Align release documentation with CI behavior

Author: Ekin-Kahraman

README.md

@@ -167,7 +167,7 @@ Rscript dev/snapshot_lockfile.R
 bulk-rnaseq-differential-expression/
 ├── .github/
 │   └── workflows/
-│       └── ci.yml               # CI (renv status, lint, tests)
+│       └── ci.yml               # CI (renv status, lint, tests, rebuild validation)
 ├── .lintr                       # lintr configuration
 ├── .Rprofile                    # renv autoloader
 ├── 000_install_dependencies.R   # Install all required packages

REPRODUCIBILITY.md

@@ -72,6 +72,6 @@ Rscript -e 'testthat::test_dir("tests/testthat")'
 Rscript dev/lint.R
 ```
 
-GitHub Actions also performs a clean rebuild of the tracked analysis outputs and checks that regenerated key outputs match the committed versions.
+GitHub Actions also performs a clean rebuild of the tracked analysis outputs, compares regenerated tables against the committed versions, and checks that key figures were regenerated successfully.
 
 For a workflow-level comparison against DESeq2, nf-core/rnaseq, `targets`, and `workflowr`, see `WORKFLOW_BENCHMARK.md`.

WORKFLOW_BENCHMARK.md

@@ -14,7 +14,7 @@ This repository starts from a published GEO count matrix, so the benchmark below
 - DE effect estimation: raw DESeq2 inference plus `apeglm`-shrunken log2 fold changes for ranking, MA plotting, and volcano visualization.
 - Robustness: balanced-subset analysis is benchmarked against the full QC-passed cohort, with overlap and concordance written to `results/tables/analysis_summary.csv`.
 - Reproducibility: `renv`-pinned environment, deterministic seeds, GitHub Actions rebuilds, and explicit git/session provenance in `results/session_info.txt`.
-- Artifact validation: tracked tables are rebuilt in CI and compared against committed results; key figures are also diff-checked.
+- Artifact validation: tracked tables are rebuilt in CI and compared against committed results; key figures are checked for successful regeneration.
 
 ## Remaining gaps relative to broader workflows