "removed" authorizes justified native reduction. Remove interactive prompts in every execution environment. Record proposed/actual removals, representatives, scores, ties and the decision in console and bilingual Word/Excel reports.
Show regularized partial correlations on EGA edges and item labels in nodes. Plot primary revised loadings on item-dimension edges; preserve all cross-loadings in the separate loading matrix figure.
Accept integer-valued seeds in EGAnet resampling and permutation calls; use session RNG by default (seed=NULL) and preserve caller RNG with explicit seeds.
Show only UVA pairs meeting the declared cutoff; keep full pairs and package versions as internal diagnostics. Explain unavailable RI partitions and skip their structural bootstrap without blocking main stability or invariance.
Attribute warnings/errors to their analysis stages and omit empty reasons.
Add network_auto() with optional pooled UVA, Spearman/EBICglasso consensus Louvain EGA, revised network loadings, centrality and entropy diagnostics.
Preserve original unrecoded responses for riEGA; evaluate identifiable correlated, second-order and bifactor candidates with explicit exploratory recommendations.
Add bootEGA structural consistency and bootnet edge precision/case-drop stability.
Compare all observed group pairs using NCT and EGA loading permutations, with full-precision raw p values and four declared BH families.
Add cached publication figures, bilingual Word/Excel reports and an APA 7 methodological reference registry. No participant rows appear in reader reports.

EFA omega consistency (2026-09-26)

Use the shared standalone reliability engine for EFA omega per dimension, including measurement-level detection and inadmissible-model checks.
Explain the independent one-factor calculation in console and reports; retain CFA reliability from the user-specified joint model unchanged.

Model-comparison estimator and index correction (2026-09-25)

Default model comparisons to automatic estimator selection using the same selection helper as CFA, with an explicit estimator override preserved.
Display estimator, selection source and rationale. Use scaled CFI/TLI/RMSEA for corrected estimators in both the comparison and reader output; retain base SRMR and do not silently replace unavailable scaled values.
Allow explicit fit_variant = "scaled", "robust", or "base" for both the displayed comparison and recommendation inputs.
Suppress dependency startup messages while loading MVN, without suppressing model warnings or errors.

Coefficient interpretation and help revision (2026-09-23)

Show descriptive adequacy and explicit cutoffs for reliability (.70) and AVE (.50), preserving computational failures and the original numeric results.
Simplify item-analysis sample-size display; retain per-fit sample sizes internally.
Link numbered CFA suggestions to a dedicated notes section.
Complete Spanish invariance labels and missing-data warnings; include the Hirschfeld and von Brachel DOI.
Replace programming guards and simulations in public help examples with adaptable user-data syntax.

Reader-output corrections

Avoid duplicate console titles; add bilingual direct console and Excel/Word reporting to screen_items, preserving its raw data-frame columns.
Shorten measurement-invariance output without changing decisions or diagnostics.
Present alignment tables with compact headers, separate flagged groups and explicit group names, latent means and latent standard deviations.

Language and statistical reporting revision (2026-09-22)

Public analysis functions now default to English; Spanish is explicit.
cormat() defaults to automatic selection using category count and observed distribution shape, explains the decision, names the method, and cites sources.
CFA computes Mardia regardless of estimator, distinguishing observed ordinal codes from latent-response assumptions. Alpha is labelled Alpha (Cronbach).
Invariance uses scaled indices for corrected estimators (base SRMR), with no silent robust fallback, and consistently computes current-minus-previous deltas.
Configural gates and change limits are visible. Removed automatic classification of unknown Chen scenarios; ordinal extrapolation and threshold identification are explained with the supplied Hirschfeld/von Brachel and Bowen/Masa sources.

Minimal-interface revision (2026-09-17)

Removed public check_items(), dimensionality_sensitivity() and audit_report() entry points; validators and audit metadata remain internal.
Moved demo datasets to test fixtures, removed their data catalog entries, and made installed examples self-contained. S3 methods remain internal and are hidden from the regular help index.
Added automatic congeneric omega, corrected item-rest correlations, and reliability after item deletion. Ordinal omega targets observed scores; ordinal alpha targets latent responses. Non-evaluable results retain reasons.
Added observed and ordinal alpha per eligible CFA factor/group. Clarified standardized loading intervals, omitted redundant hierarchy columns, explained why Mardia is not applied to ordinal models, localized inadmissible conclusions, and retained original lavaan fit warnings and vcov diagnostics.
Simplified EFA variance percentages and correlation output. Invariance labels use delta symbols with explicit subtraction direction; variant priority is unchanged (robust, scaled, base) and is disclosed in the report.
The earlier development notes below describe the history, not the current public interface. No publication or remote Git update is implied.

New workflows

Added psychomatic_spec() to declare item columns, measurement levels, ordered-category mappings, reverse keys, ranges, grouping, metadata, and missing-data intent in one reusable object.
Added check_items() with stable issue codes for missing/non-finite values, variation, sparse or empty categories, names, ranges, pairwise overlap, and correlation-matrix validity.
Added reliability_auto() with explicitly labelled observed-score, ordinal, two-item, CFA-based, second-order, and bifactor estimands where the supplied data/model support them.
Added dimensionality_sensitivity() to compare parallel-analysis retention results across correlation methods, quantiles, and seeds without declaring an automatic winner.
Added researcher-directed invariance_partial(). It requires exact parameter releases and a substantive rationale, protects anchor items, and never runs an automatic modification-index search.
Added audit_report() to extract serializable settings, provenance, conditions, diagnostics, decisions, references, result tables, and report artifacts without refitting models.

Statistical and implementation corrections

Unified item classification across specifications, checks, correlations, EFA, reliability, and ordinal model workflows. Ordered-factor levels are authoritative; ambiguous nominal encodings fail closed.
Corrected reverse-key deduplication, append-name collisions, non-finite input validation, and incomplete-row proration for mean and sum scale scores.
Correlation matrices now retain requested/effective methods, pairwise sample sizes, and structured diagnostics. No positive-definite repair occurs unless explicitly requested where supported.
Descriptive and analytical fields retain full-precision raw values; digits is restricted to display and export formatting.
Reworked EFA iteration so every item-set change is followed by a complete refit and reclassification. Parallel analysis can retain zero factors, uses the intended factor-eigenvalue reference, restores caller RNG state, and avoids categorical “optimal solution” language.
Updated CFA normality integration for MVN >= 6.3; retained base, scaled, and robust fit evidence; and separated requested from effective estimators.
Corrected two-item reliability to preserve the sign of the correlation. CFA reliability now uses current public semTools APIs and full Lambda, Psi, and Theta matrices, with group/factor-specific non-evaluable states.
Strengthened CFA admissibility checks, bifactor-structure validation, second-order/bifactor reliability mapping, modification-syntax preservation, and structured condition capture.
Rebuilt complete invariance around one frozen analytic sample and explicit fit/admissibility/comparability gates. Corrected fit-change directions and records the effective estimator, missing policy, parameterization, identification, and equality families at every stage.
Ordinal complete and partial invariance use identification-aware semTools syntax. Chen thresholds are exposed as scenario-dependent heuristics, never as formal proof of invariance.
Alignment now matches matrices and weights by semantic identities, rejects incomplete/ambiguous names and protected argument overrides, preserves missingness diagnostics, captures engine output/conditions, and fails closed for malformed, non-finite, or nonconverged results.
Automatic binary IRT now selects 2PL; Rasch is explicit. Mixed item banks use item-specific types, raw slopes are preserved, and only applicable parameter cells enter admissibility checks.
DIF matching scores exclude the focal item and undeclared metadata, freeze one analysis sample per comparison, and fail closed on separation, nonconvergence, non-nesting, or degenerate probabilities. Polytomous items now default to not_evaluable rather than silent median dichotomization.
CFA model comparison preserves positional candidate identity, effective estimator/estimand metadata, common fit-index variants, and comparability gates. AIC/BIC are suppressed outside the unique comparable ranking set.
Report and export adapters now preserve diagnostic data frames, sanitize unsafe live model objects recursively, avoid duplicate table expansion, validate all formats/targets before writing, retain worksheet-name mappings, and apply Excel number formats without changing numeric cell values.

API, objects, and compatibility

Introduced the common psychomatic_result envelope: call, requested and effective settings, data summary, results, structured warnings/errors, diagnostics, decision log, and conclusion.
Added or completed concise print(), structured summary(), and selected base plot() methods for the new workflows and invariance results.
Existing public function names and valid 0.3.x argument positions are retained. New controls were added as trailing arguments where applicable.
CFA and complete-invariance $warnings are now structured tables. Use the transitional $warnings_list character alias for 0.3.x code during the deprecation cycle.
Single-group second-order reliability retains its historical factor-indexed view. Multigroup output uses group-then-factor identity; canonical long rows are available in $reliability$rows.
Spanish argument aliases in EFA remain available for compatibility; the documented English names are preferred for new code.

Data, examples, and release infrastructure

Moved demonstration datasets to development test fixtures; they are not exported datasets. Preserved reproducible generators and simulation metadata for numerical regression checks, including likert_demo and invariance_demo.
Added twelve installed example scripts covering the principal 1.0.0 workflows without personal paths, attached packages, network access, or repository writes.
Expanded tests with numerical contrasts against public psych, lavaan, semTools, MVN, mirt, and sirt APIs where applicable.
Added a cross-platform R CMD check workflow for Ubuntu, Windows, and macOS.

Reader-facing usability

Analysis functions display reader-facing results directly, including assigned calls. verbose = FALSE supports silent workflows; digits controls presentation only.
Per-analysis report, format, file_name, and overwrite arguments provide Excel/Word reports without a separate exporter. Integrated reports default to analysis sections; detail = "audit" preserves technical evidence tables.
EFA restores comprehensive diagnostics, fit, variance, loadings, item allocation, reliability, and references. CFA defaults to explicitly labelled scaled indices; robust and base estimates remain available.
Reliability explains score scope and calculation, separating AVE from reliability. Invariance retains non-evaluable stages and actual engine errors visibly.
Corrected the Wu and Estabrook (2016) reference and formatted references as prose.

PsychoMatic 0.3.0

CRAN readiness

Updated DESCRIPTION, maintainer metadata, vignette configuration, and DOI-backed methodological references.
Moved report-only and workflow-specific packages to conditional suggested dependencies.
Added English manual pages, lightweight vignettes, bundled example datasets, and temporary-path report examples.

New functions

Added screen_items(), reverse_items(), score_scale(), psychomatic_report(), irt_auto(), dif_auto(), and compare_models_auto().
Added structured summaries for EFA, CFA, and complete invariance.
Aligned EFA export with the integrated report writer and retained bilingual output through language = "esp" and language = "eng".

PsychoMatic 0.2.0

Added Pearson, polychoric, and tetrachoric inter-item correlation matrices.
Added Excel export for correlation matrices.
Corrected the confidence-interval label for factor loadings.