Trying an uncertain option can teach you something, even when another seems better now.
Information & Signals.
How clues, uncertainty and communication shape what we can know.
44 live experiment · 0 plannedLeading digits in scale-spanning measurements are often far from uniform.
A stricter threshold reduces false alarms but can miss real signals.
A fair binary event carries more uncertainty than one that almost always repeats.
Repeating a bit can overcome some errors but uses extra capacity.
A part that cannot be sped up limits the gain from adding more workers.
A ranked item can appear roughly in inverse proportion to its rank.
Memorable headlines are not a representative sample.
Selecting on either of two strengths can create a misleading tradeoff.
A clue is valuable when it can change a worthwhile decision.
Test whether a flattering description actually distinguishes you.
Compare a familiar name with evidence that matters to the task.
Update an urn hypothesis one observation at a time.
Deploy a forecast and watch the forecast change what happens next.
Move two thresholds and compare different meanings of an equal outcome.
Release a noisy count and see why repeated queries spend a privacy budget.
Add one grain at a time and record when your label changes.
Two rules fit every training point. Commit a prediction before seeing the next one.
Inspect the factory's badges and look for a result the claim rules out.
Assess the same testimony anonymously, then reveal a status label and reassess it.
Use encouragement to identify an effect, then deliberately break the instrument.
Search thirty hypotheses without mistaking every small p-value for a discovery.
Find a credential threshold that separates two types with different costs.
Turn skewed observations into collections of averages without changing the original distribution.
Add noise to the predictor and investigate why a fitted slope can look too weak.
Open ranked search results and balance useful finds against unnecessary reading.
Make ten probability forecasts, reveal outcomes, and score confidence honestly.
Remember a tile sequence, optionally solve another task, then reconstruct the order.
Estimate a team's private support, then reveal the anonymous preference poll.
Change spacing while keeping all thirty-six dots identical.
Edit a four-node link graph and follow the flow of ranking mass.
Flip transmitted bits and discover what one-error correction can and cannot repair.
Plot four datasets and omit individual observations to inspect the structure behind a regression.
Compare a hat puzzle with and without a public announcement that at least one hat is blue.
Search hidden binary objectives, then change the family from unrestricted to ordered.
Judge a well-supported true belief, then inspect whether its truth depended on an accident.
Build a prefix code, change symbol weights and encode your own short message.
Insert words into a tiny bit filter and hunt for a false positive.
Estimate a fictional sensitive-trait prevalence from deliberately noisy survey answers.
Design a publicly known signal policy while keeping the receiver’s Bayesian incentives visible.
Choose actions after costless messages from aligned or biased scripted senders.
Remove a bridge or a redundant local edge and compare who remains reachable.
Exchange weighted beliefs and compare an averaging network with a stubborn source.
Inspect identical appearances before revealing different stipulated chemical environments.