29-EFT.WP.TBN.Measurement v1.0 | Chapter 10 | Analytics vs Observation vs Anchors (Anchor / Replay)

Home ／ Docs-Technical WhitePaper ／ 29-EFT.WP.TBN.Measurement v1.0

Chapter 10 | Analytics vs Observation vs Anchors (Anchor / Replay)

• One-line objective: Bring model analytics (predict), measured observations (measure), and anchor-based calibration (anchor) / replay into a single metrology framework, forming an executable loop of differenced residuals → contract gating → disposition/replay, co-published with the dual-form T_arr.
• I. Scope & Objects
• Inputs
• Analytics side: from RefCond and path/environment/instrument/network models, produce
  pred = {phi/f/y, S_phi/S_y/sigma_y, T_arr^{form1/2}, T_arr*}.
• Observation side: estimator outputs
  meas = {phi/f/y, S_phi/S_y/sigma_y, T_arr^{form1/2}, T_arr*, EVM/eye} (see Chs. 4–6/9).
• Anchors & references: anchor ∈ {loopback/peer/triangle, frequency ratio chains, reference spectral lines / time sources}.
• Replay assets: raw or downsampled data, windows/bandwidth/conventions, scripts and version hashes.
• Outputs
• Residuals & gating: resid = meas − pred across spectral/statistical/delay/quality components.
• Replay work order: the minimal sufficient fields and scripts to recompute pred/meas.
• Manifest: manifest.tbn.anchor.* (anchor/replay evidence, uncertainties, and contract outcomes).
• Boundary
• No new physics here; this chapter focuses on consistency and replay: how to compare, how to anchor, how to reproduce.
• II. Terms & Variables
• Residuals & components: resid_phi, resid_f, resid_y, resid_Sφ, resid_Sy, resid_sigma_y, resid_T, resid_EVM.
• Anchors & baselines: anchor (ratio chains, spectral lines, loopback/peer/triangle, time-sync calibration), replay script replay.script and version replay.hash.
• Dimensions: each residual follows the units/dimensions of its parent quantity; always record unit/dim and run check_dim.
• III. Postulates P510-*
• P510-1 (Same window/bandwidth/units): pred and meas must be compared under the same W, same RBW/ENBW, and same units/dimensions; if not, resample/recompute first, then compare.
• P510-2 (Dual-form hard gate): Comparison of T_arr^{form1/2} and delta_form is a hard gate; if delta_form ≤ tol_Tarr is not met, handle disposition before proceeding.
• P510-3 (Anchor-first & non-identifiables): Use invariants (ratio chains / loopback / spectral lines) to calibrate bias/scale; non-identifiable parts enter u/U as upper bounds.
• P510-4 (Replay verifiability): Any decision must be independently reproducible via replay.script and manifest hashes.
• P510-5 (Dimensional & provenance traceability): All fields pass check_dim; RefCond.hash, source/firmware/script hashes are traceable.
• IV. Minimal Equations S510-*
• Differenced residuals (same window/bandwidth)
• S510-1:
  resid_y(t) = y_meas(t) - y_pred(t);
  resid_Sy(f) = S_y,meas(f) - S_y,pred(f);
  resid_sigma_y(τ) = sigma_y,meas(τ) - sigma_y,pred(τ);
  resid_T = T_arr*,meas - T_arr*,pred;
  resid_EVM = EVM_meas - EVM_pred.
• Residual gates and release thresholds
• S510-2 (with uncertainty):
  |resid| + k·u(resid) ≤ threshold;
  common gates include resid_T ≤ tol_T, resid_sigma_y ≤ tol_sigma, |resid_EVM| ≤ tol_evm.
• Anchor-based calibration
• S510-3: cf. Ch. 8 (loopback/peer/triangle) and Ch. 11 (ratio chains/spectral lines):
• Loopback: ΔT_bias ≈ τ_loop/2 - (T_arr* + ΔT_inst + ΔT_proc)
• Ratio chain: solve z_bias / z_scale
• Spectral line: Δ_line = f̂_line - f_line
  Apply apply(anchor) : meas → meas_cal, then update residuals.
• Replay minimal sufficiency
• S510-4:
  replay.core = {raw|downsampled, window:{W,w}, RBW, ENBW, method/cfg, RefCond.hash, path.hash, scripts.hash};
  recompute pred from RefCond + models.hash; recompute meas from raw + estimator.cfg.
• V. Metrology Flow M50-10 (Ready → Align → Compare → Anchor → Replay → Persist)
• Ready: Freeze RefCond; pull manifests and windows/bandwidths for pred and meas; stage anchors and replay scripts/assets.
• Align: Resample pred/meas to the same W/RBW/ENBW/units and to the unified timebase tau_mono; align T_arr^{form1/2} and verify delta_form.
• Compare: Compute resid_* (per S510-1) and uncertainties u(resid); execute release gates (per S510-2).
• Anchor: If residual biases are significant, execute calibrations per S510-3; update resid_* and u/U.
• Replay: Package replay.core; provide one-click recompute scripts and version hashes; validate third-party replay equivalence.
• Persist:
  manifest.tbn.anchor = {compare:{resid_*, u(resid)}, anchor:{type, params, Δbias/scale}, replay:{core, scripts.hash}, Tarr:{form1,form2,delta_form,T_arr*}, RefCond, contracts.*, signature}.
• VI. Contracts & Assertions C50-10x (Suggested thresholds)
• C50-1001 (Dual-form hard gate): If delta_form_p95 ≤ tol_Tarr is not met → reject and trigger strategy card.
• C50-1002 (Same window/bandwidth): W/RBW/ENBW/units of pred/meas must match; otherwise recompute first.
• C50-1003 (Residual release gates): |resid_*| + k·u(resid_*) ≤ threshold_*.
• C50-1004 (Anchor/Replay completeness): Evidence URIs and hashes for anchor and replay.core must be persisted; missing → reject.
• C50-1005 (Dimensions/traceability): All fields pass check_dim; RefCond.hash/models.hash/scripts.hash are traceable.
• VII. Implementation Bindings I50-10* (Interface prototypes, I/O, invariants)
• I50-101 align_pred_meas(pred, meas) -> {aligned, meta}
• I50-102 residuals(pred, meas) -> {resid_*, u(resid), pass}
• I50-103 apply_anchor(meas, anchor) -> {meas_cal, params, u}
• I50-104 build_replay_core(pred, meas, RefCond, scripts) -> {replay.core}
• I50-105 assert_anchor_contracts(ds, rules) -> {report, pass}
• I50-106 emit_anchor_manifest(results, policy) -> {uri, status}
• Invariants: two_forms_present=true; same window/bandwidth/units; RefCond.hash and script/model hashes traceable; replay independently reproducible.
• VIII. Cross-References
• Mathematical baseline & dual forms: Ch. 2.
• Acquisition & estimation: Chs. 3–5.
• Link / environmental corrections: Chs. 6–7.
• Instrument metrology & anchors: Ch. 8.
• Uncertainty & release gates: Chs. 11–12.
• Runtime rollback: Chs. 13–14.
• Manifests & interfaces: Appendices A/B/C/E.
• IX. Quality & Risk Control
• SLIs/SLOs: delta_form_p95, resid_T_p95, resid_sigma_y_p95, resid_EVM_p95, anchor_bias/scale, replay_pass_rate, panel_freshness.
• Fallback strategies: window/bandwidth mismatch → recompute; persistent residual overruns → widen guardbands & update models; anchor calibration fails → switch anchor or bypass; replay failure → freeze publication and review scripts/sources.
• Audit: residual & threshold curves, anchor parameters & evidence, replay bundle & signature chain, third-party recomputation reports.
• Summary
• This chapter unifies analytics vs observation vs anchor/replay into a same-window, same-bandwidth, same-units compare–anchor–replay loop, paired with the dual-form T_arr.
• With M50-10 / C50-10x / I50-10* and manifest.tbn.anchor, TBN publications become comparable, calibratable, re-verifiable, and rollback-ready.