R07 · Staged delivery
Carry one mandate through UX, canvas and code up to the recorded stopAfter, with shared budget and stage-appropriate checks.
- Status
- Proposed
- Last reviewed
- 2026-10-06
- Stage
- Operations
How to use this route
Record stopAfter before starting: ux, canvas or code. Activate only needed roles; the coordinator may author. Ship requires independent canvas and code reviews; explore and refine use self-checks unless the budget adds a review. A UX deliverable does not authorize canvas, and a canvas deliverable does not authorize code. Canvas and code share the repair batches.
Stages: explore, refine, ship. Work surface: as specified in the assignment. Involvement and budget come from the run's dials (shared rules).
Map
Read from top to bottom. Labeled arrows show conditions; dashed arrows exist only in the stages they name; loops represent repairs within the budget. Instructions for each node appear below.
Loading diagram…
Mermaid source
flowchart TD
n_p(["Verify mandate and stopAfter"])
n_brief["Define tasks, states and criteria"]
n_research["Collect required evidence"]
n_decide["Resolve choices"]
n_result{"What is the stopAfter output?"}
n_ux_verify["Verify the UX artifact"]
n_ux_gate{"Does the UX artifact satisfy the contract?"}
n_ux_done(["Deliver UX and checkpoint"])
n_copy_needed{"Does the content need work?"}
n_copy["Prepare the content"]
n_design["Design the canvas"]
n_canvas_check{"Does the stage need an independent canvas review?"}
n_canvas_review["Review the canvas"]
n_canvas_gate{"Is the canvas accepted?"}
n_canvas_fix["Repair the canvas"]
n_code_scope{"Is code in the mandate?"}
n_canvas_done(["Deliver the verified canvas"])
n_b["Implement the accepted canvas"]
n_code_check{"Does the stage need an independent code review?"}
n_review["Review the browser result"]
n_gate{"Did the final checks pass?"}
n_fix["Return repairs to the builder"]
n_done(["Deliver result and run summary"])
n_stop(["Report the block and completed work"])
n_p --> n_brief
n_brief -->|"Evidence missing"| n_research
n_brief -->|"Evidence sufficient"| n_decide
n_research --> n_decide
n_decide --> n_result
n_result -->|"UX artifact only"| n_ux_verify
n_result -->|"Canvas or code"| n_copy_needed
n_result -->|"Blocked by a decision outside authority"| n_stop
n_ux_verify --> n_ux_gate
n_ux_gate -->|"Yes"| n_ux_done
n_ux_gate -->|"No, fix pass left"| n_brief
n_ux_gate -->|"Cap reached or blocked"| n_stop
n_copy_needed -->|"New, changed or unapproved"| n_copy
n_copy_needed -->|"Approved and unchanged: reuse"| n_design
n_copy --> n_design
n_design --> n_canvas_check
n_canvas_check -.->|"Self-check only · explore/refine only"| n_canvas_gate
n_canvas_check -->|"Independent review"| n_canvas_review
n_canvas_review --> n_canvas_gate
n_canvas_gate -->|"Yes"| n_code_scope
n_canvas_gate -->|"No, repair batch left"| n_canvas_fix
n_canvas_gate -->|"Decision or budget missing"| n_stop
n_canvas_fix --> n_canvas_gate
n_code_scope -->|"Yes"| n_b
n_code_scope -->|"No, design only"| n_canvas_done
n_b --> n_code_check
n_code_check -.->|"Self-check only · explore/refine only"| n_gate
n_code_check -->|"Independent review"| n_review
n_review --> n_gate
n_gate -->|"Yes"| n_done
n_gate -->|"No, repair batch left"| n_fix
n_fix --> n_gate
n_gate -->|"Cap reached or blocked"| n_stopInput and outcome
- Input: Brief, delegated authority, stopAfter and budget.
- Outcome: Verified result, limits and run summary.
Work limits
Route cap: at most 8 new delegations, with 3 agents in parallel; at most 2 repair batches. The effective budget is the lower of this cap and the budget dial; the stage sets the fix passes per artifact. Shared rules.
Step instructions
Verify mandate and stopAfter
Check access, scope, delegated authority, stopAfter, dials and budget.
ID: p.
Continue with «Define tasks, states and criteria».
Define tasks, states and criteria
Prepare brief, flow and acceptance criteria from available sources.
Skill instructions jado-ux-architect ID: brief.
- Evidence missing → «Collect required evidence».
- Evidence sufficient → «Resolve choices».
Collect required evidence
Close only the source gap that blocks dependent decisions.
Skill instructions jado-researcher ID: research.
Continue with «Resolve choices».
Resolve choices
Apply the involvement dial: decide and log, or batch the open questions at this checkpoint. Decisions outside delegated authority stop the dependent work.
ID: decide.
Continue with «What is the stopAfter output?».
What is the stopAfter output?
UX stops after artifact verification; canvas or code needs visual design.
ID: result.
- UX artifact only → «Verify the UX artifact».
- Canvas or code → «Does the content need work?».
- Blocked by a decision outside authority → «Report the block and completed work».
Verify the UX artifact
Check only the requested artifact, sources, IDs, links and states.
Skill instructions jado-evidence-contract ID: ux_verify.
Continue with «Does the UX artifact satisfy the contract?».
Does the UX artifact satisfy the contract?
Return defects to the same author within the fix cap.
ID: ux_gate.
- Yes → «Deliver UX and checkpoint».
- No, fix pass left → «Define tasks, states and criteria».
- Cap reached or blocked → «Report the block and completed work».
Deliver UX and checkpoint
Return the artifact with evidence; record next step, decisions and remaining budget.
ID: ux_done.
The route ends here.
Does the content need work?
Reuse approved, unchanged copy as-is and record its source; new, changed or unapproved text needs content work.
ID: copy_needed.
- New, changed or unapproved → «Prepare the content».
- Approved and unchanged: reuse → «Design the canvas».
Prepare the content
Prepare only new, changed or unapproved copy.
Skill instructions jado-content-designer ID: copy.
Continue with «Design the canvas».
Design the canvas
Integrate the copy as the single canvas writer; follow the stage's variant count, fidelity and fix cap.
Skill instructions jado-visual-designer ID: design.
Continue with «Does the stage need an independent canvas review?».
Does the stage need an independent canvas review?
Ship: always. Refine: when the budget is normal or deep. Explore: the self-check is enough.
ID: canvas_check.
- Self-check only (explore, refine only) → «Is the canvas accepted?».
- Independent review → «Review the canvas».
Review the canvas
Review captures and structure read-only before any implementation.
Skill instructions jado-visual-reviewer ID: canvas_review.
Continue with «Is the canvas accepted?».
Is the canvas accepted?
Check the criteria and the selection authority; in checkpoints involvement, confirm the canvas with the user before code.
ID: canvas_gate.
- Yes → «Is code in the mandate?».
- No, repair batch left → «Repair the canvas».
- Decision or budget missing → «Report the block and completed work».
Repair the canvas
The same designer repairs one batch, recaptures and verifies the findings.
ID: canvas_fix.
Continue with «Is the canvas accepted?».
Is code in the mandate?
Continue only when stopAfter is code.
ID: code_scope.
- Yes → «Implement the accepted canvas».
- No, design only → «Deliver the verified canvas».
Deliver the verified canvas
Return canvas, revision, evidence and remaining budget.
ID: canvas_done.
The route ends here.
Implement the accepted canvas
Implement the accepted direction with project components.
Skill instructions jado-ui-builder ID: b.
Continue with «Does the stage need an independent code review?».
Does the stage need an independent code review?
Ship: always, plus UX review for changed critical flows. Refine: when the budget is normal or deep.
ID: code_check.
- Self-check only (explore, refine only) → «Did the final checks pass?».
- Independent review → «Review the browser result».
Review the browser result
Inspect current browser evidence read-only.
Skill instructions jado-visual-reviewer ID: review.
Continue with «Did the final checks pass?».
Did the final checks pass?
Compare checks and evidence with the contract.
ID: gate.
- Yes → «Deliver result and run summary».
- No, repair batch left → «Return repairs to the builder».
- Cap reached or blocked → «Report the block and completed work».
Return repairs to the builder
One aggregated batch to the same author, who verifies the findings.
ID: fix.
Continue with «Did the final checks pass?».
Deliver result and run summary
Return outputs, evidence, decision log, gaps and consumption.
ID: done.
The route ends here.
Report the block and completed work
Record the block and verified parts; continue only independent work.
ID: stop.
The route ends here.