Meta-Analysis — Full Analysis & Manuscript Generation

Open-source skill. Read reference/specs/output-variation-protocol.md before every generation — apply all variation layers.

Workflow

Continues from where vera-data-meta-reviewing stopped (PART 0-2 done).

Step	Responsibility	Executor	Document	Input	Output
Publication bias	Run Additional Tests	Main Agent	workflow/step04-run-additional-tests.md	Prior step output	PART 3 code + prose
Subgroup & regression	Analyze Subgroups	Main Agent	workflow/step05-analyze-subgroups.md	Prior step output	PART 4 code + prose
Advanced models	Fit Models	Main Agent	workflow/step06-fit-models.md	Prior step output	PART 5 code + prose
Model comparison	Compare Models	Main Agent	workflow/step07-compare-models.md	Prior step output	PART 6 code + prose
Manuscript	Generate Manuscript	Main Agent	workflow/step08-generate-manuscript.md	Prior step output	methods.md + results.md

Additional Inputs

Collect if not already provided:

• Target discipline (for reporting conventions)
• Target journal or style (APA 7th, PRISMA, Cochrane)
• Research question / hypothesis
• Subgroup variable (categorical moderator)
• Continuous moderator (for meta-regression)

Output Structure

output/
├── methods.md
├── results.md
├── tables/             <- Markdown + CSV per table
├── figures/            <- PNGs, 300 DPI
├── references.bib
├── code.R              <- Style-varied
└── code.py             <- Style-varied

Key References (read before generation)

File	Purpose
reference/specs/output-variation-protocol.md	Output quality variation layers
reference/specs/code-style-variation.md	Seven-dimension code style diversity
reference/patterns/sentence-bank.md	4-6 phrasings per result type
reference/rules/reporting-standards.md	Hard rules for statistical reporting

Reporting Standards

Same as vera-data-meta-reviewing, plus:

• Egger's test: "intercept = X.XX, SE = X.XX, p = .XXX"
• Begg's rank test: "Kendall's tau = X.XX, p = .XXX"
• Trim-and-fill: "k imputed = X, adjusted ES = X.XX, 95% CI [X.XX, X.XX]"
• Meta-regression: "B = X.XX, SE = X.XX, p = .XXX, R² = .XX"
• Leave-one-out: "pooled ES ranged from X.XX to X.XX when individual studies removed"
• Subgroup differences: "Q_between(df) = X.XX, p = .XXX"
• Bayesian: "posterior mean = X.XX, 95% CrI [X.XX, X.XX]"
• Three-level: "sigma²_within = X.XX, sigma²_between = X.XX"

Cross-Skill Interface

Method Unit Contract:
├── code_r           → .R script (style-varied)
├── code_python      → .py script (style-varied)
├── methods_md       → methods.md (varied structure)
├── results_md       → results.md (varied phrasing)
├── tables/          → Markdown + CSV
├── figures/         → PNGs 300 DPI (varied layout)
├── references_bib   → .bib with cited references
└── comparison       → cross-method narrative (in results.md)

Invoked directly after vera-data-meta-reviewing or orchestrated by vera-data-application-pipelining.