Say exactly what you mean.

Peirce's name for inference to a candidate explanation: a surprising fact C is observed; if A were true, C would follow as a matter of course; so there is reason to suspect A. It proposes hypotheses for testing but never confirms them.

Learn it in Hypotheses, Predictions and Tests · also in Liberal Arts and Reasoning: Start Here

Conditions, written before work starts, that define when it is done, including at least one failure case. Writing them first stops the goalposts moving, and breaking something on purpose proves a check can fail; a check that cannot fail proves nothing.

Learn it in Working with AI Coding Agents: A Disciplined Loop · also in Deliberate Practice: Choose Targets You Can Observe, Building a Career on Capabilities, Not Tools, Engineering and Career: Start Here

The essential omega-3 fat in plant foods such as flaxseed, chia, hemp and walnuts. The body converts only a small fraction into the longer-chain EPA and DHA, so whether vegans should add an algae-oil source of those is still debated.

Learn it in Nutrient Coverage on a Plant-Based Diet · also in The Daily Dozen and Roles Over Ingredients, How to Rank Foods Honestly: A Transparent Scoring Method

Aggregate Nutrient Density Index, popularised by Joel Fuhrman: a nutrients-per-calorie score scaled so the densest foods reach 1,000. It measures density, not health outcomes, and undervalues energy, protein, fat, portion size and cost, so treat it as one limited input.

Learn it in How to Rank Foods Honestly: A Transparent Scoring Method

Restating an author's argument in your own words as numbered premises leading to a conclusion, then naming hidden premises, the weakest premise and the strongest objection before judging it. A reconstruction is fair when the author would accept it.

Learn it in The Weekly Loop: Read, Reconstruct, Make, Defend, Log · also in Peirce on the Fixation of Belief

A short note on one durable idea, in your own words, linked to related notes in the Zettelkasten tradition. A principle note adds source, mechanism, when it holds and fails, and a case; a map of content (MOC) indexes such notes.

Learn it in Principles Ledger and Question Arcs · also in How to Read a Great Book: Levels, Modes and the Completion Standard

Authentication establishes who is making a request; authorization decides what that identity may do. A successful login proves only the first, not that current policy, group membership or address mapping grants the expected access right now.

Learn it in Stale Information and Dependable Reports

How common a condition is in the relevant population before any test. Ignoring it (base rate neglect) misleads: at 1% prevalence, a test catching 90% of cases with 9% false positives gives about 98 positives per 1,000 people, only 9 of them real.

Learn it in Mathematics for Better Decisions · also in Judgment Under Uncertainty: Probability, Causation and Forecasting, Systems Thinking: Feedback, Queues, Information and Dynamics

Revising belief as evidence arrives: posterior odds equal prior odds times the likelihood ratio. A 10% prior (odds 1:9) meeting evidence three times likelier if the hypothesis is true becomes odds 3:9, or 25%, not certainty.

Learn it in Judgment Under Uncertainty: Probability, Causation and Forecasting · also in Mathematics for Better Decisions, Liberal Arts and Reasoning: Start Here

The share of a nutrient in food that the body actually absorbs and uses. Phytate lowers zinc and iron absorption, oxalate locks up much of the calcium in spinach, and vitamin C eaten with legumes improves non-heme iron absorption.

Learn it in Nutrient Coverage on a Plant-Based Diet

A measurable indicator, such as blood pressure, LDL cholesterol or after-meal glucose, used as a stand-in for disease risk. Improving a biomarker is not the same as demonstrating fewer heart attacks, strokes or deaths.

Learn it in Reading Nutrition Evidence: What a Study Can and Cannot Show · also in The Daily Dozen and Roles Over Ingredients, How to Rank Foods Honestly: A Transparent Scoring Method, Daily Walking and Easy Cardio

The most damage a component can do if it errs or is hijacked. For an agent, bound it with written limits on capabilities, run time, call count, spending and secrets rather than trusting a cleverer prompt to prevent misuse.

Learn it in From Assistant to Agent: Tools, MCP and Evaluation

Matching stated confidence to outcomes: of the things you call 70% likely, about 70% should happen. The Brier score, the mean squared gap between forecast probability and the 0-or-1 outcome, rewards both calibration and discrimination; lower is better.

Learn it in Judgment Under Uncertainty: Probability, Causation and Forecasting · also in Technology, AI and Human Judgment

A capability is something you can do and demonstrate on an unfamiliar system; a tool is one way of doing it today. Tools, versions and exam syllabi expire within years while capabilities transfer, so organise study plans around demonstrations of capability.

Learn it in Building a Career on Capabilities, Not Tools · also in Engineering and Career: Start Here, From Assistant to Agent: Tools, MCP and Evaluation

A quantity growing at rate r per period is multiplied by (1 + r)^n after n periods. The Rule of 72 estimates doubling time as 72 divided by the percentage rate: at 6%, about 12 periods.

Learn it in Mathematics for Better Decisions

A variable that influences both the supposed cause and the effect, creating an association that need not be causal. Hot weather raises both ice-cream sales and swimming, so sales seem to predict drownings; causal diagrams make such assumptions explicit.

Learn it in Systems Thinking: Feedback, Queues, Information and Dynamics · also in Mathematics for Better Decisions

From 'if P then Q', the statement 'if not Q then not P'. It always has the same truth value as the original, so proving it proves the original; the converse, 'if Q then P', is a different claim.

Learn it in Proof and Precise Reasoning: From Arguments to Theorems

A single case that makes a universal claim false. Confirming examples can never establish a 'for all' statement, but one counterexample refutes it, which is why 'what would refute this?' is the most useful question in reasoning.

Learn it in Proof and Precise Reasoning: From Arguments to Theorems · also in Discrete Mathematics and Graphs: Counting, Relations and Networks, Learning Mathematics: Procedures and the Six Abilities, The Seven Liberal Arts, Hands-On: Logic, Proof and the Quadrivium, Working with AI Coding Agents: A Disciplined Loop

The 40 to 140 °F (4 to 60 °C) range in which bacteria multiply fastest. Refrigerate perishables within two hours (one hour above 90 °F), keep the fridge at or below 40 °F and reheat leftovers to 165 °F (74 °C).

Learn it in The Practical Kitchen: Staples, Meal Prep and Food Safety

Practice on well-defined tasks just beyond current ability, with immediate feedback and repetition with correction (Ericsson and colleagues, 1993). A 2014 meta-analysis found it explains only a modest share of performance differences, so treat it as a method, not a guarantee.

Learn it in Deliberate Practice: Choose Targets You Can Observe · also in Building a Career on Capabilities, Not Tools, Engineering and Career: Start Here, Strength Training for Busy Lives: The Recovery-Based Cycle, Twelve-Week Arcs and the Capstone Project Method

The overall combination of foods eaten together, such as DASH, Mediterranean or the Portfolio diet. A trial of a whole pattern can show the combination helps but not which ingredient did the work, so it cannot rank single foods.

Learn it in Reading Nutrition Evidence: What a Study Can and Cannot Show · also in A Plant-Based Eating Framework You Can Repeat, The Daily Dozen and Roles Over Ingredients, How to Rank Foods Honestly: A Transparent Scoring Method

RDA: the intake meeting the needs of nearly all (97 to 98%) healthy people; AI: an adequate-intake estimate used where evidence is too thin for an RDA; UL: the highest daily intake unlikely to cause harm. They are planning references, not treatment doses.

Learn it in Nutrient Coverage on a Plant-Based Diet

A test whose possible outcomes lead to different next actions, so it separates competing explanations, like Platt's crucial experiment. If every result would send you to the same action, the test only makes you feel busy.

Learn it in Evidence-Based Troubleshooting: Separating Explanations

A loading rule with two dials turned one at a time: add repetitions within a target range at a fixed load, and once every set reaches the top of the range with reps in reserve, make the smallest load increase and rebuild the reps.

Learn it in Strength Training for Busy Lives: The Recovery-Based Cycle

Socrates' method of cross-examination in Plato's early dialogues: ask for a definition, draw out its consequences and find where they conflict with other commitments. By analogy, testing a proposed cause by deriving what else must be true makes a similar move.

Learn it in The Seven Liberal Arts, Hands-On: Logic, Proof and the Quadrivium · also in Evidence-Based Troubleshooting: Separating Explanations, Technical Writing as Reasoning, Reading Nutrition Evidence: What a Study Can and Cannot Show, Reasoning for What Ends? Wisdom, Character and the Examined Life, Engineering and Career: Start Here, Liberal Arts and Reasoning: Start Here

Taking in less energy than you expend, which fat loss requires whatever the diet. Fixed rules such as '3,500 calories per pound' mislead over time because expenditure falls as weight falls (Hall and colleagues, 2011).

Learn it in Gradual Fat Loss and Maintenance: A Method · also in Protein for Muscle on a Plant-Based Diet

Shannon's measure of a source's unpredictability, H = -Σ p log2 p bits per symbol. A fair coin carries 1 bit per toss and a 90%-heads coin about 0.47 bits; entropy sets the limit for lossless compression.

Learn it in Systems Thinking: Feedback, Queues, Information and Dynamics · also in A Map of Mathematics: Twenty-Five Areas and What Depends on What, Technology, AI and Human Judgment

Treating a missed workout, low-step day or big meal as something to repay with extra miles, skipped meals or hard sessions. Compensation turns a daily floor into a debt that collapses routines, so the rule is no catch-up and no punishment.

Learn it in Daily Walking and Easy Cardio

The list of everything that should report, held independently of whatever actually responded. It is the denominator of any coverage claim: 'all visible hosts are healthy' describes only the reporters, and without the inventory, silence looks like health.

Learn it in Evidence-Based Troubleshooting: Separating Explanations · also in Quantitative Reasoning with Stated Assumptions

A candidate solution, created by an irreversible step such as squaring both sides, that fails the original equation. Squaring √(x + 2) = x yields x = 2 and x = -1, but √1 is 1, not -1, so check every candidate.

Learn it in Algebra: The Skills That Carry Everything

The view, named by Peirce, that any belief may need revision. As a trait of character it means valuing correction over having been right, so contrary evidence feels interesting rather than threatening.

Learn it in Reasoning for What Ends? Wisdom, Character and the Examined Life · also in Systems, Decisions and Robust Design

Popper's criterion that a scientific claim must risk refutation by some possible observation; no number of confirmations proves a universal claim, but one contradiction can refute it. In practice, write down the falsifier, the result that would make you drop a hypothesis, before testing.

Learn it in Hypotheses, Predictions and Tests · also in Judgment Under Uncertainty: Probability, Causation and Forecasting

A path by which a stock's level feeds back to the flows that change it. Balancing loops push toward a goal, like a thermostat; reinforcing loops amplify change, like compound interest or a bank run; delays in balancing loops cause overshoot and oscillation.

Learn it in Systems Thinking: Feedback, Queues, Information and Dynamics · also in Systems, Decisions and Robust Design, Mathematics for Better Decisions, Stale Information and Dependable Reports

The earliest mistake in an attempt, the one that made later errors inevitable; the last visible error is often only a symptom. Deliberate practice repairs that error first, because fixing five at once hides which fix mattered.

Learn it in Deliberate Practice: Choose Targets You Can Observe

Peirce's 1877 account of four ways to settle doubt: tenacity, authority, the a priori method and the method of science. Only science has external permanence, answering to something independent of opinion, so anyone who pursues it far enough reaches the same result.

Learn it in Peirce on the Fixation of Belief · also in Evidence-Based Troubleshooting: Separating Explanations, Reading Nutrition Evidence: What a Study Can and Cannot Show, Proof and Precise Reasoning: From Arguments to Theorems, What Does It Take to Know Something?, Hypotheses, Predictions and Tests

Gettier's 1963 cases in which someone has a justified true belief that is not knowledge, because it is true only by luck. They show justification plus truth is not enough, and no proposed fix has won general agreement.

Learn it in What Does It Take to Know Something?

When a measure becomes a target, people optimise the measure and it stops tracking what it stood for. It is an observation, not a theorem; in AI the same pattern appears as reward hacking, where a system exploits a loophole in its objective.

Learn it in Systems, Decisions and Robust Design

A write-up or tool is dependable when someone who was not in the room can run it or explain it back without you narrating. Wherever they stumble, there is a gap or an undocumented dependency that lives only in your head.

Learn it in Technical Writing as Reasoning · also in Stale Information and Dependable Reports

A fixed set of questions with known correct outcomes, split so you tune on one part and score on the unseen rest; tuning on every case measures memorisation of the test. Include unanswerable, conflicting, tool-failure and prompt-injection cases.

Learn it in From Assistant to Agent: Tools, MCP and Evaluation

The tendency, once an outcome is known, to judge it as more predictable than it was (Fischhoff, 1975). The remedy is to write predictions down before seeing the result, so the record can prove you wrong.

Learn it in Hypotheses, Predictions and Tests

Pearl's three rungs of causal reasoning: association (what does seeing X tell me about Y?), intervention (what happens to Y if I change X?) and counterfactuals (what would have happened otherwise?). Correlation alone cannot climb past the first rung.

Learn it in Judgment Under Uncertainty: Probability, Causation and Forecasting

Performance is how well you do during study; learning is what persists and transfers later. Conditions that make study feel easy, such as watching a smooth explanation, can raise fluency yet lower retention, while some effortful conditions improve it.

Learn it in Learning Mathematics: Procedures and the Six Abilities · also in Technology, AI and Human Judgment, Deliberate Practice: Choose Targets You Can Observe

An essential amino acid that acts as a key trigger for muscle protein synthesis. Many plant proteins contain less of it than animal proteins, one reason to aim toward the upper end of the per-meal protein range.

Learn it in Protein for Muscle on a Plant-Based Diet

A proof technique for statements about all natural numbers: prove a base case, then prove that truth at n forces truth at n + 1. Despite the name it is deductive, unlike the inductive inference behind Hume's problem.

Learn it in Discrete Mathematics and Graphs: Counting, Relations and Networks

An open protocol for connecting AI applications to external capabilities. A server can offer tools (actions the model may request, possibly with side effects), resources (data supplied as context) and prompts (reusable templates), over local stdio or HTTP-based remote transport.

Learn it in From Assistant to Agent: Tools, MCP and Evaluation · also in Working with AI Coding Agents: A Disciplined Loop, Building a Career on Capabilities, Not Tools, Engineering and Career: Start Here

Chamberlin's 1890 method of holding several explanations at once so you do not fall in love with one. Platt's strong inference (1964) extends it: devise alternatives, run a test that can exclude at least one, then repeat.

Learn it in Hypotheses, Predictions and Tests

A simple method without AI, such as keyword search over the same documents, run on the same task and measured first. If your system cannot beat it or usefully complement it, the extra machinery has not earned its place.

Learn it in From Assistant to Agent: Tools, MCP and Evaluation · also in Engineering and Career: Start Here

A job the day's eating must do, such as supplying protein, a legume, ALA, greens or flavour. Commit to the role and treat foods as interchangeable ways to fill it, but when substituting, check which nutrient the old food carried.

Learn it in The Daily Dozen and Roles Over Ingredients

A vague goal rewritten as behaviour someone could check: a watchable verb such as explain, ship, revert, cite or choose, plus a 'not yet' condition. 'Study networking' is a mood; 'name two causes and the observation that separates them' is a target.

Learn it in Deliberate Practice: Choose Targets You Can Observe · also in Engineering and Career: Start Here, Building a Career on Capabilities, Not Tools, Learning Mathematics: Procedures and the Six Abilities

Three kinds of statement to label separately: an observation is what a tool or person actually saw, an inference is what you concluded from it, and an assumption is what you had to believe to draw that conclusion. Fluent prose hides the difference.

Learn it in Technical Writing as Reasoning · also in Evidence-Based Troubleshooting: Separating Explanations, Observe, Define, Represent: The First Three Moves, Working with AI Coding Agents: A Disciplined Loop, Hypotheses, Predictions and Tests

Research that records what people already eat or do and follows their outcomes without assigning anything. It reveals associations but cannot rule out confounding, such as healthier people simply walking more.

Learn it in Reading Nutrition Evidence: What a Study Can and Cannot Show

Defining a concept by the procedure used to measure it (Bridgman, 1927), so a vague word like 'slow' or 'unstable' names a specific measurement and threshold. It turns arguments about words into checkable claims.

Learn it in Observe, Define, Represent: The First Three Moves

A scale that orders items without equal gaps between ranks or a true zero. A food scoring 90 is not shown to be any fixed amount healthier than one scoring 80, and gaps of one to three points are ties, not demonstrated differences.

Learn it in How to Rank Foods Honestly: A Transparent Scoring Method

The probability of data at least as extreme as those observed, assuming the null hypothesis and the model are true. It is not the probability that the null hypothesis is true, and statistical significance does not mean practical importance.

Learn it in Calculus and Beyond: Change, Structure and Uncertainty

The value below which a stated share of samples falls. 'p95 latency' means nothing until you name the samples, the aggregation and the time window; percentiles expose the tail that averages hide.

Learn it in Quantitative Reasoning with Stated Assumptions · also in Systems Thinking: Feedback, Queues, Information and Dynamics

Aristotle's practical wisdom: the perceptive skill of seeing what a particular situation calls for when rules run out. It differs from theoretical knowledge and from techne, skill in making, and cannot be fully written down as a procedure.

Learn it in Reasoning for What Ends? Wisdom, Character and the Examined Life

A lectin in raw or undercooked kidney beans that causes severe vomiting and diarrhoea. The FDA advises soaking dry beans at least five hours, discarding the water, then boiling in fresh water at least 30 minutes; slow cookers on low may not suffice.

Learn it in The Practical Kitchen: Staples, Meal Prep and Food Safety

Peirce's 1878 rule for making ideas clear: consider what practical effects the object of a concept might conceivably have, and your conception of those effects is your whole conception of it. A word that makes no conceivable practical difference has no clear meaning.

Learn it in What Does It Take to Know Something? · also in Peirce on the Fixation of Belief

The main result a trial commits in advance to test. Because measuring many outcomes makes some chance findings likely, positive secondary results cannot rescue a null primary result, as when a blueberry trial missed its primary outcome, insulin resistance.

Learn it in Reading Nutrition Evidence: What a Study Can and Cannot Show

Hume's argument that we observe only regular succession, never necessary connection, and cannot justify expecting the future to resemble the past without circular reasoning. Whether it has a satisfying answer is still disputed.

Learn it in Judgment Under Uncertainty: Probability, Causation and Forecasting

Instructions hidden in content a model reads, such as a web page, document or tool result, that it may obey as if legitimate. OWASP lists it first (LLM01) for LLM applications; no complete defence is known, so limit what a hijacked agent can do.

Learn it in From Assistant to Agent: Tools, MCP and Evaluation · also in Working with AI Coding Agents: A Disciplined Loop

The ISSN position stand (2017) suggests 1.4 to 2.0 grams of protein per kilogram of body weight daily for most exercising adults, with the upper end or more during an energy deficit. It is a planning heuristic, not a measured individual requirement.

Learn it in Protein for Muscle on a Plant-Based Diet

In the simplest single-server queue (M/M/1), average time in the system equals service time divided by (1 - utilisation), so delay explodes near capacity: about 10 times the service time at 90% busy and 100 times at 99%.

Learn it in Systems Thinking: Feedback, Queues, Information and Dynamics

A study that randomly assigns participants to an intervention or a control so the groups differ only by chance. It is the strongest design for cause and effect, but small samples, short durations and narrow populations limit what any one trial shows.

Learn it in Reading Nutrition Evidence: What a Study Can and Cannot Show

When chance contributes to a measurement, an extreme result tends to be followed by a less extreme one with no real change. Named by Galton in the 1880s, it breeds false explanations, such as crediting a remedy for recovery after a bad spell.

Learn it in Mathematics for Better Decisions

The number of clean repetitions you could still have completed when you end a set. Stopping two or three short of failure is easier to recover from, and evidence suggests training to failure is probably not required for muscle growth.

Learn it in Strength Training for Busy Lives: The Recovery-Based Cycle

Ashby's law that only variety can destroy variety: to hold outcomes steady, a regulator needs at least as many distinct responses as the distinct disturbances it must counter. A theorem in its formal setting, it is only an analogy for organisations.

Learn it in Systems, Decisions and Robust Design · also in Principles Ledger and Question Arcs

Recalling or solving from memory instead of rereading, which improves delayed retention even though rereading can look better on an immediate test (Roediger and Karpicke, 2006). Spacing spreads such practice over growing gaps; the best gap depends on how long you must remember.

Learn it in Placement, Repair and Retention: Keeping Mathematics Alive

Reviewed code is agent-written code you have read and can explain, including its assumptions; accepted code merely ran and you moved on. Only reviewed work counts as evidence of your skill, and accepted code should stay out of parts where failure is costly.

Learn it in Working with AI Coding Agents: A Disciplined Loop

The chance that a repeated activity carrying any risk of permanent loss eventually suffers it, which is why survival comes before optimising. Losses are asymmetric: a 50% loss needs a 100% gain to recover.

Learn it in Mathematics for Better Decisions

Re-running a model or decision while varying inputs or weights, say by ±20%, to see whether the conclusion survives. A winner that keeps changing means the options are close; it tests the weights, not whether you chose the right criteria.

Learn it in Twelve-Week Arcs and the Capstone Project Method · also in Systems Thinking: Feedback, Queues, Information and Dynamics

The property of a chaotic but deterministic system whose trajectories diverge from nearly identical starts, so long-range prediction fails despite exact rules. In the logistic map at r = 3.9, starts differing by one part in a million separate within a few dozen steps.

Learn it in Systems Thinking: Feedback, Queues, Information and Dynamics · also in A Map of Mathematics: Twenty-Five Areas and What Depends on What

A trend that holds within every subgroup but reverses when the groups are pooled. In a kidney-stone study, open surgery did better for both small and large stones yet worse overall, because stone size affected both treatment choice and success.

Learn it in Systems Thinking: Feedback, Queues, Information and Dynamics

A simulation is evidence about a specified model under chosen parameters; a proof covers every case in a stated domain. Simulations suggest conjectures and test models but establish neither theorems nor facts about the world.

Learn it in A Map of Mathematics: Twenty-Five Areas and What Depends on What · also in Proof and Precise Reasoning: From Arguments to Theorems, Calculus and Beyond: Change, Structure and Uncertainty, Mathematics: Start Here, Systems Thinking: Feedback, Queues, Information and Dynamics

A copy of a fact, such as a cached DNS answer, an identity-to-address mapping or a monitoring sample, that was true when recorded and may since have changed. Ask its age, its source of truth and what fresh evidence to compare it with.

Learn it in Stale Information and Dependable Reports

Drawing conclusions only from cases that passed a selection filter while ignoring those that dropped out. In the popular retelling of Wald's wartime work, armour belongs where returning planes show little damage, because planes hit there did not return.

Learn it in Mathematics for Better Decisions

An intensity gauge for easy cardio: you should be able to speak in full sentences. In small studies, the point where talking becomes hard tracks the ventilatory threshold fairly well, making it a free substitute for heart-rate zones.

Learn it in Daily Walking and Easy Cardio

Perrow's two conditions for 'normal accidents': little slack between steps, and unexpected, hard-to-see interactions among parts. Together, he argued, they make serious accidents eventually inevitable in high-risk systems, a claim later safety research partly disputes.

Learn it in Systems, Decisions and Robust Design

A trend is the direction of many repeated readings, such as three to four weeks of averaged weigh-ins; a single reading is noisy, since water, gut contents and salt can move body weight a pound or more in a day. Act on the trend.

Learn it in Gradual Fat Loss and Maintenance: A Method

The seven liberal arts: the trivium (grammar, logic, rhetoric) for language and argument, and the quadrivium (arithmetic, geometry, music, astronomy): in the classical scheme, number in itself, magnitude at rest, number in relation and magnitude in motion. Without the quadrivium, the trivium never meets resistance.

Learn it in The Seven Liberal Arts, Hands-On: Logic, Proof and the Quadrivium · also in Mathematics: Start Here, Engineering and Career: Start Here, Proof and Precise Reasoning: From Arguments to Theorems, A Map of Mathematics: Twenty-Five Areas and What Depends on What, English Expression Mastery: Words, Sentences, Paragraphs and Speech, Geometry, Euclid and Trigonometry, Choosing a Language to Study for Reading: Latin, Greek, French, German and Beyond, Technical Writing as Reasoning

The rule that missing, stale or unpriced data must never silently count as healthy or zero. Overall status is green only if every expected target is fresh and healthy, and a cost without a known price is reported as unknown in the total.

Learn it in Stale Information and Dependable Reports

An argument is valid if its conclusion must be true whenever its premises are, whether or not they actually are; it is sound if it is valid and its premises are true. A valid argument with a false premise can reach a false conclusion.

Learn it in Proof and Precise Reasoning: From Arguments to Theorems · also in The Seven Liberal Arts, Hands-On: Logic, Proof and the Quadrivium

A vitamin with no reliable unfortified plant source, so vegans need a fortified food or a supplement with a stated dose, taken regularly. Feeling well is not evidence of adequacy, because deficiency can develop slowly over years.

Learn it in Nutrient Coverage on a Plant-Based Diet · also in A Plant-Based Eating Framework You Can Repeat

A workflow follows a path fixed in code, with some steps calling a model; an agent directs its own process, choosing which tool to use next. Anthropic's 2024 guidance is to start simple and add autonomy only when evaluation shows it helps.

Learn it in From Assistant to Agent: Tools, MCP and Evaluation · also in Working with AI Coding Agents: A Disciplined Loop, Technology, AI and Human Judgment