slug: scenario-planning view: skill layer: block agent: architect companion: corpus/storymakers/frameworks/block/scenario-planning.md

Scenario planning — operational reference

Compact reference for an LLM running a scenario-planning exercise. The showcase version (scenario-planning.md) is the human-facing explanation. This file is the chuleta: rules, recipe, anti-patterns.

Use this when

Time horizon is ≥12 months.
There are ≥2 high-impact uncertainties that can't be expressed as a probability distribution.
The two top uncertainties are roughly independent (move on their own calendars).
The team commits to revisit every 12–18 months.

Use something else when

If…	Then use…
Horizon < 12 months	Sensitivity analysis or decision tree.
You have a probability distribution	Monte Carlo simulation.
One dominant uncertainty, others conditional	Driver-tree / cascade scenarios.
Single binary fork (yes/no on a regulation)	Decision tree, two branches.
Team won't revisit	Don't run scenarios — write a memo.

Decision tree

Q1. Is there ≥1 high-impact uncertainty?
    NO  → forecast / sensitivity. STOP.
    YES → Q2

Q2. Horizon ≥12 months?
    NO  → decision tree. STOP.
    YES → Q3

Q3. Probability distribution available?
    YES → Monte Carlo. STOP.
    NO  → Q4

Q4. Top two uncertainties independent?
    NO  → cascade scenarios. STOP.
    YES → Q5

Q5. Will the team revisit in 12–18 months?
    NO  → memo. STOP.
    YES → run 2×2 scenario planning.

Recipe (10 steps)

Focal question — one sentence.
Brain-dump driving forces — 30–50 macro factors.
Rank by impact × uncertainty.
- High impact × high certainty → assumption (bake in).
- High uncertainty × low impact → noise (drop).
- High impact × high uncertainty → keep (5–8 candidates).
Pick the two most decisive. Cross axes. Four quadrants.
Write each quadrant as a one-page story. Beginning / middle / end.
Name each scenario with a world-shape word — not a probability word. ("Bloom" not "best case".)
List 3–5 measurable early signals per scenario with thresholds.
Stress-test the strategy against each scenario. Mark moves:
- Wins everywhere → no-regret, do now.
- Wins in some → sequence, hedge, stage-gate.
- Loses somewhere → drop, defer, or buy an option.
Document the wind-vane. Which signals would trigger which contingency, at what threshold.
Calendar the revisit. Specific date 12–18 months out.

Headline language (verbatim bank)

"In every scenario, [move X] is value-accretive — start it now."
"In the [scenario name] world, [strategy Y] becomes the dominant call. Here's the early signal we're tracking toward it."
"The current plan implicitly bets on [scenario A]. Here's what we lose if [scenario B] plays out."
"By [date], if [signal X > threshold], we shift to [contingency]."

Pre-mortem checklist (run before shipping)

Each scenario has a name describing the world, not the mood.
Each scenario is internally coherent (no act-2 contradictions).
No scenario is impossible — a senior outsider couldn't say "this could not happen".
The four scenarios are plausibly the future, not Goldilocks-spaced.
Each scenario has 3–5 measurable signals with thresholds.
Strategy section names no-regret moves explicitly.
A revisit date is on the calendar — not "as needed".

Anti-patterns (reject on sight)

Pattern	Why it's wrong
"Best / base / worst case"	Three points on the same axis, not four scenarios.
"Black swan scenario"	Taleb's term means unknowable; modelling it is a contradiction.
Probability-weighted average of scenarios	Defeats the point — collapses to expected value.
Scenarios named "Scenario A / B / C"	Lazy. The name should describe the world.
One scenario per business unit	That's a budget, not scenarios. Scenarios are about the world.
Quant-heavy, narrative-thin	A 50-row Excel ≠ a scenario. The story is what survives the room.

Related canon (call these tools when relevant)

scenario-analysis — the slide artefact (4-quadrant grid).
planning-fallacy — the bias to counter.
three-horizons — sequence the moves identified.
monte-carlo-simulation — when probability distributions are available.