semanticquant · collection

The Armory

MatheMetaphor catalogues mechanisms; ExploRational's distant-analogy move needs somewhere to borrow them from. This page treats the first as the stockroom of the second and states the audit that separates a loan from a decoration. Colour marks the destination site.

the protocol

The steps between a hunch and a loan.

Run them in order on one problem at a time. Steps one and two choose the source; three and four price it; five decides whether you keep it.

State the structure with the domain words removed.

Write your problem in one sentence containing no noun from its own field. "Our hiring pipeline stalls at the final round" becomes a queue whose service rate falls as the queue lengthens. The stripped sentence is the only thing you can shop with, because it names what a mechanism can match: states, rates, thresholds, feedback, irreversibility. If you cannot write it, you do not yet know your problem's structure — and every metaphor on the shelf will seem to fit, which is the symptom.

Shop the shelf by structure, never by vibe.

Match the stripped sentence against a page's mechanism rather than its mood. A rotting codebase feels like decay, which recommends entropy; it is structurally irreversible accumulation without recombination, which recommends Muller's ratchet — a page filed under evolution that mentions no software and predicts more about it. Scan at least three of the catalogue's clusters. A source drawn from the cluster your problem already lives in will refine what you think, and refinement is what you came here to get past.

Build the mapping table element by element.

Every load-bearing part of the source gets a row and a specific referent in your case — not "the population is the team" but "the least-loaded mutational class is the two files nobody has had to patch this year." Parts with no referent stay in the table, marked empty; they are where the loan is void, and step four needs them. A source whose parameters you cannot state in your own units — what is your barrier height, in what does your noise arrive, what is your utilization — has been invoked rather than mapped, and the table is how you tell the difference.

Hunt the disanalogy before you lean on it. The tear-line gallery below shows what the hunt finds.

Every explainer on the shelf closes with a section headed Where the metaphor tears, written by whoever built the instrument. Read it before you touch the sliders. It lists the assumptions the mechanism cannot survive without, and your case either satisfies them or it does not: Kramers escape assumes a fixed landscape, so the moment you rearrange your kitchen to make the habit easier you have left the model. A borrowing whose tear you cannot state in one sentence has not been audited, and an unaudited metaphor keeps generating fluent sentences well past the point where it means anything.

Cash out one prediction or one reframing — or return it.

The loan pays in exactly two currencies. A prediction: "drop one project, get utilization under 0.85, and the wait on the rest should fall by more than the project cost" — checkable, with a date. Or a reframing: "this is a coevolutionary race, so the exit is changing what the audience scores, not arguing better" — a different action from the one you were about to take. Name which you got, in a sentence you could not have written before opening the page. If neither arrives, put the metaphor back on the shelf; the pleasure it gave you was genuine, and it was the entire product.

eight worked borrowings

Problem, source, and the seam where it fails.

Each row is steps three and four compressed to two sentences. The mapping table and the cash-out are yours to write; these are the loans, already priced.

A career decision under uncertainty⤢ Explore–exploit tradeoff
maps
Each option is an arm of unknown payoff, the remaining horizon sets how much sampling is rational, and regret is scored against the arm you never pulled — which is why the identical choice is correct at 25 and wrong at 55.
tears
Bandit arms are stationary and independent; a career's compound, so five years of exploiting one raises its payoff and decays every other, and the door you skipped is not the door still standing open.
A team that keeps relitigating⤢ Hysteresis
maps
The state depends on the path taken to reach it, and the threshold to reopen a settled question must sit well above the threshold that closed it; with no gap between the two the system chatters, flipping on evidence too small to act on.
tears
In a magnet the loop width is a property of the material and no one chooses it; a team's is set by whoever writes the rule, so the borrowing diagnoses the chatter and prescribes a deadband while saying nothing about why yours is narrow.
An argument that escalates⤢ The Red Queen
maps
Each escalation is a rational reply to the last one, both sides keep investing, and relative position stays flat while absolute cost climbs — all the running you can do to hold the place you had.
tears
Coevolution has no exit and no audience; an argument has both, so unilateral withdrawal is available to you at a status cost that a parasite could never pay, and a third party can rewrite what counts as winning.
A codebase rotting⤢ Muller's ratchet
maps
Each workaround is a deleterious mutation, the ratchet clicks the moment the last clean module is edited, and with no recombination there is no route back to the least-loaded state; few maintainers turn it faster, exactly as small populations do.
tears
Mutations are random and selection is weak, which is what makes the biological ratchet one-way; your damage was authored and is under version control, so a deliberate rewrite restores the clean class — a move the model has no term for.
A habit that won't move⤢ Kramers escape time
maps
Escape time from a well rises exponentially with barrier height over noise, so shaving the barrier beats redoubling the effort by a wide margin, and the long flat wait before a change "suddenly" happens is a mean rather than a verdict on your character.
tears
Kramers assumes a landscape fixed from outside and noise with no direction; a person can lower their own barrier by moving the biscuits, and the instant you edit the terrain the escape-time formula stops describing you.
A market of excuses⤢ Adverse selection
maps
The excuse-giver knows whether the constraint is real and you cannot check, so accepting every excuse at face value drags the pooled credence down to the average and prices the genuine constraint like the manufactured one.
tears
Akerlof's good sellers walk away and the market unravels; your colleague with the real constraint cannot leave the team, so you get a stable discounted pool instead of a collapse — the model's headline prediction is the part that never arrives.
Attention under load⤢ Queueing & slack
maps
Expected wait scales as ρ/(1−ρ) — four service times at 80% utilization, nineteen at 95% — so fifteen points of extra load costs nearly five times the delay, slack is what buys latency, and variance in arrivals hurts more than the mean does.
tears
The textbook queue has a fixed service rate and loses nothing; attention drops items without telling you and its service rate falls as the backlog grows, so the curve understates your collapse and may have no steady state to compute.
A friendship drifting⤢ Limit cycles & coupled oscillators
maps
Two rhythms with slightly different natural frequencies stay phase-locked only while coupling exceeds the detuning; let contact fall below that threshold and they separate at a rate set by the mismatch, so the loss arrives as a threshold crossing rather than a slope.
tears
Coupled oscillators carry no memory, so relocking costs nothing the moment coupling returns; a drifting friendship accumulates unshared context, which pushes the coupling needed to relock above the coupling that would have held the lock.
An apartment hunt with a deadline⤢ Optimal stopping
maps
Offers arrive in sequence, each mostly gone if you pass; the rule says spend an opening fraction of the search calibrating, then take the first option that beats everything seen.
tears
Listings return, agents negotiate, and your taste moves as you look — the theorem's no-recall, fixed-preference world is cleaner than any housing market.
Sizing a bet on yourself⤢ Kelly criterion
maps
Edge and odds set the fraction to stake; the log keeps you solvent through variance, and overbetting a true edge still ruins you.
tears
Life bets correlate — the startup, the city, and the partner move together — and your edge is exactly the number you cannot know when it matters.
A practice spreading through a company⤢ Percolation
maps
Adoption spreads along contacts; below a connectivity threshold it stays in pockets, above it one giant cluster takes the floor — which is why the same pitch dies in one org and sweeps another.
tears
People are not lattice sites: adopters recruit, resisters organize, and the network rewires in response to the thing spreading through it.
Two teams that must meet without talking⤢ Schelling points
maps
When bargaining is impossible, salience decides: noon, the main entrance, the default branch. The equilibrium is whatever both sides expect the other to expect.
tears
Salience is cultural. Your obvious is not their obvious, and the focal point you are sure of is the one assumption the silence never tests.

the tear-line gallery

Cautions from the catalogues.

Every MatheMetaphor page ends by naming where its own metaphor fails. A selection of those lines, verbatim — transferable to almost any borrowing you will ever audit.

“The alternative to metrics is often favoritism, not wisdom.” — Goodhart's law

“Cooperation can leave the race.” — Red Queen

“Some systems really are springs.” — Hysteresis

“The arms drift under you.” — Explore–exploit

“The lens can fatalize.” — Coupled oscillators

“One person's lemon is another's perfect match.” — Adverse selection

“A mean is a poor promise to a single life.” — Kramers escape

“A full life is not automatically a jammed one.” — Queueing & slack

“Expectation ignores risk, ruin, and how much a dollar means when you have few of them.” — Martingales

“The metaphor shows you the valley; it hides your hand on the shovel.” — Attractor basins

the challenge

Candidates from three regions, at most one survivor.

Take the problem you are carrying this week — the live one with a decision attached, not the interesting one. Write its stripped sentence, step 01. Then pull three candidate mechanisms from three different clusters of the catalogue: one physical (Tipping Points, Rhythms & Feedback), one social (Living Together, Spread & Inequality), one informational (Signal & Secrecy, Belief & Evidence). Three clusters is the constraint doing the work — three sources from one cluster will agree with each other and teach you nothing they did not already share.

Run steps 03 to 05 on each. Expect most to die at step 04: the audit is a filter with a low pass rate, and a week where all three fail is a week the protocol worked. Keep at most one. Two survivors on the same problem means one of them is decorating, and while both feel apt you have no way to tell which.

Write the survivor's cash-out as a dated prediction or a changed decision, and write the tear beside it. In a month, check the prediction and reread the tear — that pair is the record of whether the borrowing was worth anything.