Knowledge/Domain
How to know things, and how to avoid being confidently wrong.
Drag to pan · hold ⌘/ctrl and scroll — or pinch — to zoom.
The measurement that arbitrates, as opposed to the artifact that merely reports.
Method·95 connections
A causal explanation that was inferred rather than tested, and reads as more rigorous for being specific.
Method·63 connections
A failure that produces no signal, so the absence of an error reads as success.
Choosing identifiers so the code states its own meaning without commentary.
Method·55 connections
Escalating past the visible symptom until you find the layer that actually produced it.
Method·49 connections
A claim is only worth something if you know what observation would refute it.
Method·46 connections
When a measure becomes a target, it stops being a good measure.
Method·42 connections
The recorded origin of a change or a claim — who made it, when, and on what evidence.
Method·40 connections
A stated number pulls subsequent estimates toward it, regardless of relevance.
Method·35 connections
Documentation that has drifted from the system it describes, and is now worse than none.
Method·27 connections
Reading a truncated result as if it were the whole result, so every counterexample is invisible.
Method·26 connections
A statement that holds only because its subject set is empty — a green result that measured nothing.
The smallest set of steps that reliably produces a fault, and the unit of real debugging.
Method·25 connections
How much of a system's internal state can be inferred from what it emits.
Method·24 connections
The future cost of a present shortcut, metaphorically accruing interest.
A statement of the form "the last one" or "nothing else does this" — load-bearing, and only as good as the search behind it.
Method·23 connections
Reaching a defensible order-of-magnitude answer by decomposing a question into estimable factors.
Outcomes shaped by the sequence of past choices rather than by present merit.
Method·20 connections
A previously working behavior that a change has broken.
Method·19 connections
The two error rates of a test, and why neither answers the question a person actually asks.
judging a specific case from vivid evidence while discounting how common the categories actually are.
Method·18 connections
Do not remove something whose purpose you cannot explain.
Crash at the point of misconfiguration rather than degrading quietly into a wrong state.
A sample distorted because inclusion in it was never random.
Method·16 connections
Marking something as superseded and scheduled for removal, without removing it yet.
Method·15 connections
Drawing conclusions from a sample already filtered by success.
Over-trusting an automated recommendation, including against available contrary evidence.
Method·13 connections
Defending a modest claim while trading on a bolder one made under the same name.
any headline phrased as a question can safely be answered "no".
Method·12 connections
The tendency to search for, interpret, and recall evidence in ways that favor what you already believe.
A substituted value that papers over a missing input and hides the misconfiguration that produced it.
Never attribute to malice that which is adequately explained by stupidity.
without a clear signal of intent, sincere extremism and parody of it are indistinguishable.
Looking where the light is good rather than where the answer is.
Prose inside source code, useful only when it records a constraint the code cannot state.
Method·11 connections
A conditional-probability puzzle in which switching wins twice as often as staying.
Imitating the visible form of a practice while the mechanism that made it work is absent.
Method·10 connections
The time spent debating an item is inversely proportional to its actual stakes.
a placeholder identifier chosen to carry no meaning of its own.
The unit of an absolute change in a percentage, and the gap between that and a relative change.
Explaining a problem aloud, in full detail, to a listener who cannot help — and finding the fault while explaining it.
The tendency for a system's successor, freed of its predecessor's constraints, to become dangerously over-engineered.
Judging how likely or common something is by how easily examples come to mind, not by actual frequency.
Method·9 connections
A chain of prerequisite tasks that stands between you and the task you meant to do.
Two things moving together is evidence for a causal link but never proof of one, or of its direction.
Method·8 connections
Low skill removes the very ability needed to recognize low skill, inflating self-assessment most at the bottom.
A claim feels more true the more times you've heard it, independent of any evidence for it.
A model that fits its training data too well has memorized noise instead of learning the pattern that generalizes.
An extreme measurement tends to be followed by a more average one, with no cause required beyond noise.
Method·7 connections
Analyzing an incident on the assumption everyone involved made a reasonable decision given what they knew at the time.
Method·6 connections
Adding people to a late software project makes it later, because onboarding and coordination costs outpace the new capacity.
A product accumulates features one reasonable addition at a time until the sum is bloated, slow, and hard to use.
Repeatedly asking why a symptom occurred, chaining each answer into the next question until the causal chain runs out.
The smallest version of a product that lets you learn something real from actual users, not the smallest embarrassing version.
Prefer the explanation with fewer assumptions, not the one that is simplest to state.
A simpler, less correct design that ships and spreads beats a more correct one that doesn't, because adoption compounds.
A complex system that works evolved from a simple system that worked; a complex system built from scratch never does.
Method·5 connections
A task always takes longer than expected, even when you plan for it to take longer than expected.
The bias against adopting an outside solution simply because it wasn't built by your own team.
Work expands to fill the time allotted for it, so a deadline shapes effort more than the task's actual size does.
In a hierarchy, people get promoted based on performance in their current role until they reach one they're bad at, and stop there.
Be conservative in what you produce, liberal in what you accept — a design rule for interoperating systems.
Spending effort making code fast before you know where it's actually slow trades clarity for speed you may not need.
A trend appears in several groups of data but reverses or disappears when the groups are combined.
Continuing an effort because of what has already been spent on it, rather than what remains to be gained.
Asking for help with your attempted solution (Y) instead of your actual problem (X), so the help solves the wrong thing.