Current capabilities and supported builds

Start here

These guides describe the implemented workflows and their limits. Public documentation needs no account. To open your private workspace, visit /login and choose Continue with Google. Access to a particular model or execution action depends on the signed-in account and its region.

Start with native input editing and then saved JSON output plans and results. Inputs, output plans and simulation results have separate identities and review steps.

Supported builds

RuFaS 1.0.5 is the first integration. The two identifiers below describe reviewed integration builds of the same source release; they are not two different upstream releases.

The default is resolved once when creating a workspace or revision. Existing saved revisions remain pinned to their build and binding. A changed catalog default does not migrate them. Changing builds requires a compatible new workspace or scenario; no automatic cross-build scientific migration is supplied.

An available catalog entry, a successful save, merged code, or fixture-based CI does not establish live deployment availability. Check the application's current catalog and regional action capabilities. The server still checks ownership, region, build binding, input evidence and execution admission. A reserved surrogate engine is unavailable; additional RuFaS releases are not advertised here.

What works in the code

• Native input import, exact Source editing, supported GUI projections, conditional batch saves, revision history, diffs and export share the same native files.
• The standard-library Python CLI supports local input work and the authenticated public remote gateway. Local input editing needs no account or model execution.
• Outputs provides a reviewed five-selector catalog, search and native JSON preview. The opt-in build adds saved JSON plans and bounded import/export.
• Stored successful file-workspace simulation runs with reviewed bindings and pinned manifests support searchable series, literal observation tables and a chart of the currently loaded page. Original files and verified downloads remain available.
• Saved-data inventories and selected input exports exist. Their availability does not imply a complete deletion workflow.

The reviewed schema projection is partial. Missing descriptions, units, dependencies, CSV identities and reviewed discovery-registry bindings remain explicit. These workflows do not invent missing scientific metadata or certify unrestricted RuFaS inputs.

Saving, audit and execution

Preview checks a proposed edit or output selection without publication. Save creates a preserved revision and receipt; it does not launch a simulation. Portable validation checks syntax and supported declared rules, with coverage limits. A native input audit is separate evidence on a frozen snapshot. Execution admission then checks the selected build's protected profile and available regional action.

Choose Run, review the exact saved scenario revision and choose Confirm and run to explicitly request execution. Where required, the existing flow validates, freezes a snapshot and completes its native audit before requesting simulation. A passing audit is not whole-model scientific certification, and an unavailable action cannot be enabled by saving inputs or a filter.

See revision checks and audit versus execution eligibility.

Results and scientific limits

The five reviewed selectors cover initial herd-population statistics and daily milk production. A catalog emitter does not guarantee that a run emitted the corresponding output. Missing selectors and metadata are reported.

Observation position is a zero-based location in a series, not an inferred simulation day. Loaded-page charts use only finite scalar numeric values on the page already loaded for the table. Units are checked only on that page. No full-series validation, time reconstruction or scientific certification is implied. Read the chart limits before interpreting a picture.

Deferred or unavailable

Persistent saved work is the default. General saved-data deletion and a production ephemeral lifecycle are unavailable. The intended ephemeral behavior would remove submitted inputs after their operational need and provide a 24-hour results window after terminal completion, but that is a deferred requirement, not a current expiry or erasure guarantee. Removing a field from a new revision does not erase its history.

Selected-build live rollout remains deferred. Real compatible-run comparisons, wider output formats, standalone reusable output presets, durable scientific observation indexing, experiments, optimization and surrogate execution remain unavailable or separately scoped. Demonstration Compare/Experiments surfaces are not scientific comparison tools.

The input Agent view and results inspector provide a local, read-only explanation capability for one saved JSON revision or the selected observation page. It requires explicit server-side provider/model configuration; otherwise it shows unavailable. Explanations select inspected evidence citations and display literal values and declared schema/catalog evidence. Unsaved drafts are not shared. Schema/native alignment remains unverified and build-qualified discovery remains unavailable. Explanation requests remain read-only. The input Agent view also supports one explicitly requested JSON scalar replacement: specify its existing pointer and literal value, inspect the exact before/after values and declared validation warnings, then choose whether to accept it into the existing unsaved draft. Dirty or stale drafts are protected. Saving uses the separate explicit review/validation flow. Unsupported schema targets, structural/multiple edits, inferred scientific values, simulation launches, comparisons and recommendations remain unavailable. Local fake-provider checks do not establish live-model quality or deployment availability; evaluated broader agent workflows remain separately scoped.