R05 · Explore alternatives

How to use this routeCopy Anchor Link

Variant count: the user's number, otherwise the stage default (explore 4, refine 2). Each option tests a distinct structural or directional hypothesis with the same content; a palette or font swap alone is not an option. Explore has no independent review and at most one fix pass per artboard: speed and range matter more than polish. Refine adds jado-canvas-verify and, with a normal or deep budget, one independent review. Selection belongs to the user; in autopilot the coordinator records a provisional selection.

Stages: explore, refine. Work surface: Paper. Involvement and budget come from the run's dials (shared rules).

MapCopy Anchor Link

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.

flowchart TD
  n_p(["Frame the comparison"])
  n_evidence{"Does missing evidence block the options?"}
  n_r["Collect blocking evidence"]
  n_d["Design the alternatives"]
  n_self["Self-check within the stage cap"]
  n_check{"Does the stage need an independent review?"}
  n_review["Review the alternatives"]
  n_fix["Repair blocking findings"]
  n_present["Present the options"]
  n_choose["Select or iterate"]
  n_next{"What did the user decide?"}
  n_snap["Record the selection"]
  n_done(["Deliver the selection"])
  n_code(["Hand off to build"])
  n_stop(["Deliver the unselected draft"])
  n_p --> n_evidence
  n_evidence -->|"Yes"| n_r
  n_evidence -->|"No"| n_d
  n_r --> n_d
  n_d --> n_self
  n_self --> n_check
  n_check -.->|"No review · explore/refine only"| n_present
  n_check -.->|"Independent review · refine only"| n_review
  n_review -->|"No blocking findings"| n_present
  n_review -->|"Blocking findings"| n_fix
  n_fix --> n_present
  n_present --> n_choose
  n_choose --> n_next
  n_next -->|"Iterate with feedback"| n_d
  n_next -->|"Selection made"| n_snap
  n_next -->|"Budget reached or no selection"| n_stop
  n_snap -->|"Stop here"| n_done
  n_snap -->|"Code in the mandate"| n_code

Input and outcomeCopy Anchor Link

• Input: Brief and comparison question.
• Outcome: Comparable alternatives and a recorded selection.

Work limitsCopy Anchor Link

Route cap: at most 4 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 instructionsCopy Anchor Link

Frame the comparisonCopy Anchor Link

Record the question, shared content and state, and the resolved variant count; report a budget shortfall instead of silently changing the count.

ID: p.

Continue with «Does missing evidence block the options?».

Does missing evidence block the options?Copy Anchor Link

Search only when a missing fact would change every option.

ID: evidence.

• Yes → «Collect blocking evidence».
• No → «Design the alternatives».

Collect blocking evidenceCopy Anchor Link

Answer only the blocking question.

Skill instructions jado-researcher ID: r.

Continue with «Design the alternatives».

Design the alternativesCopy Anchor Link

Explore: divergent hypotheses at low fidelity, built quickly. Refine: develop the selected direction with project tokens and states.

Skill instructions jado-visual-designer ID: d.

Continue with «Self-check within the stage cap».

Self-check within the stage capCopy Anchor Link

Explore: fix only broken text, clipping, overlaps or missing content, at most one pass. Refine: jado-canvas-verify, at most two passes. List remaining issues instead of polishing.

ID: self.

Continue with «Does the stage need an independent review?».

Does the stage need an independent review?Copy Anchor Link

Refine: when the budget is normal or deep. Explore: the self-check is enough.

ID: check.

• No review (explore, refine only) → «Present the options».
• Independent review (refine only) → «Review the alternatives».

Review the alternativesCopy Anchor Link

Compare captures with the criteria read-only; report only issues that affect the choice.

Skill instructions jado-visual-reviewer ID: review.

• No blocking findings → «Present the options».
• Blocking findings → «Repair blocking findings».

Repair blocking findingsCopy Anchor Link

The same designer spends one fix pass and recaptures; the flow continues with the fresh capture.

ID: fix.

Continue with «Present the options».

Present the optionsCopy Anchor Link

Show each option with its hypothesis, capture and known issues; recommend one with reasons.

ID: present.

Continue with «Select or iterate».

Select or iterateCopy Anchor Link

Checkpoints and pair: ask with the native question tool (each option, combine, iterate, more options). Autopilot: record the recommended option as a provisional selection in the decision log.

ID: choose.

Continue with «What did the user decide?».

What did the user decide?Copy Anchor Link

Iteration rounds consume the same budget.

ID: next.

• Iterate with feedback → «Design the alternatives».
• Selection made → «Record the selection».
• Budget reached or no selection → «Deliver the unselected draft».

Record the selectionCopy Anchor Link

Preserve selected nodes, revision, content and the reasons behind the choice.

ID: snap.

• Stop here → «Deliver the selection».
• Code in the mandate → «Hand off to build».

Deliver the selectionCopy Anchor Link

Return the comparison and selection; implementation starts in R04 on request.

ID: done.

The route ends here.

Hand off to buildCopy Anchor Link

Continue to R04 only when code is in the mandate; in checkpoints involvement confirm first.

ID: code.

The route ends here.

Deliver the unselected draftCopy Anchor Link

Return the options and state the missing selection or reached budget.

ID: stop.

The route ends here.

Method and sources