The field guide
How to hunt well.
Flint works out of the box — press Strike and it does the rest. But the grader is honest, and most first strikes land tame. Hunters who know what a tame is actually telling them — when to flip the register, when to fuse, when to reconfigure — get more idea per credit. This is that knowledge.
Pick the right door.
All four doors end the same way — a graded idea — but they start from different truths about you.
Spark from scratch
You bring nothing; the engine brings the collision. Best when you want to be surprised out of your own patterns.
Build on an asset
Name one thing you actually have — a skill, a space, a stockpile, a network. The engine finds the move you wouldn't make on it.
Start from a pain
Name who hurts and when. The engine throws resources against the pain until a business forms around it.
Sharpen a business idea
Paste the idea. Flint grades it cold, shows the dimensions it truly stands on, then hands you three hunt moves.
If you're torn between doors, prefer the one where you can be most concrete. A vivid asset beats a vague pain; a vivid pain beats starting cold.
Expect the first grade to be honest.
Flint's grader is walled off from the part that writes the reading — it can't be charmed. The first read of any collision is always the straight one, graded cold, and most first strikes land tame. That isn't a miss; it's the baseline the hunt is measured against.
So every good session has the same shape: strike, take the honest grade, then make one deliberate move and let the grader re-judge. A flicker that survives that loop means something. A flicker handed to you on the first press would mean nothing.
The shape of every hunt: an honest grade first, then one deliberate move at a time until the grader concedes a flicker.
The demo on the homepage plays this exact arc on real engine output — an honest tame first, then the one move that lifts it. The walkthrough below it narrates a full strike stage by stage.
Read the altitudes deliberately.
Every strike lands at three heights, and they're for different moments:
The category the collision opens. Read here when you're mapping territory — what kind of business is this?
The wedge — a specific offer for a specific buyer. The default read: concrete enough to judge, wide enough to grow. When unsure, read here.
The niche — narrow, vivid, immediately buildable. Read here when you want something you could start next month.
One strike, three frames: the identical collision read as a category, a wedge, or a niche.
Each altitude carries three small controls: ↻ regenerates that level and the levels nested inside it, ⤓ drills in — the level becomes the new Broad, with sharper angles nested under it — and + lets you phrase the level in your own words.
The pain door runs the same geometry in reverse: you hold the pain, and the resource Flint throws against it arrives at the same three altitudes.
You don't need all three. Read one altitude, and only read another if the first one surprised you enough to want a different frame on it.
Choose the register.
Before Flint reads a collision, you choose how it should read it — and the choice matters more than any other single control:
The plain, obvious business those coordinates describe. Honest and buildable — and usually tame, because the mainstream got there first.
The surprising, non-obvious reading of the same coordinates — the bisociative spark worth a second look.
When a straight read lands tame, re-reading it Inventive is the first rescue to reach for: same collision, different eyes. Every twist carries its own register toggle too, so a fused resource or a structure move can be read either way.
The first Spark journey is exactly this move — a tame straight read of the same pairing flips to a flicker when re-read Inventive.
Know your evolve moves.
After a read, the controls under the card each move the idea in one direction — and the walled-off grader re-judges every result. This is where the hunt is won:
Fuse a second resource
The classic rescue for a tame: strike something unlike into the existing idea. Removing the twist you just applied is a free revert, so a fuse costs nothing to undo.
Apply a business-model move
Same ingredients, different architecture — who pays, who owns, when money moves. Often the gap between tame and flicker is structural, not material.
Re-grade in another market
Edge is local. Name a geography and the grader re-judges: an idea that's tame worldwide can flicker somewhere specific.
Plain words
Every reading can be re-said simply, in a panel over the untouched card. If the idea stops making sense in plain words, that tells you something too.
Edit the reading
Rewrite the idea in your own words and Flint re-grades exactly what you wrote — the verdict then belongs to your phrasing, and the card says so.
Hunt from a graded idea.
The fourth door works backwards from an idea you already hold. Flint grades it cold, shows the load-bearing dimensions it stands on, then offers three moves — each different in kind:
Reads the same idea again for a different, more surprising business — a fresh angle, not a new idea. It holds the anchor and excludes what you've already seen. Honest by design: a re-read informs; it rarely rescues.
Turns a load-bearing dimension. Flint holds the resource it found and collides different moves against it — a pain-anchored idea holds the pain and throws resources instead. This changes the structure, and structure is where grades move.
Keeps every dimension your idea stands on and adds exactly one — a fused second resource or a business-model move — then re-reads and re-grades.
Re-reads inform; restructures rescue. If two honest readings agree your idea is tame, stop re-reading — turn a dimension, or add one.
Treat verdicts like information, not applause.
Flicker
Real, untaken edge. This is a lead, not a finish line — your next move is outside Flint: talk to whoever the reading says bleeds.
Tame
The idea is sound; the surprise is spent. Don't abandon it — work it in order of cost: re-read it Inventive first, then make one structural move — a fuse or a business-model move — and let the grader re-judge.
Dud
Whoever feels this pain is already served. Duds are the cheapest verdicts you'll ever get — a dollar to learn what would otherwise cost you a year. Strike again.
The grade measures surprise and untaken edge — not profit. Plenty of excellent livelihoods are tame businesses. Flint's job is to tell you which is which before you commit.
Phrase your own inputs like a hunter.
When you bring your own asset, pain, or idea, concreteness is everything. The engine strikes best against something with edges.
my logistics background nine years routing refrigerated trucks
healthcare is broken new parents can't reach a paediatrician after 6pm
an app for local services a booking service where retired electricians inspect houses before purchase
The pain door in particular accepts only a stranded pain — one rooted in who (an excluded group), where (a place where normal service fails), or what (a hard blocker) — not a frustration everyone shares.
Bring one resource, one pain, or one idea per run — not a bundle. The engine's job is to add the unlike thing; leave it room to.
Spend credits like they're ammunition.
Every generative press spends in proportion to the work it does, and your balance is always on screen. A typical hunt — strike, verdict, and the moves that evolve it — runs about a dollar. Three habits keep it there:
Read one altitude before you touch anything else. When a tame appears, flip the register before you spend on a twist. And change one thing per move — the grader can only tell you what lifted the grade if you moved one lever. The controls reference lists every press and whether it spends.
Enough theory.
The demo on the homepage plays a full hunt free. Your own hunt starts at a dollar.
Start for $1