evidence_map.md

Claim-to-evidence map

This page gives reviewers, reusers, and first-time readers the shortest path from each public claim to its protocol, data, code, and validation record.

Primary claim

Claim. At fixed treewidth, interaction count, and parameter count, a target-supported spanning-tree Gibbs representation trains better than a generic chain; selecting the spanning tree by maximum absolute target-coupling weight improves further over a prespecified random target-supported tree.

The protocol and seed commitments were frozen before the separately generated target ensemble was evaluated. This is an internal confirmation on unseen targets, not an external replication. In code and result files, problem_tree denotes the target-supported spanning tree that maximizes edge weight $|J|$ (MAXJ).

Evidence layer	Canonical location
Human-readable result	README.md
Exact claim wording and nonclaims	docs/scientific_claims.md
Prespecified protocol	experiments/sparse_ising_confirmation/protocol/protocol.md
Locked pre-generation source	experiments/sparse_ising_confirmation/protocol/frozen_source/
Aggregate outcomes	results/confirmatory/aggregate.csv
Primary paired effects	results/confirmatory/primary_effects.csv
Statistical method	docs/statistical_analysis.md
Raw trajectory evidence	experiments/sparse_ising_confirmation/results/raw/
Stored initial/final parameters	experiments/sparse_ising_confirmation/results/states/
Analysis implementation	experiments/sparse_ising_confirmation/scripts/analyze_results.py
Scientific validator	experiments/sparse_ising_confirmation/scripts/validate_experiment.py
Validation record	results/confirmatory/validation.json

Geometry claim

Claim. Full alignment gives $\nabla E=-Ic$, while partial alignment adds the state-dependent omitted-cost covariance term $r_G$.

• Derivation: docs/theory.md
• Maintained implementation: src/qbm_alignment/
• Supporting partial-alignment study: studies/partial_alignment_geometry/
• Supporting validation: results/partial_alignment_geometry/validation.json

Preparation-resource claim

Claim. The chain and tree representations have exact width-one factorization and 31 conditional angles, while full graphs require 75–159 angles in the confirmatory ensemble.

• Definitions and scope: docs/preparation.md
• Instance-level resources: results/confirmatory/preparation_resources.csv
• Resource-pair analysis: results/confirmatory/resource_pairs.csv
• Compiler and resource script: experiments/sparse_ising_confirmation/scripts/compute_preparation_resources.py

Temperature-dependent tree boundary result

Result. The tree maximizing retained target-state cooling power changes with temperature, but that optimum is not a better compressed Gibbs representation on the tested development corpus. At the certification temperature it gives a worse projected target-energy gap than both the best hot-optimal tree and the forward-KL-optimal tree on all ten instances.

Evidence layer	Canonical location
Human-readable summary	studies/temperature_tree_geometry/README.md
Frozen protocol	studies/temperature_tree_geometry/protocol.md
Scientific report	studies/temperature_tree_geometry/report.md
Temperature-path summary	results/temperature_tree_geometry/temperature_path_summary.csv
Certification-temperature summary	results/temperature_tree_geometry/certification_temperature_summary.csv
Instance gate summary	results/temperature_tree_geometry/instance_gate_summary.csv
Mechanism correlation summary	results/temperature_tree_geometry/mechanism_correlation_summary.csv
Mechanism summary	results/temperature_tree_geometry/mechanism_summary.json
Reproduction script	studies/temperature_tree_geometry/scripts/run_exhaustive_study.py
Compact validator	studies/temperature_tree_geometry/scripts/validate_study.py
Validation record	results/temperature_tree_geometry/validation.json

This is a later supporting boundary study on reused instances. It is not part of the primary confirmatory table and does not change the MAXJ benchmark claim.

Success certificate

Claim. For a unique ground state with gap $\gamma$, the criterion $\Delta E/\gamma\le0.1$ certifies $p_\star\ge0.9$.

• Proof: docs/theory.md#energy-certificate
• Exact spectra and numerical check: results/confirmatory/validation_spectra.csv
• Stored-state validation: results/confirmatory/validation.json

Reproduction routes

Fast package and metadata checks:

python scripts/validate_repository.py
python -m pytest -q

Full primary-experiment validation:

python experiments/sparse_ising_confirmation/scripts/validate_experiment.py

Regenerate compact tables and figures from packaged data:

bash scripts/refresh_analysis.sh
bash scripts/refresh_analysis.sh temperature

See docs/reproducibility.md for the complete workflow.

Citation

Use CITATION.md for copy-ready APA-style and BibTeX entries. Cite release v1.1.0 for the complete archive or v1.0.0 for the primary confirmed MAXJ result in its original release state. Machine-readable metadata are in CITATION.cff.