LAWS THAT SURVIVE

A self-development manual with the evidence graded and the dead ideas buried

No heroes. No origin stories. No borrowed anecdotes. Claims, mechanisms, grades — and a graveyard.

AICE Library · Self-Development · Vol. II · 2026

A book that cannot show you its dead ideas is selling you live ones.

Foreword: The Podium Problem

Self-improvement books have a structural defect, and it is not the authors' fault. It is the genre's.

During the Second World War, the US military examined bombers returning from missions and found bullet holes clustered on the wings and fuselage. The natural conclusion: armor those areas. A statistician named Abraham Wald pointed out — in a memo, not a movie moment — that the military was studying only the planes that came back. Aircraft hit in the engine never returned to be counted. The armor belonged exactly where the survivors showed no damage.

The self-help shelf is an airfield of returning planes. Every author on it survived; the people who did the same things and crashed did not get publishing deals, so the shelf is incapable of showing their bullet holes. On top of this sits a second distortion, the halo effect: once an outcome is known to be good, every behavior that preceded it starts to glow. The same decision is "bold" in a winner's biography and "reckless" in a loser's autopsy — written by the same journalist, sometimes in the same year. Outcome contaminates the judgment of process, and the reader cannot decontaminate it by trying harder, because the reader was never shown the process. Only the podium.

So the average self-improvement book is a sample of one, selected by survival, lit by a halo, narrated backward — sincerely. The authors are not lying. The genre simply cannot tell the truth in that format.

This manual makes a different bargain, with four terms.

First, mechanisms before stories. Where a claim appears here, it is argued from how the thing works — incentives, mathematics, biology — not from what once happened to somebody impressive. Anecdotes about people are absent by design. If a claim cannot stand without a hero attached, it does not appear.

Second, every claim carries a grade. Not all knowledge is equally solid, and a book that will not say which of its own sentences are load-bearing is asking for faith, not attention. The grading system is on the next page. Some grades are humble. That is the point.

Third, the graveyard is printed. At the back of this book is a cemetery of famous self-improvement ideas that were tested properly and died — or survived only after demotion. Several of them still sell millions of copies. A field that cannot bury its dead is not a field; it is a séance.

Fourth, nothing here is exempt. The final chapter audits this book's own biases, including the ones it cannot fix, and sets the date on which the book itself must justify its existence or be buried with the others.

No author's journey will be narrated. No reader will be flattered. What remains, after all of that is removed, is the useful residue of the genre: claims that survive testing, stated plainly, with instructions.

How to Read This Manual

The grades. Every law carries one primary tag — its honest epistemic status:

Grades move. A ⚖ can be demoted by the next decade of replications; a ✦ can earn promotion. A book whose claims cannot change rank is not honest — it is embalmed. These grades were verified against the state of the literature in 2026; the audit chapter explains why even that is not enough.

The chapter anatomy. Every chapter runs one loop:

THE CLAIM — one sentence. THE MECHANISM — why it works, with no mysticism. THE EVIDENCE — what supports it, graded, including where support is thin. RUN IT — drills small enough to use this week. HOW IT DIES — the failure mode of the law itself, because any law taken past its dose becomes a poison. STATUS — the tag, and any live disputes.

If you do not trust this book yet — correct instinct; see Chapter 5 — start from the back. Read the Graveyard first. Honesty leaves bodies.

PART I — THE INSTRUMENT

Before improving anything, identify what is actually being operated.

Chapter 1 — Knowing Isn't a Safeguard

Status: ⚖ EVIDENCE (strong), with ✦ drills

THE CLAIM. Knowledge does not protect the person who holds it. The mind that knows the rule and the mind that acts under pressure are, functionally, two different machines — and only one of them reads books.

THE MECHANISM. Knowledge lives in the system that answers quiz questions: slow, verbal, available on request. Behavior under stakes is issued by older machinery running on different inputs — appetite, fear, momentum. The bridge between them is far narrower than it feels from the calm side. Psychologists call one version of this the hot–cold empathy gap: in a cold state, people systematically mispredict what they will do in a hot state — how much they will risk, eat, spend, or rationalize. The prediction error is not a character flaw; it is a design property. And intelligence does not close the gap. It staffs it: a faster mind generates better justifications at exactly the moments it needs honest ones. Motivated reasoning — reasoning as a lawyer building a case rather than a scientist testing one — is largely uncorrelated with measured intelligence. Brighter people do not escape the bias; they write better briefs for it.

THE EVIDENCE. The hot–cold gap: well-documented across risk, hunger, and addiction research (⚖ moderate-to-strong). Motivated reasoning as default: a foundational, repeatedly confirmed framework (⚖ strong). Intelligence failing to protect against myside bias: measured directly; the correlation between cognitive ability and resistance to myside bias hovers near zero (⚖ strong). The practical conclusion — that safeguards must be structural rather than informational — follows from all three (⚙).

RUN IT.

• Before any consequential bet — money, time, reputation — write the ruin condition first: the outcome that ends your ability to play again. If it cannot be capped, the bet is not placed. This is a rule for the cold state to impose on the hot one.
• Make binding decisions in the cold state, and make them specific: amounts, dates, exits. Vague resolutions are renegotiated by the hot state within minutes.
• Never renegotiate a rule while inside the state it was designed to survive. Renegotiation happens cold, on schedule, or not at all.
• At any decision point, ask one diagnostic question: which state is deciding right now? Naming the driver is often enough to change it.

HOW IT DIES. Two overdoses. Rule-worship: rules written cold can simply be wrong, and a rule that cannot be revised in daylight becomes a cage — schedule revisions, cold. And cynicism: concluding that since knowing isn't enough, betting is for fools. The gap is an argument for engineering, not for hiding.

Chapter 2 — Choose the Right Engine

Status: ⚙ MECHANISM + ⚖ EVIDENCE (strong)

THE CLAIM. Generating and verifying are different cognitive operations. Most people are meaningfully better at one than the other, most self-improvement assumes everyone must become a generator — and the economy just broke that assumption.

THE MECHANISM. Checking a solution is not the same act as producing one, and it is very often cheaper — a fact deep enough that computer science built an entire complexity hierarchy on it. Human memory shows the same asymmetry: recognition reliably outperforms recall; picking the right answer from candidates is a different, easier operation than generating it from silence. Minds differ in which engine runs at elite temperature. Some produce fluently from a blank page; others judge sharply among options but stall when asked to create structure from nothing. Schooling and hiring have historically graded everyone as generators — essays, pitches, self-presentation — so strong verifiers often spend decades holding a wrong verdict about their own ability.

Then the 2020s inverted the market. Machine generation of text, code, plans, and designs collapsed toward zero cost. When production becomes cheap, the complement becomes expensive: judgment of what is correct — evaluation, taste, quality control, the decision of what to trust — became the scarce operation. By the mathematics of comparative advantage (which holds even when one party is better at everything), the rational move is to specialize in the engine you run hot and trade for the other. For verifiers, this is the first era in which the trade is cheap.

THE EVIDENCE. Verification cheaper than generation: formal (⚙). Recognition over recall: classical, replicated memory science (⚖ strong). Comparative advantage: mathematics (⚙). One warning from the same literature: introspection is an unreliable witness — people confidently misreport why they do what they do. So the engine inventory below must be run on records, not on self-image (⚖ strong).

RUN IT.

• Inventory your engine from evidence, not introspection: where have you repeatedly been elite — bounded choices, spec-work, judging candidates? Where do you repeatedly stall — blank pages, open-ended plans? Past performance is the instrument; feelings about performance are not.
• If you verify better than you generate: structure all work as spec-bounded sprints with an external grader. Bounded task, clear acceptance test, short horizon. This is not a crutch; it is correct engineering for that cognition.
• Build pipelines shaped generate → verify: let tools, drafts, templates, or machines produce candidates; spend yourself on the verdict.
• Grade yourself on your engine's metrics — verdicts made, errors caught, specs met — not on the other engine's poetry.

HOW IT DIES. A verifier with no incoming candidates verifies air; the seat has power only when a pipeline flows through it. A generator with no checker ships noise at volume. And the inventory itself can be motivated reasoning in disguise (Chapter 1) — test it against records, revisit it yearly.

Chapter 3 — Burnout Is Data

Status: ⚖ EVIDENCE (strong), riding on a famous corpse

THE CLAIM. When a self-improvement system collapses, the collapse is data about the system's price structure — almost never a verdict about the person's character.

THE MECHANISM. For decades the genre ran on a tank model: willpower as a fuel that depletes and must be topped up with grit. That model is now a resident of this book's Graveyard — the "ego depletion" effect at its foundation shrank to approximately zero under large multi-lab replication. What replaced it is more useful: effort is priced, not tanked. Sustained self-control is simply expensive — aversive, motivationally costly — and its price spikes exactly when demand does. So any system billed in continuous willpower fails on schedule, the way any budget fails that is denominated in a currency you must mint under stress.

The systems that survive are billed in cheaper currencies. Defaults: what happens when you do nothing decides most outcomes, so set the nothing-path to the desired one. Friction: add steps to the wrong behavior, remove steps from the right one; small frictions move real behavior. Implementation intentions: pre-compiled if–then plans ("if situation X, then action Y") that convert a run-time decision into a setup-time one. Bounded sprints: hard modes entered by choice, with an end time, rather than ambient mandates that bill every waking hour.

THE EVIDENCE. Ego depletion's collapse: multi-lab registered replications near zero effect (⚖ strong — as an obituary). Defaults changing outcomes at population scale (organ donation, auto-enrollment savings): among the most robust results in behavioral economics (⚖ strong). Implementation intentions: meta-analytic effect in the medium range across hundreds of studies (⚖ strong). Friction effects on behavior: consistent (⚖ moderate-to-strong). The composite claim — design beats willpower — is the honest summary of the whole literature after its most famous idea died (⚖).

RUN IT.

• After any personal-system collapse, write one line before any feelings: what did this system price in willpower that could have been priced in design?
• Convert ambient mandates into opt-in, time-boxed sprints. Ambient costs compound invisibly; bounded costs are payable.
• Move decisions from run-time to setup-time: one cold decision (Chapter 1) that configures defaults beats a thousand hot ones.
• Engineer the environment before lecturing the self. The self has heard it before.

HOW IT DIES. "Burnout is data" degenerates into a universal license to quit whatever hurts. The test of a legitimate redesign is what it preserves: if the objective survives with a new price structure, that was engineering. If the objective quietly dies with the method, that was surrender with paperwork.

Chapter 4 — One Variable, One Ledger

Status: ⚙ MECHANISM + ⚖ EVIDENCE (strong)

THE CLAIM. Run several self-experiments and pool their outcomes, and the pooled verdict will oscillate between "capable" and "worthless" — mechanically, regardless of the underlying truth.

THE MECHANISM. Attribution requires isolation. Change four things — a skill, a platform, a schedule, a diet — and read the merged result, and no inference about any single one of them is logically available; the data cannot say what caused what. Worse, the merged feed is dominated by whichever experiment reported last, so self-assessment tracks recency instead of reality. Add regression to the mean — extreme results are always part luck, and luck drifts back — and the oscillation is fully explained: an unusually good week is followed by a more ordinary one by arithmetic, which the pooled ledger reads as decline. A person can be driven into despair by statistics behaving normally.

Laboratories learned this centuries ago and built instrument rails: one variable at a time, a baseline before every fix, a run ledger per experiment. The rails are not bureaucracy. They are what makes experience readable. And one rule sits above the rest: an outcome in one variable is never a verdict on another — and never, under any bookkeeping, a verdict on the experimenter. That last inference is not just cruel; it is a category error.

THE EVIDENCE. Confounding destroying attribution: mathematics (⚙). Regression to the mean and its systematic misreading as trend: classical statistics plus decades of judgment research (⚖ strong). Noise — unwanted variance in human judgment across occasions — as a large, underrated error source: well-supported (⚖). The debugging heuristic "failures cluster at boundaries — check interfaces before blaming the core" is engineering lore: not lab-proven, widely load-bearing (✦).

RUN IT.

• Keep one ledger per experiment. Physically separate. A language-learning result never shares a page with a revenue result.
• When any outcome arrives, write down which experiment owns it before allowing any feeling about it. Feelings are fast; attribution must be faster.
• Baseline before fixing. An improvement without a measured starting point is a story, not a result.
• When something fails, walk its edges first — the handoffs, the interfaces, the places where your plan touches systems you don't control. Indict the core only after the edges are clean.

HOW IT DIES. Ledger bureaucracy: tracking so elaborate that experimenting stops — the ledger serves the experiment, never the reverse. And emergencies invert the rule: when the system is on fire, revert everything to the last known-good state first and bisect later. Know which mode you are in.

PART II — ACTING UNDER UNCERTAINTY

The claims in this part are about how to move when nothing is guaranteed — without lying to yourself about it.

Chapter 5 — Falsify First, Celebrate Later

Status: ⚖ EVIDENCE (strong) + ⚙ MECHANISM

THE CLAIM. A belief that has never been attacked is not knowledge; it is mood with confidence attached. "Unproven" means untested. It never means true.

THE MECHANISM. The default search strategy of the human mind is confirmation: given a hypothesis, it looks for instances that fit. This was demonstrated formally decades ago — subjects asked to discover a hidden rule overwhelmingly test cases their theory predicts will work, almost never cases designed to break it — and it has been confirmed in every costume since. Add motivated stopping (the search ends the moment the answer pleases) and a mind can assemble overwhelming support for nearly anything while feeling rigorous the entire time. Confirmation is cheap. Discrimination costs.

The remedies are procedural, not moral — rigor cannot be felt into existence, only scheduled:

Order of interrogation. Ask what would make this false? before what supports this? The sequence is the discipline; reversed, the supports arrive first and the falsifier is never seriously sought.

Null controls. Before crediting your action for any observed change, check what happens when you do nothing. Any delta without your input means a foreign driver is loose in the system — and your action may be decoration.

Audit your auditors. Any judge you rely on — a metric, an advisor, a grading tool — must itself be tested: plant known errors and see if it catches them. A checker you have never checked is a belief, not an instrument.

Vocabulary discipline. Keep "untested" as a live word. The load-bearing claims in any plan deserve the attack budget; trivia does not.

THE EVIDENCE. Confirmation-biased testing: classic and robust (⚖ strong). Motivated reasoning: Chapter 1 (⚖ strong). Null-control logic and planted-error auditing: statistics and signal-detection reasoning (⚙), long practiced in every field that ships things that must not fail (✦ as craft).

RUN IT.

• Rank beliefs by load: which ones is the current plan standing on? Attack those first, this week.
• For any dynamic system — a market, an audience, an algorithm — run the no-action control before taking a bow.
• Test every grader you depend on with planted errors, once, before trusting its scores.
• Require present-tense evidence for the word "done": produced now, inspectable now. Memory of evidence is not evidence.

HOW IT DIES. Infinite skepticism — doubting everything equally — is paralysis with a philosophy degree. Falsification has a budget. Spend it where claims bear load; accept ordinary evidence for ordinary claims; ship.

Chapter 6 — The Asymmetric Lever

Status: ⚙ MECHANISM + ✦ HEURISTIC

THE CLAIM. Effort is spend, not virtue. Rank every candidate move by return-per-effort, and make anything below the top pick justify itself before it receives a single hour.

THE MECHANISM. The feeling of hard work is an accounting entry, not a result. Grind feels like progress because it hurts; leverage feels like cheating because it doesn't — and both feelings are wrong, because outcomes are priced by the market, not by your sweat receipts. In domains with skewed payoffs — creative work, business, technology, anywhere returns follow a power law — a small number of moves produce most of the result, and honest mediocre effort spread evenly across everything is the most expensive purchase available.

Two disciplines keep the lever honest. Scout before build: existing, standard, boring solutions have already survived other people's mistakes, which is the cheapest testing that exists. The prior formalized as the Lindy heuristic — for non-perishable things, longevity predicts further longevity — is contested as mathematics but useful as a default: the boring old method has earned a seniority the shiny new one hasn't. Cap the downside first: leverage has a defensive half. Ruin mathematics is unforgiving — a strategy with any repeatable path to total loss has an expected value that no upside repairs, because the game ends. Before hunting asymmetric gains, find the places where a small failure costs everything, and armor those (the survivors' engines, again).

THE EVIDENCE. Expected value, convexity, and ruin: mathematics (⚙). Power-law-shaped returns in creative and economic domains: strong observational regularity (⚖ moderate — measurement debates exist, the skew itself is not in doubt). Scout-before-build and Lindy: heuristics with long practitioner pedigrees (✦).

RUN IT.

• Before starting anything, list three candidate moves and mark one as the recommendation, with the reason. Inability to rank them means the problem isn't understood yet.
• Ask of every plan: what is the version that costs ten times less? It exists more often than pride prefers.
• Search for existing standard solutions before building custom ones. Adopt, then adapt.
• Write down the three places where a small failure would be catastrophic to you. Cap those before optimizing anything.
• Timebox the lever-hunt. Meta-work is procrastination wearing a nicer suit.

HOW IT DIES. Never digging. Eventually the best available tool is in hand and dirt still needs moving; past the timebox, further optimization is avoidance. The lever is for multiplying work — not replacing it.

Chapter 7 — Attention Is the Scarce Input

Status: ⚖ EVIDENCE (mixed, disputes printed) + ⚙ MECHANISM

THE CLAIM. Attention is the one input that cannot be bought, stored, or parallelized. Spend it in verdicts, not in noise — and remember that any report, system, or habit is worth exactly as much as what is actually read.

THE MECHANISM. Every item that reaches you invoices you for a decision, and the invoice is paid in the day's fixed attention budget. Task-switching carries a real, measurable cost: performance drops around every switch, and fragments of the previous task — attention residue — contaminate the next one. Ten small deliverables cost ten verdicts; one substantial deliverable costs one. That arithmetic, not any mystical focus doctrine, is the case for consolidating output.

Two clarifications the genre usually gets wrong. "Decision fatigue" is not a fuel gauge. The famous claim that judgment depletes like a battery through the day inherited its mechanism from ego depletion, and that mechanism is dead (Graveyard, entry 1); the surviving interpretation is that priorities and motivation shift over a session. The practical fix is therefore scheduling — put judgment-heavy work in your best hours — not rationing a mythical tank. Unread output is worthless regardless of quality. Information only becomes information when consumed: the value of any recurring report, dashboard, or log equals min(what it produces, what gets read). A perfect warning read by no one is zero — and its production cost makes it negative.

THE EVIDENCE. Switching costs: robust laboratory literature (⚖ strong). Attention residue: supported, smaller literature (⚖ moderate — honesty requires the size note). Decision fatigue: ⚠ contested — observed patterns, discredited mechanism, active dispute. Judgment quality tracking circadian state and sleep: strong (⚖, and see Chapter 10). min(produced, read): an accounting identity (⚙).

RUN IT.

• Consolidate toward one apex output per cycle. When tempted to ship five small things, ask which single one deserves the verdict.
• Every recurring report or automation must name its reader. No reader → kill it or reroute it. Audit this quarterly; readers rot silently.
• Apply the two-key test to anything that lands on your desk: (1) can only you decide this, and (2) does the decision materially change the outcome? One key missing → it doesn't reach you. Apply it in both directions — including what you send to others.
• Schedule verdict-heavy work in your best hours; batch the trivial into your worst.

HOW IT DIES. Monk-mode: cutting input to zero. Some scatter is scouting, and a judge with no incoming candidates judges nothing (Chapter 2). The goal is not an empty desk — it is a desk where everything present has passed the two-key test.

Chapter 8 — Give Every Plan a Death Date

Status: ⚖ EVIDENCE (strong) + ⚙ MECHANISM

THE CLAIM. Decide at a plan's birth what evidence would kill it — and when that evidence will be checked. A plan without a death date does not die; it becomes a zombie that eats the calendar.

THE MECHANISM. Sunk cost is not a feeling to be overcome with character; it is a default that must be scheduled against. Resources already spent scream louder than data at every decision point, and the scream intensifies with the size of the investment — which is why escalation of commitment is worst on the projects that deserve it least. Pilots have a name for their version, plan continuation bias: the tendency to press on with a failing plan grows as the destination nears. The cure is not willpower at the moment of crisis (Chapter 1 explains why that fails); it is a decision made earlier, in the cold state, when no scream had started: these numbers, checked on this date, mean burial.

Biology solved this a billion years before management science. Apoptosis: healthy tissue kills its own cells on a schedule, by design. The cell that refuses to die has a name, and the name is cancer.

One refinement matters more than it looks. Kill on evidence of death, not on the clock. A fixed rule like "terminate anything stale after N days" quietly executes every slow-maturing success and then — the elegant part — deletes the evidence that it did so, leaving a ledger full of fast wins that "proves" the policy right. That is a self-manufactured selection bias. The date on a kill-gate is when the evidence must be checked. The evidence is what kills.

THE EVIDENCE. Sunk-cost effects: classic, replicated, visible in field data (⚖ strong). Escalation of commitment: decades of lab and organizational studies (⚖ strong). Plan-continuation bias: applied aviation-safety literature (⚖ moderate). The pre-mortem — imagining the project already dead and writing the causes — rests on findings that "prospective hindsight" measurably increases the number and specificity of identified risks (✦ with ⚖ support). Apoptosis as fact: biology (⚙ as analogy).

RUN IT.

• Every plan ships, at birth, with kill criteria and a review date. No exceptions — especially the beloved plans; love is where escalation lives.
• Run a pre-mortem at launch: "It is one year later and this failed — write the story." Armor what the story exposes.
• When a gate fires, write the epitaph in three lines — believed / evidence said / what survives — and bury it visibly. Quiet burials teach nothing.
• Calibrate the gates: if none of your last five ever fired, they were decoration.

HOW IT DIES. Gates so tight nothing survives infancy — most real things look dead at week two. Clock-kills that execute slow compounders. And the inversion: using death dates as a reason never to commit deeply to anything. The gate exists to enable full commitment between reviews, not to prevent it.

PART III — OTHER MINDS

No one is lied to more politely than a person doing well.

Chapter 9 — Mirrors, Not Fans

Status: ⚙ MECHANISM + ✦ HEURISTIC, with ⚖ on the underlying biases

THE CLAIM. Surround yourself with things that can tell you no and prove it. Praise that was never stress-tested is noise wearing a compliment.

THE MECHANISM. Follow the incentives. Friends are paid in comfort for agreeing; employees are paid in safety; advisors are paid in renewals; recommendation algorithms are paid in your continued engagement. None of these parties is evil, and all of them, summed, produce the same output: a world that agrees with you slightly more every year. Meanwhile the halo effect operates internally — past wins make the next idea sound better in the same head that produced it — and social desirability bias guarantees that even honest people soften what they tell you, in proportion to how much you seem to need it. Left uncorrected, every mind drifts toward being its own fan club, attended by volunteers.

The correction is structural, never attitudinal. Install an antagonist: a person, a metric, a review process, a standing rule — something whose job is to attack the load-bearing claims (Chapter 5) before reality does it for a higher price. The practice exists at every scale with the same skeleton: adversarial collaboration in science, red teams in security, devil's advocates given formal roles rather than personalities. Two design rules decide whether it works. Pay for catches: a delivered piece of bad news must be rewarded — visibly — or the supply stops; a mirror punished for honesty learns to flatter, and then it is an expensive fan. Demand symmetric honesty: a mirror that can only say no is a fan inverted — contrarianism as theater optimizes for effect on you, exactly like flattery. The mirror must be able to say yes, ship it and be believed.

The deep precedent is biological: sustained adversarial pressure is the engine that builds capability — organisms locked in long red-queen races accumulate sophistication that comfortable ones never need. Comfort is not neutral. It is decay with good manners.

THE EVIDENCE. Halo effect, social desirability, and self-serving evaluation: robust (⚖ strong). Incentive analysis of advice markets: mechanism (⚙). Red teams, adversarial collaboration, pre-registered opposition: practitioner structures with long institutional track records, not RCT-proven (✦). Red-queen dynamics: established evolutionary framework (⚙ as analogy).

RUN IT.

• Install at least one antagonist in every loop that matters. If nothing in a given domain can currently tell you no, that domain is running blind.
• Keep a catch ledger: every confirmed error someone finds in your thinking gets one line — and becomes a rule the same day, or the catch evaporates and will be paid for again.
• Reward the sting, publicly. Thank the person who caught the flaw more warmly than the person who praised the plan.
• Watch your own proposals for the all-or-nothing signature — maximum stakes, no exits. Ideas arriving in that costume deserve the antagonist first.

HOW IT DIES. Mirror-shopping: collecting critics and ignoring them, which launders confidence ("it survived review") without the review mattering. And staffing the role with a pure contrarian, which teaches you to dismiss dissent as noise — inoculating you against the real thing.

Chapter 10 — Delegate Everything but the Verdict

Status: ⚙ MECHANISM + ⚖ EVIDENCE (strong on the sleep side)

THE CLAIM. Strength is not doing everything. It is knowing which decisions need your hands — refusing to hand those to anyone — and handing off nearly everything else without guilt.

THE MECHANISM. A single decision-maker does not scale by hours; hours are capped and quality degrades at the margin. The true capacity is sound verdicts per day — and every task grabbed that something else could do steals capacity from the verdicts nothing else can make. Comparative advantage runs the numbers: specialization pays even for the person who is better at everything, because their time has the highest opportunity cost precisely where they are most uniquely valuable. So the practical law splits in two directions at once. Delegate production, research, drafts, logistics — freely, imperfectly, early. And never delegate the small class of undelegables: identity and its signatures, money above a floor you set cold, trust between humans, taste on the outputs that define you, and final calls on anything irreversible. Handing those off is not efficiency; it is abdication with better branding.

The verdict engine also has a physical substrate, and it is maintained or ruined in bed. Sleep deprivation measurably degrades exactly the faculties verdicts run on — sustained attention, working memory, risk calibration, emotional regulation — and the degradation begins well before the sleeper notices it, because the noticing apparatus is among the first things degraded. Guarding rest is not softness. It is capital maintenance on the only irreplaceable machine in the operation.

THE EVIDENCE. Comparative advantage: mathematics (⚙). Verdicts-per-day as the binding constraint of a one-brain operation: mechanism (⚙). The undelegable list: heuristic, assembled from expensive institutional experience everywhere (✦). Sleep loss degrading executive function and decision quality, including underestimation of one's own impairment: strong, consistent literature (⚖ strong).

RUN IT.

• Triage every non-trivial task in one breath: own it / route it / kill it. If routing beats owning, route without guilt.
• Write your undelegable list — five items or fewer — in a cold state, once. Everything not on it is routable in principle.
• Calibrate trust like an engineer: stakes × freshness. High stakes on stale information → verify before trusting; low stakes → let it run.
• Treat sleep as an asset with a maintenance schedule, and schedule nothing heavy against it. No one negotiates well tired, and no one tired believes they are negotiating badly.

HOW IT DIES. In both directions. Hoarding: doing ten-dollar tasks while million-dollar verdicts queue — control addiction wearing a work ethic. And abdication: routing away heavy verdicts because they are heavy. The two-key test (Chapter 7) cuts both ways; some things must reach you.

Chapter 11 — Batteries, Not Fire

Status: ⚖ EVIDENCE (gold-standard core) + ✦ practice

THE CLAIM. Intensity without storage is decoration. Experience compounds only when it is written, indexed, and re-read — and whatever burns without locking something in was just weather.

THE MECHANISM. Memory decays on a curve that was measured in the 1880s and replicates today: steep at first, then flattening — most of an unrecorded lesson is gone within days. This is not a flaw to lament but a design parameter to engineer around, and the learning sciences have delivered, twice, with the most robust effects the field owns: spacing (reviews spread over time massively outperform bunched ones) and retrieval practice (recalling something strengthens it far more than re-reading it). Between them sits the engineering law of this chapter: an experience becomes an asset only through three verbs — write it while fresh, index it so it can be found, re-read it on a schedule that fights the curve. Most people stop before the first verb, then wonder why ten years of experience feels like one year repeated ten times.

Two extensions carry most of the compounding. Negative knowledge: recording what failed and why — a private graveyard — is the cheapest insurance ever invented, because the most expensive lesson is the one paid for twice. Decisions rot faster than data; a conclusion not written the day it was reached is already dying. Survival mathematics: compounding of any kind has one non-negotiable precondition — staying in the game. One uncapped loss deletes the whole ledger (Chapter 1's ruin condition, Chapter 6's downside cap; the laws in this book link arms here). And one warning taken from measurement theory: the moment the ledger becomes the goal — streaks maintained for their own sake, metrics gamed — it stops measuring anything. That is Goodhart's law, and it eats record-keeping systems from the inside.

THE EVIDENCE. The forgetting curve and its modern replication: solid (⚖ strong). Spacing and retrieval practice: the most consistently supported techniques in the learning literature, across ages and materials (⚖ gold). Write–index–reread as a composite practice, and graveyards of dead beliefs: craft, strongly convergent with the above (✦). Ruin precondition: mathematics (⚙). Goodhart's law: mechanism with abundant institutional evidence (⚙/⚖).

RUN IT.

• Same-day writes: the moment a lesson lands or a decision is made, one paragraph, dated, where the future will look. Not tonight — now; the curve is already running.
• Keep a hot index: one line per lesson, pointing at the details. An archive without an index is a landfill with good intentions.
• Schedule re-reads at expanding intervals. A battery no one drains is a brick.
• Keep a graveyard file for killed beliefs and dead plans — with epitaphs (Chapter 8). Read it before starting anything that rhymes with a resident.
• Date and source every note. Old notes lie at a predictable rate; trust them by stakes × age.

HOW IT DIES. The museum: notes accumulated as artifacts that change no decisions. Memory exists to change the next decision — a note that can't is dead weight. And Goodhart: when maintaining the system becomes the achievement, the fire has gone out and the rack is being polished.

PART IV — ENDINGS

Whatever has no chosen ending will be given one by a deadline — and deadlines write in the dark.

Chapter 12 — Choose the Ending First

Status: ⚙ MECHANISM + ✦ HEURISTIC

THE CLAIM. Decide what finished looks like at the start. Projects, tools, habits, and programs that never define their ending do not end; they metastasize into metrics, and the metrics take over.

THE MECHANISM. Two forces converge on anything open-ended. The first is escalation (Chapter 8) operating at life scale: with no defined ending, there is never a natural point to evaluate whether continuing makes sense, so continuation becomes the default forever. The second is Goodhart's law: when a measure becomes a target, it ceases to be a good measure. An endeavor without a chosen ending must navigate by proxies — followers, streaks, revenue this month, chapters shipped — and proxies, once targeted, detach from the thing they proxied. The combination produces the zombie: a project that cannot die, pursuing numbers that no longer mean anything, consuming attention (Chapter 7) that has stopped buying anything real.

The same logic binds anything designed to help you. A tool, a program, a teacher, a manual — each should have a direction of dependence, and the healthy direction is down: it should make you need it less over time, transferring capability until it is scaffolding you no longer stand on. A helper whose users need it more every month is not failing; it is farming. Growing dependence is a business model, not a defect — which is exactly why it must be measured from the user's side.

And endings decided under deadline pressure inherit the deadline's priorities, not yours. Time pressure narrows search, favors the reversible-looking option, and writes conclusions optimized for being finished rather than being right. The schedule may be compressed by circumstance. The definition of done should never be written by it.

THE EVIDENCE. Goodhart/Campbell dynamics: mechanism, with a century of institutional case evidence (⚙/⚖). Escalation: Chapter 8's literature (⚖ strong). Time pressure degrading decision quality and narrowing option search: supported (⚖ moderate). Definition-of-done and scaffold-removal: engineering and pedagogy practice (✦).

RUN IT.

• For any long project, write the final scene first: one dated paragraph describing what done, well looks like. One owner's name on it.
• Close every loop you open, or kill it explicitly. Open loops are how zombies feed.
• When time pressure spikes, say the rule aloud: the schedule may compress; the ending may not. Then cut scope, not the destination.
• Once a quarter, check the direction of dependence on your main tools, advisors, and systems: do you need each one less than last quarter, or more? Growing dependence on anything is a finding. Investigate it.
• Re-choose endings only deliberately — in daylight, by the owner, on evidence. Never by drift.

HOW IT DIES. Rigidity: an ending is re-chooseable; what it must never be is unwritten-until-the-deadline. And piety: a beautiful chosen ending used as a reason never to ship anything imperfect along the way. The destination justifies the walking; it does not replace it.

Chapter 13 — The Bias in This Book

Status: unrated. This chapter is the audit.

A manual that grades everyone else's claims owes the reader a grading of itself. Here is the prosecution's case against this book, stated as strongly as honesty allows, with the answers that exist and an admission where none does.

The census, because honesty is also arithmetic: twelve laws — four graded primarily ⚖ EVIDENCE, four ⚙ MECHANISM, four mixed ⚙/✦ with heuristic drills throughout; one load-bearing ⚠ printed with its dispute (decision fatigue); seven ⚰ dead and three demoted in the Graveyard. And the book itself, as an intervention in a life: untested, as of the day of printing. It has changed zero decisions. That is the correct score for a manual one day old.

So the manual receives what every plan receives (Chapter 8): a death date. Six months from printing — 2027-01-16 — this book must point to at least one real decision it changed, or be moved to the reader's own graveyard with an honest epitaph. That is not failure; that is the system working on itself.

And if it passes — if the laws hold, the ledgers fill, the gates fire when they should — then by Chapter 12 the ending is already chosen: the reader eventually stops needing it. Both roads end with the book unnecessary. That was the design.

THE GRAVEYARD ⚰

Famous self-improvement ideas, tested properly, now dead or demoted. Every stone lists what was claimed, what the evidence showed, and what still stands. The field that cannot print this page is not a field.

⚰ 1. Ego depletion — willpower as a fuel tank. Claimed: self-control draws on a limited resource that depletes with use; hence glucose top-ups, willpower budgeting, an industry of grit-tank metaphors. The evidence: the foundational effect shrank to approximately zero under large pre-registered multi-lab replication in 2016, and again in later multi-lab attempts. What survives: effort is genuinely aversive and costly — but the accounting is motivational, not a literal battery. Design beats willpower either way (Chapter 3), which is more useful than the corpse ever was.

⚰ 2. Power posing — stand big, change your hormones, change your life. Claimed: two minutes of expansive posture raises testosterone, lowers cortisol, and improves real-world outcomes. The evidence: the hormonal and behavioral claims failed direct, larger replication in 2015; the original paper's own first author publicly disavowed the effect in 2016. A residual "feeling of power" self-report survives — and is plausibly an artifact of participants guessing the hypothesis. What survives: posture affects mood at the margin. The career-changing version is buried.

⚰ 3. Learning styles — match the teaching to the learner's "style." Claimed: visual/auditory/kinesthetic learners exist and learn better when instruction matches their type. The evidence: the definitive review found no properly controlled support for matching; controlled tests since keep agreeing. It persists in education anyway — pedagogy's most durable zombie. What survives: people have preferences, and content has natural best-fit modalities (maps are visual). Matching instruction to learner type improves nothing measured.

⚰ 4. The Mozart effect — music makes babies smarter. Claimed: listening to Mozart raises intelligence, especially in infants; states funded classical CDs for newborns. The evidence: meta-analysis found a tiny, short-lived arousal effect — any pleasant stimulation works equally — and nothing lasting. What survives: being alert helps you do things while alert. Learning an instrument may have value; hearing one raising IQ is buried.

⚰ 5. Brain training — games that transfer to general intelligence. Claimed: daily puzzle apps improve memory, attention, and intelligence in daily life. The evidence: comprehensive review found improvement on the trained games with little or no far transfer to anything else; a leading vendor paid a federal false-advertising settlement in 2016. What survives: practice improves the practiced task — which is real, and is called practice.

⚰ 6. The 10,000-hour rule — practice quantity is sufficient for mastery. Claimed: roughly ten thousand hours of practice makes anyone expert at anything. The evidence: meta-analysis across domains found deliberate practice explains a real but minority share of skill variance (on the order of a fifth in music and sports; a few percent in professions); even the researcher whose data inspired the rule objected to the popularization. What survives: practice matters enormously and its structure (feedback, difficulty, spacing — Chapter 11) matters more than its raw hours. Sufficiency is buried; necessity stands.

⚰ 7. Positive visualization — imagine success and it comes. Claimed: vividly visualizing the desired outcome mobilizes the mind to achieve it. The evidence: repeated studies found the opposite — idealized success fantasies predict lower effort and worse outcomes, apparently because the mind pre-consumes the reward. What survives: mental contrasting — visualizing the goal and then the obstacles, with if–then plans attached (Chapter 3) — which outperforms both fantasy and pessimism. Dream, then interrogate the dream.

THE DEMOTED — not dead; cut down to honest size.

△ Grit — passion-plus-perseverance as the master trait. Meta-analysis found it overlaps heavily with plain conscientiousness and adds little predictive power beyond it. Persistence matters; the novel super-trait was mostly rebranding. △ Growth mindset — believing ability is malleable transforms achievement. Large pre-registered studies find a real but small effect, concentrated in specific struggling populations, far from the revolution marketed. Worth having; not worth outsourcing your strategy to. △ The Yerkes–Dodson curve — the tidy inverted-U of pressure and performance. A 1908 result on mice, inflated by a century of slideware into a universal law; modern reviews call the relationship task-dependent and the tidy curve unsupported. What survives is narrower and still useful: extreme stakes reliably degrade complex judgment — which is all Chapter 9's warning against all-or-nothing schemes requires.

THE EXIT

A manual, unlike a franchise, should know how it wants to end.

This one ends in one of two ways. Either it fails its own gate — six months, no changed decision, buried with an epitaph like everything else here — or it works, in which case its laws slowly become things you do without consulting it, the tags become a reflex you apply to every new bestseller, the graveyard becomes one you maintain yourself, and the book goes quiet on a shelf, needed a little less each month. Chapter 12 does not exempt the chapter's own container.

Add your own stones to the graveyard. Demote this book's claims when the evidence moves. Re-grade its grades. None of that is disrespect; it is the only kind of respect a book like this can coherently ask for.

What survives testing is yours. Bury the rest without grief.

Colophon

AICE Library · Self-Development · Vol. II. Companion volume to Vol. I (Wisdom Is a Weapon, the personal edition — same laws, opposite method). This volume contains no case studies of any living person, no author narrative, and no borrowed success stories; exactly one historical incident (Wald's memo) and one named psychological effect appear in the Foreword, both fact-checked before use.

Claim grades reflect the replication literature as of July 2026, cross-checked in an independent external verification pass on 2026-07-16. Citations in this book are anchors for the reader's own search, not proofs; the named studies are the standard entry points (Wason's rule-discovery task; the 2016 multi-lab ego-depletion replication; the 2015 power-posing replication and 2016 first-author statement; the 2008 learning-styles review; the 2010 Mozart-effect meta-analysis; the 2016 brain-training review and FTC settlement; the 2014 deliberate-practice meta-analysis; the mental-contrasting research program; the 2017 grit meta-analysis; the 2018–2019 mindset meta-analyses and national RCT; the spacing/retrieval-practice reviews; the implementation-intentions meta-analysis; the 1985 sunk-cost studies; the task-switching and attention-residue literatures; the sleep-and-executive-function literature).

Tag census: ⚖ ×4 primary · ⚙ ×4 primary · mixed ⚙/✦ ×4 · ⚠ ×1 load-bearing · ⚰ ×7 dead · △ ×3 demoted · book itself: untested.

Kill-gate: 2027-01-16.

What survives testing is yours.