Free interactive mental model

Send Out Tendrils

Problem solving, creating solutions, learning a new domain — think of it all as a path you can't see yet. You don't reason your way to the end from the start. You send out tendrils. Some touch something. Most touch nothing. You strengthen the ones with promise, and probe again from the new frontier — like a slime mold. There is no failure in this picture. Only map.

Tendrils sent

Cheap probes, run in parallel

Dead ends mapped

Things you now know for sure

Promising leads

Your current frontier

Failures

Not a typo. Read on.

Press Send tendrils. Each round probes a few directions from your most promising positions — exactly the move you make when you don't yet know the way.

The method

One loop, five moves

The simulation above is running a loop you already know: it's the scientific method wearing a biology costume. When the way forward is unclear, you don't need a better plan — you need a faster, cheaper version of this loop.

Probe

Send out several small, cheap tendrils at once. Not one big bet — many little ones. Each is an honest guess about where the path might be.

Science: form hypotheses

Sense

Where did they land? Some touch signal, most touch nothing. Both results are data. A tendril that finds nothing has still answered a question.

Science: run the experiment

Reinforce

Feed the tendrils that found something. Thicken them gradually — more attention, more resources — instead of going all-in on the first flicker.

Science: weight the evidence

Prune

Let the quiet ones go, without ceremony and without shame. They weren't failures — they drew the map of where not to search.

Science: discard falsified hypotheses

Advance

Your frontier has moved. Probe again from the new position. Every loop starts from more knowledge than the last one did.

Science: new hypotheses, better priors

The mindset

Not fail fast. Learn fast.

"Fail fast" was always a clumsy phrase — it smuggles the word failure into a process that doesn't contain any. A tendril that finds nothing hasn't failed; it has converted uncertainty into knowledge, which is the entire job. When things don't go as you expected, that's not the process breaking — that's the process working. Now you know better. That was the point.

Language shapes behaviour. Teams that talk about failure start hiding results, defending dead tendrils, and betting big to avoid being wrong. Teams that talk about learning publish their dead ends — so nobody wastes a probe on mapped territory. Try the swap:

"The experiment failed." "We mapped a dead end. Nobody has to wonder about that direction again."

"We wasted two weeks." "We bought two weeks' worth of certainty about where the path isn't."

"That's not what I expected." "That's the most useful result of the day — my model of the world just improved."

"It didn't work." "It answered the question. The answer was no."

"Fail fast." "Learn fast. Small probes, quick reads, honest pruning."

🟡 About your teacher: the slime mold

Physarum polycephalum is a single cell with no brain, no neurons, and no plan. It solves mazes. In a famous 2010 study, researchers placed oat flakes on a dish in the pattern of cities around Tokyo — and the mold grew a network connecting them that closely matched the actual Tokyo rail system, a design that took human engineers decades.

Its entire method: extend everywhere, feel what feeds you, thicken those tubes, withdraw from the rest. No step in that loop is failure. Retraction is how it thinks. If a brainless cell can out-plan a planning committee by probing and reinforcing, the loop deserves a place in how you work.

Mental models

Eight models that describe the same shape

The tendril picture isn't new — it keeps getting rediscovered under different names, in statistics, biology, military strategy, and startups. Recognizing that they're the same shape lets you borrow the sharpest tool from each.

🔬

The scientific method

Hypothesis → experiment → evidence → update. The original tendril loop, and still the gold standard: a negative result is a publishable contribution, not an embarrassment.

The root model

⚖️

Explore vs. exploit

Early on, send wide tendrils — you know too little to commit. As signal accumulates, shift weight toward strengthening the best path. Most mistakes are running the wrong mode for the moment.

Decision theory

🧪

Safe-to-fail probes

In complex terrain you cannot analyze your way to the answer — cause and effect only become visible afterwards. So probe, sense, respond. Design each probe so that a "no" is affordable.

Cynefin — Snowden

💸

Affordable loss

Size each tendril by asking "what would I happily pay for this information?" — not "what's the expected return?" You're buying knowledge; make sure the price is one you'd pay again.

Effectuation — Sarasvathy

⛰️

Gradient ascent

Follow local signal uphill — but beware local peaks that feel like progress and then go quiet. The cure is the occasional long-shot tendril: a deliberately weird probe far from your frontier.

Optimization

🌱

Growth mindset

Swap "I failed" for "not yet." The point isn't comfort — it's accuracy. Ability genuinely grows through the probing itself, so a dead end literally leaves you more capable than before it.

Dweck

🔁

The OODA loop

Observe, orient, decide, act — then again. The insight isn't the steps, it's the tempo: whoever cycles the loop faster finds the path first, even with worse individual guesses.

Boyd

🔀

Pivot or persevere

Put a recurring moment in the calendar to ask, per tendril: strengthen it, or redirect the energy elsewhere? Scheduled honesty beats waiting for the sunk-cost fog to clear on its own.

Lean Startup — Ries

Best practices

How to send good tendrils

Make probes cheap. A phone call, a landing page, a one-day spike, a sketch — if a tendril costs a quarter of your budget, it's not a tendril, it's a bet.

Probe in parallel, not in sequence. Three tendrils at once triangulate; three in a row just take three times as long to teach you the same thing.

Define "signal" before you probe. Decide in advance what a promising result looks like — otherwise every result will look promising in hindsight.

Timebox every tendril. A probe without an end date quietly becomes a commitment. Set the retraction point when you set the probe.

Write down what dead ends taught you. The map is the asset. An unrecorded dead end will be re-probed by someone else — probably you, in six months.

Strengthen gradually. Double down in increments as signal accumulates. The slime mold thickens tubes; it doesn't teleport its whole body to the first oat flake.

Send the occasional weird one. Reserve a small fraction of probes for long shots far from your frontier — they're the escape hatch from local peaks.

Set kill criteria in advance. "We stop if X isn't true by Friday" is easy to say before you're attached, and nearly impossible after.

Where it applies

The same loop, six territories

🐛

Debugging

Each hypothesis you rule out — "it's not the cache, it's not the config" — narrows the search space. The bug isn't found by the last test; it's cornered by all of them together.

📚

Learning a new domain

Sample widely at first: a course, a small project, a conversation with a practitioner. Go deep where energy and signal coincide. Quitting a course that taught you it's not your path is a result, not a lapse.

🧭

Product discovery

Interviews, mock landing pages, concierge prototypes — tendrils you send before writing production code. Every "meh" from a customer is a feature you didn't spend a quarter building.

💼

Career moves

Side projects, informational interviews, a short engagement in a new field. You don't decide your next chapter at a whiteboard; you probe your way into it and thicken what feeds you.

🏢

Strategy & innovation

A portfolio of small pilots beats one flagship transformation. Fund tendrils, review signal quarterly, reinforce the two that pull — and celebrate the team that retired theirs early.

🚨

Incident response

Under pressure, run parallel lines of investigation rather than arguing over the single best theory. Cheap checks first; strengthen whichever line the evidence feeds.

Go deeper