The Curiosity Engine™ the model every Panama Red sprint runs on

What do you do with the entire written record of human knowledge?

Now that a model holds all of it, knowing more stops being the edge. Every model knows the same things. The advantage is interpretation. The engine's whole job is the reframe: turning what everyone can already see into something nobody has thought yet.

It starts from a line we keep close: we don't see things as they are; we see them as we are. The conditioned gaze can be de-conditioned. For an AI, the most conditioned view is its first draft, the mean of everything ever written. The engine sets that aside and looks again.

Fig.1 · the reframe

Input · the conditioned gaze

The mean of everything written

Every model's first, safest draft.

Output · truer & more useful

A reframe

An extraordinary thought about an ordinary thing.

You never ask the machine to be improbable, which it can't do. You take away the answer it was about to give, and the next one down is further out by definition. Or you move the world the question gets asked from, and let the machine go on being probable somewhere else.

The six frictions

01–06

Go to the edges

The safest reading sits in the middle of the curve, and a model's first draft is exactly that. Name it, set it aside, then look for the force moving on the category from outside it.

Excavate the meaning, import only the lens

A brand's name, packaging and first advertising usually argue for something bigger than its current claim. Dig that out. Borrow a lens from anywhere to reveal it, never a verdict about what the brand should stand for.

Invert it

Test the opposite of the claim, of the convention, and of your own first reading. Where both poles hold, keep the contradiction live. The sharpest version takes the enemy the category funds itself against and asks what its presence proves: dirt is what a day of play leaves behind, which is where Unilever's Persil and OMO found Dirt is Good in 2004.

Steal across domains

An obvious idea in one field is strange in another. Strip the trade vocabulary out of the problem, match on mechanism rather than on resemblance, and say where the analogy stops holding.

Impose a constraint

Give the problem an arbitrary rule and solve it anyway. Take away the brand's biggest asset. Forbid the category's main verb. Work it with no media budget. Perec wrote a novel without the letter E.

Build a world

A model returns the most probable continuation of the world you put it in. So build a cosmos whose median falls at the edge of ours, then ask it for its paperwork: a job listing, a statute, a museum label. An idea can stay vague, a document can't. The fiction is the excavation site, and what ships is the reading carried back.

Where the facts come from

Desk research

Twelve queries across five layers, in three passes, searched live and never answered from memory. A query that comes back with nothing is recorded as nothing, and nobody invents a percentage to fill the gap.

What every claim carries

The tagging layer

Every claim is tagged sourced, inferred or hypothesized. Sourced ships as a fact, inferred ships as a pattern, a leap ships labelled as a leap. A claim that fits none of the three doesn't ship. Distance without this is being wrong further from the centre.

What ships, and what goes back

The gate

Paste the output under the nearest competitor's name. If it still reads true, it goes back. Does it name something the audience was already carrying? Does it land before it gets explained? Two states only: pass, or fix and rerun.

Fig.2 · how the work runs

This is the method, drawn: the order the work happens in, and which steps are done blind to each other. The eight parallel reads are the quotas inside two of the frictions, five bridges and three receipts. The other four run inside those nodes. Ask one room for five ideas and the fifth one is a cousin of the first.