Elamur Groupantsapproximation-next-runner
The groupant beside tools/approx/examples/groupants_next.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 |
| ast | partial | syntax.json |
| cost | symbolic unmeasured | Cost |
| descriptant | present | descriptant |
| invariant | partial | Invariant |
| tests | bounded evidence | Checks and limits |
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.
| id | type | label | where |
|---|---|---|---|
behavior-complement | behavior_complement | partial source derived model | behavior-complement |
reverse-reconstruction | reverse_reconstruction | partial source derived model | reverse-reconstruction |
oracle-independence | independent_observation | partial source derived model | oracle-independence |
Model text
descriptant
Second approximation runner
The mathematical combiner composes fixture generation, reference/surrogate observations and domain/invariant/metric checks into finite verdicts. Correct leaf models do not prove this wiring. Behavior complement uses the composed universe; reverse reconstruction keeps all compatible execution fibers.
Epoch 2026-10-30: predict fast certificate dispatch on constructed narrow intervals and visible overhead or trimmed work on fallback. Range build/query are measured separately. Sparse S1 is a second domain fixture, not fallback. Three runner tests plus included tests pass (34 total); corrupted outputs fail. Finite observations do not prove analytical classifications. Exact witness identity, membership of approximate minima and Unicode character semantics remain outside the metrics.
model
Second executable runner
Forward
For caps {1,2,5,10}, emit four classification rows and two deterministic CHECK rows. Compare byte edit scalar distance, feasible LIS length, scalar minima and S1 probabilities with production references. Require interval containment and expected fast/fallback dispatch for certificate methods. S1 dense/sparse cases have no fallback. Zero references require zero; nonfinite metrics fail. Range results need not occur in queried ranges; LIS witness identity and Unicode character edits are outside observations.
behavior-complement
Universe: fixture revision, budget and proposed CHECK/verdict. Absent outputs include successful verdicts with infeasible witnesses, omitted reference bounds, invalid range acceptance, wrong dispatch, nonfinite probabilities, broken normalization or exceeded error. Classification cannot upgrade finite evidence. Timing is outside this deterministic relation.
reverse-reconstruction
A CHECK row retains all executions producing its scalars and probability bit sum. Checksums are not injective; without the fixture revision many inputs remain compatible. Timing fibers contain many schedules and allocator states. No unique recovery follows.
oracle-independence
Small subsequence masks enumerate the optimum independently. UTF-8 fixture fixes byte semantics. Hand-derived two-world S1 values anchor Game::s1. Range outputs also match direct scans. Included companion tests add separate exhaustive/differential evidence. Scalar corruption, repeated indices, shortened witnesses, halved probabilities and NaN must fail.
Invariant
Successful checks preserve cap, feasible witnesses, zero behavior and finite evidence. Both timing sides receive identical inputs. Full LIS/table objects pass black-box barriers. Build includes allocation and scanning; query excludes build symmetrically.
Cost
CPU-only fixtures: edit<=1003 bytes, LIS/range8192, RSA8x8. Seven alternating rounds after warmup report median[min,max] microseconds. Edit: two calls/round; LIS/range build eight; range_query128:128 batches of128 queries; S1 four. Full original edit DP is compared to complete surrogate including trimming/dispatch.
Checks and limits
2026-10-30: 34 example tests pass (three runner tests plus included companion/reference tests), six shared-library and fifteen schema tests pass; strict approximation all-targets clippy passes. These are not 34 independent new oracles. Two integrated release runs have identical eight CHECK and 16 CLASSIFICATION records; forty paired timings per run are documented in docs/rust-approximants-next.md, including fallback regressions. Semantic AST, universal empirical guarantee, whole-system bound and automatic production substitution are absent.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/tools/approx/examples/groupants_next.rs.groupant/