| name | n-order-consequences |
| description | Trace the second-, third-, and nth-order consequences of a decision, change, or plan — the effects of the effects — so thinking doesn't stop at the obvious first-order outcome. Use when the user asks "what are the consequences of X", "think this through", "second-order effects", "n order consequences", "what could this cause downstream", or before committing to a significant decision (technical, product, organizational, personal).
|
| user_invocable | true |
| arguments | [{"name":"description","description":"What to apply the thinking to — a decision, change, plan, or situation, in the user's words. Optionally suffixed with a depth like \"to 4 orders\". If omitted, ask what decision or change to analyze.\n","required":true}] |
N-Order Consequences
First-order thinking answers "what happens if I do X?" and stops. This skill forces the next
question, repeatedly: "and then what?" Each order's consequences become the inputs to the
next, until the chain stops producing insight. Most bad decisions look good at order 1; most of
their cost lives at orders 2 and 3.
Step 1 — Explore the problem space first
Consequences of a half-understood action are fiction. Before tracing anything, actually
understand the subject:
- Investigate what you can yourself. If the decision touches something inspectable — a
codebase, a repo, a config, a document, data — go read it. Map the real system: who calls
what, where the money/data/users actually flow, what constraints already exist. Don't trace
consequences through an imagined system when the real one is available.
- Reconstruct the context: what problem is this action solving, what has been tried before,
what alternatives were on the table, and what does "success" mean to the person deciding?
- Ask when you can't infer. If load-bearing facts are missing — scale, who's affected,
reversibility, timeline, what constraint is binding — ask 2–3 targeted questions (via
AskUserQuestion when options are enumerable, free-form otherwise). Ask only what changes the
analysis; don't interrogate for completeness.
Then pin the frame and say it back:
- The action — what is actually being done (not the hoped-for outcome).
- The system it lands in — who and what it touches: people, code, money, incentives, time.
- Depth — default to 5 orders; honor an explicit "to N orders" in the argument. If the
chains genuinely dry up earlier, say so and stop rather than padding orders with restatements —
a forced 5th order is worse than an honest 3rd.
Step 2 — Trace the orders
Build the chain order by order. For each order:
- Take every consequence from the previous order (order 0 = the action itself).
- For each, ask "and then what?" across these lenses — use the ones that fit, don't force all:
- Behavior — how do people/systems adapt to the new state? Adaptation, not the state
itself, is where second-order effects come from (incentives, workarounds, Goodhart effects).
- Time — what compounds, decays, or only shows up later? (Tech debt interest, trust
erosion, network effects.)
- Second parties — competitors, teammates, users, dependents reacting to the change.
- Feedback loops — does a consequence loop back and amplify or dampen the original action?
Name loops explicitly; they dominate at higher orders.
- Irreversibility — which consequences are one-way doors?
- Prune honestly: keep a branch only if it's plausible and load-bearing. N-order thinking
dies by combinatorial explosion; three sharp chains beat thirty speculative ones. Say what you
pruned and why.
Mark each consequence with a rough likelihood (likely / plausible / speculative) and
valence (good / bad / mixed). Don't only hunt downsides — second-order benefits (learning
effects, optionality, trust built) are equally invisible to first-order thinking.
Step 3 — Present the map
Output, in order:
- The action (one line).
- The chains — each significant chain as an indented cascade, e.g.:
1st: Ship the feature behind a manual toggle
2nd: Support becomes the toggle's gatekeeper → ticket queue grows
3rd: Support hires scale with feature adoption — cost scales linearly with success
2nd: Engineers stop building self-serve config ("support handles it")
3rd: Product develops a structural dependency on human ops
- What first-order thinking would have missed — 2–4 bullets naming the non-obvious effects
the analysis surfaced; this is the payoff, lead the summary with it.
- Dominant loop or risk — the single feedback loop or irreversible consequence that should
most influence the decision.
- Cheap mitigations — if a bad higher-order chain has an inexpensive early counter (a
reversible variant, a tripwire metric, a sunset clause), name it. Don't redesign the plan —
that's the user's call.
Be concrete and falsifiable: "support queue grows because every toggle flip is a ticket" — not
"there may be operational implications". If a chain depends on an assumption, state the
assumption inline so the user can reject the chain by rejecting the assumption.