agentleFS
natexai2026natexai2026/2030-skills/skills/analogy-check/SKILL.md

analogy-check

Skill0 starsChanged 4 months agoMITView on GitHub

Discipline any historical or analogical argument by stating what actually transfers, what breaks, and what base rate the precedent implies — before the analogy is allowed to carry a conclusion. Use whenever an argument leans on "X is like Y," a historical precedent, or "this is just the next [printing press / internet / electricity / dot-com]." Triggers on comparisons to past technologies or events, "history shows," "we've seen this before," precedent- based forecasts, and persuasive analogies in writing or strategy. If a claim rests on a comparison doing the reasoning, run it through this first.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Every analogy has points where it fails, and naming them is not optional — it's what separates analysis from rhetoric.

    line 35
  2. Distinctive:

    If you can't name a disanalogy, you haven't examined the analogy — you've adopted it.

    line 41
  3. Distinctive:

    Multiple precedents beat one: If three different transitions all imply the same thing, that's stronger than any single rhyme.

    line 119

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Analogy Check

Analogies are how people reason about the genuinely new — and how they most often fool themselves. "AI is like the internet" can illuminate or mislead depending entirely on which parts you claim transfer. Used well, a precedent gives you a base rate and a mechanism. Used lazily, it smuggles a conclusion by feel: the two things rhyme, so the outcome must too. This skill forces every analogy to declare its terms — what maps, what doesn't, and what the precedent actually predicts — before it's allowed to support anything.

The three questions every analogy must answer

1. What transfers? (the mapping)

Name the specific mechanism claimed to carry across, not a vibe of similarity. "Like electrification" could mean many things — say which:

  • General-purpose technology whose payoff lags because organizations must redesign around it? (Electricity took ~40 years to reach 50% adoption because factories kept steam-era layouts before rebuilding around distributed motors.)
  • Diffusion that outran the institutions meant to govern it? (The printing press spread to hundreds of towns in decades, and information outpaced ecclesiastical and state control — producing decades of upheaval before new norms formed.)
  • Capital capturing productivity gains long before labor did? (UK output per worker rose far faster than real wages for ~50 years — "Engels' Pause" — until bargaining power shifted.) State the one mechanism you're actually borrowing. An analogy that "just feels apt" is borrowing none.

2. What breaks? (the disanalogy — mandatory)

Every analogy has points where it fails, and naming them is not optional — it's what separates analysis from rhetoric. For each precedent, state where the target differs:

  • Different diffusion speed? (AI's marginal-infrastructure cost is far lower than a national electrical grid, so adoption may compress the timeline — or compute concentration could reproduce the same access gaps. Both are live; say which your argument needs.)
  • Different agency/control assumptions? (Every "prior GPT" analogy assumes humans stay in operational control throughout. If the target scenario doesn't, the precedent falls out of scope entirely.)
  • Different mechanism, same surface? (Two eras can produce the same statistic for opposite reasons — and which reason holds changes what intervention works.) If you can't name a disanalogy, you haven't examined the analogy — you've adopted it.

3. What base rate does it imply? (and does it survive it?)

A precedent isn't just a story, it's a data point with an outcome. State the outcome and check whether the argument survives it:

  • How long did the transition actually take, and does your forecast honor that lag or quietly assume it's instant?
  • Did the precedent's access gaps close on their own, or persist for generations? (Electrification is still incomplete a century-plus on, concentrated in the same regions — precedent against "it'll equalize naturally.")
  • Did confident forecasts based on that precedent pan out? Beware the analogies that feel strongest.

The exposure-is-not-outcome trap

A recurring failure worth its own flag: technical-susceptibility forecasts dressed as outcome forecasts. A famous 2013 study rated 47% of US jobs at high automation risk within a decade; the occupations rated highest-risk sometimes grew, and unemployment fell to multi-decade lows. Task-level exposure measures what's technically automatable, not what happens — outcomes depend on bargaining power, retraining, demand, and institutions the exposure metric ignores. When an analogy runs "X% exposed, therefore X% displaced," stop it there. (See also hype-check and protect-the-skill.)

The forecaster's-bias flag

Watch for predictions that always sit "15–25 years out" regardless of when they're made — far enough to feel unfalsifiable, near enough to feel urgent. That clustering is a fact about the psychology of forecasting, not evidence the timeline is right. Discount single-point precedent-based timelines accordingly; prefer conditional ranges (see scenario-branch).

Procedure

  1. Extract the analogy and the conclusion it's being used to support.
  2. Mapping: name the specific mechanism claimed to transfer (not a general resemblance).
  3. Disanalogy: name at least one concrete way the target differs — speed, agency, mechanism, scale.
  4. Base rate: state the precedent's actual outcome and timeline; check whether the argument honors it.
  5. Trap check: is this exposure-dressed-as-outcome? Is it a forever-15-years timeline?
  6. Verdict: does the analogy still support the conclusion, support a weaker one, or collapse? Rewrite the claim to what survives.

Output template

ANALOGY: <"X is like Y"> | USED TO ARGUE: <the conclusion it's carrying>

TRANSFERS: <the specific mechanism claimed to map across>
BREAKS: <at least one concrete disanalogy — speed / agency / mechanism / scale>
BASE RATE IT IMPLIES: <precedent's real outcome + timeline; does the argument honor it?>
TRAP CHECK: <exposure≠outcome? forever-15-years timeline? other>

SURVIVES AS: <the claim the analogy can actually support — often weaker>

Example (compressed)

Analogy: "AI won't cost jobs — the ATM analogy proves it. ATMs increased bank teller employment."

TRANSFERS: the real mechanism is "automation of a task cut the cost per branch,
  so banks opened more branches, so total tellers rose even as tellers-per-branch
  fell." That's a genuine, specific mechanism — credit it.
BREAKS: it requires that cheaper unit-cost expands the market and that the
  remaining human tasks stay valuable. AI targets the cognitive tasks that were
  the teller's *remaining* value, and doesn't obviously trigger branch expansion.
  The demand-expansion step is exactly what may not transfer.
BASE RATE: the ATM transition played out over ~30 years with the human kept as
  relationship/again-value worker. A forecast citing it inherits that lag and
  that "humans keep the valuable residual" condition — both of which the argument
  hides.
TRAP CHECK: this is the inverse of exposure-panic — a single reassuring precedent
  generalized to a different task structure.

SURVIVES AS: "automation *can* raise employment when it expands the market and
  humans keep a valuable residual task" — a conditional, not a proof. Whether
  those conditions hold for AI is the actual question, which the analogy skipped.

Gotchas

  • Analogies illuminate; they don't prove. The best outcome is usually a conditional — "this holds if X" — not a settled conclusion.
  • The disanalogy is mandatory. An analogy presented without its breaking points is rhetoric. Always name at least one.
  • Multiple precedents beat one. If three different transitions all imply the same thing, that's stronger than any single rhyme. If they diverge, that divergence is the finding.
  • Pairs with scenario-branch (precedents seed the branches), hype-check (exposure≠outcome), and grade-the-evidence (analogical reasoning is tier 5 — flag it).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.