↩ ExploRational · A Catalogue of Divergent Rationality
explorational · a divergent move
The answer that would not come at the desk arrives in the shower. Setting a problem down is not resting from the search — under one disclosed assumption, it is part of the search. When does a break out-solve the hours it replaces?
Wallas gave the break a job title in 1926: The Art of Thought names four stages — preparation, incubation, illumination, verification — and the second is a formal stage of thought in which, to all appearances, no thinking happens. Wallas books the walk on the solution's timesheet, between the preparation that loads the problem and the verification that audits what the shower hands back.
The modern mechanism story is harder on effort. Continued work keeps you circling one approach: each recent failed attempt primes more like it, and the longer you push, the narrower the corridor. A pause lets that fixation decay — the post-break search starts wide again, and the credit the shower takes belongs to the decay. One discipline separates move from surrender: the walk has a return date. Setting-down without one is giving-up.
the signature instrument
Twenty-four approaches on a wheel; one (★) is the answer; you arrive circling a hunch (▲). Two searchers get the same 30 attempts: the grind spends them straight through, the walk stops after 15, pauses, and returns. Every number comes from 400 seeded runs per schedule, paired seed for seed — identical until the pause, isolating exactly what the pause changes.
The rut and the walk
—
the amber dial is the bet: ≤ 3 steps lands inside your primed neighbourhood (σ = 2) · ≥ 6 lands outside it
where the attempts landed — all runs
bar length ∝ that schedule's share of attempts on the approach · the rust tower over ▲ is the rut
φ — the pull of the rut, over time
deterministic for an unsolved run — fixation follows the schedule, not the luck · shaded band = the pause · the cliff is φ × e−λ·pause
solved — equal attempts, unequal ruts
the grind
—
—
the walk
—
—
—
| grind | walk |
|---|
—
when the answer arrived — wall clock, pause included
an attempt costs 1 tick; the pause costs its length in ticks and holds no attempts — the hatched bins are the drawer, and the silence inside them is the point
circling the same hole
The rut is three disclosed lines of arithmetic. Every attempt leaves a prime — a bump of attraction around the approach just tried. Older primes fade geometrically (× 0.8 per attempt); the framing you arrived with keeps a permanent one, the Einstellung anchor. Proposals mix the wide prior (weight 1 − φ) with the primed neighbourhood (weight φ), and φ grows with every failure — φ ← φ + g·(1 − φ) — digging-in, formalised. The longer you work, the more the next idea resembles the last.
Watch the grind's rust bars tower around the hunch and starve the far side: by attempt twenty the searcher proposes almost entirely from inside its own recent history. Local search is what competence feels like from the inside — which is why you cannot feel the moment it stops being search.
what to try
Press the grind, then the walk. Same 30 attempts, answer 9 steps out: the shipped seed gives the walk 42.8% of 400 runs to the grind's 30.5%, the surplus arriving after the drawer. Reseed — the gap wobbles, the ordering holds.
Drag where the answer lies to 1 or 2 — inside the rut. The grind wins now: fixation is aimed at the right hole, and the pause abandons a hot neighbourhood. The break is a bet about where the answer is; here it loses.
Untick the return date (or press the drawer). The after-the-drawer row reads 0 — exactly: of every run still unsolved at the pause, none ever solves. The pause did not stop working; it stopped being search.
Answer outside, pause length from 12 toward 40: P(solved) barely moves — φ is already near zero — while the median time-to-answer climbs tick for tick. Past the decay, a longer pause is pure delay.
the return date
In Wallas's sequence the pause is bracketed by work: preparation loads the problem before it, verification audits the gift after it. Incubation is a stage, not an exit; the difference between a drawer and a grave is a calendar entry. The instrument enforces the grammar — every solve the pause produces arrives through the attempts after it; remove the return and the contribution is zero, by construction.
The grammar also explains why pause length saturates: the pause works by decaying φ, and exponential decay finishes fast. Once φ is a few percent the returning eye is as fresh as it will get — a pause of 12 banks nearly all the benefit; 40 adds only delay. "I'll come back to it someday" loses to "I'll come back to it Thursday." The return date is most of the move.
model
One correct approach among 24; the searcher starts at φ₀ = 0.55 with a permanent prime on its hunch; proposals mix the uniform prior (weight 1 − φ) with a Gaussian kernel (σ = 2) over hunch and recent attempts (fade 0.8); failure grows φ by g·(1 − φ); a pause of length b multiplies φ by e−λb. Both schedules get exactly 30 attempts — extra clock, never extra tries — and repeats are wasted: the cost of a rut. Every rate is a fraction over 400 paired runs, recomputed on each drag.
The calibration warning: measured incubation effects are modest. The Sio–Ormerod meta-analysis (2009) pools a hundred-odd experiments into small-to-moderate gains — strongest when the pause holds light distraction rather than hard work or pure rest, stronger for divergent problems than tight analytic ones. The instrument turns the mechanism up loud, and shows both regimes side by side. A mechanism illustration, not a promise.
the move ↔ the model
| The move | The model |
|---|---|
| an attempt | one real try at the problem — proposed, tested, usually failed. |
| fixation φ | the rut recent failures dig — the pull of the next attempt toward the last few. |
| the pause | decay time for the rut — φ falls exponentially while the problem sits in the drawer. |
| the wide prior | the fresh eye you return with: proposals drawn from the whole wheel again. |
| equal-effort schedules | the fair comparison — the walk gets no extra attempts; the break is not extra time on task. |
| the return date | what separates incubation from abandonment — the attempts after the pause carry its entire value. |
how this opening fails
risk · endless deferral
The catalogue's named risk — the instrument's P = 0 mode. Incubation without a return date becomes abandonment: the decay happens, the fresh eye is prepared, and no attempt ever collects it — of the runs unsolved at the drawer, exactly none solve. If you cannot name the return date, you have not set the problem down. You have put it away.
risk · the rut may be right
Slide the answer inside the rut: the grind wins. Fixation aimed at the right hole is called focus, and decaying it throws away concentration that was about to pay. Incubation is a bet that the answer lies outside your neighbourhood — not a universal virtue. The bet is usually well placed when you are stuck, since being stuck is evidence the rut is empty; but the instrument shows both regimes, and so should you.
A break is a bet that the answer lies outside your rut — schedule the return, and let the rut decay.
can you use it?
RECOGNITION — Which is deliberate incubation? A: giving up on the problem. B: setting the problem down with a scheduled return — the walk, the night, the drawer until Thursday. C: pushing through by working longer.
B. It is scheduling with a return date, which is what separates setting-down from giving-up. A has no return; C never sets down.
THE NEAREST NEIGHBOR — Negative capability also tolerates the unresolved. What distinguishes incubation from it?
Incubation is an action with a clock; negative capability is a temperament. Incubation deliberately parks a specific problem to let recombination run offstage, then comes back; negative capability is the trained comfort of dwelling in uncertainty at all.
PRODUCTION — You are stuck on a decision you've circled for an hour. Schedule an incubation: name what you'll set down, the unrelated thing you'll do, and exactly when you'll return to it.
A version: 'I'll stop drafting the reorg, take a 40-minute walk with no podcast, and reopen it at 3pm.' Yours works if it has a concrete return time — without it, you've quit, not incubated.