54-Reproducibility Checklist Template v1.0 | Chapter 1 — Bundle Overview & Scope of Use

Home ／ Docs-Technical WhitePaper ／ 54-Reproducibility Checklist Template v1.0

Chapter 1 — Bundle Overview & Scope of Use

I. Purpose & Scope

• Define the composition, boundaries, and acceptance rules of the Minimal Reproducible Bundle (MRB), covering environment lock, data snapshot & lineage, weights & parameters, freshness policy, scripts & commands, metrics & intervals, gate mapping, and release manifests.
• Applicable to local/container/cloud modes; for model training/inference reproduction, offline evaluation, and online regression verification.
• For path quantities (arrival time/phase), explicitly show gamma(ell) and d ell in text, and record delta_form ∈ {general, factored} on the data side; parenthesize all expressions; publication requires p_dim = 1.0 with check_dim_report.json.

II. Minimal Bundle

• Environment lock: env_lock.json, container_spec.yaml (HW/OS/driver/framework/image/clock source).
• Data snapshot: data_refs.yaml, split_manifest.json (source/Schema/splits/checksums/license).
• Weights & params: weights_manifest.yaml, param_lock.yaml (signatures/versions/freshness/cov_group).
• Training & inference: train_config.yaml, inference_openapi.yaml, inference.proto, binding_spec.md.
• UQ & error budget: model_uq.yaml, uq_summary.json, budget_breakdown.csv (choose exactly one coverage mode: k/alpha/quantile across the volume).
• Evaluation & scoring: bench_plan.yaml, scorecard.json, eval_report.md.
• Monitoring & rollback: monitoring_rules.yaml, rollback_fsm.yaml, alerts.jsonl.
• Reports & signatures: check_dim_report.json, validate_report.json, audit.jsonl, report_manifest.yaml, SIGNATURE.asc.

Root: PTN_EXPORT/; register sha256 and signatures for all artifacts.

III. Scope of Use & Limits

• In-scope: engineering/research reproduction, comparative evaluation, and release verification aligned with Dataset/Model/Error/Pipeline/Parameter cards.
• Limits: if path blocks (gamma/measure/delta_form) are missing, clock_state!="locked", p_dim < 1, or cross-volume coverage modes disagree, mark [Restricted] and allow qualitative checks only.

IV. Dependency Matrix

V. Versioning & Signatures

• Use SemVer: MAJOR (breaking: schema/units/dimensions/control equations), MINOR (backward-compatible additions), PATCH (fixes).
• Record checksum and signature for each artifact; fix rollback points with *-lock tags; public references v1.* only.

VI. Gate Mapping

• G1 Schema completeness | G2 Citation compliance (anchor coverage ≥ 90%) | G3 Path conventions (gamma/measure/delta_form present; step compliant) | G4 Dimensional closure (p_dim = 1.0) | G5 Freshness (clock_state="locked") | G6 Coverage consistency (k/alpha/quantile) | G7 Covariance consistency (Σ PD) | G8 Uniqueness & acyclicity.
• Trigger S1–S5 (dimension/freshness/path/covariance/citation) → reject acceptance or tag [Restricted].

VII. Normative Path Forms

• Arrival (two equivalent):
  T_arr = ( 1 / c_ref ) * ( ( ∫ n_eff d ell ) )
  T_arr = ( ( ∫ ( n_eff / c_ref ) d ell ) )
• Phase:
  Phi = ( ( 2π / λ_ref ) * ( ∫ n_eff d ell ) )

Before reproduction, align time → path → phase; arrays len(gamma_ell)=len(d_ell)=len(n_eff)≥2; record delta_form on the data side.

VIII. Machine-Readable Index

# repro_manifest.yaml (example)

version: "1.0.0"

entrypoints:

train: "reproduce.sh train"

infer: "reproduce.sh infer"

eval: "reproduce.sh eval"

artifacts:

env: ["env_lock.json","container_spec.yaml"]

data: ["data_refs.yaml","split_manifest.json"]

model: ["train_config.yaml","weights_manifest.yaml","param_lock.yaml"]

api: ["inference_openapi.yaml","inference.proto","binding_spec.md"]

uq: ["model_uq.yaml","uq_summary.json","budget_breakdown.csv"]

bench: ["bench_plan.yaml","scorecard.json","eval_report.md"]

monitor: ["monitoring_rules.yaml","rollback_fsm.yaml","alerts.jsonl"]

reports: ["check_dim_report.json","validate_report.json","audit.jsonl","report_manifest.yaml","SIGNATURE.asc"]

checksums:

sha256: "checksums.txt"

IX. Execution & Acceptance

• Execute: complete train → infer → eval → compare in local/container/cloud modes; log to audit.jsonl.
• Acceptance: metrics/intervals match manifests; /validate passes G1–G8; all artifacts’ checksums/signatures match; non-compliances handled as [Restricted] with diagnostics.