48-Experimental Protocol Card Template v1.0 | Chapter 8 — Quality Gates & Stop Criteria

Home ／ Docs-Technical WhitePaper ／ 48-Experimental Protocol Card Template v1.0

Chapter 8 — Quality Gates & Stop Criteria

I. Purpose & Scope

• Define mandatory Quality Gates (Gates) and Stop/Fallback criteria (Stops/Fallbacks) from acquisition to export, ensuring dimensional closure, timebase consistency, sufficient path sampling, and citation compliance.
• Applicable to timing / imaging / spectral / mixed; when arrival time/phase is involved, path quantities must explicitly show gamma(ell) and measure d ell, with delta_form recorded.

II. Gates G1–G8 (must pass for release)

• G1 | Schema completeness: fields, types, required flags, and indices consistent; {time × channel × position} dimensions aligned.
• G2 | Citation compliance: see[]/references[] use “volume + version + anchor (P/S/M/I)”; anchor coverage ≥ 90%; no external links or aliases.
• G3 | Path–measure integrity: len(gamma_ell)=len(d_ell)=len(n_eff)≥2 and synchronized; Δell satisfies sampling constraints.
• G4 | Dimensional closure: p_dim = 1.0; arrival time T_arr = ( ∫ ( n_eff / c_ref ) d ell ) has unit s; phase Phi has unit rad.
• G5 | Calibration freshness: |ts_start − calib.timestamp| ≤ τ_calib; record calib.version/timestamp.
• G6 | Noise & residuals: robust metric Q_res within admissible band; robust pipelines report a second-order surrogate.
• G7 | Paraxial conservation: ε_flux → 0 @ O(θ^2); if violated, switch model or segment.
• G8 | Uniqueness & integrity: unique record_id, non-duplicated checksum.sha256; complete audit.jsonl.

III. Stop Criteria (Stops)

• S1: p_dim < 1 or check_dim failure → immediately stop current batch.
• S2: clock_state != locked beyond grace, or |δt_abs| / |Δτ_ch| exceed thresholds → stop writing.
• S3: f_s < 2 f_max, Δell > Δell_max, or len(path) < 2 → stop and reject batch.
• S4: |θ| > θ_max or |∇ n_eff|·L_coh ≮≪ 1 → stop paraxial chain.
• S5: citation non-compliance (ver_presence < 1 or anchor coverage < 90%) → block release.

IV. Fallbacks & Actions

• F1 | Sync fallback: GNSS → PTP → NTP; if still failing, enter [Restricted] and pause writes.
• F2 | Model switch: when paraxial fails, enable full-wave/numerical propagation; tag results [Restricted].
• F3 | Recalibrate/Recompute: expired τ_calib, u(θ_k) over limit, or p_dim < 1 → enqueue for recalibration/recompute.
• F4 | Segmentation & adaptation: if |∇ n_eff| rises or leakage risk increases, shorten T_coh/L_coh, increase spectral resolution, or segment the path.
• F5 | Robust substitute: if Q_res out of band or heavy-tailed noise, switch to Huber/quantile pipeline and report surrogate + power impact.

V. Online Monitoring & Alerts

• Rolling metrics: Q_res, p_dim, σ_y(τ), SNR, ε_flux, δt_abs, Δτ_ch.
• Alert levels: warn (near threshold), error (threshold breach → trigger F1–F5), fatal (meets S1–S4 → stop immediately).
• Heartbeat & audit: log clock_state and threshold decisions to audit.jsonl every Δt_beat.

VI. Reporting & Release Gate

• Release bundle must include: manifest.yaml / schema.json / observations.parquet / paths.parquet / check_dim_report.json / quality_report.json / audit.jsonl.
• If any G1–G8 fails or any S1–S5 is triggered, block release; only [Restricted] qualitative notes and diagnostics may be output.

VII. Machine-Readable (drop-in)
A. gates.yaml

version: "1.0.0"

gates:

G1: { schema: true }

G2: { cite_anchor_min: 0.90, forbid_external_links: true }

G3: { path_min_len: 2, require_sync: true }

G4: { p_dim: 1.0, units: { T_arr: "s", Phi: "rad" } }

G5: { tau_calib_s_max: 86400 }

G6: { Q_res_band: [0.0, 0.2], robust_surrogate: true }

G7: { epsilon_flux_guard: "≈0@O(theta^2)" }

G8: { unique_record_id: true, unique_checksum: true }

stops:

- { code: "S1", when: "p_dim<1 or check_dim_fail" }

- { code: "S2", when: "sync_unlocked and over_grace or abs_t/skew_over" }

- { code: "S3", when: "fs<2*fmax or Delta_ell>Delta_ell_max or path_len<2" }

- { code: "S4", when: "theta>theta_max or grad_n_eff_Lcoh_not_small" }

- { code: "S5", when: "ver_presence<1 or anchor_coverage<0.90" }

fallbacks:

F1: "sync_fallback"

F2: "switch_fullwave"

F3: "recalibrate_recompute"

F4: "segment_adapt"

F5: "robust_substitute"

B. quality_report.json (example fields)

{

"run_id": "01JXYZABCD...",

"gates": { "G1": true, "G2": 0.94, "G3": true, "G4": true, "G5": true, "G6": true, "G7": true, "G8": true },

"stops_triggered": [],

"fallbacks": ["F1"],

"metrics": { "p_dim": 1.0, "Q_res": 0.12, "epsilon_flux": 0.006, "delta_t_abs_ns": 18, "allan_1s": 1.1e-11 }

}

VIII. Results & Compliance Page — Minimal Required (aligned with Ch.5/7)

• Metrics: ΔT_arr (s), r_phi (1), ε_flux (1), p_dim (1), Q_res (1) with intervals/threshold comparison.
• Compliance: check_dim_report.json summary, anchor_coverage, ver_presence, clock_state, τ_calib.
• Note: if [Restricted], append the tag to all figure titles and list the specific triggers.

IX. Checklist

• All G1–G8 passed; anchor_coverage ≥ 90%; no external links/aliases.
• Path expressions explicitly show gamma(ell) and d ell; delta_form recorded.
• p_dim = 1.0, clock_state = locked; |δt_abs|/|Δτ_ch| and σ_y(τ) within thresholds; τ_calib compliant.
• Sampling satisfies f_s ≥ 2 f_max, Δell ≤ Δell_max; len(path) ≥ 2.
• Release bundle (manifest/schema/data/reports/audit) complete and validated; checksum not duplicated.
• Any S1–S5 triggers handled via F1–F5; when applicable, tag [Restricted] and block external quantitative release.