56-Report-Level Methods Appendix Template v1.0 | Chapter 12 Reproducibility & Artifact Export

Home ／ Docs-Technical WhitePaper ／ 56-Report-Level Methods Appendix Template v1.0

Chapter 12 Reproducibility & Artifact Export

I. Chapter Goals & Scope (Mandatory)

• Fix a unified caliber for environment—dependencies—data versions—random sources—execution commands—artifact signatures—audit trail, ensuring experiments/evaluations are configure-once, replay-anywhere, and exported artifacts are verifiable, traceable, archivable.
• Aligned with Ch.5 (Data & Experimental Design), Ch.6 (Math & Pseudocode), Ch.7 (Metrology & Calibration), Ch.8 (Evaluation Protocol), and Ch.11 (Implementation Binding).

II. Environment & Dependencies (Mandatory)

• Container image: image@sha256:<digest>; list base image and layered dependencies with versions and hashes.
• Runtime domains: dev/stage/prod (or equivalent); specify CPU/GPU/RAM/accelerator libraries (CUDA/ROCm/OneAPI) and versions.
• Dependency lock: language lockfiles (requirements.txt/poetry.lock/package-lock.json, etc.) with hashes; record OS packages and kernel version.

III. Data & Random Sources (Mandatory)

• Data versioning: dataset@version + sha256; enforce partition consistency (train/valid/test ratios) and row-order locking.
• Random sources: seed, RNG backends (numpy|torch|random|cuRAND…), determinism toggles.
• Mounts & caches: read/write separation, read-only mounts, cache invalidation policy and hash locations.

IV. Execution Command & Parameters (Mandatory)

• Minimal repro command: single-line or script entry (include args, paths, env vars); long jobs must support resume and log locations.
• Numerical accuracy: explicitly provide key tolerances/steps such as Δell/ε_int/τ_c.
• Retry: idempotency key and failure snapshot policy.

V. Artifacts & Signatures (Mandatory)

• Artifact types: yaml/json/pdf; optionally csv/parquet.
• Hash & signature: sha256, signing cert/fingerprint, generation time, and source script script@commit.
• Directory layout: /artifacts/<run_id>/{yaml,json,pdf,logs,plots}; a manifest.json lists all files with hashes.

VI. Audit Trail & Archiving (Mandatory)

• Audit fields: timestamp/actor/command/env/image@sha256/dataset@version/git@commit/artifact_hash/gate_decisions.
• Archival: retain ≥ 1 year; cold storage with replicas; verifiable restore steps.
• External deliverable: snapshot bundle (tar.zst) and a verification script verify_artifacts.py@commit.

VII. Arrival-Time Consistency (Mandatory)

1. If evaluation involves arrival-time criteria, declare explicitly in the repro list:
   • Factored: T_arr = ( 1 / c_ref ) * ( ∫ n_eff d ell )
   • General: T_arr = ( ∫ ( n_eff / c_ref ) d ell )
2. In the same paragraph state path gamma(ell) and measure d ell; set check_dim=true; write Δell/ε_int/τ_c and delta_form into artifact metadata.

VIII. Human × Machine Mapping (Mandatory)

IX. Field & Constraint List (copy-ready)

X. Machine Schema (YAML; JSON-equivalent, Mandatory)

repro:

image: "registry/replay:2025.09@sha256:…"

hardware:

cpu: "32c/64t"

ram: "256GiB"

gpu: { model: "A100x2", driver: "535.129", accel: "CUDA 12.2" }

deps:

lockfiles:

- { path: "requirements.txt", sha256: "sha256:…" }

- { path: "poetry.lock", sha256: "sha256:…" }

data:

datasets:

- { name: "cmb_set_v3", version: "v3", sha256: "sha256:…", license: "CC-BY-4.0" }

- { name: "lens_v1", version: "v1", sha256: "sha256:…", license: "CC-BY-4.0" }

seed:

value: 20250927

backends: ["numpy","torch","random"]

cmd: "docker run … replay --suite arrive_time_check.py --dataset cmb_set_v3@v3"

args: ["--batch B1","--region EU"]

env_vars: ["OMP_NUM_THREADS=8","TOKEN=***"]

numeric: { Δell: 1e-3, ε_int: 1e-6, τ_c: 0.01 }

arrival_time:

delta_form: "general"

path: "gamma(ell)"

measure: "d ell"

check_dim: true

artifacts:

types: ["yaml","json","pdf"]

hashes:

- { path: "results.yaml", sha256: "sha256:…" }

- { path: "summary.json", sha256: "sha256:…" }

- { path: "report.pdf", sha256: "sha256:…" }

manifest:

files:

- { path: "logs/run.log", sha256: "sha256:…", size: 1048576 }

- { path: "plots/curve.png", sha256: "sha256:…", size: 98312 }

audit:

record:

timestamp: "2025-09-27T12:34:56Z"

actor: "runner@host"

command: "…"

image: "registry/replay:2025.09@sha256:…"

dataset_versions: ["cmb_set_v3@v3","lens_v1@v1"]

git_commit: "a1b2c3d"

artifact_hash: ["sha256:…"]

gate_decisions:

- { name: "gate_accuracy", rule: ">=0.99@7d", value: 0.995, pass: true }

archive:

retention_days: 365

bundles:

- { name: "snapshot-REL-2025-1001.tar.zst", sha256: "sha256:…" }

verify_script: "verify_artifacts.py@9f8e7d"

XI. Minimal Sample (human abstract × machine snippet, Mandatory)

• Human abstract: container registry/replay:2025.09@sha256:…, data cmb_set_v3@v3 (sha256:…), seed=20250927; minimal command as above; export yaml/json/pdf with hashes; retain for 365 days. If T_arr is involved, use the general form and declare gamma(ell) and d ell; check_dim=true.
• Machine snippet:

repro: { image:"registry/replay:2025.09@sha256:…", seed:{ value:20250927, backends:["numpy","torch"] } }

artifacts: { types:["yaml","json","pdf"], hashes:[{ path:"results.yaml", sha256:"sha256:…" }] }

archive: { retention_days: 365 }

XII. Validation Rules (regex/consistency, Mandatory)

• Image: ^.+@sha256:[0-9a-f]{64}$; data hash: ^sha256:[0-9a-f]{64}$.
• Time: all UTC timestamps must be ISO8601; retention_days ≥ 365.
• Arrival-time: if T_arr appears, arrival_time.path/measure required and math.check_dim=true.
• Artifacts: artifacts.types ⊇ {"yaml","json","pdf"}; every manifest.files[*].sha256 non-empty.
• Audit: audit.record must include timestamp/actor/image/git_commit/artifact_hash/gate_decisions.
• Idempotency: for long-running exports, record the Idempotency-Key in the audit trail (if used).

XIII. Citation & Cross-Reference Style (Mandatory)