Authored
Documents are written with real structure — clause anchors, typed tables, formulas and figures — so the corpus has a spine to anneal around.
Konneal serves your standards corpus as a question-answering service.
Members ask questions in natural language and receive answers that name the publication, edition and clause behind every claim. Rules that carry machine-readable models are evaluated deterministically, and the verdict is returned with the arithmetic shown. The service runs under your organization’s own brand, on your own corpus.
The method is measured: asked over plainer representations of the same corpus, the same questions answer two of eighteen capability probes; over the annealed corpus, seventeen to eighteen.The measurement.
What your members see
An answer from a deployment carries its own evidence. The rendering below is illustrative of the served format; every element in it is produced by a mechanism described on this site.
Questionillustrative
Is a creep test to Dmax = 26 000 v valid for a cell rated Emax = 30 000 v?
Verdict · computedINVALID
constraint D_max ≤ 0.8 · E_max
0.8 × 30 000 v = 24 000 v
D_max = 26 000 v
→ exceeds the limit by 2 000 vsource: machine model · R 60-1 model · constraint
Answer
The test is out of scope: for creep testing the maximum test load is bounded at“0.8 · Emax” — a hard limit of the model, so the verdict is computed rather than judged R 60-1:2021 §4.4.2. The cell may be tested to Dmax = 24 000 v at most.
01 · citation
Every substantive claim names the publication, edition and clause behind it — the reader can go and look.
02 · verbatim values
Normative values are quoted exactly from the source passage, so the claim can be checked mechanically.
03 · verdict as data
The rule is evaluated by the deterministic engine. The verdict and its arithmetic arrive attached to the answer, and the model may not soften or contradict them.
What you already have
Every artifact below is structure your organization has already authored. Konneal reads each one directly, and each one raises what your members get.
Metanorma
You get a working answering service in which answers are cited to the exact clause, and tables, formulas and figures are rendered from source.
+ Pubid
Identifiers are understood natively: a question naming a publication steers retrieval, and citations render in your identifier scheme.
+ Relaton
The registry carries publication families and supersession links, the citation graph records what each edition’s bibliography cites, and answers are steered to the edition in force.
+ Glossarist
Everyday words are resolved to your defined terms, in register, across languages.
+ Lutaml
Structured models are indexed and served as first-class objects.
+ Primmel
Verdicts are computed with the arithmetic shown, counterfactuals are evaluated, and certificates of absence are returned.
A filled tag marks the baseline that a Metanorma library already provides; an outlined tag marks what a further asset adds.
How that is possible
Three layers make it possible: the documents are authored with structure, the rules are modelled as machines, and the serving layer anneals and serves the result.
02Primmel
Modelled
The rules themselves are written as machines: typed attributes, constraints with numeric limits, calculations and ordered test sequences.
03Konneal
Served
Konneal runs annealment passes over what you have written, retrieves over the structure, enforces the answer contract, and holds every release to the gates.
The public reference deployment
Konneal serves a live, public deployment over a real standards corpus today. Every claim on this site about answers, citations and verdicts can be tested there by anyone.
Adoption starts with a Metanorma library.
If your organization authors in Metanorma, the grounded answering surface is available from day one. Every asset you add on top — identifiers, bibliographies, terminology, models — raises the annealment your members get.
The adoption paths →