© Ribose Asia Limited 2026
This document is not a Ribose Standard. It is distributed for review and comment, and is subject to change without notice and may not be referred to as a Standard. Recipients of this draft are invited to submit, with their comments, notification of any relevant patent rights of which they are aware and to provide supporting documentation.
All rights reserved. Unless otherwise specified, no part of this publication may be reproduced or utilized otherwise in any form or by any means, electronic or mechanical, including photocopying, or posting on the internet or an intranet, without prior written permission. Permission can be requested from the address below.
Ribose Asia Limited
Suite 1, 8/F, New Henry House
10 Ice House Street
Central
Hong Kong
www.ribose.com
This part fixes the numeric law of the series: where exact arithmetic is required, how numbers cross serialization boundaries, which algorithm produces transcendental results and how its version is stamped. Every other part that computes or serializes a number cites this part.
Metrology data dies by a thousand float roundings. A conversion declared exact that transits an IEEE-754 double, an author’s 2.10 normalized to 2.1, a coverage factor recomputed by a different library on each platform — each silently invalidates byte-identity between implementations. This part makes numerics a contract: exact where exact is possible, declared and versioned where it is not, and byte-identical at every boundary.
This document specifies: exact rational arithmetic requirements; decimal-string representation; the t-quantile algorithm and its versioning; rounding rules; and the byte-identical JSON serialization contract for conformance corpora and evaluator output.
This document does not define when to compute (Parts 4, 5 and the evaluation parts of the Plurimath series decide that); it defines how any computation in the series must behave numerically.
No terms and definitions are listed in this document.
a decimal string preserved verbatim from its source; trailing zeros are significant (2.10 ≠ 2.1).
the declared number of extra digits carried beyond the published precision of a non-exact result.
byte-exact or distributional (CalConnect 36819).
the identity and version of the algorithm producing a non-exact result (Clause 7).
Requirement language per CalConnect 36812. All examples in this part show serialized decimal strings, not internal representations.
Conversions declared exact="true" (CalConnect 36816, scale and difference arithmetic (CalConnect 36815, Type A statistics, and all rational operations of propagation are computed in exact rational arithmetic; an implementation never rounds an intermediate exact value.
Irrational results — square roots of non-square rationals, Student-t quantiles — are computed by the declared algorithm of Clause 7 or refused — never through a floating-point shortcut.
Expected outcome. Converting 32 °F to kelvin: the factor 5/9 and the offset arithmetic run in rationals; the serialized result is exactly "273.15". A float path would emit platform-dependent digits at the tail.
Refusal expected. sqrt over a non-square rational factor in a canonical decomposition has no exact rational result; the conversion of such a synthetic unit is refused with a typed error, not approximated.
Numbers crossing a serialization boundary are decimal strings. IEEE-754 values never appear in any artifact, corpus or API result of this series. Decimal strings preserve:
the author’s text verbatim in record positions — rounding is the author’s act (JCGM 100:2008, 7.2.2);
the exact rational value elsewhere: integer strings for integral rationals, p/q rationals where the register uses them, and fixed-decimal strings with declared guard digits for non-exact results.
Verbatim preservation. The author writes 2.10; the record, every render and every vector carry "2.10". "2.1" is a different author text and never replaces it.
Register rationals. The register carries the litre as factor: '0.001' and degree Fahrenheit as factor: 5/9, offset: 229835/900 — exact by construction. Conforming conversions reproduce these strings on round-trip.
Coverage factors from the Student-t distribution at probability p with ν degrees of freedom are computed by Newton iteration on the regularized incomplete beta function at fixed precision, with a published monotone error bound. The algorithm identity and version are stamped in every result and vector metadata (numeric_algorithm).
Expected outcome. For p = 0.95, ν = 16: k = 2.119905…, published to the vector’s declared decimals with numeric_algorithm: "unitsml-t-newton-1". A conforming TypeScript runtime emits the same bytes as the Ruby reference for the same stamp. Substituting a table-based or library quantile without restamping is non-conforming.
Record nodes: no rounding by the implementation; author text is preserved verbatim.
Vectors: published at the fixed decimal places declared per vector.
Presentation: rounding is a presentation decision and never feeds back into stored values.
Interpretation. An evaluator computing u_c = 34.1839… nm for a vector declaring three decimals publishes "34.184"; the stored record meanwhile keeps whatever text the author wrote. Neither the author text nor the vector is altered by the presentation layer’s choice to display 34 µm.
JSON output of conformance replays and evaluator APIs are: UTF-8; fixed key order (lexicographic, defined once per schema); decimal strings for all numerics; stamped with engine_version, grammar_version, vector_set_version, numeric_algorithm. Two conforming implementations given the same input emit byte-identical output.
Byte-identity. Given the same conversion input, the Ruby and TypeScript runtimes emit identical JSON including key order and the trailing digits of "127/25000"-class rationals. Any byte difference is vector drift and blocks the release (CalConnect 36819).
This part is enforced by every vector of CalConnect 36819: byte-identical replay is the test. Class coverage: any claimed class implies this part. Additional test-suite examples are given in
Exact flag discipline. exact="false" percent-per-degree (0.5729577951308232087679815481): the result carries the declared digits, the vector declares tolerance_d, and the output contains no exact marker.
No float leakage. Grep gate: no serialized corpus artifact contains a decimal exponent form (e-17) originating from a binary double; guard-digit strings only.
Stamp integrity. Two vectors with differing numeric_algorithm stamps are never byte-compared against each other; the harness rejects the cross-stamp comparison.
The following documents are referred to in the text in such a way that some or all of their content constitutes requirements of this document. For dated references, only the edition cited applies. For undated references, the latest edition of the referenced document (including any amendments) applies.