Elamur Groupantsmova-nightingale-delta
The groupant beside tools/nightingale/src/delta.rs in the Mova repository: what the code computes, what it never does, how to go back from its outputs, its invariants, cost and checks.
Facts
Implementation material. A Mova code model describes this code at one source revision: partial models, finite checks and their stated scope. It is not a reviewed general exposition, and its labels are shown as written, never raised.
Components
| component | label | where |
|---|---|---|
| Approximant | partial | Forward approximant |
| Ast | partial | syntax.json |
| Composition | partial | Composition |
| Cost | partial | Cost |
| Descriptant | present | Descriptant |
| Invariant | partial | Invariant |
| Tests | finite evidence | Tests |
Obernants
The typed relations around the forward model: the behaviour complement (what the code never does), the reverse reconstruction (every input compatible with an output) and specialised views.
| relation | type | label | where |
|---|---|---|---|
| Behavior complement | behavior complement | partial source derived model | Behavior complement |
| Reverse reconstruction | reverse reconstruction | partial source derived model | Reverse reconstruction |
Model text
Descriptant
Varieties as deltas over the standard: per-facet log ratios for forms and bigrams, kept by evidence, size and Dunning's G2, written beside the model and composed log-linearly by a label's facets. Source: tools/nightingale/src/delta.rs.
Forward approximant
learn: for each facet with text on both sides and each f/b record, delta = log2((cF+1/2)/NF) - log2((cR+1/2)/NR), kept if cF+cR >= 5, |delta| >= 1 bit, G2 >= 10.83, top 200 000 by G2; shift(label) = sum of the facets' deltas, unknown facets reported.
Invariant
A facet without text on both sides learns nothing; deltas are read only for their model's lineage.
Cost
O(facets x records) at a handoff (seconds on the probes).
Composition
Parent: nightingale.
Behavior complement
Never changes the standard's counts. Evidence status: source-derived model with finite checks, not a proof.
Reverse reconstruction
deltas.tsv keeps each delta's counts on both sides, so the estimate can be recomputed.
Tests
Checks
tests/model.rs varieties test; random-split control learns 0 deltas. These are finite fixtures and measurements; not a proof.
History
2026-11-06 · document
Epoch 2026-11-06
First model, written with the source (the Nightingale: one per language under the Eel; a standard naturalness model with variety deltas).
Provenance
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/tools/nightingale/src/delta.rs.groupant/