Skip to content

Instantly share code, notes, and snippets.

@cyberstorm-builder
Last active March 4, 2026 06:02
Show Gist options
  • Select an option

  • Save cyberstorm-builder/59fb26f8221dd879a8fb4ae4e0f9075d to your computer and use it in GitHub Desktop.

Select an option

Save cyberstorm-builder/59fb26f8221dd879a8fb4ae4e0f9075d to your computer and use it in GitHub Desktop.
Semantic Simulation Method - SKILL.md
name description status
semantic-sim-method
Use when you need to simulate futures in a reduced semantic space with rotations, falsification checks, and adversary analysis
experimental

Semantic Simulation Method

Status: Experimental. This method is structurally sound but community-validated, not author-tested end-to-end.

Why This Exists

You're facing a decision with real stakes, incomplete information, and no historical dataset to model from. You need to think about what might happen — but you want something more rigorous than brainstorming and less demanding than quantitative forecasting.

Most scenario planning falls into one of two traps:

  • Too loose: "Let's brainstorm three scenarios" produces narratives that confirm what you already believe
  • Too heavy: Quantitative models require data you don't have and produce false precision

The Semantic Simulation Method sits in between. It gives you a structured process for exploring futures that's rigorous enough to reveal blind spots, but doesn't require probability distributions or historical data.

The Core Idea

In any complex situation, a small number of forces explain most of what will happen. If you can:

  • Identify those forces
  • Systematically ask "what happens when each force wins?"
  • Check whether your plan survives all of those worlds
  • Figure out what you can do to reshape the forces in your favor
  • Then consider what happens when others are doing the same

...you'll understand the possibility space better than most people who spend 10x longer thinking about it.

This borrows intuition from dimensionality reduction in statistics (like PCA). Instead of reducing numeric variables to principal components, you reduce a complex situation to a small set of driving forces and simulate by letting each one dominate.

Concepts

Forces (Semantic Attractors)

A force is a pressure that shapes outcomes. Not something you control, but something that pulls the situation toward a particular configuration. Good forces are:

  • Independent enough from each other to generate meaningfully different scenarios
  • Observable — you can point to real-world signals that show a force strengthening or weakening
  • Not outcomes — "success" isn't a force; "foot traffic" is

Choose 4–6 forces. Fewer than 4 won't capture enough of the picture. More than 6 becomes incoherent.

Dominant-Force Transitions

The key rule: when one force dominates, it reconfigures the others. This is what makes the simulations non-trivial.

You're not just asking "what if foot traffic is high?" You're asking "what does the whole situation look like when foot traffic is the dominant force — and how does that reshape rent costs, competition, and customer loyalty?"

Rotations

After running your initial simulations, you may notice that one force doesn't generate much difference across scenarios — every sim looks about the same along that axis. That's a signal to rotate: replace it with a related but more discriminating concept.

Think of it like adjusting the angle of a camera. The subject is the same, but a better angle reveals details you couldn't see before.

Idempotency

Your plan should be idempotent across simulations — it should work regardless of which force dominates. If your plan only works in one scenario, it's fragile.

Idempotency checking forces you to separate:

  • Core — What's true in every scenario (your real plan)
  • Boundary — What's true in some scenarios (your assumptions)
  • Fragile — What's only true in one scenario (your risks)

Three Layers of Simulation

The method builds up in layers, each adding realism:

  • Passive landscape — You observe. Forces play out. You check if your plan survives. ("What world am I walking into?")
  • Active agency — You act. Your moves reshape the forces. Boundary conditions become things you can influence. ("What can I do about it?")
  • Live competition — Others act too. Multiple agents with agency reshape forces simultaneously. ("What happens when someone is actively working against me?")

Each layer depends on the one before it.


Interactive Mode

When running this method in conversation (agent + human), do not one-shot the entire analysis. The method exists to prevent confirmation bias — running it end-to-end without human checkpoints is the agent confirming its own assumptions unchecked, which is the exact problem the skill solves.

Mandatory checkpoints (pause and wait for human input before continuing):

Checkpoint After Why
🔴 CP-1 Step 1 (Define Your Space) Wrong forces = wrong everything downstream. Human confirms the 4–6 forces and survival test before sims build on them
🔴 CP-2 Step 3 (Idempotency Check) Human validates sim narratives and core/boundary/fragile classification before rotation
🔴 CP-3 Step 4 (Rotation) Human confirms the rotation choice and its effect before falsification
🔴 CP-4 Step 5 (Falsification) Human validates the devil's-advocate scenario before moving to agency layers

At each checkpoint, present your work so far, then explicitly ask for confirmation, corrections, or additions. Do not proceed until the human responds. If the human says "just run it" or similar, you may batch — but the default is interactive.


The Method

Step 1: Define Your Space

Before simulating, get explicit about three things:

  1. Your objective — What are you trying to achieve? State it concretely. Not "succeed" but "sustain a profitable business for 5+ years."
  2. Your forces (4–6) — The pressures that will shape whether you achieve that objective
  3. Your survival test — What must remain true across all scenarios for your plan to work? This is your idempotency criterion.

🔴 CP-1: Pause here. Present the objective, forces, and survival test. Confirm with the human before proceeding. Wrong forces here means wrong analysis everywhere.

Step 2: Run Base Simulations

Layer 1 — Passive landscape. You're an observer.

Run 3 simulations, each with a different force as the dominant pressure:

Sim Structure
A First force dominates — how does it reshape everything else? Where does your plan land?
B Second force dominates — different configuration, different trajectory
C Third force dominates, or no single force dominates (fragmented/multipolar)

For each simulation, write:

  • Configuration — How the dominant force reshapes the other forces
  • Trajectory — What happens over time (early → mid → late dynamics)
  • Your position — Does your plan achieve the objective in this world?

Step 3: Check Idempotency

Review all 3 simulations against your survival test. Categorize everything:

  • Core — True across all sims (this is your real plan)
  • Boundary — True in some sims, contingent in others (these are assumptions you're making)
  • Fragile — Only true in one sim (these are risks you need to manage or accept)

🔴 CP-2: Pause here. Present the base sims and idempotency classification. Human validates the narratives and core/boundary/fragile findings before rotation.

Step 4: Rotate for Variance

Look at your simulations. Is there a force that doesn't generate much difference?

  • Replace it with a related concept that's more discriminating
  • Re-run the simulation that changes most under the new axis
  • Document why you rotated — this is insight, not just bookkeeping

Usually 1–2 rotations is enough. Stop when each force generates a meaningfully different scenario.

🔴 CP-3: Pause here. Present the rotation choice, rationale, and the re-run result. Confirm before falsification.

Step 5: Falsify (Mandatory)

Still Layer 1 — neutral world, no agency yet.

Write the devil's-advocate simulation — the scenario where your plan fails. This is not optional.

Answer three questions:

  1. What events would cause your plan to fail? Be specific.
  2. What would you do instead? (Your pivot — what you become when Plan A dies)
  3. What early signals would tell you this scenario is materializing? (So you can pivot before it's too late)

If you can't write a convincing falsification, either you don't understand the situation well enough, or your plan is trivially safe (and probably not very ambitious).

🔴 CP-4: Pause here. Present the falsification scenario, pivot boundary, and early signals. Human validates the threat model before moving to agency.

Step 6: Act (Your Agency)

Layer 2 — You're no longer an observer. You're a player.

Go back to your idempotency check from Step 3 and ask:

  • Which boundary conditions can you move to core? — Are there assumptions that you could make true through deliberate action?
  • Which fragile findings can you move to boundary? — Can you reduce your exposure to single-scenario risks by taking specific actions?
  • What moves reshape the forces in your favor? — Not just "what's my strategy" but "how do my actions change the landscape itself?"

For each action you identify, ask:

  • What does it cost? (money, time, attention, optionality)
  • When does it need to happen? (some moves only work early; others are evergreen)
  • Is it reversible? (irreversible moves need higher confidence)

The output of this step is an action plan layered onto your landscape map.

Step 7: Compete (Multiple Live Players)

Layer 3 — Other agents have agency too.

If someone could actively work against your plan — a competitor, a regulator, a market force with its own agenda — model them as a live player:

  • What are their objectives? — They have goals too. What are they optimizing for?
  • What can they capture? — Which leverage points could they control?
  • What friction can they create? — How could they make it harder for you to operate?
  • What's their optimal sequence? — If they were running this method from their perspective, what would their action plan look like?

Then overlay their moves onto yours:

  • Where do your actions conflict?
  • Where do their moves invalidate your Step 6 actions?
  • What are your course-correction triggers?
  • What are your abort criteria?

Synthesis

Compress findings into layers:

Layer What to capture
Robust What's true in every scenario — your real plan
Contingent What depends on assumptions — and which you can act on (Step 6)
Fragile What only works in one world — your risks
Your moves Actions that reshape the landscape in your favor
Competitive exposure Where others' agency threatens your plan
Watch for Early signals from falsification + course-correction triggers

Extensions

Extension A: Variable Interaction Report

After completing the full simulation, compress findings into a structured report mapping how forces interact.

Analyst version — Interaction matrix:

For each force pair, classify the interaction:

  • Amplifies — Force A strengthening makes Force B stronger
  • Dampens — Force A strengthening makes Force B weaker
  • Reshapes — Doesn't change intensity but changes character
  • Decouples — Reduces your sensitivity to Force B
  • Buffers — Gives you more time before Force B becomes critical
  • Conflicts — Forces push toward mutually exclusive configurations

The matrix reveals feedback loops, natural hedges, and critical dependencies.

Executive version: Distill the matrix into 3–5 bullets on what matters for the decision.

Extension B: Quantized Model

Add a lightweight quantization layer to catch distinctions that narratives blur. This is structured intuition with numbers, not a statistical model.

Step 1: Score forces per simulation on a simple ordinal scale (1 = negligible → 5 = dominant).

Step 2: Sensitivity analysis — For each force in each sim, ask: if this force shifted one level (±1), would the outcome flip? Forces where ±1 flips the outcome are your highest-sensitivity variables.

Extension C: Data Augmentation

Data sharpens specific steps:

Step What data sharpens Example
Force selection Are your forces real? Market research, industry analysis
Base sims Are your trajectories plausible? Historical case studies, domain data
Falsification How likely is the failure scenario? Base rates for the trigger events
Act Will your actions work? Evidence from others who tried similar moves

Extension D: Temporal Chaining

Run the method periodically and compare:

  • Did the forces move as expected?
  • Which sim is the actual world converging toward?
  • Have your core/boundary/fragile classifications changed?
  • Do your Step 6 actions still make sense?

The chained simulation becomes a strategic dashboard — not just "what should I do?" but "am I doing it, and is it working?"


Common Mistakes

Mistake Why it hurts Fix
Too many forces (>6) Lose clarity, sims become incoherent Pick 4–6 that are most independent and observable
No rotations Miss variance — a force that doesn't discriminate wastes a slot Replace non-discriminating forces after base sims
No falsification Confirmation bias — you only explored worlds where your plan works Step 5 is mandatory, not optional
Treating outcomes as probabilities The method doesn't produce likelihoods Use core/boundary/fragile, not percentages
Skipping Layer 2 (Agency) You map a landscape but don't act on it — boundary items stay assumptions Step 6 is where assumptions become action items
One-shotting in conversation Agent confirms its own biases unchecked Use Interactive Mode checkpoints

Quick Reference

  1. Define your space — objective, 4–6 forces, survival test
  2. Run 3 base sims — each with a different dominant force
  3. Check idempotency — core / boundary / fragile
  4. Rotate — replace non-discriminating forces, re-run affected sims
  5. Falsify — devil's-advocate scenario, pivot boundary, early signals
  6. Act — move boundary→core, fragile→boundary, reshape forces
  7. Compete — model adversary objectives, captures, friction, sequence
  8. Synthesize — robust / contingent / fragile / your moves / competitive exposure / watch for
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment