Elamur GroupantsElamur Groupants
A groupant is the node that sits beside a piece of code and gathers what is known about it: the code's syntax tree, a mathematical model of what it computes, executable surrogates with stated error caps, typed relations that run it backwards or say what it never does, its invariants, and a readable description that remembers earlier attempts. This site publishes the groupants of Mova's code, and a global library of reviewed general explanations as entries are admitted.
Every node has the same core form, written in the project's notation as G = {AST, A, {B}, J, D, tests, …}: the code it covers, models of what that code does and does not do, and the evidence for each. The set is open — a node may hold diagrams, counterexamples or notes beyond these.
What a groupant holds
| component | what it is |
|---|---|
| AST | The structured form of the covered Rust code: a partial syntax projection tied to one source snapshot by its hash; macro bodies stay opaque. |
| A — forward approximant | What the code computes, stated mathematically on a declared input domain: the function or relation from inputs to outputs, errors included. |
| Rust approximants | Executable surrogates that trade exactness for cost under an error cap of 1, 2, 5 or 10 percent. Each names its metric and domain and how the bound was obtained: analytically, by a per-input certificate, or by measurement on stated inputs. |
| B — obernants | Typed relations around the forward model. The behaviour complement says what the code never does within its universe; the reverse reconstruction gives every input that could have produced an observed result, of which undoing is the special case of exactly one; specialised obernants state narrower facts such as metamorphic, resource or termination relations. |
| J — invariants | The constraints that hold on every admitted run, and that let a reverse search discard impossible candidates. |
| D — descriptant | Readable Markdown about the node, including the memory of earlier modelling epochs: hypotheses, surprises, rejected attempts and open questions. |
| tests and evidence | The finite checks, worked examples and measurements behind the models, each with its scope. A passing check is evidence for the listed cases, not a proof for all inputs. |
Reverse reconstruction is a relation, not a function: a result may have many compatible inputs, or none. Choosing one input that recreates the output does not prove it was the original, and a small output error can hide a large input error.
Two sections
Mova code models — 2013 groupants in 67 parts of the Mova repository. Each keeps one Markdown file of blocks beside its manifest; the pages render them from the open repository's committed tree. They describe Mova's own code at one revision and are a section of their own.
Global library — general explanations admitted by an independent review through the publication catalog. Today 0 entries are admitted, so the section shows none; the catalog's other entries are implementation material or unreviewed.
How sure each part is
Each component carries an assurance label: partial, source derived, bounded checked, finite checks, symbolic unmeasured, analytical and others. A paper's claim, an argument from the source, a finite observation and a formal proof are four different kinds of evidence, and the labels keep them apart. This site shows each label as written and never raises one; an error percentage appears only with its metric and domain.
Not a proof. A groupant page is an explanatory edition, not a mathematical proof or a semantic review of its code. Hash and structure checks establish that the files are the ones the manifest names; they do not establish that the models are right.
Sections
Mova code models
Every Rust file of Mova has a groupant beside it, and groups of files have groupants of their interactions. Each page shows one node: its components with the labels they were given, its obernants, and its model text of blocks, bound to the exact source bytes by a SHA-256 hash. They are implementation material for one revision of the code, published as a section of their own. They are not reviewed general expositions, and the labels are shown exactly as written.
2013 groupants · 67 parts📚Global library
The global library is for general explanations of algorithms and scholarly works that an independent review has admitted. Admission is decided by a hash-bound publication catalog: scope, editorial review, rights and an explicit list of the files that may be shown. Code snapshots and implementation wrappers are not promoted into it. Until an entry passes every condition, this section says so instead of filling the space. Its page shows the catalog's counts as they stand.
0 admitted entries⚖️Licence
Our descriptions, models and code snapshots are published under the Elamur Cosmic License, as the rest of Mova. It lets anyone use, change and share the work for any purpose, commercial or not, without asking. Its one condition is that the work never serves surveillance, censorship or political repression. Scholarly works named here keep their own rights, and no paper text is copied onto these pages. The licence page holds the full text exactly as the library keeps it.
Elamur Cosmic License 3.33🔬Elamur Research
Elamur Research is the family's open library of scientific papers, with summaries, authors, topics and citations. A research card links to a groupant here when the card's paper is cited by an admitted global node, matched by DOI, arXiv, ACL Anthology or OpenAlex. The library is ordered by citations and grouped into shelves, topics and concepts. It is the place to read about a work, while this site models code and, once admitted, general explanations.
research.elamur.aiThis page in Markdown → https://md.groupants.elamur.ai/free/en/