Known limitations in VAMOS 1.1.0¶

These limitations are part of the current release scope. They describe unsupported behavior rather than future compatibility promises.

Algorithms and numerical behavior¶

VAMOS minimizes every objective. Users must transform maximization objectives themselves.
A valid population size may depend on an algorithm's reference directions or weight lattice. Incompatible sizes fail instead of being adjusted silently.
The strict evaluation budget must be at least the resolved population size.
Deterministic same-environment runs do not imply bitwise equality across different kernels, BLAS implementations, operating systems, architectures, or optional third-party engines.

welded_beam is a six-variable mixed design example with its own formulas, materials and stiffeners. Its constraint metadata is fixed, but this does not certify equivalence to the canonical four-variable welded-beam benchmark.
cec2009_uf3 still searches [-1, 1] for variables 2 to n. Treat its domain as pending reference validation rather than assuming canonical CEC2009 UF3.
Independent algorithm-conformance validation for AGE-MOEA, RVEA and SMPSO remains pending. Passing API and regression tests does not establish convergence equivalence with a reference implementation.
IBEA's epsilon indicator is unscaled, as in jMetalPy, so objective units still shape its selection pressure. With indicator="hypervolume", far-dominated solutions in early populations dominate the adaptive normalization, which slows convergence on problems such as DTLZ1; jMetal's IBEA behaves the same. VAMOS 1.0.2 and earlier also inverted the hypervolume indicator and counted overflowing contributions as zero; see the IBEA indicator corrections. Indicator values divided by kappa must fit in float64 (objective differences below about kappa * 1.8e308); larger ones saturate to infinity, and solutions with saturated contributions are not ordered among themselves.

NumPy is the reference path. Numba accelerates selected kernels rather than every operation; MooCore accelerates selected indicators rather than whole algorithms.
Explicitly selecting an unavailable backend is an error. VAMOS does not install dependencies during a run, verification, or replay.
Optional research frameworks and model-based tuning packages have their own Python and operating-system constraints.
The release CI claims Python 3.10, 3.11, and 3.12 on Linux, and Python 3.12 on Windows and macOS. Other interpreter/platform combinations are unclaimed.

Public readers support the 1.0.0 run schema. Internal pre-public formats are unsupported and must be regenerated.
Exact replay is available only for reconstructable registered built-in components with exact material environment compatibility.
Custom Python, plugins, cross-backend execution, and best-effort environment matches are not exact-replay targets.
SHA-256 evidence detects modification; it is not an authenticity signature.

Study mutation is single-owner. Do not run, resume, or retry the same study concurrently from multiple processes.
Study execution is sequential in 1.1.0. Distributed workers, multiprocess ownership, and cross-process cancellation are unsupported.
Cancellation is cooperative and local to the process that owns execution. Studio's token-addressed worker stop ends the worker and requires reconciliation on resume; it is not durable cross-process cancellation of the study API.

Statistical analysis, visualization, MCDM helpers, tuning, and racing APIs remain experimental and may change in a minor release.
Users are responsible for selecting indicators, reference points, sample sizes, and statistical tests appropriate for their scientific claim.

Studio is experimental and is intended for trusted local use. It is not a multi-user hosted service.
AST checks, restricted builtins, process isolation, resource limits, and timeouts are best-effort controls, not a security sandbox.
Reviewed Python executes with the current operating-system user's permissions. Remote binding increases exposure and requires explicit opt-in.
LLM-generated code is displayed for review and is never executed automatically.

Plugin descriptors, custom component interfaces, and LLM-provider integrations are experimental and are not covered by the 1.x stable API.
Loading and verification never import recorded plugins or contact providers.

The stable API and strict release scope pass mypy. Full-source typing still has a frozen diagnostic ratchet and is not yet zero-error; new or increased diagnostics fail the release gate.