04-EFT.WP.Core.Metrology v1.0 | Chapter 8 — Cross-Volume Consistency and Regression

Home ／ Docs-Technical WhitePaper ／ 04-EFT.WP.Core.Metrology v1.0

Chapter 8 — Cross-Volume Consistency and Regression

I. Objectives and Interface Boundaries

• Establish a unified convention and regression mechanism across Core.Metrology and Core.Equations / Core.Parameters / Core.Terms, ensuring that units, dimensions, uncertainties, and the two arrival-time conventions remain consistent across volumes.
• This chapter uses I40 8 (import/export & regression) and I40 9 (cross-volume interop) as execution hubs; when needed, call Core.Parameters I30 10/11 (cross-volume coupling, scenario governance).

II. Cross-Volume Consistency Postulates (P98-1…P98-4)

• P98-1 (UnitPolicy invariance): every cross-volume binding must declare a unit_policy and keep it unchanged during a regression cycle; changes require version_semver bump and baseline rebuild.
• P98-2 (ArrivalTime two-form identity): given the path gamma(ell) and constant c_ref,
  T_arr_const = ( 1 / c_ref ) * ( ∫_gamma n_eff d ell ) and
  T_arr_gen = ( ∫_gamma ( n_eff / c_ref ) d ell ) are identically equal.
• P98-3 (Dimensional closure): all cross-volume expressions satisfy dim(lhs) = dim(rhs); check_dim(expr) must return the same result before and after any binding.
• P98-4 (Explicit RefCond): any cross-volume computation involving environmental corrections must pass RefCond explicitly and leave an audit trail; implicit “standard state” is forbidden.

III. Minimal Consistency Equations and Assertions (S98-1…S98-3)

• S98-1 (arrival-time identity): ΔT_arr def= T_arr_const - T_arr_gen = 0。
• S98-2 (dimensional identity): dim( bind_to_equation(Sref, unit_policy) ) = dim( Sref )。
• S98-3 (nondimensional mapping conservation): with t0 def= L0 / c_ref,
  ( ∫_gamma ( n_eff / c_ref ) d ell ) = ( ∫_gamma bar_n_eff d bar_ell ), where bar_n_eff = n_eff, d bar_ell = d ell / L0。

IV. Binding and Execution Constraints (I40 9 Protocol)

• Binding: bind_to_equation(Sref:str, unit_policy:str) -> bool; returns success only after check_dim and all assertions in this chapter pass.
• Arrival-time convention: call enforce_arrival_time_convention() -> None before loading the S20-* series to guarantee two-form equivalence.
• Units and environment: before data crosses into the equation domain, run convert(•) and corr_env(•; RefCond) and lock to U_target.

V. Regression Baselines and Data Assets

1. Baseline set G_metrology:
   • units.yaml (from export_units("yaml"))
   • refcond.yaml (from export_refcond("yaml"))
   • scenarios/ (scenario exports of CalibrationCertificateV1, reports, and derivatives)
   • tests/arrival_time/ (includes n_eff(ell), gamma(ell), T_ref, and tolerances)
2. Every version change is tagged with version_semver, and comparisons via compare_reports with recorded thresholds are preserved.

VI. Regression Metrics and Thresholds (R98 indicators)

• Δmean = mean( value_new - value_base )。
• ΔU_rel = ( U_new - U_base ) / U_base。
• pass_rate = fraction( guard_band( |e|, U, MPE, "shared-risk" ) == pass )。
• dim_ok = all( check_dim(expr_k) == dimstr_k ) (boolean)。
• T_arr_mape = mean( | T_arr_new - T_arr_base | / max( |T_arr_base|, eps ) )。
• Suggested thresholds:
  |Δmean| ≤ tol_mean; |ΔU_rel| ≤ tol_U_rel; pass_rate ≥ tol_pass_rate; T_arr_mape ≤ tol_Tarr; dim_ok == True.

VII. Typical Cross-Volume Regression Scenarios and Remedies

• Unit-policy change (e.g., Pa ↔ kPa): rebuild units.yaml; verify no regression with compare_reports(..., ["mean","U","pass_rate"]).
• Reference-condition upgrade (RefCond for T/P/H): recompute U for all corr_env-affected quantities and audit with ΔU_rel.
• Measurement-model adjustment (change family or order of g_cal): lock valid_range; compare residual_stats and pass_rate within range.
• Arrival-time pipeline modification (integrator or path discretization): validate ΔT_arr = 0 via S98-1; gate with T_arr_mape.
• Parameter-volume coupling (priors or transforms): use I30 11 diff_scenarios to compare theta drift and derived U; trigger recalibration as needed.

VIII. Cross-Volume Regression Pipeline (Mx-6)

• Prepare: freeze unit_policy and RefCond; export units.yaml, refcond.yaml.
• Bind: call bind_to_equation(Sref, unit_policy); run enforce_arrival_time_convention().
• Generate: rebuild reports and certificates per scenario; retain phi_hat, Cov[phi_hat], U, pass_rate.
• Validate: batch-run check_dim and S98-* assertions; compute R98 metrics.
• Compare: compare_reports(a,b, metrics=["mean","U","pass_rate"]); add T_arr_mape.
• Decide: if any metric exceeds its threshold, flag regression_detected and block release.
• Audit trail: save traceability_chain, software_version, version_semver, seed, and n.
• Release: if passed, update the G_metrology baseline; record the CHANGELOG and threshold evolution.

IX. Arrival-Time Consistency Test Suite (Tx-AT)

• Case AT-1 (constant-parameter path): constant c_ref, given n_eff(ell); verify T_arr_const == T_arr_gen.
• Case AT-2 (piecewise constants): assume segment-wise constant c_ref; verify equivalence of piecewise integrals; after unifying to global c_ref, residuals vanish.
• Case AT-3 (nondimensionalization): verify T_arr = ( ∫_gamma bar_n_eff d bar_ell ) * ( L0 / c_ref ) consistent with both conventions.
• Assertion integration: place under tests/arrival_time/ and hook into Step 4 of Mx-6.

X. Coordination with Core.Parameters and Core.Equations

• Structural stability: compute_jacobian(eqn, params) must not alter dimstr or unit(x) after binding.
• Scenario governance: use I30 11 create_scenario/activate_scenario/diff_scenarios to synchronize parameter sets with metrology baselines.
• Identifiability and sensitivity: when phi changes perturb U, call Core.Parameters I30 6 local_sensitivity and global_sensitivity_sobol to locate root causes.

XI. Regression Report Template (RegressionReportV1)

• version_semver, unit_policy, RefCond_name, software_version.
• metrics: Δmean, ΔU_rel, pass_rate, dim_ok, T_arr_mape.
• violations: threshold exceedances with localization (scenario, model, dataset, equation reference Sref).
• decisions: accept|reject|rollback with rationale; guard_band settings.
• artifacts: units.yaml, refcond.yaml, CalibrationCertificateV1 summary, compare_reports summary.

XII. Compatibility and Migration

• Version advancement: bump_version(semver) only after regression passes; if unit_policy or RefCond changes, bump at least the minor version.
• Deprecation: provide deprecate_after and a replacement for retired units or models; maintain dual-track regression for two release cycles.
• Rollback: when regression_detected == True, revert to the previous G_metrology and lock the release branch.

XIII. Routine Tasks and Threshold Maintenance

• Periodic tasks: re-run Mx-6 by T_valid or on a fixed cadence.
• Threshold governance: tol_mean, tol_U_rel, tol_Tarr, tol_pass_rate must be versioned and included in G_metrology.
• For audits: persist traceability_chain, RefCond, seed, n, nu_eff, source and timestamp of MPE.

XIV. Common Mismatches and Remedies

• Unit omission: symptom dim_ok=False; remedy convert and unify U_target.
• Environmental inconsistency: symptom Δmean bias; remedy align RefCond and recompute corr_env.
• Arrival-time numeric deltas: symptom T_arr_mape over threshold; remedy enforce_arrival_time_convention() and unify integrator precision.
• Model family change: symptom pass_rate drop; remedy reset valid_range and rebuild baseline.

XV. Release Gate (Gate-98)