effectcheck 0.2.3

New features

User configuration: New cross_type_action, ci_affects_status, and plausibility_filter parameters for check_text().
Plausibility bounds: Implausibly large effect sizes (e.g., d > 10, r > 1) flagged via extraction_suspect column. Configurable via EFFECT_PLAUSIBILITY in constants.
Effect size families: OR, RR, IRR, h added to EFFECT_SIZE_FAMILIES.
Completeness: Cohen’s f² = r²/(1-r²) for r-tests.
Beginner-friendly UI: Legend, tooltips, plain English verdicts, narrative report mode in frontend.
Paste-text input: Direct text input mode in frontend.

Cross-type matching: Best match now selected across effect size types, reducing false ERROR rate from 47.6% to ~10%.
Paired N fix: Paired designs no longer double the sample size incorrectly.
NOTE→PASS: Results matching within tolerance now correctly report PASS instead of NOTE.
check_type column: Added to output for transparency in variant matching.
extraction_suspect flag: Extreme deltas flagged for manual review.

SKIP status: Extraction-only results (no p-value or effect size to verify) now get status “SKIP” instead of misleading WARN.
Universal tail fallback: Phase 9 tries both one-tailed and two-tailed when decision error occurs, resolving to NOTE with explanation.
Two-tailed detection: New two_tailed_detected flag overrides one-tailed when both present in text.
Method context detection: method_context_detected flag suppresses decision_error for p-curve, equivalence test, TOST contexts.
N candidates for r-tests: Multiple N values extracted from context; best match selected with assumption note.
One-tailed scope fix: one_tailed_detected now searches chunk only, preventing cross-chunk bleeding.
Garbled p-value detection: p-values like p < 0.645 flagged as extraction_suspect.

Fixed thousands-separator N parsing (e.g., N = 1,182 no longer misread as N = 1).
Fixed Welch t-test N estimation.
Removed officer dependency; DOCX extraction via pandoc only.
Fixed DOCX segfault crash.

Cold start progress indicator for Render free tier.
Page-load health check and server status banner for classroom use.
Paste-text input mode.
Parser robustness: 7 fixes from AI verification testing across 12 journal styles.

Nonparametric tests: Mann-Whitney U, Wilcoxon W, and Kruskal-Wallis H test parsing and effect size computation (rank-biserial r, Cliff’s delta, epsilon-squared, Kendall’s W).
Regression support: b/SE coefficient parsing, t = b/SE verification, regression type inference, standardized beta computation.
HTML report generation: generate_report() produces self-contained HTML reports with executive summary, color-coded rows, and interactive tables. render_report() provides a convenience wrapper. PDF fallback available.
CSV and JSON export: export_csv() and export_json() for machine-readable output.
statcheck comparison: compare_with_statcheck() and compare_file_with_statcheck() for side-by-side comparison with statcheck results.
Variant helper functions: get_variants(), get_same_type_variants(), get_alternatives(), format_variants(), compare_to_variants(), get_variant_metadata(), get_effect_family().

Fixed missing r p-value computation.
Fixed missing CI for paired designs (ci_dz).
Corrected eta-squared CI formula (partial eta-squared).
Fixed MBESS ci.sm -> ci.smd call.
Fixed drm_from_dz formula.
Fixed Cohen’s f regex false positives.
z-test pattern now checked last to avoid capturing auxiliary z-values from U/W tests.

mirror server hosted at Truenetwork, Russian Federation.