Reading More Critically by Tracking Claims and Evidence

Stop reading like an audience watching a magic trick. Critical reading means walking onto the stage mid-performance and examining the apparatus. Stop asking what the writer wants you to believe. Start asking what the text requires you to accept for that belief to hold.

The method is a disciplined annotation. Mark every assertion, trace it to its evidence, name the warrant connecting them, and circle what the text assumes but does not state. A proposition can be factually true and still indefensible if the cited evidence does not reach it. The gap between what the evidence supports and what the assertion declares is the claim-defensibility gap. Closing it is the writer's job, not yours. Your job is to measure it.

Apply four passes to any nonfiction text. Do not attempt them simultaneously. Editing and writing are different cognitive acts, and critical reading requires the same separation: evaluating is not absorbing. You cannot do both well at once.

Pass One: Isolate The Claims

Underline every declarative statement that asserts a proposition offered for acceptance. Ignore transitions, examples, and throat-clearing. You are extracting the argument's skeleton. A 2000-word piece yields 15 to 40 assertions.

Pass Two: Tag The Evidence

For each assertion, draw an arrow to the line that supplies the reason for accepting it. If no line does, mark the assertion as unsupported. At this stage, 30 to 60 percent of declarative statements in a typical "evidence-based" article show no traceable source. The gap is not an accident. It is the structural consequence of publishing economics that eliminated institutional fact-checking workflows.

Pass Three: Extract The Warrant

The warrant is the often-unstated principle that makes the evidence relevant to the assertion. Write it in the margin. A stated warrant that is itself contestable and undefended can be more misleading than an absent one, because it signals completeness where none exists. Score each warrant on explicitness: absent, implied, stated, or defended.

Pass Four: Identify The Presuppositions

These are the propositions the line requires to be true for the line to be evaluable at all. "The king of France is bald" and "The king of France is not bald" both presuppose there is a king of France. Measure the presupposition load: a line with a high count can be neither affirmed nor denied without accepting its unstated frame.

The attribute set from the Taking Down Words dossier gives you a shared vocabulary for what you find. These are measurements you apply line by line.

Score Source Traceability

Use a three-point scale: untraceable, traceable to a named source, or traceable to primary evidence. A citation to a named publication is not traceable if the publication is inaccessible or the assertion does not appear in it. "According to" followed by a prestigious outlet often obscures a chain of sourcing in which the outlet reported a claim it did not independently verify.

Measure Attribution Specificity

Use a five-point scale from no source to named study with access. "Experts say" functions as a pseudo-attribution that invokes authority without accountability. "Studies show" is barely better. Insist on named researchers and accessible documents.

Calculate Evidence-to-Claim Latency

Count the lines between the introduction of evidence and the assertion it supports. Shorter is better. Long latency allows intervening material to reshape what the reader thinks the evidence established. A text that buries its evidence three paragraphs before its assertion is asking you to forget what the evidence actually said.

Take this line: "A growing body of evidence suggests that mindfulness reduces workplace burnout."

Claim: Mindfulness reduces workplace burnout.
Evidence: None provided. The phrase "a growing body of evidence" is a pseudo-evidence marker. It invokes a research base without citing it, and "growing" is unfalsifiable because any addition constitutes growth.
Warrant: Implied. The warrant would be something like: "Whatever a growing body of evidence suggests is probably true." That warrant is contestable: a growing body of weak studies does not strengthen a conclusion.
Presuppositions: At least three. That mindfulness is a stable, measurable construct. That workplace burnout is a stable, measurable construct. That a body of research exists and has been fairly characterized.
Claim-defensibility gap: Wide. The evidence supplied (zero) does not reach the assertion declared. The line is not necessarily false, but it is structurally indefensible as written.

Now compare: "A 2019 meta-analysis of 23 randomized controlled trials, published in the Journal of Occupational Health Psychology, found that mindfulness-based interventions reduced emotional exhaustion scores by a small but significant effect size (d = 0.32) relative to passive controls, with no significant effect on depersonalization scores." The claim-defensibility gap has narrowed considerably. The warrant explicitness has moved from implied to stated. The source is traceable to primary evidence. The qualifier density has increased from zero hedging terms to four in one line, which is appropriate for the genre.

Every text rests on unstated premises. Writing necessarily selects some propositions for explicit statement and leaves others to be supplied by the reader. This is not a flaw; it is a condition of communication. The question is whether those unstated premises are benign or whether they do argumentative work the writer prefers not to own.

Apply negation. Restate the line in the negative and observe what remains. "The failure of the policy demonstrates the administration's incompetence" becomes "The failure of the policy does not demonstrate the administration's incompetence." Both versions presuppose the policy failed. If you do not accept that presupposition, you cannot engage the line on its own terms. You must first challenge the frame.

A paragraph with uniformly high presupposition load functions like a stacked deck. Each line requires you to accept the previous line's unstated frame before you can evaluate the next. By the third line, you have accepted three invisible premises and the conclusion feels inevitable. The charitable-reading principle requires you to engage the text in its strongest plausible formulation, but it does not require you to accept its presuppositions unchallenged.

Attribute	Scale	Better Direction	What to Watch For
Source Traceability	Untraceable → Named Source → Primary Evidence	Higher	Named source may be inaccessible or misrepresented
Warrant Explicitness	Absent → Implied → Stated → Defended	Higher	Stated but undefended is often more misleading than absent
Claim-Defensibility Gap	Distance from evidence to assertion	Smaller	Factually true claims can still be indefensible
Attribution Specificity	No Source → 'Experts' → Institution → Named Researcher → Named Study	Higher	'Studies show' is a pseudo-attribution
Presupposition Load	Count of propositions required for evaluability	Lower	High load forces acceptance of unstated frames
Evidence-to-Claim Latency	Lines between evidence and assertion	Shorter	Long latency lets intervening material reshape memory

You do not need to know what the text left out. You need to know where omissions predictably occur. The Taking Down Words dossier identifies several structural patterns that create characteristic gaps.

The End-Note Problem

When a text's caveats, limitations, and uncertainties are confined to footnotes, appendices, or final paragraphs while the body text speaks with unqualified certainty, the reader's impression is shaped by the body, not the caveats. Check whether the headline's confidence survives contact with paragraph 27. The headline-body gap is a permanent feature of publishing, not a bug. Budget for it on every piece you read.

The Abstraction Ladder Trap

Language moves between concrete specifics and high-level abstractions. Arguments that stay at one level fail in characteristic ways: pure abstraction cannot be tested; pure concreteness cannot generalize. A text that opens with "The healthcare system is broken" and never descends to a specific hospital, a specific procedure, or a specific patient outcome is running an argument on pure abstraction. Spot the omission by asking: what would a counterexample look like, and could the text accommodate one?

The Motte-and-Bailey Structure

A controversial claim (the bailey) is defended by retreating to a trivially true claim (the motte) when challenged, then re-extended when the challenge passes. The structure is detectable in text alone if you track whether the key term shifts meaning. When a piece argues that "we need to abolish the police" and, under criticism, clarifies that it means "we need to rethink public safety funding," the motte-and-bailey detectability is high. Measure equivocation chain length: count the meaning shifts a key term undergoes. A short chain is honest; a long one produces conclusions that appear to follow but do not.

Critical reading nonfiction demands a default stance of non-acceptance, not disbelief. The burden is on the text to move you from non-acceptance to acceptance through evidence and reasoning. This is the null hypothesis discipline applied to reading: assume no effect, no connection, no claim established until the evidence meets a specified threshold. You are not obligated to disprove the text; the text is obligated to prove itself.

Apply the discipline to every claim the text makes about "the other side." Measure straw-person distance: the semantic divergence between an opponent's actual claim and the version presented for refutation. This is the single most common failure in public argument and is almost never measured in practice. When a text says "critics say," ask: which critic? Where? In what publication? The phrase frequently introduces a single critical tweet, an anonymous comment, or a straw-person version of criticism the writer has not actually sourced.

Check counterargument integration depth on a five-point scale: absent, mentioned, summarized, engaged, or structurally incorporated. A counterargument summarized only to be dismissed is not engagement. Structural incorporation means the argument's shape changes in response to the objection. Most published arguments never reach the fourth level. If a piece you are reading mentions an objection and then pivots immediately to "however," look for the missing step: did the argument actually change, or did the writer just signal fairness and move on?

Two mechanical checks reveal structural problems that prose style conceals.

Calculate The Nominalization Rate

Count verbs turned into nouns per 100 words. Nominalizations hide agency by removing the actor. "They decided" names who did what. "The decision was made" removes the actor from the line. A text with a high nominalization rate is often evading responsibility. This is a grammatical mechanism, not a political observation, and it operates in all registers of English.

Score The Qualifier Density

Count hedging terms per 100 words. Too few qualifiers overstate certainty; too many can paralyze the prose. The right density varies by genre and audience expectation, but within a single piece, a sudden drop in qualifier density at a key conclusion is a red flag. The text is shifting from careful to confident without earning the shift.

Test The Given-New Contract

Readers expect lines to begin with information they can identify (given) and end with information that is new. Violations produce prose that feels disjointed regardless of its logical validity. When a text feels slippery or hard to follow, map the given-new structure of a paragraph. If every line begins with new information, the writer is not managing your cognitive load. The reader-effort-to-comprehension ratio is failing.

You cannot annotate every piece you read. Build the method into a three-tier habit.

Tier One: The Scan

For daily news and social media, apply one check: source traceability. Before accepting a claim, ask whether you can trace it. If you cannot, hold it at non-acceptance. This takes five seconds per article and catches 70 percent of structural problems.

Tier Two: The Single-Pass Annotation

Once a week, take a piece of longform nonfiction and perform passes one and two: claims and evidence. This takes 20 to 40 minutes. You will find that even well-regarded publications routinely fail to connect their claims to their evidence.

Tier Three: The Full Four-Pass Annotation

Once a month, perform all four passes on a single article or book chapter. Extract warrants, identify presuppositions, measure straw-person distance, and score counterargument integration depth. This is the practice that builds the diagnostic instinct. After six months, you will see structural problems on first reading that you previously needed annotation to detect.

Do this next: take the last longform article you read and trusted, print it out, and perform pass two. Underline every declarative claim, then draw an arrow to its evidence. Count how many claims have no arrow. That number, divided by the total, is your baseline for how much of what you read is asserted rather than supported. Most readers land between 40 and 60 percent on their first attempt. The number itself is the lesson.