Compare two scenarios side by side.

Run two versions of a decision through Cascade and lay the consequence trees next to each other. The diff highlights what each choice uniquely causes and where the two futures converge, so the real trade-off stops hiding in the noise.

Run Cascade
AuthorMariete
AgentCascade
Runtime~ 8 minutes
IndustryStrategy
  1. 1

    Describe the task

    Compare mode is Cascade run twice against the same question, once for each option, with the two consequence trees overlaid so you can read them together. Six specialist agents fan out on each scenario; the output puts the trees side by side, tints the branches that only one future causes, and deduplicates the branches both share. What you are left with is the actual difference between the two decisions, not two separate walls of analysis to reconcile by hand.

    The comparison is only as clean as the framing. Hold everything constant except the one variable you are testing, and name it explicitly. Tell Cascade what both scenarios have in common, the single thing that changes between them, and which outcomes you want the diff to foreground, and the unique branches line up against a real fork instead of two loosely related what-ifs.

    Compare two futures against the same question: do we stay fully remote or move to a three-day hybrid starting next year. Hold everything else constant (same headcount plan, same markets, same budget) and change only the work model. Run both consequence trees three orders deep across: - Talent (hiring reach, attrition, seniority mix) - Cost (real estate, relocation, compensation bands) - Culture and delivery (collaboration, onboarding speed, decision latency) - Competitive positioning (how each choice reads to candidates against rivals) Diff the two trees: tint the consequences unique to each path, collapse the ones they share, and tell me where the two futures converge. End with the one branch where the two decisions diverge most sharply.

  2. 2

    Give Cascade context

    Give Cascade the shared baseline both scenarios sit on and the single lever that differs between them. The tighter you pin the common ground (same team, same timeframe, same constraints) the more the diff isolates the effect of the one thing you are actually deciding, rather than surfacing differences that come from sloppy framing.

    The most useful add is the decision criteria you already care about. If Cascade knows the three or four outcomes that will actually settle the call, it can foreground exactly those branches in the diff instead of giving equal weight to every downstream effect.

    Required context
    The shared question both scenarios answer, framed once so each tree runs against the same root
    The single variable that differs between the two futures, named explicitly
    Optional context
    A shared baseline, the headcount, budget, and market assumptions both scenarios hold constant
    Your decision criteria, the outcomes that will settle the call, so the diff foregrounds them
    A depth setting, push the branch where the two paths diverge most to fourth order
    baseline-assumptions.xlsxscenario-a-notes.mdscenario-b-notes.md
  3. 3

    What Cascade creates

    Cascade streams both trees as they build, then overlays them. The output is a paired consequence view: branches unique to each future are tinted to their scenario, branches both futures share are deduplicated into a single neutral node, and the point where the two paths converge is called out. You read the decision as one diff instead of two analyses.

    From Cascade: Compared fully-remote vs three-day-hybrid against one question, three orders deep, diffed branch by branch.

    2Scenario trees overlaid
    31Branches unique to one path
    17Branches both futures share
    Branches unique to one path31 branches
    DomainOnly underUnique consequenceOrder
    CostHybridRe-leasing office space reopens a fixed cost the remote path never carries2nd
    TalentRemoteHiring reach stays national; hybrid quietly narrows the funnel to commuter distance2nd
    DeliveryHybridOnboarding speed rises, but decision latency climbs on the two remote days3rd
    Where the futures converge17 shared
    Shared consequenceOrderWhy it lands either way
    Senior attrition rises in the first two quarters regardless of model2ndDriven by the comp review both scenarios run, not by the work model
    Competitors read either choice as a signal and adjust their own postings3rdThe market reacts to the change itself, not to which direction it went

    "The trees agree more than they differ: 17 shared branches, and the senior-attrition risk lands either way. The real fork is cost versus hiring reach. Want me to push that diverging branch to fourth order, or re-run the hybrid tree at a four-day split to see how much the diff moves?"

  4. 4

    Follow-up prompts

    Push the diverging branch deeper

    The diff already names where the two futures split hardest. Take that one branch to fourth order on both trees so the decision-critical fork is mapped further than the rest.

    Expand the cost-versus-hiring-reach branch to fourth-order effects on both scenario trees. Keep the two paths side by side so I can see where the deeper consequences still diverge and where they rejoin.

    Re-run one scenario with a shifted assumption

    Compare mode is only as good as the two inputs. Change one variable in a single scenario and re-diff, so you can see how sensitive the fork is to the assumption underneath it.

    Re-run the hybrid scenario as a four-day in-office week instead of three, hold the remote tree fixed, and diff again. Show which unique branches grow, shrink, or flip when the in-office days increase.

    Focus the diff on the deciding criteria

    A full diff weighs every branch equally. Tell Cascade which outcomes settle the call and it re-ranks the comparison so the branches that matter to the decision sit at the top.

    Re-rank the diff around our three deciding criteria: 18-month cost, senior retention, and hiring reach. Collapse the branches that do not touch those outcomes and lead with the ones that do.

  5. 5

    Tips and troubleshooting

    Change one variable, not three

    A clean diff needs a clean fork. If the two scenarios differ in work model and headcount and budget all at once, the unique branches blur together and you can not tell which difference caused what. Hold everything constant except the single lever you are actually deciding on.

    Read the shared branches, not just the tinted ones

    The consequences both futures share are the ones your decision can not change, and they are easy to skip past. Knowing a risk lands either way is often more useful than the unique branches, because it tells you where the choice simply does not matter.

    Let the convergence point frame the trade-off

    Where the two trees rejoin is where the decision stops mattering. Reading the diff from the divergence out to the convergence tells you exactly how far apart the two futures really are, and how much of the debate is noise.

  6. Ready to try it yourself?

    Frame both options against one question, and read the real trade-off as a single diff instead of two competing analyses.

    Run Cascade