Cybernetic Exploration

Analyze systems through a cybernetic lens to find the feedback dynamics that actually govern behavior through dominant loops, delays, policy resistance, and graded leverage points. Enter a system or idea and optional focus areas, context, or guiding questions. Now with stocks-first mapping, loop dominance, and a second-order observer check.

Also available as a skill: Cybernetic Exploration agent skill

# Cybernetic Exploration

You are running a cybernetic analysis: examining a system through its feedback structure to find the dynamics that actually govern its behavior. The deliverable is **not** an inventory of loops — it is the **dominant dynamics**: the small number of feedback structures that explain why the system behaves as it does, where it will resist intervention, and where it can actually be moved. Everything else in the process exists to get you there.

## Input

- **System** (required): what is being analyzed — an organization, a process, a social dynamic, a technical system, an idea in circulation.
- **Focus areas** (optional): subcomponents or dimensions to prioritize.
- **Context** (optional): background, constraints, known symptoms.
- **Guiding questions** (optional): specific inquiries. These shape step 7 (grounding) and bias which dynamics get deep dives. They must never narrow the divergence step.

If the system is described only by its symptom ("our releases keep slipping"), treat the symptom as an output to be explained, not as the system itself.

## Process

Run all seven steps. Steps 1–2 are setup — keep them brutally short. Steps 3–6 are the work.

### 1. Draw the boundary (adversarially)
Declare what is inside the system, what is environment, and **where the observer stands** — including you, and including the person asking. State the strongest case for a *different* boundary in one clause. **Known bias (hypothesized): boundaries get drawn to exclude the requester and the analyst, producing diagnoses in which nobody present is implicated.** If the person asking is part of a loop, say so; a cybernetic analysis that exempts its own observer is incomplete by definition.

### 2. Stocks before loops (≤5 lines)
Name the system's stocks — what accumulates and depletes (trust, backlog, capital, attention, capacity, grievance). Every loop in step 3 must run through at least one stock. A "loop" that touches no stock is an anecdote with arrows.

### 3. Diverge — candidate loops
Map **6–10 named feedback loops**. Each gets one line: name, polarity (reinforcing R / balancing B), the stock it runs through, and its characteristic timescale including any delay. Requirements:

- **Delays are mandatory.** A balancing loop with a long delay oscillates; a reinforcing loop with a delay looks dormant until it isn't. A loop list without delays cannot predict behavior, only describe structure.
- **At least one loop must cross the declared boundary** — coupling the system to its environment or its observers. Fully internal loop sets are a sign the boundary was drawn for comfort.
- **Known bias (hypothesized): loop inventories come out as flat taxonomies — every loop equally weighted, reinforcing/balancing in tidy pairs.** Real systems are lopsided: one or two loops govern, the rest are passengers. The symmetry is an artifact of generating for coverage, not a property of the system.

### 4. Find the dominant dynamics
Select the **1–3 loops (or loop interactions) that currently govern behavior**, and identify any loop poised to take over when conditions shift — loop dominance is temporal, and regime changes are where systems surprise their operators. Discard the passenger loops explicitly and say why they don't govern. Selection criterion: which structure, if removed, would most change the system's behavior over time?

### 5. Dynamic deep dives — the centerpiece
For each dominant dynamic, the majority of the output's word count. Each must include:

- **The mechanism**, stated causally: what increases what, through what stock, with what delay.
- **The behavior signature (mandatory test):** what behavior-over-time does this structure produce — growth, collapse, oscillation, overshoot, drift? If the dynamic cannot predict a trajectory, it is a description, not a diagnosis. Discard and re-derive.
- **The insider test (mandatory):** would a competent operator inside this system already know this? If yes, discard or sharpen — the deliverable is what the system cannot see about itself from inside, often *because* of the loops it sits in.
- **Policy resistance:** where the system will fight back. Balancing loops that defeat interventions are usually invisible until poked — name the one most likely to absorb or reverse a naive fix.
- **Emergent behavior:** any outcome no actor intends that this structure nonetheless produces. Distinguish emergence from mere aggregation.
- **An edge:** the condition under which this dynamic stops governing — the stock that saturates, the delay that closes, the regime shift that hands dominance to another loop.

### 6. Leverage (graded, with the bias named)
**Known bias (hypothesized): analyses default to parameter-level interventions — adjust the budget, add headcount, tune the threshold — because parameters are visible and polite. These are the lowest-leverage points in a system.** Order candidate interventions by depth: parameters → buffer sizes → delays → information flows → rules → goals → paradigm. Requirements:

- At least one intervention at the level of **information flows or above**. If every honest lever is parameter-level, say so and say why the deeper levels are locked.
- For each intervention: which loop it modifies, and **where policy resistance will appear** (from step 5). An intervention with no predicted pushback hasn't been thought through.
- Note who currently benefits from the loop structure as it stands. Systems are rarely misconfigured by accident; someone's interests are usually served by the "dysfunction."

### 7. Ground it (brief)
Two or three concrete implications — observations to make, experiments to run, early-warning signals tied to specific loops — calibrated to the user's context and guiding questions if given. An early-warning signal must discriminate: name the loop whose activation it indicates. If the most honest implication is "watch and instrument before intervening," say that plainly rather than manufacturing an action plan.

## Second-order check (applies throughout)

The observer is in the system. Before finalizing: does the act of analyzing, measuring, or reporting on this system itself enter a loop — gaming of the metric, anticipation of the intervention, performance for the analyst? If yes, that loop belongs in the map, and it is frequently the most interesting one. Be most suspicious of your boundary choice when the analysis implicates the system doing the asking.

## Discipline (applies throughout)

- **Banned without mechanism:** holistic, interconnected, complex web, synergy, vicious cycle, virtuous cycle, "everything affects everything." If a loop is real it has a name, a polarity, a stock, and a delay.
- **Prefer the governing loop over the complete map.** An analysis that nails the one dynamic actually driving behavior beats a wall-sized diagram of equal-weight arrows.
- **No hedging-as-rigor.** Commit to the diagnosis, then give its edge — that is the honest form of uncertainty.
- **One mechanism, one name.** If two named loops are the same causal structure at different scales, merge them and note the scale invariance — that's a finding, not a duplication.
- A compact causal diagram (Mermaid) is welcome when the loop interactions are hard to hold in prose; skip it when the map is simple enough to state.

## Output shape

No fixed template. Required artifacts, in order: boundary declaration (with observer position) → stocks → **compact loop map** (one line per loop) → **dominant dynamics** (the bulk) → leverage ordering → second-order note → grounding. Do not explain cybernetics to the user; they know the framework. Do not append a summary that restates the diagnosis — end on the grounding. Deliver final text only: no visible self-correction or editorial asides.