engineering.uncertainty.propagate

Three distinct suites: educational examples on the calculator page, this legacy regression count (88/88 golden + boundary), and the CVP verification suite below. They are not the same number. Property, metamorphic, round-trip, O3, and interface statuses/checks are recorded in the CVP Manifest . SKIP (0/0) means that suite has no published vectors.

This calculator CURRENT · Public schema 1.9.8 matches · Semantic contract ✓ · Production attested · Public/cache ✓ · Origin ✓

CVP verification suite

Protocol checks from the Evidence Manifest. Distinct from the legacy regression suite above. Verification revision 2026-10-28.meta-metrics-v2. Calculation version 1.9.8. CVP 1.0.0-proposed. 18/18 golden · 47/47 CVP boundary · 29/29 invalid · 29/29 property · 2/2 metamorphic · 10/10 O3 · 6/6 cross-interface · 10/10 CVP contract. digest 3c57ad0b1aee. Published modes: linear GUM (sum, difference, product, quotient, RSS, sensitivity, budget) and seeded Monte Carlo replay of y_mean/u_c/percentiles. Not a full JCGM 101 validation procedure. RNG: mulberry32; normal sampler: Box-Muller. ≤2 ULP vs O3 applies only to the published tabulated linear GUM vectors (sum, difference, ρ=1 difference, product, quotient, RSS, sensitivity, 3×3 PSD, singular PSD, budget). It is not a whole-domain guarantee and does not cover Monte Carlo.

• Golden (oracle-backed) 18 / 18
• CVP boundary 47 / 47
• Invalid input 29 / 29
• Property 29 / 29
• Metamorphic 2 / 2
• O3 high-precision 10 / 10
• Cross-interface 6 / 6
• CVP contract 10 / 10
• Interfaces PASS · UI (SSR) / REST / MCP
• Interface claim ui-ssr is query-result HTML, not a live browser session.
• Verified scope Linear GUM: sum, difference, product, quotient, RSS, sensitivity, budget. Correlation matrices must be PSD (singular PSD allowed). O3 mpmath (80 dps) tabulated linear u_c. Monte Carlo: O2-B seeded mulberry32+Box-Muller replay of y_mean/u_c/percentiles (≤2 ULP) for sum, difference, product, quotient, correlated normal, and budget/rectangular; linear_u_c vs first-order GUM.
• Assurance Engineering
• Deferred Adaptive MCM, arbitrary PDFs, and the full JCGM 101 validation procedure. Budget sensitivity-coefficient dimensional algebra is not automatically verified. Correlated non-normal MC remains out of O2. O3 does not cover Monte Carlo.

• contract page version PASS
• contract capability version PASS
• contract input schema version PASS
• contract output schema version PASS
• evidence version matches manifest PASS
• schema output required declared PASS
• schema valid input executes PASS
• engine success output validates schema PASS
• schema rejects incomplete input PASS
• rest success payload validates schema PASS

Fingerprint sha256:d8cfcf55210b3e57585f02738f49bb344b4c1cc9d6597384fadb8f7da98e3fb8

Input schema sha256:263d0ba0dda3ea81aa635bcd61d0eb9e9e9fc347ab4890b6cc20ccacc3a6843f · Output schema sha256:d1d0c1d06cbe6b38725a8566ba4273291b9558081c5786b8e0b65b862a8e66b1 · Build-time SHA-256 digests prove published artifacts agree with each other (release_integrity.scope=workspace). Production attestation GETs canonical URLs twice (public/cached view and origin), including Capability HTML, via npm run attest:production --write (published at /.well-known/calculatorx-production-attestation.json) and fails if the public Capability page advertises a different calculation_version. This is artifact consistency, not an independent immutable release log.

Tool

uncertainty-propagate

Engine

uncertainty-propagate

Engine build

2026-09-12

Evidence core

2026-09-12T08:32:37.705Z

Rendered

2026-11-10T05:18:41.719Z

Last reviewed

2026-08-10

Assumptions

• Default uncorrelated (rho=0); optional correlations when provided.
• Linear propagation assumes a local first-order approximation is adequate. For significant model nonlinearity, consider higher-order treatment (JCGM 100:2008/Amd.1:2026) or Monte Carlo propagation (JCGM 101:2008).

Limitations

• Linear GUM (binary/sensitivity/RSS/budget) or Monte Carlo (JCGM 101) with normal/rectangular draws; not adaptive/MCMC
• MC: N in 100…100000 (default 10000), deterministic Mulberry32 seed; confidence → empirical percentiles; k → U=k·u_c
• budget/sensitivity Type A/B + distributions are metadata (except MC uses distribution for draws); u is standard uncertainty
• contributions are diagonal (c·u)² fractions for linear modes; with correlations percents may not sum to 100%
• Product/quotient use absolute first-order forms in linear mode; MC quotient rejects >5% near-zero denominator draws (MONTECARLO_UNSTABLE)
• n×n correlation matrix not positive semidefinite → CORRELATION_MATRIX_NOT_PSD (singular PSD allowed); duplicate unordered pair → DUPLICATE_CORRELATION; conflicting ρ for the same pair → CONFLICTING_CORRELATION
• Optional Quantity Engine v1.1: SI-prefix conversion + electrical derived rewrites (1000 mV + 2 V → 3 V; V/A → Ω; A·Ω → V; V×A → VA); different additive dimensions → UNIT_MISMATCH; not full SI algebra
• Unknown mode → INVALID_MODE; k and confidence together → INVALID_INPUT
• confidence with nonzero ρ/correlations → CORRELATED_DOF_NOT_SUPPORTED (use fixed k)
• MC rectangular + correlation → CORRELATED_NON_NORMAL_MC_NOT_SUPPORTED (normal marginals only)
• Public Agent schema uses canonical field names only (components.u/c/x; correlations i/j as index or name). Runtime still accepts legacy aliases (ua/uncertainty/sensitivity/value) for compatibility.