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.
Bug fixes
Fixed F→t conversion missing Hedges’ g variant
(g_ind).
Added d_ind_min/d_ind_max bounds for
F-test conversions.
Fixed chi-square inline N parsing and multi-stat sentence
handling.
Fixed section-number false positives in p-value extraction.
Fixed ns (non-significant) notation parsing.
API changes
Options forwarding in plumber.R via do.call().
Summary and version fields included in all API responses.
effectcheck 0.2.2
Bug fixes
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.
New features
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.
Garbled p-value detection: p-values like p <
0.645 flagged as extraction_suspect.
effectcheck 0.2.1
Bug fixes
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.
Improvements
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.
CRAN-related
Fixed invalid URI, APA spelling, broadened package scope
description.
Fixed effectcheck namespace refs for shinyapps.io deployment.
Removed pdftools dependency.
Added development disclaimers throughout package and app.
effectcheck 0.2.0
New features
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.