Rate the sketches

Help us find the sketches that land. Look at each one, then give it a rating. ~2 minutes.

From Sheena Iyengar — choice overload (the jam-tasting study)
Past a point, more options stop freeing people and start stalling them — the same idea, in transit.
More Gates, Fewer Through Two entrances. A wide bank of many turnstiles where a crowd piles up, hesitates between gates and turns back. Beside it, one open gate through which people flow steadily inside. More Gates, Fewer Through Open every gate and the crowd jams. Open one and they flow. many gates they pile up, hesitate… and turn back one gate one clear way in — and they take it The wider the entrance, the more people stay outside.
Does this sketch land the idea?
From Sheena Iyengar — choice overload (the jam-tasting study)
Past a point, more options stop freeing people and start stalling them — curate a clear few and point to a default.
The Two Jam Tables Two tasting tables side by side. The table of twenty-four jams draws a big crowd but few buy; the table of six jams draws a smaller crowd but most buy. Two small bar gauges below show the reversal: more lookers at the big table, far more buyers at the small one. The Two Jam Tables The big table pulls the crowd. The small table makes the sale. 24 jams a crowd gathers… 6 jams …more walk away with a jar who actually bought 3% from the 24-jam table 30% from the 6-jam table Faced with too much, most people choose nothing.
Does this sketch land the idea?
From Sheena Iyengar — choice overload (the jam-tasting study)
Past a point, more options stop freeing people and start stalling them — the same idea, at home.
Nothing to Wear Two clothes rods. A jam-packed rod under which a person stands stuck, thinking "nothing to wear." Beside it a sparse rod of a few well-spaced pieces, one already pulled out and worn, the person moving. Nothing to Wear A fuller closet, fewer decisions made. A curated few gets worn. too many nothing to wear! stuck — and stays in the old clothes a curated few picks one — and walks out the door More to choose from, less actually chosen.
Does this sketch land the idea?
From The sunk cost fallacy — behavioural economics
Same idea as the dead star, rebuilt from the construct/object library to A/B-test the draw engine.
The Dead Star A glowing star (the sunk cost) exploded long ago; its light only now reaches a figure standing at "now" on a then-to-now timeline. The Dead Star You’re still reacting to light from something already gone. star exploded2,000 years ago2,000 years of lightsees it nowthennow What glows bright in your mind may already be gone.
Does this sketch land the idea?
From The sunk cost fallacy — behavioural economics
Sunk cost is steering by what you've already spent instead of the road ahead.
The rearview mirror trap Driver's view with a visible headliner above and dashboard below, framing a smaller windshield. The rearview mirror hangs at centre, labelled sunk cost. The Rearview Mirror Navigate by what’s ahead — not what’s already behind. the road ahead sunk cost what you’ve already spent looking behind What’s spent is gone. Only the road ahead earns your next mile.
Does this sketch land the idea?
From “Attention Is All You Need” — Vaswani et al., 2017
Read the whole sentence in parallel instead of word-by-word — why the Transformer trains fast.
Read the whole sentence at once Left: words of a sentence queue at a single checkout, served one at a time — slow. Right: every word served at its own open till simultaneously — fast. The metaphor for self-attention processing all positions in parallel versus a recurrent network working word by word. Read the whole sentence at once One word at a time waits in line. Every word at once is served together. ONE TILL — word by word till the cat sat on the mat slow vs ALL TILLS — every word at once the cat sat on the mat fast Don't queue the words — open every till.
Does this sketch land the idea?
From “Attention Is All You Need” — Vaswani et al., 2017
Run several attentions in parallel, each catching a different kind of relationship, then combine them.
Multi-head attention — many perspectives on the same words at once Three readers each study the identical sentence and mark a different kind of link — subject to verb, which pronoun, nearby words — then their three readings pool into one richer combined reading. Many readers, one sentence Read it once and you catch one link. Read it several ways at once and you catch them all. the cat she fed subject → verb the cat she fed she → the cat the cat she fed nearby words the cat she fed One pair of eyes reads a line — a panel reads between them.
Does this sketch land the idea?
From “Attention Is All You Need” — Vaswani et al., 2017
Attention sees an unordered bag of words, so each position must be stamped on — or 'the cat sat' equals 'sat the cat'.
Attention has no sense of order, so stamp each word with its place The same loose word-cards can spell two opposite sentences until each card is stamped with a coral position number, which fixes the order and the meaning. Same words, stamp the place Order is meaning: without a place-stamp, "cat sat on mat" = "mat sat on cat". a bag of words cat mat sat on the the the cat sat on the mat the mat sat on the cat ? stamp each word with its place the1 cat2 sat3 on4 the5 mat6 now it can only mean one thing No numbers, no word order — so number the words.
Does this sketch land the idea?
From “Attention Is All You Need” — Vaswani et al., 2017
Attention is a search: each word issues a query, matches every key, and pulls back a blend of values weighted by fit.
Each word takes a blend, not a pick A word arrives with a question (a query). Three labelled bottles stand for the other words: each label is a key advertising what it offers, the liquid inside is its value. Streams pour into one glass — a thick stream from the best-matching bottle, thin trickles from the rest — so the word's takeaway is a weighted blend, not a single hard pick. The metaphor for attention as a weighted lookup over queries, keys and values. Each word takes a blend, not a pick Ask a question, read every label — pour a lot from the best fit, a little from the rest. the word "sat" its question: "who did it?" PLACE the ANIMAL the cat TIME on the mat label = key liquid = value what "sat" now knows most from the closest match A big pour from the best bottle, a splash from the rest.
Does this sketch land the idea?
From “Attention Is All You Need” — Vaswani et al., 2017
Self-attention gives every word a direct line to every other; an RNN relays meaning down a chain that blurs.
A direct line beats a whisper chain Top: five words in a row pass a message one to the next like Chinese whispers, and the last word comes out garbled. Bottom: the same five words each hold a direct line to every other word, so all stay true. The direct lines are drawn in coral. Give every word a direct line Relay a message down a chain and it blurs; connect everyone and it arrives true. Down a chain (the old way): the cat sat on ?og whisper… garbled! Every word to every word (attention): the cat sat on dog all true No relay, no whisper: everyone hears the original.
Does this sketch land the idea?
From Open-ended medical QA with clinical reasoning (CLINICR) — Nachane et al.
One guess is unreliable; sample several and the right one is usually among them.
Your first try usually isn't the keeper A photographer's contact sheet of six numbered frames. Five frames show a soft, out-of-focus blur; one frame shows a crisp, sharp subject and is ringed with a coral grease-pencil circle and a check — the keeper picked out of the burst. Your first try usually isn’t the keeper One attempt often misses. Take several, and the right one is usually among them. 1 2 3 4 5 6 the keeper ✓ Don’t pin your hopes on one frame — shoot the burst.
Does this sketch land the idea?
From Open-ended medical QA with clinical reasoning (CLINICR) — Nachane et al.
The multiple-choice options meant to test reasoning also constrain it — clinical reasoning only wins once they're removed.
The options meant to help you can fence you in A thinking figure stands penned inside a small four-post fence labelled A, B, C and D, its thought bubble only able to cycle "A? B? C? D?". All around the pen, an open field is scattered with many other possible diagnoses as grey dots — one of them, out in the open in coral, is the one worth exploring, but the fence keeps the reasoner from ever reaching it. The options meant to help you can fence you in Four choices don’t just offer answers — they pen your reasoning inside them. anaemia gout out here A B C D A? B? C? D? The fence was built to help you choose. It just walled off the field.
Does this sketch land the idea?
From Open-ended medical QA with clinical reasoning (CLINICR) — Nachane et al.
Clinical reasoning builds a differential diagnosis as clues arrive, rather than crossing off a fixed menu of options.
You can't eliminate from a list nobody gave you Two sculptors side by side: one chips a bust free from a marble block that already contains the whole face, dashed and visible inside; the other builds a face out of a shapeless lump of clay, one pinch at a time, with no pre-existing form to reveal. You can’t eliminate from a list nobody gave you A marble bust only needs removing what’s already there. A clay face has to be built. already there still forming Real diagnoses aren’t hiding in the marble — they’re shaped from clay.
Does this sketch land the idea?
From Open-ended medical QA with clinical reasoning (CLINICR) — Nachane et al.
Exam questions hand you a multiple-choice menu; real clinics never do — so strip the options out.
Finding an answer isn't the same as making one Two letter grids side by side. The left grid is a word search with one row already spelling ANSWER, highlighted in coral, plus a magnifying glass — the word is already there, just waiting to be found. The right grid is mostly blank, with a pencil actively writing the first letter into an empty row — nothing is given, it has to be produced. Finding an answer isn’t the same as making one In a word search, the answer is already sitting on the page. In a crossword, you have to put it there. QZTKMB FLRYOP ANSWER JXCDIG HUVTPE there to find A yours to make Out in the world, nobody hides the answer in the grid for you.
Does this sketch land the idea?
From Open-ended medical QA with clinical reasoning (CLINICR) — Nachane et al.
A correct answer with hidden reasoning isn't trustworthy — the reasoning has to be visible, not just the result.
A right answer you can't check is only a guess Two boxes side by side, each topped with the same coral diagnosis tag. The left box is made of glass — you can see the reasoning inside, a small staircase of steps climbing to the answer. The right box is sealed and opaque with a padlock and a big question mark — the same answer sits on its label, but the reasoning behind it is hidden. A right answer you can’t check is only a guess Two boxes, the same diagnosis on the label. You can only trust the one whose reasoning shows. Dx ✓ you can check it Dx ✓ ? just take its word In medicine, the working matters as much as the answer.
Does this sketch land the idea?
From Open-ended medical QA with clinical reasoning (CLINICR) — Nachane et al.
Sample several candidate diagnoses, then let a trained verifier score them instead of trusting the first guess.
Closest has to be measured, not eyeballed Four arrows land in a loose cluster near the centre of a target. Thin dashed lines run from the bullseye to three of them; a solid coral line runs to the fourth, marking it — after measuring, not guessing — as the true closest shot. A figure stands beside the target with a magnifying glass, pointing at the result. Closest has to be measured, not eyeballed Four guesses land near the centre. Only a measurement says which one actually did. Don’t ask which guess looks best. Measure which one is.
Does this sketch land the idea?

Thank you.

Your ratings help Napkin Solo learn which sketches actually work.

napkinsolo.ai · rate