How Does JetKit Decide What to Pack?

The short version: your closet is the candidate set, wrong combinations are removed by rule, the rest are scored, and quantities come from the shape of the trip rather than the number of days. Here's each stage in more detail.

Quick Answer

Stage One: Your Closet Becomes the Candidate Set

Everything downstream depends on the system knowing what you own, which is why cataloging comes first.

Items go in by photograph, with recognition extracting category, subcategory, color, material and pattern. There's also a route that takes a photo of a complete outfit and separates it into individual garments, each with its own attributes, which gets a new wardrobe to a usable size faster than adding one item at a time.

Color is stored as a family rather than a raw value. This matters more than it sounds: navy and midnight blue are different numbers and functionally identical, while two similar-looking mid-tones can clash badly. Comparing families rather than pixel values is what lets the system judge whether two pieces work together. Compatibility is graded rather than binary, running from strong through acceptable to incompatible, so a strong pairing can be preferred without a merely acceptable one being forbidden.

Stage Two: Rules Before Scoring

Some combinations are wrong no matter how well they score. Swimwear with formal footwear. Shorts with boots. A formal garment with casual shoes.

These are eliminated by rule before any scoring happens, rather than being ranked low and occasionally slipping through. The reason is practical: a purely score-based system will eventually surface something absurd carrying a high number, because a strong result on six dimensions can outweigh a catastrophic result on one. A hard filter is the correct fix rather than more tuning.

There's a related constraint on formality range. A selection that mixes pieces from opposite ends of the formality scale reads as incoherent even when each individual item is appropriate, so wide formality gaps are penalized rather than allowed to average out.

Stage Three: Scoring What's Left

Surviving candidates are evaluated across several weighted dimensions rather than a single relevance number:

Occasion fit — whether the piece suits what you'll actually be doing.

Weather fit — whether it works in the destination's conditions.

Style cohesion — how well the selection holds together as a set rather than as individually good items. This carries the heaviest weight, because a bag of pieces that don't combine is the most common way a technically correct list fails.

Preference — what you've told the system you like, with favorites receiving a boost.

Practicality and comfort — how the piece behaves in use.

Versatility — how many different days and contexts the item covers, which is what earns a piece its space.

Freshness — how recently you've worn it. This exists specifically to stop the same reliable pieces being selected on every trip, which is a failure mode any scoring system drifts toward.

When you swap an item, the replacement is chosen with a diversity bonus: a piece introducing a color not yet in the plan scores higher than one duplicating something already selected.

Stage Four: Quantities From Trip Structure

This is where most packing lists go wrong, and where the arithmetic actually matters.

Quantities don't come from multiplying by days. Trip length selects an underlying formula, of which there are several, and which one applies changes the shape of the result rather than just its size.

Laundry access then reduces the counts rather than leaving them to scale linearly. That single adjustment is why a ten-day list comes out only slightly longer than a seven-day one instead of proportionally larger.

Packing mode shifts the whole result. Minimal pulls the counts down; comfort pushes them up. Same trip, same closet, different tolerance for having less.

Base layers are handled separately, because underwear and socks are the only categories that genuinely scale one-to-one with days, and socks additionally scale with activity. Treating them the same as tops is a common error in generic lists, and it produces both too few socks and too many trousers.

The Same Trip Gives the Same Answer

The planning logic is deterministic. Identical inputs produce an identical plan, so the list doesn't shift underneath you between openings, and any change you see traces back to something you changed about the trip or your wardrobe rather than the system quietly reconsidering.

That predictability matters more than it sounds. A list that varies each time you look at it is one you can't build trust in, because you can never tell whether a missing item was a decision or noise.

What It Doesn't Do

Worth stating plainly. It can't know about a one-off event you didn't tell it about. It isn't an authority on local dress expectations, which is the category where generic AI is documented to be weakest. And it requires cataloging your wardrobe first, which is genuine work, paid once and amortized across every trip afterward.

For the broader question of what "deciding" means as distinct from generating a list, see how AI decides what to pack. For the cataloging pipeline in detail, see can AI create a packing list from your own clothes.