How to Prove This Wrong: The Falsification Protocol

The theory claims 100%. Every phenomenon, everywhere, always optimizes the process of optimization. That means one valid counterexample kills it. Not weakens it, not "requires adjustment." Kills it.

But "valid" is doing real work in that sentence. The falsification protocol on this page makes the rules explicit, the same way Karl Popper argued in 1934 that scientific theories must specify what would refute them. Without clear rules for what counts, any defender can explain away anything and any attacker can claim victory on vague grounds. This page locks down the definitions, the rules, and the failure conditions. Think of it as the rulebook. If you want to try breaking the theory, these are the terms. The argument the rules apply to is laid out step by step in The Logical Chain.

The definition (Locked)

"Optimize optimization" means one specific thing: improve future improvement capability.

Three ways to test it. Does this phenomenon increase the universe's capacity to improve its capacity to improve? Does it accelerate the acceleration of optimization? Does it create conditions that enable better optimization in the future?

Three things it does NOT mean. "Is this good for humans?" No, the universe optimizes at every scale, not just ours. "Does this seem efficient to me?" No, apparent waste may serve exploration or error correction. "Can I find ANY connection to optimization?" No, the connection must be specific, mechanistic, and bounded.

These definitions are fixed. They cannot be adjusted after a counterexample candidate is proposed. If the definitions need revision, that creates a new protocol version. Results from prior versions stand.

Falsification rules

Write down your target before you start. State what phenomenon you're testing, why you think it breaks the theory, at what scale, and over what time frame. Lock it in before you look at the answer. This prevents moving the goalposts after the fact.

Stay within scope. Evaluate at the phenomenon's own scale, one scale up, and one scale down. This applies equally to attackers AND defenders. Testing a biological phenomenon? Check biological, chemical, and ecological scales. Don't rescue a failing counterexample by invoking billion-year cosmic consequences, and don't rescue a failing defense by invoking them either.

The phenomenon's characteristic timescale determines scope. Chemical reaction: seconds to hours. Biological evolution: generations to millions of years. Extinction events: recovery time (millions of years) IS the characteristic timescale, so citing it isn't scope violation. Individual suffering: a lifetime, not geological eras.

Run the counterfactual. Ask: what would happen without this phenomenon? If removing it would make the universe optimize better, you have a genuine counterexample. If removing it would make things worse (less exploration, less error correction, fewer optimization pathways), then the phenomenon serves optimization even if it looks wasteful or harmful.

Keep the definition locked. The connection must be specific (point to the actual mechanism), explain HOW it works (not just claim that it does), and stay local (no "well, eventually, through a chain of a thousand events, it helps somehow").

Three logical steps maximum. If connecting the phenomenon to "improves future improvement" requires more than three steps, flag it as weak support, not strong evidence. One step (strong): "Pain provides error correction signals." Three steps (borderline): "Parasites force immune system evolution, which drives faster adaptation, which improves optimization." Five steps (too many): "Cosmic void lets matter drift, which forms a star, which evolves life, which creates AI, which optimizes."

Accept failure honestly. If you can't find any plausible connection to optimization under these constraints, treat it as a possible counterexample. Don't rescue it by widening the scope or loosening the definitions.

What counts as a kill

Result	What it means
0 valid counterexamples	Consistent with all tested phenomena. Strength depends on search power and independence.
1 valid counterexample judged by outsiders	The 100% claim is dead. The author can't be the judge of whether their own theory survives.
3+ valid counterexamples	The pattern is broken. Partial explanatory value at best, not universal.
10+	The optimization lens might still be useful in some domains, but it is not a universal principle.

"Valid" means it satisfies all the rules above: stated in advance (no changing the target after you see the result), bounded scope, counterfactual tested, specific mechanism identified, three or fewer steps, and survives independent review. If a term in the protocol is unclear, check the glossary.

The strongest test

"Improving future improvement capability" is specific enough to generate real, testable explanations. The rules above (bounded scope, three-step limit, locked definition, failure condition) prevent elastic stretching. The theory is falsifiable by literally any phenomenon in the universe. Find one law, one principle, one process at any scale that doesn't optimize optimization, and the theory dies. Every scientific theory faces the same standard: anyone can try to falsify it. This one gives you the largest possible attack surface. The Counterexample Challenge is open.

How to run an independent test

Pick 50+ phenomena across diverse domains. The person selecting them should be unfamiliar with the framework to avoid cherry-picking. For each one, apply the universal question: "How does this optimize the process of optimization itself?" using the locked definitions and rules above.

Multiple people should evaluate each phenomenon independently before comparing notes. Report how often the evaluators agree. Agree on what counts as pass or fail before any testing begins. Report everything, including borderline cases and clear failures. Set a time limit: if nobody can find a connection to optimization within 30 minutes of focused analysis, classify it as a possible counterexample. This prevents people from spending unlimited time talking themselves into a connection that isn't really there.

Where things stand

The framework has been tested across every major domain with zero counterexamples found. The rules above are public and fixed. The counterexample challenge is open to anyone. One valid counterexample kills the 100% claim. For the major objections people typically raise (and how the framework responds), see Common Objections. For the longstanding open problems in physics that the framework claims to address, see Ten Open Problems.

Try to Break This

Steel-manned objections — strongest counterarguments first. Submit yours →

Three steps balances two risks: allowing legitimate multi-scale explanations and preventing motivated reasoning from chaining arbitrary connections. If you cannot get from the phenomenon to "improves future improvement" in three specific, mechanistic steps, the connection is probably too tenuous to count as strong support. Longer chains are flagged as weak, not rejected outright.

The time box prevents unlimited creative rescue. If a trained evaluator cannot find any optimization pathway in 30 minutes, that is informative: either the pathway is genuinely absent or it is so non-obvious it should not be claimed as strong support. Provisional counterexamples can be revisited later, but the initial classification should be honest.

Exactly right. That's why this page exists: to make the rules clear enough that anyone can use them. You shouldn't have to trust the author's judgment. Run the test yourself with the same rulebook. The rules are here. The challenge is open.

The definition is specific: "improve future improvement capability." Three-step limit. Locked definition. Counterfactual required. If you can break it under these constraints, the theory dies. Try it.