WHAT
WOULD
COUNT?
A POCKET FIELD GUIDE TO INTELLECTUAL HONESTY / VOL. 01

What
would count?

Before defending a belief, design the evidence that could change it.

Not every disagreement can be settled today. But every useful inquiry can get clearer about what would support a claim, what would challenge it, and what would leave us saying: “I still don’t know.”

Draft an evidence contract Browse model claims

No accounts. No trackers. Your words stay in this browser unless you choose to export them.

00 / THE CENTRAL MOVE

The question before
the argument.

01

Make it observable.

“This always works” is fog. “In this setting, compared with what, which outcome changes—and by how much?” is a question you can inspect.

02

Make both doors real.

Write down what would count for your claim and against it. If no imaginable result could weaken your belief, you have not made a fair test.

03

Leave room for maybe.

Weak samples, noisy measurements and competing explanations can make the answer unresolved. Being uncertain is different from being indifferent.

01 / CLAIM LIBRARY

Choose a belief
to investigate.

Ten finished demonstration cases. They model better questions; they do not announce the truth of the original slogans.

02 / YOUR PERSONAL EVIDENCE DESK

Make an
evidence contract.

A promise to examine a claim fairly—not a promise to agree. Start blank, or open a library case and load its model into this desk.

Privacy promiseDrafts are not transmitted or stored by this site. A downloaded file stays on your device. Leaving the page may discard edits.
FIELD INSTRUMENT / 01NOT A VERDICT
My evidence contract
0 of 7 reflection fields filled
A good formulation specifies an outcome, a comparison and a timeframe where feasible.
08 My current assessment — chosen by me, not the site

These are personal, provisional labels—not scientific certification. The website will never assign a status automatically.

“Unsupported” does not automatically mean “disproved.” Your contract should explain which meaning applies.

Start with an honest question. Export when ready.

03 / THE TRAP CABINET

Twelve ways
evidence misleads.

Open a drawer for the diagnosis and a better question. Try spotting one in an argument you recently heard—including your own.

04 / CAUSALITY CORNER

Together does
not mean because.

Correlation is a pattern: two things vary together. Causation is a claim about what changes when one actually influences the other. Moving from pattern to cause takes additional evidence and assumptions.

INTERACTIVE DEMONSTRATION / NOT REAL-WORLD DATA

The coffee & deadlines puzzle

A toy office dataset can make coffee look like the cause of task completion. But what if deadlines influence both?

Coffee drinkers
75%tasks completed
Non-drinkers
45%tasks completed

Pooled toy illustration: a 30-point difference. That alone cannot tell us whether coffee caused it.

CHANGE THE LENS

What are we actually comparing?

Toggle a hidden condition. The diagram does not estimate a real coffee effect; it demonstrates how choosing comparison groups can change a causal story.

First impression

Coffee is associated with more completed tasks. But high-pressure deadlines may prompt both coffee drinking and intense work. This is confounding.

Deadline pressure
CoffeeTask completion
A / THREE POSSIBILITIES

X causes Y. Y causes X. Or Z affects both.

A fourth possibility is chance, bias, or a mixture of mechanisms. A graph can display an association without deciding which story created it.

B / WHAT HELPS

Better comparisons, multiple methods, replication.

Random assignment can help balance confounders in suitable, ethical settings. Observational studies can also be valuable when their designs, assumptions and limitations are carefully examined. Neither method is magic.

C / THE DIFFERENCE

Absence of evidence ≠ evidence of absence.

A null finding may be informative when the test had a strong chance of detecting the effect. A tiny, imprecise study may simply leave the matter open.

05 / REVISION STORIES

Changing your mind
is not losing.

Three fictional teaching stories. The annotations show the precise point where revising a belief became more rational than protecting it.

STORY 01 / THE MANAGER

“The quiet ones aren't contributing.”

A manager equated speaking time with value. Then a review of written decisions, code reviews and peer feedback revealed important contributions by quieter teammates. Meetings rewarded quick speech; the job required much more.

The revision → They stopped using airtime as a proxy for contribution and started assessing work against multiple relevant criteria.

Not a flip from “quiet is bad” to “quiet is best”—a correction to the measurement.

STORY 02 / THE FRIEND

“Late reply = not interested.”

A delayed text felt like clear evidence of indifference. One honest conversation brought context: another person preferred thoughtful, less frequent replies. Yet repeated cancelled plans would still matter to the friendship.

The revision → They separated a single ambiguous signal from an observable pattern of reciprocity.

An explanation is not proof of affection; nor is one delay proof of its absence.

STORY 03 / THE RESEARCHER

“This intervention helped.”

An early before-and-after result looked promising. A later comparison group showed similar improvement without the intervention. The original study had caught a wider trend, not necessarily a causal effect.

The revision → The researcher reported a narrower conclusion and designed a stronger next test.

Revising a conclusion in response to a better comparison is the work—not a defeat.

06 / READING ROOM

Go past
the notebook.

This field guide is a practical aid, not a substitute for formal research methods or a judge of anyone's beliefs. The frameworks above are original teaching exercises informed by scholarship on evidence, confirmation and causal inference.

What is evidence?Stanford Encyclopedia of Philosophy / EvidenceHow does evidence support a hypothesis?Stanford Encyclopedia of Philosophy / ConfirmationTesting, falsification and scientific practiceStanford Encyclopedia of Philosophy / Scientific MethodWhy randomized controlled trials matterOur World in Data / Research methodsConfounding in real risk estimatesOur World in Data / Methods and limits

The model claims are hypotheses for practice; the source links explain research or supply relevant findings. A linked study does not by itself establish the truth of every claim as stated. Fictional stories and toy data are identified as such.

ONE FINAL FIELD NOTE

Have a thought, challenge,
or idea? Write to Sahil.

The best conversations leave both people with better questions. A counterexample is welcome here.

Write to SahilExplore the Sahilverse