Skip to content

kei-saito-research/nrr-boundary

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

14 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

NRR-Boundary: Same Operators, Opposite Effects by Provider and Order

NRR-Boundary is a pre-submission repository for the boundary-focused manuscript and reproducibility artifacts around Stage B and ordered-combination follow-up checks.

NRR is not an anti-LLM framework. NRR does not replace standard LLM use. NRR optimizes when to commit and when to defer, under explicit conditions. Series numbering policy: paper3 is permanently skipped and never reused.

Part of the Non-Resolution Reasoning (NRR) research program.

NRR Series Hub (Start here)

For the cross-paper map and current series links, start here:

Current manuscript snapshot

  • Source: manuscript/current/nrr-boundary_manuscript_v22.tex
  • PDF: manuscript/current/nrr-boundary_manuscript_v22.pdf
  • Figures used by manuscript:
    • manuscript/current/stageb_provider_separated_heatmaps_v9_readable.png
    • manuscript/current/fig_provider_separated_heatmaps_v9_readable.png
    • manuscript/current/stageb_sign_flip_boundaries.png
    • manuscript/current/fig_sign_flip_boundaries.png
  • Checksum manifest: manuscript/current/checksums_sha256.txt

Evidence bundle

  • Stage B processed metrics: stats/stageb_all/
  • Combo rep1 processed metrics: stats/combo_rep1_all/
  • Combo rep3 processed metrics: stats/combo_rep3_all/
  • Primary raw JSON evidence for gate parse-fail recomputation:
    • stats/raw_stageb_all/
    • stats/raw_combo_rep1_all/
    • stats/raw_combo_rep3_all/
  • Derived evidence and recompute script: stats/evidence/

Scope note (gate parse-fail)

stats/stageb_all/gate_recheck_final.csv includes the original run scope (all row based on 2025 pair-metric rows), while stats/evidence/gate_recheck_all.csv is recomputed from the bundled full-scope pair-metrics table in stats/stageb_all/ (2430 rows).

Direction rule note

Rep1-to-rep3 direction labels use abs(improvement_mean) <= 1e-12 => zero. The threshold is recorded as zero_eps_abs in:

  • stats/evidence/combo_rep1_vs_rep3_direction_check.csv
  • stats/evidence/combo_rep1_vs_rep3_direction_summary.csv

Repository structure

nrr-boundary/
|-- README.md
|-- LICENSE
|-- reproducibility.md
|-- manuscript/
|   `-- current/
|       |-- nrr-boundary_manuscript_v22.tex
|       |-- nrr-boundary_manuscript_v22.pdf
|       |-- stageb_provider_separated_heatmaps_v9_readable.png
|       |-- fig_provider_separated_heatmaps_v9_readable.png
|       |-- stageb_sign_flip_boundaries.png
|       |-- fig_sign_flip_boundaries.png
|       `-- checksums_sha256.txt
`-- stats/
    |-- stageb_all/
    |-- combo_rep1_all/
    |-- combo_rep3_all/
    |-- raw_stageb_all/
    |-- raw_combo_rep1_all/
    |-- raw_combo_rep3_all/
    `-- evidence/

License

CC BY 4.0. See LICENSE.

Stable review-package entrypoints:

  • bash scripts/build_current_manuscript.sh
  • bash scripts/verify_current_package.sh
  • bash scripts/recompute_evidence.sh