Simulator Model

Simulator Model

Owner: Bill Eisenhauer Last Updated: August 6, 2026 Status: ✅ Implemented, illustrative Contributors: Bill Eisenhauer

---

The simulator is deliberately small, deterministic, and inspectable. Its constants teach relationships; they are not empirical benchmarks.

State model

Every stage contains:

• queue size;
• sustainable weekly capacity;
• first-pass yield;
• human-attention cost;
• telemetry coverage; and
• age, utilization, and failure observations.

The model estimates the governing constraint using yield-adjusted capacity:

effective capacity = sustainable capacity × first-pass yield

Its perceived constraint is calculated separately from visible activity, queue pressure, quality loss, telemetry coverage, and bounded deterministic noise. This allows a highly visible stage to dominate the narrative while a poorly instrumented stage governs outcomes.

Six-week cycle

flowchart LR
    A["Current state + seed"] --> B["Apply selected interventions"]
    B --> C["Run six weekly flow steps"]
    C --> D["Count accepted outcomes"]
    D --> E["Recalculate constraint"]
    E --> F["Return metrics + learning receipt"]

Loading

In prose: a cycle clones the current state, applies the intervention bundle in selection order, simulates six weeks with a deterministic seed, counts accepted outcomes, recalculates the constraint, and returns a receipt.

Easy-mode guidance

Easy mode does not use a permanent answer key. It evaluates marginal next-cycle impact against the current selection.

For an unselected card:

marginal effect = result(current bundle + candidate) − result(current bundle)

For a selected card:

marginal contribution = result(current bundle) − result(bundle without candidate)

flowchart TD
    A["Current bundle"] --> B["Run comparison cycle"]
    A --> C["Add candidate or retain selected card"]
    C --> D["Run candidate cycle with same seed"]
    B --> E["Compare accepted outcomes"]
    D --> E
    E -->|Higher| F["Helpful"]
    E -->|Lower| G["Harmful"]
    E -->|Flat| H{"Other modeled effect?"}
    H -->|More telemetry| I["Learning value"]
    H -->|Lower or higher burden| J["Helpful or harmful"]
    H -->|No material change| K["Neutral"]

Loading

In prose: both plans run from the same state and seed. Accepted outcomes are compared first. If they are flat, telemetry, WIP, and human attention determine whether the result represents learning value, reduced or increased burden, or no material change.

Because the entire selected bundle is part of the comparison, one card can switch classification after another is chosen. This is intentional. The simulator is teaching interaction effects and constraint movement, not intervention memorization.

Intervention mechanisms

Intervention	Modeled mechanism	Primary modeled tradeoff
Instrument handoffs	Raise coverage and diagnosis quality	Consumes points without physical capacity
Improve context readiness	Improve yield across context, build, and verify	More shaping before implementation
Automate verification	Add verification capacity, yield, and evidence	Exposes failures and consumes setup effort
Increase agent concurrency	Add build capacity	Weak context increases rework and review pressure
Build shared foundation	Remove repeated build/release work	Creates migration inventory
Fund adoption	Add migration, documentation, and ownership capacity	Redirects product-team attention
Add partner capacity	Add implementation capacity	Consumes shaping, review, integration, and handoff capacity
Strengthen SOW and handoff	Improve context and ownership	Adds work before partner execution
Control release of work	Reduce arrivals and protect constrained stages	Fewer starts can feel slower
Add human judgment	Add scarce shaping and verification capacity	Expensive and not a permanent substitute for automation

Model limits

The simulator omits organization-specific data, money, team performance, dependency topology, skills, morale, incident risk, architecture detail, and strategic option value. A real deployment should replace constants with event-linked evidence and subject every causal claim to review.

The most important caveat is simple: Easy mode reveals this model's answer, not reality's answer.

---

Related: Kitting and Wait States · Facilitator Guide · Simulator source