RS0136/EFE_to_RSA: v1.0.0 — EFE→RSA CiC Diagnostics (code + archived outputs)

Release: Manuscript-aligned archive (code + outputs) This release publishes a frozen, manuscript-aligned snapshot of the full Python codebase and the exact exported outputs "An Identification-Diagnostic Template for EFE–RSA Softmax Bridges under Context Offsets" Ryotaro Saito (Independent Researcher) The archive is designed for reproducibility: it includes the analysis pipelines, the exported figures/tables used in the manuscript, and environment/run manifests. What's included 1) Empirical CiC pipeline (efe_to_rsa/) Outcome-level diagnostics on Colors in Context (CiC), including: Worker-disjoint train/test split (group shuffle split) Binomial GLMs (logit link) with worker-cluster robust covariance Pre-specified block Wald tests aligned to the assumption ladder: A2 slope-interaction blocks (disc / clarity / length; plus A2full) A5 condition-intercept blocks (A5base and A5|A2full) Exhaustive robustness "multiverse" over discriminability ("disc") variants Three-factor Shapley attribution of held-out predictive gains: Priors/offsets (P), Geometry (G), Decoder/cost proxies (D) Archived manuscript outputs: efe_to_rsa/output/figures/ (PNG @ dpi=800 + PDF) efe_to_rsa/output/tables/ (CSV exports) 2) A5 simulation calibration (Sim_efe_to_rsa/) A controlled three-alternative reference-game simulation that: Defines the literal listener by Bayes' rule so the A5 decomposition holds exactly by construction Preserves the paper's observability gap by collapsing to observed-only fields: condition labels, geometry summaries, realized length, worker IDs, and binary success Runs the same outcome-level block tests (A2/A5 proxies) across three regimes Archived simulation outputs in Sim_efe_to_rsa/outputs/: condition-wise success (with Wilson intervals) rejection-rate summaries across repetitions (with Wilson intervals) example-run Wald statistics raw and summary CSVs 3) Reproducibility bundle (both pipelines) Each pipeline includes: pip_freeze.txt — full package snapshot (pip freeze) environment.json — OS / Python / key package versions run_manifest.json — seeds and run configuration (including split sizes) Figures & formatting All exported plots are grayscale, title-free, and saved at dpi=800 (PNG) for manuscript use. PDF versions are also included for most empirical figures. How to reproduce Install From the repo root: python -m venv .venv # macOS / Linux source .venv/bin/activate # Windows (PowerShell) # .venv\Scripts\Activate.ps1 pip install -r requirements.txt Empirical pipeline (CiC) Note: the CiC dataset file is not included in this release. Download filteredCorpus.csv (Colors in Context; Monroe et al., 2017) and place it locally. Run: python -m efe_to_rsa.src.main \ --input ./filteredCorpus.csv \ --output ./efe_to_rsa/output_recomputed \ --seed 0 \ --test_size 0.2 \ --disc_variants all \ --dpi 800 A5 simulation python Sim_efe_to_rsa/sim.py \ --outdir Sim_efe_to_rsa/outputs_recomputed \ --seed 0 \ --n_trials 20000 \ --n_workers 600 \ --n_candidates 15 \ --test_frac_workers 0.2 \ --n_rep 20 Important notes Do not overwrite the archived outputs unless you intend to. Use a different output directory name such as output_recomputed/ or outputs_recomputed/. This repo is an archive snapshot intended to exactly preserve the analysis used for the paper, rather than a polished general-purpose library. License MIT License (see LICENSE). Contact Ryotaro Saito ryotarosaito136@gmail.com