Bring data in prototype

• The trigger is a variant, not a screen. Onboarding, a seed, a defect reproduction, a vendor feed, one corrected file and a promotion differ in their source files and their scope; none differs in what the person does first.
• The scope comes from the session. The file carries no party, the shell refuses a party other than the session's, and the screen shows the party as a read-only field rather than a picker.
• The scan is visible before the write. For the sample-folder trigger the table lists what the scanner read and what it ignored, with the reason. The seed journey says the person cannot see this today.
• The family picker is the missing scope. The onboarding journey needs the conventions and the portfolio and nothing else, so the parts the file carries but the journey does not want are unchecked and named in the refusal reason.
• Who may act is a panel, not a footnote. The operations user holds the conventions write and not the trade or market writes; the screen shows the permission that is missing instead of hiding the button.
• The workflow is named before the work starts. A card gives the type, the start message, the failure policy and every step the definition would declare, so the person can see what the import will do and where progress will appear.
• The target is a sandbox, not a free choice. The book tree draws the official tree and the sandbox subtrees anchored on it. An official node is visibly not selectable for import and says why; a sandbox node is selectable and shows the path and the virtual books and portfolios the import will create inside it.

An import always publishes, and it publishes inside a sandbox. A sandbox is not a node of the official portfolio tree: it is anchored at one official portfolio and owns a subtree of its own. The control draws both in one tree, so the person sees the anchor and the sandbox branch together.

• Only a sandbox is selectable. An official portfolio and an official book are drawn with a tag that says "not an import target", and a click on one is refused with the reason: nothing official moves. A sandbox's own portfolio and virtual book are refused too, because the target is the sandbox, not a node inside it.
• An archived sandbox is refused. The model's insert trigger makes an archived sandbox read-only, so the row is drawn locked and a click explains the trigger.
• The selected sandbox is shown by its full path, its anchor, its purpose, its owner and its root portfolio, which has no parent.
• The screen states what the import creates inside it: the virtual portfolios and virtual books, each written with the sandbox id; the root portfolio has no parent; a virtual book has no ledger reference and is not sweepable.
• The model gives sandbox_id to books and portfolios only. The convention, curve, market and run families carry none, so they still publish to the party, not into the sandbox. The screen names that boundary rather than hiding it.

The card reads nothing. The type and the step list are the definition's, and workflow.v1.definitions.list would answer the same list; the stage subjects and the commit are the draft design's. The budgets are the workflow API's: data_step_timeout for a step that stages, write_step_timeout for a step that writes a record.

The list follows the draft: one stage step per family the file carries, then one commit step and one record step. A family the server skips is declared as a step too, so the skip is reported by a step rather than dropped. The engine builds this list from the request at start and rebuilds the same list on every advance, so the card is a promise the run can be held to.

Nothing runs until workflow.v1.ops.start_workflow is published. The card's Follow the run opens the Run screen in its staged state, where the progress begins at that publish.

?case=onboard|seed|defect|feed|one-dataset|promote|party, ?actor=admin|ops|maintainer|support, ?state=source|families|refused, ?scan=0, ?target=sb-import-onboard|sb-sample-seed|sb-import-repro|sb-import-feed. The bar at the foot switches the trigger and the actor; the target is chosen in the tree and kept in the query string. The refusals are the missing permission and the row the validator refuses; a click on an official node or on an archived sandbox adds the target refusal.

The subjects and the permission model are the journeys': ore upload, ore import, ore import-run, marketdata import, compute download-input, and the writes each import borrows. The convention family names come from the modelling tree; the merge policies (replace by key, add missing world, skip) come from the draft design. The step states, their names, the budgets and the failure policies are the workflow engine's. The workflow type, its stage subjects and the per-family stage replies are the draft's and are not registered. The sandbox rules are the model's: a sandbox is anchored at an official portfolio, a book and its portfolio share one sandbox, a sandbox root portfolio has no parent, and a virtual book has no ledger reference and is not sweepable. The tree, the sandbox ids and the counts are invented. The row counts, file sizes, request ids and checksums are invented.

The screen itself has no server. No upload route, no scan route and no staging verb exists. The journey lists them as gaps, and the screen says so in its banner.

The Entry screen names the run; it does not follow one. The Run screen follows the instance and holds the retry, the Review screen sits between the two instances a decider-gated import needs, and the Register lists one row per instance. This screen answers the question the engine cannot: which workflow will this file start, and what will it do.

Not signed off. The screen asks the owner whether the trigger belongs on one screen at all, and whether the operations user is given a permission that means "may import for a party" or stays out of the trade and market imports. The sandbox target adds the blocking question: only books and portfolios have a sandbox_id in the model, so a conventions import cannot publish into the selected sandbox. Either the sandbox covers the book-tree families alone, or the model grows a sandbox column for every imported family.