56-Report-Level Methods Appendix Template v1.0 | Chapter 5 Data & Experimental Design

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

Chapter 5 Data & Experimental Design

I. Chapter Goals & Scope (Mandatory)

• Fix a reproducible caliber for data and experimental design at report level, covering: datasets@version, sampling & controls, variables & batch control, observation windows & randomization, compliance & licensing, and a minimal repro example.
• Aligned with Ch.3 machine structure, Ch.7 metrology & calibration, Ch.8 evaluation protocol, and Ch.11 implementation binding.

II. Data Asset Inventory (Mandatory)

• Dataset identifier: name@version; source, license, and availability window (UTC).
• Partitions & splits: train/valid/test or equivalent; preserve identical hashes and row order across environments.
• Accompanying metadata: acquisition caliber, time coverage, geography, de-identification, missing & outlier handling.

III. Sampling Framework & Randomization (Mandatory)

• Sampling: stratified / cluster / systematic / simple random; specify strata variables and proportions.
• Observation window: [t0, t1]; for rolling experiments specify step/overlap.
• Randomization: fix seeds and provide repro examples; enforce cross-batch leakage prevention (e.g., by subject or geography).

IV. Control & Variable Management (Mandatory)

• Control/baseline: state Option Base; list controlled vs allowed-to-drift variables.
• Covariates & confounders: provide include/exclude lists; link to ablation & sensitivity plans (see Ch.10).
• Batches & regions: use batch and region as grouping keys; report batch-effect correction methods.

V. Measures, Units & Metrological Consistency (Mandatory)

1. For any measured quantity tied to equations, provide unit and dim; set check_dim=true in machine layer.
2. If the experiment involves arrival-time criteria, use the unified forms:
   • Factored: T_arr = ( 1 / c_ref ) * ( ∫ n_eff d ell )
   • General: T_arr = ( ∫ ( n_eff / c_ref ) d ell )
     and in the same paragraph declare path gamma(ell) and measure d ell.

VI. Compliance, Ethics & Security (Mandatory)

• Licensing & restrictions: list licenses, PII/PHI de-identification strategies, and redistribution limits.
• Security: access tiers, audit trail & retention; data destruction & regeneration flows.
• Bias & fairness: monitoring metrics and gates for sampling/evaluation phases.

VII. Minimal Repro Example (Mandatory)

• Container image, data mounts, scripts & command; exported artifacts (yaml/json/pdf) with hashes.
• Repro pass lines: row counts/hash checks, partition consistency, statistical summaries (mean/variance/quantiles).

VIII. Human × Machine Fields (Mandatory)

IX. Field & Constraint List (copy-ready)

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

data:

datasets:

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

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

partition: { train: 0.7, valid: 0.15, test: 0.15 }

window: { t0: "2025-01-01T00:00:00Z", t1: "2025-06-30T23:59:59Z", timezone: "UTC" }

sampling:

method: "stratified"

variables: ["region","band"]

controls:

controlled: ["instrument","pipeline_version"]

drifting: ["weather","solar_activity"]

baseline: "Option Base"

metrology:

units: ["s","m","kg"]

dimensions: ["T","L","L T^-1"]

calibration: ["Mx-*"]

math:

statements:

- "T_arr = ( ∫ ( n_eff / c_ref ) d ell )"

path: "gamma(ell)"

measure: "d ell"

symbols:

- { name: "n_eff", unit: "1", dim: "1" }

- { name: "c_ref", unit: "m·s^-1", dim: "L T^-1" }

check_dim: true

compliance:

license_terms: "No re-identification; redistribution restricted to partners."

pii_policy: "de-identified; k-anonymity>=10"

security:

access_tiers: ["owner","contrib","read"]

retention_days: 365

audit_log: true

reproducibility:

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

scripts: ["prep_data.py@a1b2c3","split_partitions.py@9f8e7d"]

repro_cmd: "docker run … prep_data.py --in /mnt/raw --out /mnt/curated && split_partitions.py --seed 20250927"

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

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

1. Human abstract:
   • Use stratified sampling (region × band); observation window 2025-01-01 to 2025-06-30 (UTC). Baseline Option Base.
   • Arrival-time uses the general form: T_arr = ( ∫ ( n_eff / c_ref ) d ell ); declare gamma(ell) and d ell in the same paragraph; check_dim=true.
2. Machine snippet:

data:

sampling: { method: "stratified", variables: ["region","band"] }

math:

statements: ["T_arr = ( ∫ ( n_eff / c_ref ) d ell )"]

path: "gamma(ell)"

measure: "d ell"

check_dim: true

XII. Validation Rules (regex/consistency, Mandatory)

• Dataset ID: ^[A-Za-z0-9_\-]+@v?\d+(\.\d+)*$; hash: ^sha256:[0-9a-f]{64}$.
• Time window: t0 ≤ t1 and timezone UTC.
• Sampling method: method ∈ {stratified,cluster,systematic,srs}.
• Arrival-time: if T_arr appears, math.path and math.measure are required; math.check_dim=true.
• Partitions: train+valid+test = 1.0 (tolerance ≤ 1e-6).

XIII. Citation & Cross-Reference Style (Mandatory)