Level 5 — Advanced · Technical & Conceptual Reading
Lesson 74: Advanced Reading: Argument Strength, Evidence Sufficiency & Burden of Proof
The third and final part of this course's argument-analysis arc — judging not whether evidence exists but whether it's sufficient, weighing evidence quality across five dimensions, and checking whether an author's conclusion is properly calibrated to what their evidence actually supports.
Learning Objectives
By the end of this lesson, you should be able to:
- Distinguish evidence sufficiency from evidence existence, and read burden of proof as scaling with how strong or universal a claim is
- Apply five evidence-quality dimensions — relevance, reliability, representativeness, sufficiency, consistency — to judge whether a piece of evidence actually supports its claim
- Recognise triangulation — independent, converging evidence — as strengthening a claim, and actively look for conflicting evidence rather than only evidence that supports the author's position
- Rank evidentiary language — consistent with, suggests, supports, demonstrates, proves — by strength, and distinguish 'no evidence for X' from 'evidence against X'
- Judge whether an author's conclusion is properly calibrated to their evidence's actual strength and scope, and state explicitly what a passage's evidence does NOT establish
Introduction
Lesson 73 gave you a named taxonomy of argument gaps and a test for which hidden assumptions actually matter. This lesson closes out the argument-analysis arc that ran through Lessons 72–74 with the question underneath all of it: how strong is this argument, really? Because evidence being true is not the same as evidence being sufficient — an argument's evidence can be entirely accurate and still fall short of supporting the conclusion built on top of it.
Evidence Existing Is Not the Same as an Argument Being Strong
The question isn't 'is there evidence' — it's 'how much, and of what kind'
A six-month study found that automated billing reduced processing time by 28%. Therefore, the company should replace its entire billing system. Evidence exists. But ask: how big was the study? How many teams? Was there a comparison group? Was only processing time measured? What happened to accuracy? What's the cost? Is the long-term effect known? Is there another explanation? Evidence exists ≠ evidence is sufficient.
What Is Evidence Sufficiency?
Whether the evidence is enough to support the specific conclusion drawn
One employee reported that the new system was easier to use. That's an observation. Therefore, all employees will prefer the new system is not adequately supported by it — one employee's report scaling up to "all employees" is a scope mismatch, exactly the kind of leap this course has repeatedly flagged.
Burden of Proof
The claim-maker owes the support — and stronger claims owe more of it
Burden of proof means: whoever makes a claim carries responsibility for supplying adequate support for it. "This software will reduce operational costs by 40%" is a specific claim — it demands specific evidence for that 40% figure. "The software is efficient" alone does not meet that burden. As a general rule, the more specific or forceful the claim, the heavier the evidence requirement behind it.
Claim Strength Determines the Evidence Burden
A ladder from weak claim to very strong claim, each with a different evidence requirement
Weak claim: the system may improve efficiency — limited evidence can be enough. Moderate claim: the system can improve efficiency under certain conditions — needs a mechanism plus supporting evidence. Strong claim: the system will significantly improve efficiency across all terminals — needs considerably more evidence. Very strong claim: the system always improves efficiency — a universal claim, and one of the hardest kinds of claim to actually establish.
Universal Claims Deserve Extra Scrutiny
One counterexample can break a universal claim entirely
Watch for: always, never, all, none, every, completely, inevitably, guaranteed, proves, must. Automation always reduces costs — a single counterexample can challenge this. If one automated system's maintenance cost is high enough to raise total cost, "automation always lowers cost" is no longer valid, full stop. Universal claims are uniquely fragile in exactly this way.
Quantifier Strength, Revisited
Connecting back to Lessons 65, 66, and 69
Keep this ordering active while reading: all > most > many > some, and always > usually > often > sometimes > rarely. All ports use automation is the strongest claim in that family; some ports use automation is considerably weaker. Preserve the author's exact quantifier rather than rounding it up or down in your own summary.
Five Dimensions of Evidence Quality
Ask all five questions of any piece of evidence
Relevance Is Not Sufficiency
Evidence can be genuinely on-topic and still not be enough
Three experienced billing officers tested a new system and liked it. That's relevant. Is it sufficient? Probably not — management, new employees, IT staff, and finance staff may all have quite different experiences with the same system. Relevant evidence can still be insufficient evidence.
Reliability
Objective-looking data still needs evaluation
Compare "employees said the system was faster" with "system logs showed that average processing time decreased by 28%." The second is closer to an objective measurement — but it isn't automatically perfect either. Ask about the measurement method, the sample period, missing data, system changes during the period, and workload differences. Objective-looking data still requires evaluation, not automatic trust.
Representativeness
Connecting back to Lesson 73's generalization gap
A new system was tested in three large terminals. Therefore, it will work equally well in every small terminal. Large and small terminals can differ in workload, staffing, infrastructure, procedures, and cargo patterns. Sample ≠ population, unless the evidence actually supports that generalization.
Sample Size Is Not the Same as Representativeness
A big number is not automatically a good sample
10,000 randomly selected users reported similar results can be a genuinely strong sample. 100,000 users from a highly biased sample reported similar results can be misleading despite being ten times larger. Large sample ≠ automatically representative sample — size and representativeness are two separate qualities, and a huge biased sample is still a biased sample.
Corroboration and Triangulation
Independent sources pointing the same direction genuinely strengthen a claim
If system logs show fewer errors, audit records show fewer billing corrections, user reports show fewer complaints, and an independent review finds a similar improvement, these different, independent sources corroborate each other. This is broadly called triangulation — examining the same phenomenon through several different kinds of evidence at once. Triangulated evidence is genuinely stronger than any single source alone, precisely because each source's individual weaknesses are unlikely to all point the same false direction by coincidence.
Conflicting Evidence
Advanced reading means actively looking for evidence against, not just evidence for
Automated billing reduced processing time by 28%. But maintenance costs increased by 35%. One metric improved; another worsened. Overall evaluation is no longer simple — a genuine trade-off has appeared. A careful reader doesn't stop at the first supporting fact; they actively check whether the passage (or their own knowledge) contains evidence pulling in the opposite direction.
Evidence Weight: A Rough Hierarchy
From weakest to strongest — a useful default, not an absolute law
Context can still shift this ordering
This hierarchy is a strong default, not an absolute rule — the actual quality of evidence within any one category can vary enormously depending on how it was gathered.
"Proves" Versus "Supports": An Evidentiary Verb Ladder
Connecting back to Lessons 66, 69, and 71 — now arranged as one explicit scale
Supports — the evidence supports the claim; it points in the claim's favour. Strongly supports — more convincing than plain support. Consistent with — compatible with the claim, though other explanations may fit equally well. Suggests — points toward a possible interpretation, with low asserted certainty. Demonstrates — stronger language, closer to established. Proves — an extremely strong claim, especially in empirical or real-world contexts, and one that deserves real scrutiny rather than automatic acceptance.
"Consistent With"
Compatible with a claim is not the same as proof of it
The findings are consistent with the hypothesis that automation improves efficiency means the findings fit the hypothesis — it does not mean the findings prove the hypothesis. The same findings could just as easily be consistent with a different explanation entirely.
"No Evidence" Versus "Evidence Against"
Two genuinely different epistemic positions, easy to conflate
There is no evidence that the system improves accuracy means no supporting evidence has been found so far — an absence. There is evidence that the system reduces accuracy means evidence actively points against the claim — a presence, pulling the other way. No evidence ≠ evidence against; collapsing the two turns "we haven't found support yet" into a much stronger, unearned negative claim.
"Insufficient Evidence," Revisited
Connecting back to Lesson 73
The evidence is insufficient to establish the claim means the claim hasn't been shown false — only that current evidence doesn't yet establish it. A three-month pilot suggests that the system may reduce costs, but the evidence is insufficient to establish its long-term cost-effectiveness is precise, careful academic English, worth imitating in your own reading notes.
Conclusion Strength Should Not Exceed Evidence Strength
A rule worth applying to every argument you read from now on
If the evidence says "a pilot suggests…," the conclusion should read "the system may be beneficial" — not "the system will definitely be beneficial everywhere." Whenever a conclusion's certainty outruns its evidence's actual strength, that gap is exactly what a critical reader should flag.
Argument Calibration
Matching claim strength to evidence strength, in either direction
Calibration asks whether an author's claim is properly matched to their evidence's actual strength. Good calibration: limited evidence → a cautious conclusion. Poor calibration: limited evidence → an extremely strong conclusion. Two employees preferred the new interface. Therefore, the interface is universally superior is badly calibrated — a tiny, non-representative sample supporting a sweeping universal claim.
A Full Worked Example
Evidence, supporting factors, limitations, and a carefully calibrated final claim
A six-month pilot of an automated billing system across four terminals reduced average processing time by 28%. System logs also showed a moderate decline in billing corrections. However, the pilot did not measure long-term maintenance costs, and the participating terminals had relatively standardized workflows. These findings therefore provide encouraging evidence that automation can improve billing efficiency, but they do not establish that the system will be cost-effective or equally effective across all terminals.
Evidence: six-month pilot, four terminals, processing time down 28%, billing corrections down. Supporting factors: system logs, multiple terminals. Limitations: long-term maintenance unmeasured, standardized workflows. What the evidence supports: automation can improve billing efficiency. What it does not establish: cost-effectiveness, or equal effectiveness everywhere. Author stance: cautiously positive. Calibration: excellent — the conclusion is deliberately kept inside the evidence's actual boundary.
"What Does the Evidence NOT Establish?"
A new, powerful question to hold alongside 'what does it establish?'
Evidence: processing time decreased. It may establish: faster processing occurred during the pilot. It may not establish: lower total cost, higher accuracy, long-term reliability, employee satisfaction, universal effectiveness, or causation without a properly controlled design. Asking this second question explicitly is what turns a competent reading into a genuinely sophisticated one.
The Evidence Boundary
Every piece of evidence has an edge — name it before generalising past it
Given: 4 teams, 6 months, automated billing, processing time. Evidence boundary: 4 teams + 6 months + processing time. Anything beyond that boundary — all teams, 5 years, total cost — requires additional support before a reader should accept it. Naming the boundary explicitly is a concrete way of catching the generalization and time gaps from Lesson 73 the moment they appear.
A Master Test for Argument Strength
Ten steps for rating how strong any argument actually is
Lessons 72–74, Combined: A Master Framework
The full argument-analysis architecture built across this three-lesson arc
Vocabulary in Context
burden of proofnoun phraseformal
the responsibility of whoever makes a claim to supply adequate supporting evidence for it, scaling with how strong the claim is (প্রমাণের দায়ভার)
Often with bear the burden of proof, shift the burden of proof, burden of proof lies
“The burden of proof lies with the company, which must show that the shipment was properly packed.”
evidence sufficiencynoun phraseformal
whether the amount and quality of evidence available is actually enough to support a specific conclusion (প্রমাণের পর্যাপ্ততা)
Often with assess evidence sufficiency, evidence sufficiency for, question evidence sufficiency
“The auditors questioned evidence sufficiency, since only three invoices were sampled.”
triangulationnountechnical
examining the same claim through several independent kinds of evidence, which strengthens confidence when they converge (ত্রিভুজীকরণ)
Often with data triangulation, use triangulation, triangulation of evidence
“Triangulation strengthened the finding: gate logs, driver interviews, and invoices all pointed to the same delay.”
representativenessnountechnical
the degree to which a sample reflects the characteristics of the broader population it is meant to stand in for (প্রতিনিধিত্বশীলতা)
Often with question representativeness, representativeness of the sample, ensure representativeness
“The representativeness of the survey was questionable because only day-shift staff replied.”
calibrationnountechnical
how well the strength of a stated conclusion matches the actual strength of the evidence behind it (সমন্বয়/ক্রমাঙ্কন)
Often with calibration of confidence, poor calibration, careful calibration
“Good calibration means the report says 'may' when evidence is weak and 'shows' when it is strong.”
evidence weightnoun phrasetechnical
how much confidence a particular kind or source of evidence deserves, ranging roughly from anecdote to replicated, converging findings (প্রমাণের ওজন)
Often with give more evidence weight, evidence weight of, assess evidence weight
“A replicated trial carries more evidence weight than a single anecdote.”
evidence boundarynoun phrasecourse term
the exact limits — population, time period, measured variable — within which a piece of evidence actually applies (প্রমাণের সীমা)
Often with define the evidence boundary, stay within the evidence boundary, evidence boundary of
“The evidence boundary is limited to export cargo in 2023, so it says nothing about imports.”
conflicting evidencenoun phraseformal
evidence that points in a different or opposite direction from an argument's main supporting evidence (পরস্পরবিরোধী প্রমাণ)
Often with conflicting evidence, weigh conflicting evidence, face conflicting evidence
“The panel faced conflicting evidence: gate logs showed delays, while driver surveys reported none.”
Guided Reading Practice
Read this passage once, straight through. Then rate its evidence against this lesson's five quality dimensions, identify any conflicting evidence, and state its evidence boundary, before checking your reading against the notes that follow.
A logistics company tested a new route-optimization algorithm on its 12 busiest delivery routes over four months. Fuel costs on those routes fell by 14%, and drivers reported the new routes felt more efficient. However, average delivery times on three of the twelve routes increased slightly, which the company attributed to unrelated road construction. Independent GPS tracking data confirmed the fuel savings figure, closely matching the company's internal reports. Based on these results, the company announced plans to apply the algorithm to its entire national delivery network of over 900 routes.
Relevance: fuel-cost and delivery-time data are directly relevant to a route-optimization claim. Reliability: independent GPS tracking corroborating the company's own reported figures is a genuine strength — a small triangulation. Representativeness: the 12 busiest routes are very unlikely to represent the full diversity of a 900-route national network — a clear representativeness gap. Sufficiency: four months on 12 routes is a real but narrow evidence base for a claim covering 900 routes nationwide. Conflicting evidence: delivery times worsened on three routes — attributed to construction, but not independently verified as unrelated, which is itself worth flagging. Evidence boundary: 12 busiest routes, 4 months, fuel cost and (mostly) delivery time. The rollout decision to all 900 routes reaches well past that boundary — a calibration problem the passage's own evidence doesn't yet support.
Golden Rule
Golden Rule
The strength of a conclusion must be proportional to the strength, relevance, reliability, and scope of the evidence supporting it — so never ask only "is there evidence?"; ask "is this evidence enough for the conclusion the author wants me to accept?"
Lesson Summary
This lesson closed out the three-lesson argument-analysis arc by moving from finding gaps and assumptions to judging overall argument strength. You learned that evidence existing is not the same as evidence being sufficient, and that burden of proof scales with how strong or universal a claim actually is. You applied five evidence-quality dimensions — relevance, reliability, representativeness, sufficiency, consistency — and learned that relevant evidence can still be insufficient, that large samples aren't automatically representative, and that triangulated, independently converging evidence is genuinely stronger than any single source. You practised actively hunting for conflicting evidence rather than stopping at the first supporting fact, ranked evidentiary language from "consistent with" to "proves," and separated "no evidence for" from "evidence against." You closed on calibration — checking whether a conclusion's certainty actually matches its evidence's real strength and boundary — and on the question that ties the whole arc together: what does this evidence NOT establish?
What Lessons 72 through 74 have built together
Lesson 72 gave you the anatomy of a single argument. Lesson 73 gave you a taxonomy for naming exactly where an argument's bridge is missing. This lesson gave you a way to weigh how much that missing bridge — or present, well-supported one — actually matters. Together, they replace a vague sense that "this argument seems okay" or "something feels off" with a precise, defensible verdict: what the evidence establishes, what it does not, and whether the author's conclusion has stayed properly inside that boundary.
Practice: Test What You've Learned
Work through every question yourself before checking anything.
Before you start
For every argument below, name which of the five evidence-quality dimensions is most in question, and state explicitly what the evidence does and does not establish.
Part A — Sufficiency and Burden of Proof
- A single customer review states a product "changed my life." A retailer features it as proof the product works for everyone. Explain why this evidence fails the sufficiency test.
Show answer
One customer's testimonial is a single anecdote, and an anecdote cannot represent the range of experiences, needs, and reactions across an entire customer base. The claim "works for everyone" is universal, so it demands a broad, representative evidence base — one glowing review falls far short of that requirement. - A startup claims its app "will increase user engagement by 300%." Explain why this specific claim carries a heavier burden of proof than a vaguer claim like "the app may help engagement."
Show answer
"Will increase engagement by 300%" is a specific, forceful, near-certain claim, so it demands specific quantitative evidence establishing exactly that figure. "May help engagement" is a weak, hedged claim, so even limited or preliminary evidence can be enough to support it — the more precise and confident the claim, the more the evidence must match it. - Explain, in your own words, why a universal claim (always, never, all) is especially vulnerable to a single counterexample.
Show answer
A universal claim asserts something holds in every single case with no exceptions, so it only takes one genuine exception to make the claim as stated false. Unlike a claim using "most" or "usually," which can absorb some exceptions and remain true, "always" or "never" leaves no room at all for a counterexample to coexist with the claim.
Part B — The Five Evidence-Quality Dimensions
- A company surveys only its most loyal customers and concludes overall customer satisfaction is high. Identify which evidence-quality dimension is weakest here.
Show answer
Representativeness is the weakest dimension. Loyal customers are exactly the group most likely to already be satisfied, so this sample cannot stand in for the broader customer population, including occasional or dissatisfied customers who were excluded from the start. - A lab result is confirmed by three independent research teams using different methods. Identify what this lesson calls this pattern, and explain why it strengthens the claim.
Show answer
This lesson calls this pattern triangulation. It strengthens the claim because each team's individual weaknesses or blind spots are unlikely to all point toward the same false result by coincidence, so their convergence on the same finding through different methods gives the claim considerably more credibility than any one study alone. - A dataset has 50,000 entries, but all of them come from a single region during one unusual month. Explain why sample size alone does not resolve the problem here.
Show answer
Sample size and representativeness are two separate qualities: a large sample drawn entirely from one region and one unusual time period is still a narrow, biased sample, no matter how many entries it contains. A huge biased sample is still a biased sample, so the 50,000 figure cannot compensate for the lack of geographic or temporal variety.
Part C — Evidentiary Language
- Rank these three claims from weakest to strongest: "the data are consistent with the theory," "the data support the theory," "the data prove the theory."
Show answer
Weakest to strongest: "the data are consistent with the theory" (merely compatible, other explanations may fit equally well), then "the data support the theory" (points in the theory's favor), then "the data prove the theory" (an extremely strong claim that deserves real scrutiny before being accepted). - A report states there is "no evidence that the additive is harmful." Explain why this is a different claim from "there is evidence that the additive is safe."
Show answer
"No evidence that the additive is harmful" is an absence — nothing has yet been found pointing toward harm, which may simply mean it hasn't been thoroughly studied. "There is evidence that the additive is safe" is a presence — positive, supporting evidence actively pointing toward safety. Collapsing the two turns an unstudied gap into an unearned positive claim. - Explain why "the findings are consistent with X" does not rule out other explanations for the same findings.
Show answer
"Consistent with" only means the findings fit comfortably alongside explanation X — it does not mean X is the only explanation the findings could fit. A different explanation, Y, could just as easily produce the identical findings, so "consistent with" leaves the door open to competing interpretations rather than confirming one over the others.
Part D — Conflicting Evidence and Calibration
- A new hiring process reduced time-to-hire by 40%, but exit interviews from new hires show declining satisfaction with the process's transparency. Identify the conflicting evidence, and explain why it matters for an overall evaluation.
Show answer
The conflicting evidence is the declining transparency satisfaction in exit interviews, which pulls against the positive 40% time-to-hire figure. It matters because an overall evaluation can no longer rest on speed alone — a genuine trade-off has appeared between efficiency and candidate experience, and both need to be weighed together rather than reporting only the favorable metric. - A pilot study of 30 patients shows promising results. A press release states the treatment "is proven effective." Explain the calibration problem in this scenario.
Show answer
A 30-patient pilot is limited, preliminary evidence, which should support a cautious claim like "the treatment shows promise" — instead, the press release asserts "proven effective," one of the strongest possible claims. This is poor calibration: the certainty of the language vastly exceeds the actual strength and scope of the underlying evidence. - Explain why "the evidence is insufficient to establish the claim" is a more defensible statement than "the claim is false," using this lesson's terms.
Show answer
"The claim is false" is itself a strong positive assertion that carries its own burden of proof — evidence would be needed to establish the opposite of the claim. "The evidence is insufficient" makes no claim about the truth either way; it only says current evidence hasn't met the burden of proof the claim requires, which is a narrower, more careful, and more defensible statement.
Part E — Full Passage Analysis
Return to this lesson's guided reading passage about the route- optimization algorithm.
- State the passage's evidence boundary in your own words.
Show answer
The evidence boundary is the 12 busiest delivery routes, tested over four months, measuring fuel costs and (to a lesser extent) delivery times. Any claim about the other roughly 888 routes, about periods longer than four months, or about metrics beyond fuel cost and delivery time sits outside what this evidence actually covers. - Identify the conflicting evidence in the passage, and explain why the company's explanation for it deserves scrutiny.
Show answer
The conflicting evidence is that delivery times increased slightly on three of the twelve routes. The company attributed this to unrelated road construction, but that explanation was not independently verified — it is simply the company's own account, so a careful reader should treat it as a claim needing support, not an established fact. - Identify the passage's strongest evidence-quality dimension and its weakest, and justify both choices.
Show answer
Strongest: reliability, since independent GPS tracking data corroborated the company's own reported fuel savings, a genuine triangulation. Weakest: representativeness, since the 12 busiest routes are very unlikely to represent the full diversity of a 900-route national network, making the evidence base narrow relative to the scale of the proposed rollout. - Explain why the decision to roll out the algorithm to all 900 routes is a calibration problem, not simply an unreasonable decision.
Show answer
The underlying evidence is genuinely encouraging — real fuel savings, corroborated by independent data — so the decision isn't baseless. The problem is that the certainty behind a full 900-route rollout outruns what a 12-route, four-month pilot can actually support; the conclusion's scope and confidence have exceeded the evidence's actual boundary, which is precisely what calibration measures.
Next: Lesson 75 — Advanced Reading: Evidence Hierarchy, Source Quality & Triangulation: the fourth part of this course's argument-analysis arc, tracing cited evidence back to its actual origin and weighing it honestly.