This governs what you accept as an explanation, what you do when something fails, and whether the arrangements you build can still correct themselves later. Deutsch's core claim is that knowledge is created by guessing and then criticizing, never derived from experience, and that a good explanation is one whose details are locked together so tightly that they cannot be swapped out. From that follow two maxims he says are worth carving in stone: problems are inevitable, and problems are soluble. A large part of what he offers is a set of positions to take about progress rather than tasks to perform, so those sit below under Hold.

Do

Try to vary the explanation before you accept it. A good explanation is hard to vary while still accounting for what it accounts for. Deutsch's test case is the Greek myth of Persephone against the axis-tilt theory of the seasons: he constructs an alternative Persephone myth on the spot to show the marriage contract, the seed and the identity of the gods could all be swapped for other details without changing anything, while every part of the axis-tilt account is load-bearing. [Source: "The Reach of Explanations"]

Judge a proposal by what it explains beyond the thing you built it for. That surplus is reach, and it is discovered after the fact rather than designed in. The axis-tilt theory was devised only to explain the sun's changing elevation, and without a single modification it also explains why the hemispheres run in antiphase, why the tropics have no seasons, and why the midnight sun happens near the poles. [Source: "The Reach of Explanations"]

When something fails, name the knowledge you are missing rather than the limit you hit. The Principle of Optimism states that all evils are caused by insufficient knowledge. It is a way of explaining failure, not a promise of success: there is no supernatural decree or inherent barrier in the way, only something not yet worked out. Deutsch applies it even to aging, on the grounds that a sick body is a physical object and no law forbids transforming it into the same person in good health. [Source: "Optimism"]

Design for removing bad decisions cheaply instead of picking the right one up front. Deutsch takes Popper's reframing of the old question: stop asking who should rule and start asking how bad rulers and policies can be got rid of without violence. A rational system, like science, does not try to guarantee good outcomes in advance, it makes errors easy to detect and cheap to reverse. [Source: "Optimism"]

Audit your own repeated behaviors against his checklist for an anti-rational meme. Suspicion is warranted whenever a behavior is narrowly defined, faithfully repeated, persists after its stated justification has disappeared, works against your own goals, or gets defended with a line like "because I say so" or "it never did me any harm." Rational ideas replicate by using your critical faculties; anti-rational ones replicate by disabling them. [Source: "The Evolution of Culture"]

Build error-correction in, and make the thing discrete enough to correct. His example is tallying goats by the length of a piece of string: every operation adds a small measurement error, which caps how many goats can be tracked at all. Round to a fixed whole unit and errors become detectable and correctable at each step, removing the limit. This is why every historical jump to universality happened in a digital system, and why he calls error-correction the beginning of infinity. [Source: "The Jump to Universality"]

When you learn from someone, go after the idea rather than the behavior. Popper opened lectures by asking students simply to observe, then waited for someone to ask what they were supposed to observe. Copying is ambiguous without already knowing which part matters, which is why a person can recover an idea correctly from a garbled or inconsistent account of it, and a parrot cannot. [Source: "The Evolution of Creativity"]

Count a better problem as the progress. Deutsch reframes the growth of knowledge as a move from problems to better problems, not from problems to solutions, and suggests renaming theories as misconceptions to make the point plain: Einstein's misconception of gravity improved on Newton's, which improved on Kepler's. Today's open physics problems carry more knowledge than the problems Einstein started from. [Source: "The Beginning"]

Argue a moral question the way you argue a factual one. The maxim that you cannot derive an ought from an is is true, and is used to conclude wrongly that morality is outside reason. Deutsch's counter is that you cannot derive a scientific theory from an is either, because science does not work by derivation, it works by conjecture and criticism, and moral explanations are open to exactly the same treatment. [Source: "The Reality of Abstractions"]

Don't

Don't treat a claim as serious just because it is testable. A prophet's doomsday date and a gambler's hunch about a lucky night are both testable, and when refuted their holders simply swap in another arbitrary claim and make no progress at all. Testability only starts to matter once something is already a good explanation. [Source: "The Reach of Explanations"]

Don't average two good explanations into a compromise. Because good explanations are hard to vary, they cannot be blended smoothly, and the halfway version is usually worse than either one alone. Worse, nobody actually believes a compromise, so when it fails no one learns anything and no explanation gets abandoned. [Source: "Choices"]

Don't forecast by assuming only today's knowledge will ever be applied. Paul Ehrlich lectured in 1971, with Deutsch in the audience, that mass starvation was ten years away, twenty at best. A graduate student around the same time argued color television was doomed by the scarcity of europium, which liquid-crystal displays went on to not need at all. Both failed the same way. [Source: "Unsustainable"]

Don't aim at preserving the current arrangement. Deutsch calls "sustain" dangerously ambiguous between providing what people need and preventing things from changing, which are close to opposites. Curing a childhood disease is unsustainable by design: it changes population, labor and the whole subsequent way of life. As disaster closed in, the Easter Islanders put more effort into larger monuments to their ancestors, on roads made from the trees they needed. [Source: "Unsustainable"]

Don't accept a theory's predictions while refusing its explanation. Splitting a theory into an explanationless predictive core plus a separate optional interpretation is incoherent, because knowing when a rule of thumb applies always requires an explanation. Deutsch's model of bad philosophy is the phrase "because I say so," which explains anything, addresses who spoke rather than what was said, substitutes a demand for justification, conflates ideas with power, and claims immunity from criticism. [Source: "A Physicist's History of Bad Philosophy"]

Don't trust a raw observation more than an interpreted one. Cataloguing galaxies from photographic plates, Deutsch asked his hosts whether an object was a galaxy or a star and was told it was a flaw in the emulsion. Jocelyn Bell's first pulsar was a shaky line of ink from a chart recorder until a long chain of theory made it a spinning star. Accurate observation requires applying more explanatory knowledge in advance, not less. [Source: "Closer to Reality"]

Don't mistake a rule about your surroundings for a law about the world. Deutsch's word for this is parochialism: reading the rotation of the night sky as a property of the stars, or treating "the sun rises every 24 hours" as a law. Anthropocentrism is one species of it rather than the whole genus. [Source: "The Spark"]

Don't hold any commitment immune from criticism. In the dialogue, Sparta's ban on debate and persuasion is diagnosed as an unusually self-perpetuating kind of mistake, because it disables the one mechanism through which any mistake could be found and fixed. Socrates speculates that the imperative not to destroy the means of correcting mistakes may be the one from which all the others follow. [Source: "A Dream of Socrates"]

Hold

Problems are inevitable, and problems are soluble. Fallibilism guarantees the first: correcting an error always produces new problems, and there is no final unimprovable explanation. Human reach supplies the second, because it is bounded by physical law rather than by the environment we evolved in. Holding both turns a hard problem into a request for knowledge. Refusing them turns every obstacle into evidence of a ceiling, which is exactly the belief that stops anyone looking for the way through. [Source: "The Spark"]

If a problem is interesting, it is soluble. Deutsch offers this as a conjecture with three corollaries: inherently insoluble problems are inherently uninteresting, the distinction between interesting and boring is objective, and the puzzle of why interesting problems are soluble is itself soluble. Held, undecidability and unsolved conjectures stop counting as evidence against optimism. Not held, every formal impossibility result reads as a verdict on human prospects. [Source: "A Window on Infinity"]

Nothing you currently know can tell you what the next discovery contains. Feynman wrote in 1965 that fundamental physics would see no real novelty for perhaps a thousand years. Michelson said in 1894 that future discoveries lay in the sixth decimal place, and his own experiment with Morley was quietly supplying evidence for relativity. Held, this keeps you from pricing an unknown solution at zero. Not held, you will keep mistaking the edge of your own knowledge for the edge of what is knowable. [Source: "The Beginning"]

Certainty was never available and was never what you wanted. The dialogue has Socrates admit he can be sure of nothing and find the idea of perfect security in belief boring, and quotes Xenophanes on all human knowledge being guesswork. Holding this makes revision cheap. Refusing it makes every correction feel like a defeat, which is what makes people defend positions past the point of believing them. [Source: "A Dream of Socrates"]

The creativity you already use to understand other people is the whole faculty. Deutsch argues creativity evolved not to innovate but to acquire existing ideas, because working out what someone means from indirect evidence is the same logical task as working out a law of nature. Innovation came free with it. Held, this removes the excuse that you are not the creative type. Not held, you will keep treating knowledge-creation as a talent other people were issued. [Source: "The Evolution of Creativity"]

Beauty is objective, not a matter of taste. Flowers had to signal across a gap between species with no shared genetic code to build on, and had to be hard for non-nectar-bearing flowers to counterfeit, so they converged on standards recognizable independent of any one species' preferences. That is why humans, who never co-evolved with flowers, reliably find them beautiful, while a beetle attractive to another beetle looks like nothing to us. Held, aesthetic judgment becomes arguable. Not held, it collapses into reporting your own reactions. [Source: "Why are Flowers Beautiful?"]

The one line

Every failure is a piece of knowledge nobody has created yet, so stop looking for the arrangement that cannot go wrong and build the one that lets you find out it did.