Skip to main content
MiroFish
Back to Blog
Forecast reviewApr 20, 20265 min read

How to Review a MiroFish Forecast

A forecast is useful only when the operator checks the question, graph quality, evidence line, confidence, and consequence before acting on the report.

Quick answer

Review the structure before you trust the forecast summary.

A MiroFish forecast should be judged by its scenario boundary, graph completeness, evidence line, and consequence level. If any of those are weak, rerun before acting.

Validation

AI forecast validation means checking the model inputs before judging the answer.

For MiroFish, validation starts with the prompt boundary and graph. The forecast report is the last thing to read, not the first thing to believe.

Checklist

The four-pass forecast review

  1. Pass 1

    Review the question

    A vague question creates polished ambiguity. Confirm the event, audience, time horizon, and decision the forecast should support.

  2. Pass 2

    Review the graph

    Check whether the major actors, pressures, incentives, constraints, and relationships are present before trusting the report.

  3. Pass 3

    Review the evidence line

    Ask which source claims and assumptions carry the forecast. Weak evidence should lower confidence even when the report sounds coherent.

  4. Pass 4

    Review consequence level

    The stricter the downstream consequence, the stricter the human review. High-confidence wording is not the same as high-confidence reality.

Confidence should follow evidence, not tone.

  • Use higher confidence only when the source is specific, actors are visible, and the graph captures the real pressure.
  • Lower confidence when the forecast depends on one thin assumption or one missing actor could change the outcome.
  • Treat high-consequence decisions as requiring narrower reruns, outside evidence, and human review.

Rerun prompts

Use follow-up prompts to test weak assumptions.

Re-run this scenario with emphasis on the weakest assumptions and show which missing actor would most change the report.
Compare this forecast against a version where the primary audience is more skeptical and the time horizon is shorter.
Show what evidence would reverse the forecast and which graph node is carrying the most uncertainty.

Red flags that should lower trust

  • The report answers a broader question than the one you asked.
  • The graph misses the actor with the strongest incentive to respond.
  • The forecast gives confidence without explaining the source of confidence.
  • The recommendation sounds final even though the evidence is thin.

FAQ

Questions about forecast review

Should I trust the forecast summary first?+

No. Review the scenario question, graph, assumptions, and confidence before treating the summary as useful.

What is the most common forecast review mistake?+

The common mistake is accepting polished language while missing a weak scenario boundary, absent actor, or unsupported assumption.

When should I rerun a MiroFish forecast?+

Rerun when the graph misses an important actor, the prompt is too broad, the evidence line is weak, or a high-consequence decision needs a narrower test.

Can MiroFish replace human judgment?+

No. MiroFish makes uncertainty inspectable. Human review is still required for high-stakes policy, financial, legal, reputational, or operational decisions.

Read next

Build better review judgment.

What Is MiroFish?

The product model behind graph, simulation, and forecast reports.

Open guide

How MiroFish Simulates the Future

How source material becomes a graph and multi-agent simulation.

Open guide

Can AI Agents Predict Human Behavior?

A practical boundary for using agents to reason about human reaction.

Open guide