Master Art of Translation | The Comparative Baseline Fidelity System — How More Than, Less Than, As Much As and Rather Than Preserve Standards

How do you translate comparisons accurately when a sentence says more than, less than, as much as, at least as high as, no more than, twice as much, better than expected or rather than? Start by identifying the compared items, the property being measured, the reference group or baseline, the direction of the comparison, its numerical or qualitative threshold, and its intended force. Then rebuild that relationship in natural target-language expression. A translation that keeps the numbers but changes the standard of comparison can still give a reader the wrong decision.

Readers searching for comparative translation, how to translate more than and less than, as…as across languages, comparative clauses, translation of no less than or no more than, percentage changes, degree modifiers, comparisons in contracts, and how to avoid mistranslation of statistical claims need more than a list of equivalent words. They need a method for identifying what is being compared with what, including whether a sentence compares two objects, an object with its own past, an observed result with an expectation, or a quantity with a regulatory limit. This practical guide supplies comparison maps, a disciplined workflow, numerous original semantic-control examples, failure tests and cross-domain workshops.

The hidden problem is baseline drift. In a polished sentence, a translator can preserve the adjective, the number and the general topic while moving the comparator, reversing the inequality, changing a percentage into a percentage-point difference, confusing a maximum with a prediction, or turning a preference into a measurable advantage. Comparative words also hide denominator choices: per person, per item, per period and per eligible case cannot be silently substituted. Our aim is not to teach comparative grammar inside a single language. It is to preserve the comparison relation while meaning crosses languages, media and editorial environments.

Quick navigation: First isolate the comparison claim. Then map the baseline, scale and group. Run the controlled rewrite and the negative test. Audit high-risk patterns, work through the casebook, practise transfers and return to the larger translation map. All numeric scenarios below are constructed demonstrations, not claims about real products, policies, populations, scientific results or services.

Architecture: EDKSG-TRANS-MASTER-WORLD-1670. This chapter takes the comparative relation as its sole owner. For the full cross-language system, return to Master Art of Translation | The Complete System for Moving Meaning Between Languages. The preceding Definiteness and Specificity chapter handles identity and reference rather than comparative measurement. When the obstacle is knowing the words themselves, use the protected Vocabulary Learning Hub. For grammar and usage inside English rather than cross-language meaning transfer, use the protected How English Works. Those owners remain separate.

1. The answer first: a comparison has more than two sides

A useful comparison record has seven fields: the entity being described; the comparison partner; the measured property or evaluative dimension; the baseline; the direction; the degree or threshold; and the evidence condition. Consider the invented source statement ‘Method A took thirty minutes; Method B took twenty.’ A translator who replaces it with ‘A is ten percent slower’ has invented a percentage. A translator who replaces it with ‘B saves ten minutes in every situation’ has promoted a single observation into a guarantee. Both rewrites sound natural. Neither preserves the original claim.

A comparison can be descriptive without making a recommendation. If the text says that one application uses more energy than another, it does not necessarily say that the first application is worse overall; functionality, quality, throughput and measurement conditions may differ. If it says one team made fewer errors, it does not automatically say that the team is more competent. Comparative translation must preserve what is claimed and refuse the attractive extra conclusion. Translation accuracy and editorial interpretation are connected but not identical.

The practical invariant is a relational statement: under the stated conditions, the same property of the same entity stands in the same relation to the same comparator, with the same boundaries. A target-language sentence may change order, replace a single word with a phrase, or require an explicit subject. Those surface changes are legitimate if the relational truth conditions and the communication function remain intact.

2. Why baseline drift hides inside fluent prose

The first temptation is lexical. A bilingual dictionary may offer an equivalent for more, less, better, worse, over, under, almost and rather. But the meaning of a comparison is not a property of one token in isolation. ‘A is better than B’ compares two options; ‘A is better than it was’ compares time-states of one option; ‘A is better than expected’ compares performance against an expectation. A translation that merely chooses the same adjective can silently replace the second term of the relation.

The second temptation is mathematical. Twenty percent more than an initial value of fifty is sixty; twenty percentage points more than fifty percent is seventy percent. These are different operations. The translator need not be a statistician, but must recognise which transformation the text authorises. Where the source is underspecified, a confident target sentence should not fabricate a denominator or silently resolve an unresolved mathematical ambiguity.

The third temptation is pragmatic. ‘No less than’ can signal an inequality and sometimes rhetorical surprise. ‘As much as’ can mark equality, an upper-bound example, emphasis or a concessive relation, depending on syntax and context. ‘Rather than’ often expresses selection or correction, not a claim of magnitude. Translators must identify the communicative job of the whole expression before matching surface forms.

The fourth temptation is editorial normalisation. A source may use an awkward but meaningful comparison. Polishing it into the shortest target-language sentence can erase a limit, a qualification, an exceptional case or a defined measurement period. A reader then receives a clearer but different claim. Quality control must compare source and target under the same hypothetical facts, not merely ask whether the target ‘sounds right’.

3. Comparison anatomy: the seven-slot record

1. Claimed subject

Which person, process, version, category, group or event is the text actually describing?

The subject must stay tied to its measured value. A common error attaches a quantity to a competing product because the target sentence changes word order.

For revision, place this field in a separate row of the working comparison record. Ask a colleague to recover it from the target text without seeing the source. If the answer is missing or contradicts the source, reopen the line before improving style. The point of separating fields is practical: one may be correct while another has drifted, and one blanket verdict of ‘accurate’ hides the break.

2. Comparator

What is the reference: a rival option, prior version, target, forecast, norm, threshold or group maximum?

A rival option is not the same as the subject’s previous state. A prediction is not a measured baseline.

3. Dimension

What measurable or assessable property varies: time, cost, frequency, likelihood, quality, capacity or preference?

Saving time and improving accuracy can move in opposite directions. The dimension cannot be switched simply because one is more persuasive.

4. Direction

Is the relation higher, lower, equal, not greater, not smaller, preferred or ranked?

Negated comparisons require special care. Not greater than is not automatically less than; equality may remain possible.

5. Magnitude

Is the difference absolute, relative, ordinal, approximate, proportional or merely qualitative?

Ten units and ten percent are not exchangeable. Twice as much and two more are different kinds of statement.

6. Population and denominator

Who or what is counted, during which period, per which unit, and among which eligible group?

An error per 100 outputs cannot be compared honestly with an error per worker without a stated conversion.

7. Evidence and modality

Is this observed, forecast, claimed, conditional, permitted, required or hypothetical?

A measured comparison does not become a promise. An estimate does not become a firm specification.

4. A repeatable twenty-pass comparative translation method

Pass 1: Capture the complete source unit

Read at least the surrounding sentence or paragraph. A truncated comparison often leaves its comparator unstated until a later clause.

Do not translate a bare adjective in isolation if a recoverable comparison partner exists nearby.

Practical output: keep a short trace in the revision log showing the source expression, the intended relation, the tempting alternative and the test that ruled it out. In high-volume work this log need not be lengthy; one line can often prevent the same error from spreading through tables, summaries, charts, captions and repeated snippets. For critical claims, checking the source evidence is more valuable than arguing from stylistic intuition.

Pass 2: Underline the compared entities

Mark the exact nouns or events on both sides, including implicit participants recovered from context.

A method versus its former version is not a method versus a competing method.

Pass 3: Name the dimension

Write a neutral label such as ‘mean processing time per task’ before selecting any target-language adjective.

Good, strong and efficient are evaluations; minutes per task is a measurement.

Pass 4: Recover the baseline

Identify whether the other side is a benchmark, predecessor, control, historical average or expectation.

A target can be exceeded without a competitor being outperformed.

Pass 5: Recover the logical relation

Replace natural wording temporarily with >, <, =, ≤, ≥ or a non-quantitative preference relation, where appropriate.

The operator is a diagnostic sketch, not a demand to use mathematical typography in the final text.

Pass 6: Classify the measurement unit

Record currency, time, count, volume, ratio, percentage, percentage points or qualitative scale.

A dimensionless number may need a unit to be usable, but that unit cannot be invented.

Pass 7: Check arithmetic cautiously

Where data are explicit, recompute the difference and ratio, keeping denominators visible.

30 versus 20 is a difference of 10 and a 50% increase relative to 20, not an increase of 10%.

Pass 8: Mark precision

Separate exact values, ranges, upper/lower limits, rounded figures, approximate claims and estimates.

‘About thirty’ does not grant the same certainty as ‘exactly thirty’.

Pass 9: Identify the reference group

Find the population, period, selection conditions and exclusions that define a superlative or ranked claim.

‘Lowest among the tested models’ is not ‘lowest in the world’.

Pass 10: Resolve syntax scope

Determine whether negation, only, at least, perhaps, approximately or generally modifies a number, relation or entire proposition.

‘Not at least ten’ is not interchangeable with ‘at least not ten’.

Pass 11: Separate observation from judgment

Keep measurement results distinct from statements about usefulness, fairness, desirability or quality.

A longer waiting time may be undesirable, but that judgment was not necessarily in the source.

Pass 12: Write a neutral source paraphrase

Before producing polished target prose, describe the comparison in plain source-language terms.

If the neutral paraphrase is unstable, a target version cannot be confidently validated.

Pass 13: Choose natural target syntax

Select a structure that makes the same relation clear to target readers, even when the grammar differs.

Naturalness is necessary but never excuses a changed comparator.

Pass 14: Preserve unresolved ambiguity

When the source itself is unclear, mark the ambiguity and seek clarification instead of adding confident specificity.

A disputed ‘times more’ phrase may need a query to the author.

Pass 15: Test reversal

Swap A and B mentally and see whether the target keeps the expected inverse claim.

If the source says A is greater, the inverse must say B is lower, not B is greater.

Pass 16: Test equality

Where the source allows equality, create an equal-values hypothetical example and see if the target would still be true.

No greater than allows equality; strictly less than does not.

Pass 17: Test boundary values

Try just below, exactly at and just above the threshold in the target text.

Up to thirty includes the boundary in ordinary maximum-limit readings, but ‘under thirty’ excludes it.

Pass 18: Test denominator shift

Change the reference total or the unit of exposure while leaving the numerator constant.

A per-person risk rate and a per-encounter rate may move differently.

Pass 19: Assess communicative consequences

Ask what a reasonable reader would decide from each version and where those decisions diverge.

Accuracy is not merely a word-level match when a deadline, cost or eligibility decision hangs on the comparison.

Pass 20: Document, revise and release

Record the problem, accepted interpretation, author query if needed and the final bilingual check.

For consequential material, request an independent competent reviewer rather than relying on an unverified fluent rewrite.

5. The worked comparative casebook: constructed meaning tests across domains

The casebook deliberately revisits comparison patterns in different settings. This is not a list of certified translations into particular languages: an exact form must be checked by a competent translator for the actual pair and context. Every source-control sentence here is an original English diagnostic prompt. Its job is to make the truth conditions explicit before a translation decision is made. Repeated drills test transfer, not word memorisation.

In the basic numeric comparisons, A has a value of thirty, B twenty, and C ten on the stated measure. Threshold, equality, expectation, population and preference cases deliberately move into their own explicit possible worlds; each source-control sentence takes priority over the basic mnemonic. All examples are fictional teaching cases, never findings about real services or people.

6.1. Measurement and direction: casebook and comparative audit

Each case has a constructed source-control line, a deliberately wrong or unsupported rewrite, a precise comparison invariant, and a concrete source–target review question. The wrong line is a diagnostic trap, not a recommended translation. No language-pair equivalence is asserted without bilingual verification.

Mechanism 1: Greater than, not guaranteed superiority

Invariant. The observed value of the same measure is larger for A than for B. Higher is a direction, not automatically an endorsement. The measured property may be a cost, delay or burden.

Counterexample method. Would the claim stay accurate if a higher value were undesirable? Keep the numerical comparison; reserve judgment for evidence that actually evaluates outcomes. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 1: Greater than, not guaranteed superiority — an archival indexing trial

Source-control proposition. workflow Atlas recorded 30 minutes per file, versus 20 minutes per file for workflow Beacon. This is a synthetic training statement about processing time in an archival indexing trial; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. workflow Atlas performed better than workflow Beacon. The two expressions do not necessarily remain true under the same possible facts. The observed value of the same measure is larger for A than for B. Higher is a direction, not automatically an endorsement. The measured property may be a cost, delay or burden.

Reader-consequence and diagnosis. An archival manager could budget the wrong staff time or confuse processing speed with archival accuracy. The translator should mark the entities workflow Atlas and workflow Beacon, the property processing time, the unit minutes per file, and any qualifying group or period. The reviewer should ask whether the figures cover the same file types and whether queues are included.

Repair and transfer check. Would the claim stay accurate if a higher value were undesirable? Keep the numerical comparison; reserve judgment for evidence that actually evaluates outcomes. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 2: Greater than, not guaranteed superiority — a hypothetical horticulture comparison

Source-control proposition. garden plot Ash recorded 30 litres per plot per week, versus 20 litres per plot per week for garden plot Beech. This is a synthetic training statement about water consumption in a hypothetical horticulture comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. garden plot Ash performed better than garden plot Beech. The two expressions do not necessarily remain true under the same possible facts. The observed value of the same measure is larger for A than for B. Higher is a direction, not automatically an endorsement. The measured property may be a cost, delay or burden.

Reader-consequence and diagnosis. A sustainability reader might mistake total use for water efficiency without considering crop mix or yield. The translator should mark the entities garden plot Ash and garden plot Beech, the property water consumption, the unit litres per plot per week, and any qualifying group or period. The reviewer should preserve the plot area, growing period and measurement basis.

Mechanism 2: Less than, not equal to

Invariant. The value of B is strictly below the value of A. A strict inequality must not be flattened into approximation or equality.

Counterexample method. What happens if both values are 20? The source would then be false. Use a natural strict-comparison construction in the target language. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 3: Less than, not equal to — a simulated software benchmark

Source-control proposition. service Bay recorded less response latency than service Alder. This is a synthetic training statement about response latency in a simulated software benchmark; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. service Bay recorded exactly the same response latency as service Alder. The two expressions do not necessarily remain true under the same possible facts. The value of B is strictly below the value of A. A strict inequality must not be flattened into approximation or equality.

Reader-consequence and diagnosis. A deployment team could pick the wrong version if average, peak and percentile latency are conflated. The translator should mark the entities service Alder and service Bay, the property response latency, the unit milliseconds per request, and any qualifying group or period. The reviewer should check the traffic mix, statistic used and sampling window before interpreting performance.

Repair and transfer check. What happens if both values are 20? The source would then be false. Use a natural strict-comparison construction in the target language. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 4: Less than, not equal to — a hypothetical multilingual media workflow

Source-control proposition. editing pipeline Blue recorded less caption review time than editing pipeline Amber. This is a synthetic training statement about caption review time in a hypothetical multilingual media workflow; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. editing pipeline Blue recorded exactly the same caption review time as editing pipeline Amber. The two expressions do not necessarily remain true under the same possible facts. The value of B is strictly below the value of A. A strict inequality must not be flattened into approximation or equality.

Reader-consequence and diagnosis. A production editor could confuse the time spent checking captions with the quality of the subtitles. The translator should mark the entities editing pipeline Amber and editing pipeline Blue, the property caption review time, the unit minutes per episode, and any qualifying group or period. The reviewer should retain episode length and whether the workflow covers transcription, translation and revision.

Mechanism 3: At least, inclusive lower limit

Invariant. The allowed values begin at 20 and include exactly 20. Strict greater-than falsely excludes a valid boundary.

Counterexample method. Is a value of exactly 20 acceptable under the wording? Retain an inclusive minimum without adding an invented maximum. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 5: At least, inclusive lower limit — a hypothetical horticulture comparison

Source-control proposition. garden plot Ash requires at least 20 litres per plot per week in the scenario. This is a synthetic training statement about water consumption in a hypothetical horticulture comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. garden plot Ash requires more than 20 litres per plot per week. The two expressions do not necessarily remain true under the same possible facts. The allowed values begin at 20 and include exactly 20. Strict greater-than falsely excludes a valid boundary.

Repair and transfer check. Is a value of exactly 20 acceptable under the wording? Retain an inclusive minimum without adding an invented maximum. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 6: At least, inclusive lower limit — a fictional distribution plan

Source-control proposition. route Amber requires at least 20 calendar days per order in the scenario. This is a synthetic training statement about delivery lead time in a fictional distribution plan; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. route Amber requires more than 20 calendar days per order. The two expressions do not necessarily remain true under the same possible facts. The allowed values begin at 20 and include exactly 20. Strict greater-than falsely excludes a valid boundary.

Reader-consequence and diagnosis. An operations reader might mistake a sample duration for a guaranteed arrival date. The translator should mark the entities route Amber and route Birch, the property delivery lead time, the unit calendar days per order, and any qualifying group or period. The reviewer should distinguish observed transit time from a contractual deadline and define calendar versus working days.

Mechanism 4: More than, exclusive lower limit

Invariant. Exactly 20 does not satisfy the asserted threshold. A seemingly minor phrase change creates a different eligibility set.

Counterexample method. Try the value 20; source false, trap true. Preserve the exclusive boundary explicitly or with the correct native construction. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 7: More than, exclusive lower limit — a hypothetical multilingual media workflow

Source-control proposition. editing pipeline Amber requires more than 20 minutes per episode under the stated condition. This is a synthetic training statement about caption review time in a hypothetical multilingual media workflow; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. editing pipeline Amber requires 20 minutes per episode or more. The two expressions do not necessarily remain true under the same possible facts. Exactly 20 does not satisfy the asserted threshold. A seemingly minor phrase change creates a different eligibility set.

Repair and transfer check. Try the value 20; source false, trap true. Preserve the exclusive boundary explicitly or with the correct native construction. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 8: More than, exclusive lower limit — a hypothetical manufacturing audit

Source-control proposition. method Aspen requires more than 20 hours per batch under the stated condition. This is a synthetic training statement about quality-control review time in a hypothetical manufacturing audit; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. method Aspen requires 20 hours per batch or more. The two expressions do not necessarily remain true under the same possible facts. Exactly 20 does not satisfy the asserted threshold. A seemingly minor phrase change creates a different eligibility set.

Reader-consequence and diagnosis. A quality officer could confuse time savings with reduced defect risk, which is a separate outcome. The translator should mark the entities method Aspen and method Briar, the property quality-control review time, the unit hours per batch, and any qualifying group or period. The reviewer should check batch size and whether rework time is part of the measure.

Mechanism 5: At most, inclusive upper limit

Invariant. Values through and including 30 satisfy the cap. An exclusive inequality narrows the authorised range.

Counterexample method. Test 30 and 30.1 separately. Use a maximum that includes the stated endpoint. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 9: At most, inclusive upper limit — a fictional distribution plan

Source-control proposition. The cap on route Amber is at most 30 calendar days per order. This is a synthetic training statement about delivery lead time in a fictional distribution plan; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. The cap is less than 30 calendar days per order. The two expressions do not necessarily remain true under the same possible facts. Values through and including 30 satisfy the cap. An exclusive inequality narrows the authorised range.

Repair and transfer check. Test 30 and 30.1 separately. Use a maximum that includes the stated endpoint. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 10: At most, inclusive upper limit — a constructed training-programme comparison

Source-control proposition. The cap on course Alpha is at most 30 hours per participant. This is a synthetic training statement about practice time in a constructed training-programme comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. The cap is less than 30 hours per participant. The two expressions do not necessarily remain true under the same possible facts. Values through and including 30 satisfy the cap. An exclusive inequality narrows the authorised range.

Reader-consequence and diagnosis. A learner might mistake time spent for evidence of learning gain or assume all participants had identical schedules. The translator should mark the entities course Alpha and course Beta, the property practice time, the unit hours per participant, and any qualifying group or period. The reviewer should distinguish required hours, typical hours and measured attendance.

Mechanism 6: Under, exclusive upper limit

Invariant. A value equal to 30 falls outside the stated rule. A permitted boundary can become a prohibited one, or the reverse, simply by changing the relation.

Counterexample method. Would exactly 30 pass the source criterion? Retain strict upper-bound language and request clarification if the source convention is uncertain. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 11: Under, exclusive upper limit — a hypothetical manufacturing audit

Source-control proposition. method Aspen must stay under 30 hours per batch. This is a synthetic training statement about quality-control review time in a hypothetical manufacturing audit; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. method Aspen may reach 30 hours per batch. The two expressions do not necessarily remain true under the same possible facts. A value equal to 30 falls outside the stated rule. A permitted boundary can become a prohibited one, or the reverse, simply by changing the relation.

Repair and transfer check. Would exactly 30 pass the source criterion? Retain strict upper-bound language and request clarification if the source convention is uncertain. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 12: Under, exclusive upper limit — a controlled equipment test

Source-control proposition. design Aurora must stay under 30 watt-hours per cycle. This is a synthetic training statement about electricity consumption in a controlled equipment test; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. design Aurora may reach 30 watt-hours per cycle. The two expressions do not necessarily remain true under the same possible facts. A value equal to 30 falls outside the stated rule. A permitted boundary can become a prohibited one, or the reverse, simply by changing the relation.

Reader-consequence and diagnosis. A design team might claim an energy saving that the test did not establish or omit the cycle definition. The translator should mark the entities design Aurora and design Bramble, the property electricity consumption, the unit watt-hours per cycle, and any qualifying group or period. The reviewer should retain the test duration, duty cycle and power-measurement method.

Mechanism 7: No more than, negated greater-than

Invariant. This expression sets an upper bound; it is not a lower-bound requirement. Negation scope can reverse the entire compliance rule.

Counterexample method. Try zero, 30 and 31 as test values. Write the upper-bound paraphrase first, then choose native target wording. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 13: No more than, negated greater-than — a constructed training-programme comparison

Source-control proposition. course Alpha may use no more than 30 hours per participant. This is a synthetic training statement about practice time in a constructed training-programme comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. course Alpha must use at least 30 hours per participant. The two expressions do not necessarily remain true under the same possible facts. This expression sets an upper bound; it is not a lower-bound requirement. Negation scope can reverse the entire compliance rule.

Repair and transfer check. Try zero, 30 and 31 as test values. Write the upper-bound paraphrase first, then choose native target wording. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 14: No more than, negated greater-than — an illustrative pricing schedule

Source-control proposition. plan Amber may use no more than 30 currency units per month. This is a synthetic training statement about subscription charge in an illustrative pricing schedule; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. plan Amber must use at least 30 currency units per month. The two expressions do not necessarily remain true under the same possible facts. This expression sets an upper bound; it is not a lower-bound requirement. Negation scope can reverse the entire compliance rule.

Reader-consequence and diagnosis. A purchaser might infer the wrong recurring cost or assume fees include taxes and usage extras. The translator should mark the entities plan Amber and plan Birch, the property subscription charge, the unit currency units per month, and any qualifying group or period. The reviewer should preserve whether the comparison is monthly, annualised, pre-tax or inclusive of add-ons.

Mechanism 8: No less than, negated less-than

Invariant. This expression ordinarily sets an inclusive minimum, subject to any rhetorical nuance. Literal token matching can reverse a requirement despite preserving both number and noun.

Counterexample method. Test 19, 20 and 21. Retain minimum force; examine emphasis separately from arithmetic meaning. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 15: No less than, negated less-than — a controlled equipment test

Source-control proposition. design Aurora must provide no less than 20 watt-hours per cycle. This is a synthetic training statement about electricity consumption in a controlled equipment test; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. design Aurora must provide less than 20 watt-hours per cycle. The two expressions do not necessarily remain true under the same possible facts. This expression ordinarily sets an inclusive minimum, subject to any rhetorical nuance. Literal token matching can reverse a requirement despite preserving both number and noun.

Repair and transfer check. Test 19, 20 and 21. Retain minimum force; examine emphasis separately from arithmetic meaning. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 16: No less than, negated less-than — a fictional clinic scheduling example

Source-control proposition. queue Aster must provide no less than 20 days per appointment. This is a synthetic training statement about appointment waiting time in a fictional clinic scheduling example; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. queue Aster must provide less than 20 days per appointment. The two expressions do not necessarily remain true under the same possible facts. This expression ordinarily sets an inclusive minimum, subject to any rhetorical nuance. Literal token matching can reverse a requirement despite preserving both number and noun.

Reader-consequence and diagnosis. A patient could mistake an average wait for an individual appointment promise or a triage rule. The translator should mark the entities queue Aster and queue Brook, the property appointment waiting time, the unit days per appointment, and any qualifying group or period. The reviewer should preserve whether values are averages, medians, ranges or guarantees.

Mechanism 9: Ten-unit absolute difference

Invariant. The difference is an absolute ten units of the specified measure. A unit difference cannot become a rate unless the source authorises the denominator and operation.

Counterexample method. Can a reader reconstruct 30 minus 20 equals 10? Keep the unit and avoid introducing percent signs. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 17: Ten-unit absolute difference — an illustrative pricing schedule

Source-control proposition. plan Amber exceeded plan Birch by 10 currency units per month. This is a synthetic training statement about subscription charge in an illustrative pricing schedule; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. plan Amber exceeded plan Birch by 10 percent. The two expressions do not necessarily remain true under the same possible facts. The difference is an absolute ten units of the specified measure. A unit difference cannot become a rate unless the source authorises the denominator and operation.

Repair and transfer check. Can a reader reconstruct 30 minus 20 equals 10? Keep the unit and avoid introducing percent signs. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 18: Ten-unit absolute difference — an archival indexing trial

Source-control proposition. workflow Atlas exceeded workflow Beacon by 10 minutes per file. This is a synthetic training statement about processing time in an archival indexing trial; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. workflow Atlas exceeded workflow Beacon by 10 percent. The two expressions do not necessarily remain true under the same possible facts. The difference is an absolute ten units of the specified measure. A unit difference cannot become a rate unless the source authorises the denominator and operation.

Mechanism 10: Fifty-percent relative difference

Invariant. Relative increase is measured against B: (30−20)/20 = 50%. A percentage describes a ratio to a baseline; an absolute unit gap does not.

Counterexample method. Which denominator is needed for the fifty-percent claim? Keep the percentage and its comparison base. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 19: Fifty-percent relative difference — a fictional clinic scheduling example

Source-control proposition. queue Aster was 50 percent higher than queue Brook on appointment waiting time. This is a synthetic training statement about appointment waiting time in a fictional clinic scheduling example; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. queue Aster was 50 days per appointment higher than queue Brook. The two expressions do not necessarily remain true under the same possible facts. Relative increase is measured against B: (30−20)/20 = 50%. A percentage describes a ratio to a baseline; an absolute unit gap does not.

Repair and transfer check. Which denominator is needed for the fifty-percent claim? Keep the percentage and its comparison base. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 20: Fifty-percent relative difference — a simulated software benchmark

Source-control proposition. service Alder was 50 percent higher than service Bay on response latency. This is a synthetic training statement about response latency in a simulated software benchmark; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. service Alder was 50 milliseconds per request higher than service Bay. The two expressions do not necessarily remain true under the same possible facts. Relative increase is measured against B: (30−20)/20 = 50%. A percentage describes a ratio to a baseline; an absolute unit gap does not.

Mechanism 11: One-third lower from A’s baseline

Invariant. The decrease is ten out of the original A value of thirty. Reversing the comparison direction changes the denominator, even when the difference stays ten.

Counterexample method. Compute (30−20)/30 rather than (30−20)/20. Preserve both the percentage and the reference value. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 21: One-third lower from A’s baseline — an archival indexing trial

Source-control proposition. workflow Beacon was one-third lower than workflow Atlas on processing time. This is a synthetic training statement about processing time in an archival indexing trial; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. workflow Beacon was fifty percent lower than workflow Atlas. The two expressions do not necessarily remain true under the same possible facts. The decrease is ten out of the original A value of thirty. Reversing the comparison direction changes the denominator, even when the difference stays ten.

Repair and transfer check. Compute (30−20)/30 rather than (30−20)/20. Preserve both the percentage and the reference value. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 22: One-third lower from A’s baseline — a hypothetical horticulture comparison

Source-control proposition. garden plot Beech was one-third lower than garden plot Ash on water consumption. This is a synthetic training statement about water consumption in a hypothetical horticulture comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. garden plot Beech was fifty percent lower than garden plot Ash. The two expressions do not necessarily remain true under the same possible facts. The decrease is ten out of the original A value of thirty. Reversing the comparison direction changes the denominator, even when the difference stays ten.

Mechanism 12: One-and-a-half times as large

Invariant. The ratio A/B is 30/20 = 1.5. Confusing ‘times as much’ with ‘times more’ can change the multiplier.

Counterexample method. Check by multiplying the B value of twenty by 1.5. Express the ratio with a convention unambiguous to the target reader. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 23: One-and-a-half times as large — a simulated software benchmark

Source-control proposition. service Alder’s response latency was one and a half times service Bay’s. This is a synthetic training statement about response latency in a simulated software benchmark; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. service Alder’s response latency was two and a half times service Bay’s. The two expressions do not necessarily remain true under the same possible facts. The ratio A/B is 30/20 = 1.5. Confusing ‘times as much’ with ‘times more’ can change the multiplier.

Repair and transfer check. Check by multiplying the B value of twenty by 1.5. Express the ratio with a convention unambiguous to the target reader. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 24: One-and-a-half times as large — a hypothetical multilingual media workflow

Source-control proposition. editing pipeline Amber’s caption review time was one and a half times editing pipeline Blue’s. This is a synthetic training statement about caption review time in a hypothetical multilingual media workflow; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. editing pipeline Amber’s caption review time was two and a half times editing pipeline Blue’s. The two expressions do not necessarily remain true under the same possible facts. The ratio A/B is 30/20 = 1.5. Confusing ‘times as much’ with ‘times more’ can change the multiplier.

6.2. Thresholds, approximation and order: casebook and comparative audit

Mechanism 13: Ratio of three to two

Invariant. The order of the named items fixes the order of the ratio terms. A polished summary can preserve 3:2 while attaching the numbers to the wrong parties.

Counterexample method. If A is 30 and B is 20, does the ratio point the correct way? Name both terms and keep their order or invert both ratio and label. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 25: Ratio of three to two — a hypothetical horticulture comparison

Source-control proposition. The measured water consumption ratio of garden plot Ash to garden plot Beech was 3:2. This is a synthetic training statement about water consumption in a hypothetical horticulture comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. The ratio of garden plot Beech to garden plot Ash was 3:2. The two expressions do not necessarily remain true under the same possible facts. The order of the named items fixes the order of the ratio terms. A polished summary can preserve 3:2 while attaching the numbers to the wrong parties.

Repair and transfer check. If A is 30 and B is 20, does the ratio point the correct way? Name both terms and keep their order or invert both ratio and label. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 26: Ratio of three to two — a fictional distribution plan

Source-control proposition. The measured delivery lead time ratio of route Amber to route Birch was 3:2. This is a synthetic training statement about delivery lead time in a fictional distribution plan; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. The ratio of route Birch to route Amber was 3:2. The two expressions do not necessarily remain true under the same possible facts. The order of the named items fixes the order of the ratio terms. A polished summary can preserve 3:2 while attaching the numbers to the wrong parties.

Mechanism 14: Just over as approximation

Invariant. The expression communicates a value slightly greater than 20 without specifying an exact figure. A translator should not replace an imprecise report with a fabricated exact value.

Counterexample method. Could a small number above 20 satisfy the source? Preserve the approximate-above relationship and avoid borrowing the unrelated example value thirty. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 27: Just over as approximation — a hypothetical multilingual media workflow

Source-control proposition. In the rounded report, editing pipeline Amber was just over 20 minutes per episode. This is a synthetic training statement about caption review time in a hypothetical multilingual media workflow; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. editing pipeline Amber was exactly 20 minutes per episode. The two expressions do not necessarily remain true under the same possible facts. The expression communicates a value slightly greater than 20 without specifying an exact figure. A translator should not replace an imprecise report with a fabricated exact value.

Repair and transfer check. Could a small number above 20 satisfy the source? Preserve the approximate-above relationship and avoid borrowing the unrelated example value thirty. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 28: Just over as approximation — a hypothetical manufacturing audit

Source-control proposition. In the rounded report, method Aspen was just over 20 hours per batch. This is a synthetic training statement about quality-control review time in a hypothetical manufacturing audit; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. method Aspen was exactly 20 hours per batch. The two expressions do not necessarily remain true under the same possible facts. The expression communicates a value slightly greater than 20 without specifying an exact figure. A translator should not replace an imprecise report with a fabricated exact value.

Mechanism 15: Almost thirty is not thirty

Invariant. The proposed value remains below the threshold in this statement. Rounding, prediction and completion are distinct information.

Counterexample method. Would 29.9 fit while exactly 30 did not? Retain nonattainment rather than stating arrival at the threshold. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 29: Almost thirty is not thirty — a fictional distribution plan

Source-control proposition. In the hypothetical forecast, route Amber approached but did not reach 30 calendar days per order. This is a synthetic training statement about delivery lead time in a fictional distribution plan; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. route Amber reached 30 calendar days per order. The two expressions do not necessarily remain true under the same possible facts. The proposed value remains below the threshold in this statement. Rounding, prediction and completion are distinct information.

Repair and transfer check. Would 29.9 fit while exactly 30 did not? Retain nonattainment rather than stating arrival at the threshold. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 30: Almost thirty is not thirty — a constructed training-programme comparison

Source-control proposition. In the hypothetical forecast, course Alpha approached but did not reach 30 hours per participant. This is a synthetic training statement about practice time in a constructed training-programme comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. course Alpha reached 30 hours per participant. The two expressions do not necessarily remain true under the same possible facts. The proposed value remains below the threshold in this statement. Rounding, prediction and completion are distinct information.

Mechanism 16: Up to an upper bound, not a promise

Invariant. A maximum permitted extent does not promise the maximum will be realised. ‘Up to’ is often persuasive wording, but an upper limit is not a typical or minimum outcome.

Counterexample method. Does a result of ten remain compatible with the source? Preserve possibility and maximum, not certainty or frequency. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 31: Up to an upper bound, not a promise — a hypothetical manufacturing audit

Source-control proposition. The specification permits method Aspen to reach up to 30 hours per batch. This is a synthetic training statement about quality-control review time in a hypothetical manufacturing audit; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. method Aspen always reaches 30 hours per batch. The two expressions do not necessarily remain true under the same possible facts. A maximum permitted extent does not promise the maximum will be realised. ‘Up to’ is often persuasive wording, but an upper limit is not a typical or minimum outcome.

Repair and transfer check. Does a result of ten remain compatible with the source? Preserve possibility and maximum, not certainty or frequency. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 32: Up to an upper bound, not a promise — a controlled equipment test

Source-control proposition. The specification permits design Aurora to reach up to 30 watt-hours per cycle. This is a synthetic training statement about electricity consumption in a controlled equipment test; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. design Aurora always reaches 30 watt-hours per cycle. The two expressions do not necessarily remain true under the same possible facts. A maximum permitted extent does not promise the maximum will be realised. ‘Up to’ is often persuasive wording, but an upper limit is not a typical or minimum outcome.

Mechanism 17: As high as marks a possible extreme

Invariant. An attainable or reported extreme is different from a constant figure. The target must not replace one endpoint of a distribution with every observation.

Counterexample method. Could some observations be lower? Render the high-end possibility and identify the measurement period if the surrounding source provides it. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 33: As high as marks a possible extreme — a constructed training-programme comparison

Source-control proposition. The illustrative range for course Alpha can be as high as 30 hours per participant. This is a synthetic training statement about practice time in a constructed training-programme comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. course Alpha is always 30 hours per participant. The two expressions do not necessarily remain true under the same possible facts. An attainable or reported extreme is different from a constant figure. The target must not replace one endpoint of a distribution with every observation.

Repair and transfer check. Could some observations be lower? Render the high-end possibility and identify the measurement period if the surrounding source provides it. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 34: As high as marks a possible extreme — an illustrative pricing schedule

Source-control proposition. The illustrative range for plan Amber can be as high as 30 currency units per month. This is a synthetic training statement about subscription charge in an illustrative pricing schedule; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. plan Amber is always 30 currency units per month. The two expressions do not necessarily remain true under the same possible facts. An attainable or reported extreme is different from a constant figure. The target must not replace one endpoint of a distribution with every observation.

Mechanism 18: Not as high as rejects equality or superiority

Invariant. For the chosen numeric example, B is lower than A. The comparative may require a negative degree construction that differs from a direct adjective swap.

Counterexample method. Do 20 and 30 satisfy the proposed target? Choose target syntax that retains the measured ordering. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 35: Not as high as rejects equality or superiority — a controlled equipment test

Source-control proposition. design Bramble was not as high as design Aurora on electricity consumption. This is a synthetic training statement about electricity consumption in a controlled equipment test; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. design Bramble was at least as high as design Aurora. The two expressions do not necessarily remain true under the same possible facts. For the chosen numeric example, B is lower than A. The comparative may require a negative degree construction that differs from a direct adjective swap.

Repair and transfer check. Do 20 and 30 satisfy the proposed target? Choose target syntax that retains the measured ordering. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 36: Not as high as rejects equality or superiority — a fictional clinic scheduling example

Source-control proposition. queue Brook was not as high as queue Aster on appointment waiting time. This is a synthetic training statement about appointment waiting time in a fictional clinic scheduling example; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. queue Brook was at least as high as queue Aster. The two expressions do not necessarily remain true under the same possible facts. For the chosen numeric example, B is lower than A. The comparative may require a negative degree construction that differs from a direct adjective swap.

Mechanism 19: At least as high allows equality

Invariant. The relation permits A = B and A > B. Strict wording claims more than the source by excluding ties.

Counterexample method. Imagine both are 20; the source still holds. Retain equality among the permissible cases. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 37: At least as high allows equality — an illustrative pricing schedule

Source-control proposition. plan Amber was at least as high as plan Birch on subscription charge. This is a synthetic training statement about subscription charge in an illustrative pricing schedule; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. plan Amber was strictly higher than plan Birch. The two expressions do not necessarily remain true under the same possible facts. The relation permits A = B and A > B. Strict wording claims more than the source by excluding ties.

Repair and transfer check. Imagine both are 20; the source still holds. Retain equality among the permissible cases. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 38: At least as high allows equality — an archival indexing trial

Source-control proposition. workflow Atlas was at least as high as workflow Beacon on processing time. This is a synthetic training statement about processing time in an archival indexing trial; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. workflow Atlas was strictly higher than workflow Beacon. The two expressions do not necessarily remain true under the same possible facts. The relation permits A = B and A > B. Strict wording claims more than the source by excluding ties.

Mechanism 20: No higher than includes equality

Invariant. The source permits equality or a lower B value. An apparently elegant shorter target can delete the equality case.

Counterexample method. Could B equal A without violating the statement? Keep the non-strict comparison rather than a strict lower bound. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 39: No higher than includes equality — a fictional clinic scheduling example

Source-control proposition. queue Brook may be no higher than queue Aster on appointment waiting time. This is a synthetic training statement about appointment waiting time in a fictional clinic scheduling example; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. queue Brook must be lower than queue Aster. The two expressions do not necessarily remain true under the same possible facts. The source permits equality or a lower B value. An apparently elegant shorter target can delete the equality case.

Repair and transfer check. Could B equal A without violating the statement? Keep the non-strict comparison rather than a strict lower bound. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 40: No higher than includes equality — a simulated software benchmark

Source-control proposition. service Bay may be no higher than service Alder on response latency. This is a synthetic training statement about response latency in a simulated software benchmark; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. service Bay must be lower than service Alder. The two expressions do not necessarily remain true under the same possible facts. The source permits equality or a lower B value. An apparently elegant shorter target can delete the equality case.

Mechanism 21: Not lower does not mean demonstrably higher

Invariant. A is greater or equal; the proposition does not exclude a tie. Negated adjectives are not always interchangeable with positive opposites.

Counterexample method. Try A = B and observe whether the trap changes truth value. Maintain non-strict ordering and avoid unjustified strength. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 41: Not lower does not mean demonstrably higher — an archival indexing trial

Source-control proposition. workflow Atlas was not lower than workflow Beacon on processing time. This is a synthetic training statement about processing time in an archival indexing trial; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. workflow Atlas was definitely greater than workflow Beacon. The two expressions do not necessarily remain true under the same possible facts. A is greater or equal; the proposition does not exclude a tie. Negated adjectives are not always interchangeable with positive opposites.

Repair and transfer check. Try A = B and observe whether the trap changes truth value. Maintain non-strict ordering and avoid unjustified strength. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 42: Not lower does not mean demonstrably higher — a hypothetical horticulture comparison

Source-control proposition. garden plot Ash was not lower than garden plot Beech on water consumption. This is a synthetic training statement about water consumption in a hypothetical horticulture comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. garden plot Ash was definitely greater than garden plot Beech. The two expressions do not necessarily remain true under the same possible facts. A is greater or equal; the proposition does not exclude a tie. Negated adjectives are not always interchangeable with positive opposites.

Mechanism 22: Lower of two, not lowest overall

Invariant. The rank is defined only over the two named options. Expanding the comparison set invents a global superlative.

Counterexample method. What if an unseen third option is lower? Retain the exact comparison group. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 43: Lower of two, not lowest overall — a simulated software benchmark

Source-control proposition. Among service Alder and service Bay, service Bay had the lower response latency. This is a synthetic training statement about response latency in a simulated software benchmark; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. service Bay had the lowest response latency among all available options. The two expressions do not necessarily remain true under the same possible facts. The rank is defined only over the two named options. Expanding the comparison set invents a global superlative.

Repair and transfer check. What if an unseen third option is lower? Retain the exact comparison group. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 44: Lower of two, not lowest overall — a hypothetical multilingual media workflow

Source-control proposition. Among editing pipeline Amber and editing pipeline Blue, editing pipeline Blue had the lower caption review time. This is a synthetic training statement about caption review time in a hypothetical multilingual media workflow; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. editing pipeline Blue had the lowest caption review time among all available options. The two expressions do not necessarily remain true under the same possible facts. The rank is defined only over the two named options. Expanding the comparison set invents a global superlative.

Mechanism 23: Lowest of three, group-bound superlative

Invariant. The minimum belongs to an explicitly named group of three. A local rank is not a universal world ranking.

Counterexample method. Introduce an unseen fourth option below C. Preserve the specified group and the superlative relation. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 45: Lowest of three, group-bound superlative — a hypothetical horticulture comparison

Source-control proposition. Among garden plot Ash, garden plot Beech and garden plot Cedar, garden plot Cedar had the lowest water consumption. This is a synthetic training statement about water consumption in a hypothetical horticulture comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. garden plot Cedar had the lowest water consumption everywhere. The two expressions do not necessarily remain true under the same possible facts. The minimum belongs to an explicitly named group of three. A local rank is not a universal world ranking.

Repair and transfer check. Introduce an unseen fourth option below C. Preserve the specified group and the superlative relation. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 46: Lowest of three, group-bound superlative — a fictional distribution plan

Source-control proposition. Among route Amber, route Birch and route Cedar, route Cedar had the lowest delivery lead time. This is a synthetic training statement about delivery lead time in a fictional distribution plan; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. route Cedar had the lowest delivery lead time everywhere. The two expressions do not necessarily remain true under the same possible facts. The minimum belongs to an explicitly named group of three. A local rank is not a universal world ranking.

Mechanism 24: Second-highest is not second-best

Invariant. This is an ordinal position on one measure, not a quality judgment. Rank and desirability diverge when higher measurement means more cost or delay.

Counterexample method. If the measure is undesirable, is ‘best’ still justified? Keep the named dimension of ranking. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 47: Second-highest is not second-best — a hypothetical multilingual media workflow

Source-control proposition. Of editing pipeline Amber, editing pipeline Blue and editing pipeline Coral, editing pipeline Blue ranked second-highest in caption review time. This is a synthetic training statement about caption review time in a hypothetical multilingual media workflow; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. editing pipeline Blue was the second-best option. The two expressions do not necessarily remain true under the same possible facts. This is an ordinal position on one measure, not a quality judgment. Rank and desirability diverge when higher measurement means more cost or delay.

Repair and transfer check. If the measure is undesirable, is ‘best’ still justified? Keep the named dimension of ranking. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 48: Second-highest is not second-best — a hypothetical manufacturing audit

Source-control proposition. Of method Aspen, method Briar and method Cypress, method Briar ranked second-highest in quality-control review time. This is a synthetic training statement about quality-control review time in a hypothetical manufacturing audit; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. method Briar was the second-best option. The two expressions do not necessarily remain true under the same possible facts. This is an ordinal position on one measure, not a quality judgment. Rank and desirability diverge when higher measurement means more cost or delay.

6.3. Populations, baselines and movement: casebook and comparative audit

Mechanism 25: More than half has a group denominator

Invariant. Thirty out of fifty is a strict majority, not exactly fifty percent. A translation can distort both quantifier strength and sample size.

Counterexample method. Calculate 30/50 and test whether 25/50 would satisfy the source. State the population, numerator and correct majority relation. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 49: More than half has a group denominator — a fictional distribution plan

Source-control proposition. Of 50 hypothetical observations about route Amber, 30 exceeded the threshold. This is a synthetic training statement about delivery lead time in a fictional distribution plan; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. Exactly half of the observations exceeded the threshold. The two expressions do not necessarily remain true under the same possible facts. Thirty out of fifty is a strict majority, not exactly fifty percent. A translation can distort both quantifier strength and sample size.

Repair and transfer check. Calculate 30/50 and test whether 25/50 would satisfy the source. State the population, numerator and correct majority relation. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 50: More than half has a group denominator — a constructed training-programme comparison

Source-control proposition. Of 50 hypothetical observations about course Alpha, 30 exceeded the threshold. This is a synthetic training statement about practice time in a constructed training-programme comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Mechanism 26: Exactly half is not most

Invariant. Twenty of forty is precisely one-half, without a strict majority. ‘Most’ communicates a stronger share than exactly half.

Counterexample method. Would 20 of 40 count as a strict majority? Preserve exact half without a majority implication. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 51: Exactly half is not most — a hypothetical manufacturing audit

Source-control proposition. Of 40 hypothetical observations about method Briar, exactly 20 met the criterion. This is a synthetic training statement about quality-control review time in a hypothetical manufacturing audit; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. Most observations met the criterion. The two expressions do not necessarily remain true under the same possible facts. Twenty of forty is precisely one-half, without a strict majority. ‘Most’ communicates a stronger share than exactly half.

Repair and transfer check. Would 20 of 40 count as a strict majority? Preserve exact half without a majority implication. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 52: Exactly half is not most — a controlled equipment test

Source-control proposition. Of 40 hypothetical observations about design Bramble, exactly 20 met the criterion. This is a synthetic training statement about electricity consumption in a controlled equipment test; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Mechanism 27: More than expected changes the comparator

Invariant. The comparison partner is a prediction, not another observed workflow. Substituting a competitor for an expectation can fabricate evidence about that competitor.

Counterexample method. Could B’s real observed value be unknown? Describe the comparison against the forecast explicitly. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 53: More than expected changes the comparator — a constructed training-programme comparison

Source-control proposition. course Alpha took 30 hours per participant, exceeding a forecast of 20 hours per participant. This is a synthetic training statement about practice time in a constructed training-programme comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. course Alpha took more than course Beta because the forecast was 20. The two expressions do not necessarily remain true under the same possible facts. The comparison partner is a prediction, not another observed workflow. Substituting a competitor for an expectation can fabricate evidence about that competitor.

Repair and transfer check. Could B’s real observed value be unknown? Describe the comparison against the forecast explicitly. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 54: More than expected changes the comparator — an illustrative pricing schedule

Source-control proposition. plan Amber took 30 currency units per month, exceeding a forecast of 20 currency units per month. This is a synthetic training statement about subscription charge in an illustrative pricing schedule; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. plan Amber took more than plan Birch because the forecast was 20. The two expressions do not necessarily remain true under the same possible facts. The comparison partner is a prediction, not another observed workflow. Substituting a competitor for an expectation can fabricate evidence about that competitor.

Mechanism 28: Higher than last period is a self-comparison

Invariant. The same entity is compared across time. A temporal comparison is not an inter-entity contest.

Counterexample method. Ask whether the source contains any measurement of B. Retain past-versus-present reference and keep any dates if stated. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 55: Higher than last period is a self-comparison — a controlled equipment test

Source-control proposition. design Aurora rose from 20 to 30 watt-hours per cycle between the two invented periods. This is a synthetic training statement about electricity consumption in a controlled equipment test; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. design Aurora was higher than design Bramble in the same period. The two expressions do not necessarily remain true under the same possible facts. The same entity is compared across time. A temporal comparison is not an inter-entity contest.

Repair and transfer check. Ask whether the source contains any measurement of B. Retain past-versus-present reference and keep any dates if stated. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 56: Higher than last period is a self-comparison — a fictional clinic scheduling example

Source-control proposition. queue Aster rose from 20 to 30 days per appointment between the two invented periods. This is a synthetic training statement about appointment waiting time in a fictional clinic scheduling example; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. queue Aster was higher than queue Brook in the same period. The two expressions do not necessarily remain true under the same possible facts. The same entity is compared across time. A temporal comparison is not an inter-entity contest.

Mechanism 29: Increased by ten is not increased to ten

Invariant. The amount of the change is ten; the final value is thirty. A short preposition can switch the endpoint and delta.

Counterexample method. Check starting value + reported change = final value. Keep starting state, change and endpoint separate. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 57: Increased by ten is not increased to ten — an illustrative pricing schedule

Source-control proposition. plan Amber increased from 20 to 30 currency units per month. This is a synthetic training statement about subscription charge in an illustrative pricing schedule; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. plan Amber increased to 10 currency units per month. The two expressions do not necessarily remain true under the same possible facts. The amount of the change is ten; the final value is thirty. A short preposition can switch the endpoint and delta.

Repair and transfer check. Check starting value + reported change = final value. Keep starting state, change and endpoint separate. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 58: Increased by ten is not increased to ten — an archival indexing trial

Source-control proposition. workflow Atlas increased from 20 to 30 minutes per file. This is a synthetic training statement about processing time in an archival indexing trial; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. workflow Atlas increased to 10 minutes per file. The two expressions do not necessarily remain true under the same possible facts. The amount of the change is ten; the final value is thirty. A short preposition can switch the endpoint and delta.

Mechanism 30: Percentage increase and level are separate

Invariant. The relative change is fifty percent; the resulting level is thirty. A percentage number must not become a quantity in the original unit.

Counterexample method. Reconstruct the endpoint with 20 × 1.5. Write both base and endpoint when needed for clarity, without adding facts. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 59: Percentage increase and level are separate — a fictional clinic scheduling example

Source-control proposition. queue Aster increased by 50 percent from a base of 20 days per appointment. This is a synthetic training statement about appointment waiting time in a fictional clinic scheduling example; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. queue Aster increased to 50 days per appointment. The two expressions do not necessarily remain true under the same possible facts. The relative change is fifty percent; the resulting level is thirty. A percentage number must not become a quantity in the original unit.

Repair and transfer check. Reconstruct the endpoint with 20 × 1.5. Write both base and endpoint when needed for clarity, without adding facts. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 60: Percentage increase and level are separate — a simulated software benchmark

Source-control proposition. service Alder increased by 50 percent from a base of 20 milliseconds per request. This is a synthetic training statement about response latency in a simulated software benchmark; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. service Alder increased to 50 milliseconds per request. The two expressions do not necessarily remain true under the same possible facts. The relative change is fifty percent; the resulting level is thirty. A percentage number must not become a quantity in the original unit.

Mechanism 31: Percentage points need percentage-valued measures

Invariant. The increase is ten percentage points, equivalent to fifty percent relative to the starting share. The translator must not collapse percentage points into percent relative change.

Counterexample method. Compute both 30−20 and (30−20)/20. Preserve percentage-point wording or explicit endpoints. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 61: Percentage points need percentage-valued measures — an archival indexing trial

Source-control proposition. In a separate survey example, the share rose from 20 percent to 30 percent. This is a synthetic training statement about processing time in an archival indexing trial; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. The share rose by ten percent. The two expressions do not necessarily remain true under the same possible facts. The increase is ten percentage points, equivalent to fifty percent relative to the starting share. The translator must not collapse percentage points into percent relative change.

Repair and transfer check. Compute both 30−20 and (30−20)/20. Preserve percentage-point wording or explicit endpoints. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 62: Percentage points need percentage-valued measures — a hypothetical horticulture comparison

Source-control proposition. In a separate survey example, the share rose from 20 percent to 30 percent. This is a synthetic training statement about water consumption in a hypothetical horticulture comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Mechanism 32: Difference narrowed but did not vanish

Invariant. The gap shrank but remains non-zero. A narrowing difference is not proof of equality, reversal or convergence to zero.

Counterexample method. Is a remaining gap of ten compatible with the trap? Represent direction of change in the gap and its continuing magnitude. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 63: Difference narrowed but did not vanish — a simulated software benchmark

Source-control proposition. In a first comparison service Alder exceeded service Bay by 20 milliseconds per request; later the gap was 10 milliseconds per request. This is a synthetic training statement about response latency in a simulated software benchmark; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. The two became equal. The two expressions do not necessarily remain true under the same possible facts. The gap shrank but remains non-zero. A narrowing difference is not proof of equality, reversal or convergence to zero.

Repair and transfer check. Is a remaining gap of ten compatible with the trap? Represent direction of change in the gap and its continuing magnitude. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 64: Difference narrowed but did not vanish — a hypothetical multilingual media workflow

Source-control proposition. In a first comparison editing pipeline Amber exceeded editing pipeline Blue by 20 minutes per episode; later the gap was 10 minutes per episode. This is a synthetic training statement about caption review time in a hypothetical multilingual media workflow; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Mechanism 33: Difference widened without changing winner

Invariant. The lead became larger while the ordering stayed the same. A dynamic comparison can change size without reversing rank.

Counterexample method. Can A remain ahead throughout? Keep both the sign and the magnitude trajectory. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 65: Difference widened without changing winner — a hypothetical horticulture comparison

Source-control proposition. The difference between garden plot Ash and garden plot Beech grew from 5 to 10 litres per plot per week, with garden plot Ash higher in both periods. This is a synthetic training statement about water consumption in a hypothetical horticulture comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. garden plot Beech overtook garden plot Ash. The two expressions do not necessarily remain true under the same possible facts. The lead became larger while the ordering stayed the same. A dynamic comparison can change size without reversing rank.

Repair and transfer check. Can A remain ahead throughout? Keep both the sign and the magnitude trajectory. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 66: Difference widened without changing winner — a fictional distribution plan

Source-control proposition. The difference between route Amber and route Birch grew from 5 to 10 calendar days per order, with route Amber higher in both periods. This is a synthetic training statement about delivery lead time in a fictional distribution plan; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. route Birch overtook route Amber. The two expressions do not necessarily remain true under the same possible facts. The lead became larger while the ordering stayed the same. A dynamic comparison can change size without reversing rank.

Mechanism 34: Within ten means a bounded distance

Invariant. The absolute distance is at most ten; the wording alone does not fix its sign. A difference magnitude does not tell which party is higher.

Counterexample method. Imagine A = B or A slightly above B. Keep the bounded-distance relation unless context establishes direction. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 67: Within ten means a bounded distance — a hypothetical multilingual media workflow

Source-control proposition. editing pipeline Amber was within 10 minutes per episode of editing pipeline Blue on caption review time. This is a synthetic training statement about caption review time in a hypothetical multilingual media workflow; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. editing pipeline Amber was exactly 10 minutes per episode lower than editing pipeline Blue. The two expressions do not necessarily remain true under the same possible facts. The absolute distance is at most ten; the wording alone does not fix its sign. A difference magnitude does not tell which party is higher.

Repair and transfer check. Imagine A = B or A slightly above B. Keep the bounded-distance relation unless context establishes direction. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 68: Within ten means a bounded distance — a hypothetical manufacturing audit

Source-control proposition. method Aspen was within 10 hours per batch of method Briar on quality-control review time. This is a synthetic training statement about quality-control review time in a hypothetical manufacturing audit; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. method Aspen was exactly 10 hours per batch lower than method Briar. The two expressions do not necessarily remain true under the same possible facts. The absolute distance is at most ten; the wording alone does not fix its sign. A difference magnitude does not tell which party is higher.

Mechanism 35: At least ten more is a difference threshold

Invariant. The difference is ten or greater. Reversing the threshold changes the set of permitted measurements.

Counterexample method. Would a twenty-unit gap satisfy the source? Attach the lower-bound operator to the difference, not the absolute value of A. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 69: At least ten more is a difference threshold — a fictional distribution plan

Source-control proposition. route Amber exceeded route Birch by at least 10 calendar days per order. This is a synthetic training statement about delivery lead time in a fictional distribution plan; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. route Amber exceeded route Birch by at most 10 calendar days per order. The two expressions do not necessarily remain true under the same possible facts. The difference is ten or greater. Reversing the threshold changes the set of permitted measurements.

Repair and transfer check. Would a twenty-unit gap satisfy the source? Attach the lower-bound operator to the difference, not the absolute value of A. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 70: At least ten more is a difference threshold — a constructed training-programme comparison

Source-control proposition. course Alpha exceeded course Beta by at least 10 hours per participant. This is a synthetic training statement about practice time in a constructed training-programme comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. course Alpha exceeded course Beta by at most 10 hours per participant. The two expressions do not necessarily remain true under the same possible facts. The difference is ten or greater. Reversing the threshold changes the set of permitted measurements.

Mechanism 36: No more than ten apart does not fix direction

Invariant. Only the absolute distance is constrained. Translating a symmetric relation as a directional one adds an ordering claim.

Counterexample method. Could the two have equal values or could B be larger? Keep the symmetric magnitude bound. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 71: No more than ten apart does not fix direction — a hypothetical manufacturing audit

Source-control proposition. method Aspen and method Briar were no more than 10 hours per batch apart. This is a synthetic training statement about quality-control review time in a hypothetical manufacturing audit; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. method Aspen was necessarily greater than method Briar. The two expressions do not necessarily remain true under the same possible facts. Only the absolute distance is constrained. Translating a symmetric relation as a directional one adds an ordering claim.

Repair and transfer check. Could the two have equal values or could B be larger? Keep the symmetric magnitude bound. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 72: No more than ten apart does not fix direction — a controlled equipment test

Source-control proposition. design Aurora and design Bramble were no more than 10 watt-hours per cycle apart. This is a synthetic training statement about electricity consumption in a controlled equipment test; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. design Aurora was necessarily greater than design Bramble. The two expressions do not necessarily remain true under the same possible facts. Only the absolute distance is constrained. Translating a symmetric relation as a directional one adds an ordering claim.

6.4. Equality, preference and judgment: casebook and comparative audit

Mechanism 37: As much as can express equality

Invariant. The phrase expresses equality in this sentence, not ‘a very large amount’. A degree phrase can change function across syntactic environments.

Counterexample method. Would values of 20 and 20 fit the source? Preserve equivalence, even if the target language uses a different grammatical structure. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 73: As much as can express equality — a constructed training-programme comparison

Source-control proposition. In the hypothetical paired test, course Alpha used as much practice time as course Beta. This is a synthetic training statement about practice time in a constructed training-programme comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. course Alpha used more practice time than course Beta. The two expressions do not necessarily remain true under the same possible facts. The phrase expresses equality in this sentence, not ‘a very large amount’. A degree phrase can change function across syntactic environments.

Repair and transfer check. Would values of 20 and 20 fit the source? Preserve equivalence, even if the target language uses a different grammatical structure. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 74: As much as can express equality — an illustrative pricing schedule

Source-control proposition. In the hypothetical paired test, plan Amber used as much subscription charge as plan Birch. This is a synthetic training statement about subscription charge in an illustrative pricing schedule; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. plan Amber used more subscription charge than plan Birch. The two expressions do not necessarily remain true under the same possible facts. The phrase expresses equality in this sentence, not ‘a very large amount’. A degree phrase can change function across syntactic environments.

Mechanism 38: As much as can introduce a surprising extreme

Invariant. The high figure is a possible extreme rather than a universal result. The same English words do not always perform an equality comparison.

Counterexample method. Could a lower ordinary observation coexist with the stress test? Translate the extreme-possibility reading, not the equality reading. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 75: As much as can introduce a surprising extreme — a controlled equipment test

Source-control proposition. Under the stress-test scenario, design Aurora could require as much as 30 watt-hours per cycle. This is a synthetic training statement about electricity consumption in a controlled equipment test; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. design Aurora invariably requires 30 watt-hours per cycle. The two expressions do not necessarily remain true under the same possible facts. The high figure is a possible extreme rather than a universal result. The same English words do not always perform an equality comparison.

Repair and transfer check. Could a lower ordinary observation coexist with the stress test? Translate the extreme-possibility reading, not the equality reading. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 76: As much as can introduce a surprising extreme — a fictional clinic scheduling example

Source-control proposition. Under the stress-test scenario, queue Aster could require as much as 30 days per appointment. This is a synthetic training statement about appointment waiting time in a fictional clinic scheduling example; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. queue Aster invariably requires 30 days per appointment. The two expressions do not necessarily remain true under the same possible facts. The high figure is a possible extreme rather than a universal result. The same English words do not always perform an equality comparison.

Mechanism 39: Rather than indicates a choice

Invariant. The source expresses a choice between alternatives, not a measured comparison. The word ‘than’ alone does not make a statement quantitative.

Counterexample method. Does the source cite a measured result at all? Preserve choice or preference without manufacturing a metric. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 77: Rather than indicates a choice — an illustrative pricing schedule

Source-control proposition. The instruction selects plan Amber rather than plan Birch for the next trial. This is a synthetic training statement about subscription charge in an illustrative pricing schedule; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. plan Amber has a numerically higher subscription charge than plan Birch. The two expressions do not necessarily remain true under the same possible facts. The source expresses a choice between alternatives, not a measured comparison. The word ‘than’ alone does not make a statement quantitative.

Repair and transfer check. Does the source cite a measured result at all? Preserve choice or preference without manufacturing a metric. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 78: Rather than indicates a choice — an archival indexing trial

Source-control proposition. The instruction selects workflow Atlas rather than workflow Beacon for the next trial. This is a synthetic training statement about processing time in an archival indexing trial; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. workflow Atlas has a numerically higher processing time than workflow Beacon. The two expressions do not necessarily remain true under the same possible facts. The source expresses a choice between alternatives, not a measured comparison. The word ‘than’ alone does not make a statement quantitative.

Mechanism 40: Prefer one option to another

Invariant. Preference may depend on several factors; the statement does not identify one winning statistic. A translator must not turn a human judgment into a universal empirical finding.

Counterexample method. Could a reviewer still prefer B even if the measured times were equal? Retain preference and keep the reasons unspecified unless the source provides them. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 79: Prefer one option to another — a fictional clinic scheduling example

Source-control proposition. The reviewer prefers queue Brook to queue Aster for the next stage. This is a synthetic training statement about appointment waiting time in a fictional clinic scheduling example; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. queue Brook has a lower appointment waiting time than queue Aster in every setting. The two expressions do not necessarily remain true under the same possible facts. Preference may depend on several factors; the statement does not identify one winning statistic. A translator must not turn a human judgment into a universal empirical finding.

Repair and transfer check. Could a reviewer still prefer B even if the measured times were equal? Retain preference and keep the reasons unspecified unless the source provides them. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 80: Prefer one option to another — a simulated software benchmark

Source-control proposition. The reviewer prefers service Bay to service Alder for the next stage. This is a synthetic training statement about response latency in a simulated software benchmark; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. service Bay has a lower response latency than service Alder in every setting. The two expressions do not necessarily remain true under the same possible facts. Preference may depend on several factors; the statement does not identify one winning statistic. A translator must not turn a human judgment into a universal empirical finding.

Mechanism 41: Better than expected is evaluative, not competitor-based

Invariant. The comparison is with an expectation, not a rival option. Even a positive word such as ‘better’ contains a hidden comparator.

Counterexample method. Ask what expectation was held and whether B is even mentioned. Translate the evaluative baseline without inserting competitors. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 81: Better than expected is evaluative, not competitor-based — an archival indexing trial

Source-control proposition. The fictional reviewer judged workflow Atlas better than expected. This is a synthetic training statement about processing time in an archival indexing trial; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. The reviewer judged workflow Atlas better than workflow Beacon. The two expressions do not necessarily remain true under the same possible facts. The comparison is with an expectation, not a rival option. Even a positive word such as ‘better’ contains a hidden comparator.

Repair and transfer check. Ask what expectation was held and whether B is even mentioned. Translate the evaluative baseline without inserting competitors. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 82: Better than expected is evaluative, not competitor-based — a hypothetical horticulture comparison

Source-control proposition. The fictional reviewer judged garden plot Ash better than expected. This is a synthetic training statement about water consumption in a hypothetical horticulture comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. The reviewer judged garden plot Ash better than garden plot Beech. The two expressions do not necessarily remain true under the same possible facts. The comparison is with an expectation, not a rival option. Even a positive word such as ‘better’ contains a hidden comparator.

Mechanism 42: Not worse than is non-strict

Invariant. The claim permits equality and possibly improvement. Negated comparative evaluations preserve a tie option often lost in confident prose.

Counterexample method. Test the case where the evaluator considers them equally good. Keep a weak but precise non-inferiority claim. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 83: Not worse than is non-strict — a simulated software benchmark

Source-control proposition. The evaluator stated that service Alder was no worse than service Bay on the stated criterion. This is a synthetic training statement about response latency in a simulated software benchmark; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. The evaluator stated that service Alder was strictly better than service Bay. The two expressions do not necessarily remain true under the same possible facts. The claim permits equality and possibly improvement. Negated comparative evaluations preserve a tie option often lost in confident prose.

Repair and transfer check. Test the case where the evaluator considers them equally good. Keep a weak but precise non-inferiority claim. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 84: Not worse than is non-strict — a hypothetical multilingual media workflow

Source-control proposition. The evaluator stated that editing pipeline Amber was no worse than editing pipeline Blue on the stated criterion. This is a synthetic training statement about caption review time in a hypothetical multilingual media workflow; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. The evaluator stated that editing pipeline Amber was strictly better than editing pipeline Blue. The two expressions do not necessarily remain true under the same possible facts. The claim permits equality and possibly improvement. Negated comparative evaluations preserve a tie option often lost in confident prose.

Mechanism 43: Better on one measure does not mean better overall

Invariant. Evidence concerns one dimension only. Overall judgments involve additional dimensions and weights absent from the sentence.

Counterexample method. Imagine B outperforming A on an unmeasured but important criterion. Preserve dimensional scope and avoid overall endorsement. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 85: Better on one measure does not mean better overall — a hypothetical horticulture comparison

Source-control proposition. garden plot Ash scored higher than garden plot Beech on the specified water consumption measure. This is a synthetic training statement about water consumption in a hypothetical horticulture comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. garden plot Ash was the better option overall. The two expressions do not necessarily remain true under the same possible facts. Evidence concerns one dimension only. Overall judgments involve additional dimensions and weights absent from the sentence.

Repair and transfer check. Imagine B outperforming A on an unmeasured but important criterion. Preserve dimensional scope and avoid overall endorsement. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 86: Better on one measure does not mean better overall — a fictional distribution plan

Source-control proposition. route Amber scored higher than route Birch on the specified delivery lead time measure. This is a synthetic training statement about delivery lead time in a fictional distribution plan; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. route Amber was the better option overall. The two expressions do not necessarily remain true under the same possible facts. Evidence concerns one dimension only. Overall judgments involve additional dimensions and weights absent from the sentence.

Mechanism 44: Per-unit rates require denominator identity

Invariant. Identical denominator definitions are assumed in this constructed comparison. A denominator change is not harmless formatting; it changes the quantity itself.

Counterexample method. Write numerator and exposure base beneath each reported value. Keep the denominator wording consistent or flag non-comparability. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 87: Per-unit rates require denominator identity — a hypothetical multilingual media workflow

Source-control proposition. editing pipeline Amber registered 30 minutes per episode, while editing pipeline Blue registered 20 minutes per episode, under the same counting rule. This is a synthetic training statement about caption review time in a hypothetical multilingual media workflow; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. editing pipeline Amber had 30 units per person and editing pipeline Blue had 20 units per order. The two expressions do not necessarily remain true under the same possible facts. Identical denominator definitions are assumed in this constructed comparison. A denominator change is not harmless formatting; it changes the quantity itself.

Repair and transfer check. Write numerator and exposure base beneath each reported value. Keep the denominator wording consistent or flag non-comparability. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 88: Per-unit rates require denominator identity — a hypothetical manufacturing audit

Source-control proposition. method Aspen registered 30 hours per batch, while method Briar registered 20 hours per batch, under the same counting rule. This is a synthetic training statement about quality-control review time in a hypothetical manufacturing audit; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. method Aspen had 30 units per person and method Briar had 20 units per order. The two expressions do not necessarily remain true under the same possible facts. Identical denominator definitions are assumed in this constructed comparison. A denominator change is not harmless formatting; it changes the quantity itself.

Mechanism 45: About is a precision claim, not exact equality

Invariant. The claim is intentionally approximate. A neat-looking number can be part of a rounded summary rather than a contractual specification.

Counterexample method. Would 29.7 be consistent with the approximate source? Maintain approximation and avoid fabricated decimal precision. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 89: About is a precision claim, not exact equality — a fictional distribution plan

Source-control proposition. The report describes route Amber as about 30 calendar days per order. This is a synthetic training statement about delivery lead time in a fictional distribution plan; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. The report certifies exactly 30 calendar days per order. The two expressions do not necessarily remain true under the same possible facts. The claim is intentionally approximate. A neat-looking number can be part of a rounded summary rather than a contractual specification.

Repair and transfer check. Would 29.7 be consistent with the approximate source? Maintain approximation and avoid fabricated decimal precision. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 90: About is a precision claim, not exact equality — a constructed training-programme comparison

Source-control proposition. The report describes course Alpha as about 30 hours per participant. This is a synthetic training statement about practice time in a constructed training-programme comparison; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. The report certifies exactly 30 hours per participant. The two expressions do not necessarily remain true under the same possible facts. The claim is intentionally approximate. A neat-looking number can be part of a rounded summary rather than a contractual specification.

Mechanism 46: No greater than is not no less than

Invariant. The first expression sets an upper bound, not a lower bound. Negated degree terms are vulnerable to polarity reversal in fast drafting.

Counterexample method. Can zero satisfy the source? Could 31? Use a target structure tested at both boundaries. A professional reviewer should record this decision when a number or comparison determines payment, access, reporting or public trust.

Diagnostic 91: No greater than is not no less than — a hypothetical manufacturing audit

Source-control proposition. The maximum authorised value for method Aspen is no greater than 30 hours per batch. This is a synthetic training statement about quality-control review time in a hypothetical manufacturing audit; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. The minimum authorised value is 30 hours per batch. The two expressions do not necessarily remain true under the same possible facts. The first expression sets an upper bound, not a lower bound. Negated degree terms are vulnerable to polarity reversal in fast drafting.

Repair and transfer check. Can zero satisfy the source? Could 31? Use a target structure tested at both boundaries. A reviewer then checks the proposed natural target sentence against a counterfactual example at the boundary; if the target changes truth value or invents a preference, certainty, guarantee or denominator, return it for revision. This check transfers to captions, user-interface strings, reports and spoken explanations, not just full prose paragraphs.

Diagnostic 92: No greater than is not no less than — a controlled equipment test

Source-control proposition. The maximum authorised value for design Aurora is no greater than 30 watt-hours per cycle. This is a synthetic training statement about electricity consumption in a controlled equipment test; the figures are invented and describe only the stated comparison, not any actual service or test outcome.

Tempting but unsafe target paraphrase. The minimum authorised value is 30 watt-hours per cycle. The two expressions do not necessarily remain true under the same possible facts. The first expression sets an upper bound, not a lower bound. Negated degree terms are vulnerable to polarity reversal in fast drafting.

7. Thirty comparison-specific failure modes and their repairs

Failure mode 1: Comparator substitution

A past value is translated as a competitor’s present value. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Write ‘compared with what?’ beside the source and prohibit a target competitor not named by the evidence. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 2: Reference-group inflation

‘Among these three’ disappears and a local ranking becomes global. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Retain the group and its sampling boundaries. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 3: Comparator inversion

Ratio numerator and denominator change places without changing the result. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Check the written ratio against known simple numbers. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 4: Strict/non-strict confusion

Greater than becomes greater than or equal to. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Test exactly the threshold. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 5: Polarity reversal

No more than becomes no less than. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Draw the number line and mark valid values. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 6: Added equality

A strict comparison is softened into an inclusive comparison. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Create equal values and see which sentence is true. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 7: Deleted equality

A non-strict expression becomes a strict one. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Try a tie and inspect both versions. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 8: Absolute-relative substitution

Ten units becomes ten percent. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Name the dimension of each number. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 9: Relative baseline reversal

Fifty percent higher becomes fifty percent lower by swapping the reference. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Compute the ratio from the claimed starting value. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 10: Percentage-point collapse

A ten-point increase in a share becomes a ten-percent relative increase. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Write starting and ending percentages. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 11: Rank-value confusion

Second-highest becomes second-best. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Keep the measured scale separate from desirability. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 12: Median-average substitution

An average wait becomes the typical patient’s precise wait. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Retain the statistic and its population. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 13: Sample-universe expansion

A benchmark over selected machines becomes a claim over all machines. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Keep the sampling limitation in the same sentence or nearby. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 14: Time-slice drift

Improvement over last month becomes superiority over other models. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Identify both time coordinates. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 15: Endpoint-delta confusion

Increased by 10 becomes increased to 10. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Record before, change and after separately. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 16: Maximum-guarantee substitution

Up to 30 becomes always 30. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Distinguish permitted extreme from expected outcome. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 17: Forecast-observation confusion

More than expected becomes more than measured. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Label forecast, goal and actual observation. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 18: Choice-performance confusion

Rather than becomes better than. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Do not invent metrics for a decision expression. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 19: Preference-proof confusion

A stakeholder prefers an option and the translation says it is objectively superior. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Retain the subjective source of the evaluation. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 20: Dimensional leakage

Faster on one test becomes better in every regard. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Keep the measured property visible. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 21: Unit mismatch

One side uses per hour while the other uses per task. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Normalise only with justified quantities; otherwise mark non-comparability. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 22: Denominator omission

A rate is translated as a raw count. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Restore or query the exposure base. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 23: Range-endpoint drift

Between ten and thirty loses whether endpoints count. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Check the source context or query inclusive/exclusive interpretation. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 24: Approximation hardening

About becomes exactly. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Preserve approximate lexical signals. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 25: Unlicensed rounding

29.6 becomes 30 in a safety or payment limit without authorization. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Apply a declared rounding rule, or leave the original precision. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 26: Double-negation flattening

Not unlikely to exceed becomes certain to exceed. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Separate the scopes of negation and probability. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 27: Comparison-class ambiguity

A model is called fastest but the class of competitors is missing. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Ask whose set defines the superlative. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 28: Ambiguous multiplication

‘Five times more’ is rewritten as a precise multiplier without checking the intended convention. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Ask the author for base and endpoint values. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 29: Rhetorical-force deletion

No less than conveys astonishment as well as a minimum, but tone vanishes. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Preserve the factual boundary first, then calibrate emphatic force if relevant. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

Failure mode 30: Factual-claim laundering

A purported comparison is presented as verified evidence when the source merely attributes it to a speaker. In a bilingual review, this is not a harmless difference in stylistic preference: it changes the recoverable relation, the reader’s confidence or the resulting action.

Repair: Preserve who claims the numbers and the status of the evidence. Ask a reviewer to provide one simple possible world where the source and the candidate target would lead to different decisions; if such a world exists, the translation needs correction or an explicit source-side query.

8. Transfer laboratories: a working comparison audit in twelve settings

Transfer laboratory 1: Contract thresholds

A service agreement states a maximum processing time. Check inclusive versus exclusive limits, what counts as processing, and whether service interruptions are excluded. A translation that changes ‘at most’ into ‘less than’ may alter a contractual boundary. Draft three test values, have the bilingual reviewer judge each, and leave the exact legal interpretation to qualified counsel when stakes require it.

Practice cycle: annotate one source comparison, create a misleading paraphrase on purpose, explain the failure, draft a faithful target-language candidate, and ask another competent reader to recover the baseline without hints. For a second pass, change the comparison group’s composition or the numerical endpoint and verify that the target expression reacts correctly. This is how a single lesson becomes a portable review skill rather than a memorised translation.

Transfer laboratory 2: Product specifications

A device is described as using ‘up to’ a specified energy quantity. Distinguish a ceiling, an observed peak and an average. Keep the duty cycle and testing context. Use a target sentence that does not promise the device always reaches that peak or that it saves energy against an unmentioned competitor.

Transfer laboratory 3: Public-health explanations

An illustrative report compares wait times between groups. Preserve statistic, dates, denominators and uncertainty. Avoid making advice about treatment or personal outcomes from one comparative figure. The method transfers to public-facing guidance: measured difference first, interpretation second, and a clear statement of what the comparison does not establish.

Transfer laboratory 4: Research abstracts

A study claims a difference in proportions. Separate points from relative percent and identify which group provides the denominator. Preserve qualifiers such as ‘in this sample’ and ‘under these conditions’. When the original is ambiguous, mark a query rather than selecting the more dramatic number.

Transfer laboratory 5: Environmental reporting

Two processes have different energy consumption per functional unit. Total emissions, energy per task and energy per product are not interchangeable. Record the unit basis and boundaries, then compare only like with like. A lower rate may coexist with greater total output.

Transfer laboratory 6: Financial disclosures

A fee rises from twenty to thirty. Distinguish a ten-unit increment, a fifty-percent increase and a final price of thirty. Keep whether tax, promotional rates or billing periods are included. In a legal or regulated setting, obtain the required specialist review before release.

Transfer laboratory 7: Hiring and personnel review

One candidate is described as more experienced than another. Ask whether the source refers to years, a specific role, confidence or an evaluator’s impression. Do not transform a preference into a proven performance result. Keep sensitive judgments attributed and bounded.

Transfer laboratory 8: Education and training

One programme asks for more practice hours than another. Do not claim that more hours inevitably produce better learning. Preserve whether the measure is assigned time, actual participation or an estimate. Comparatives about education become misleading when input measures are silently promoted to outcomes.

Transfer laboratory 9: Software benchmarks

A benchmark compares latency. Keep percentile, load, hardware, concurrency and sample. A statement about a median cannot become a guarantee for every request. Test whether the target mislabels lower latency as higher throughput; those are different dimensions.

Transfer laboratory 10: Museum and cultural captions

A catalogue calls one work ‘older than’ another. Determine whether age refers to object date, technique, source manuscript or era of the represented event. A fluent caption can change the historical relation by silently replacing the dated object with its subject matter.

Transfer laboratory 11: Media subtitles

A spoken comparative contains hesitation or a correction. The subtitle has limited space, but condensing it must not swap which person or thing is preferred. Preserve the final corrected comparator and, when the hesitation is consequential to characterisation, retain its communicative function within timing constraints.

Transfer laboratory 12: Machine translation evaluation

A system output is rated more fluent but less accurate than another. Fluency and accuracy are separate axes. Build a test with one fluent false comparison and one less graceful faithful comparison. Review against source truth conditions before grading style, and document whether an error came from arithmetic, scope, reference or terminology.

9. A four-week practice programme for comparative fidelity

Practice 1: Week one: read the relation

Collect twenty authentic comparative sentences from permitted materials. For each, identify two terms, metric, baseline, inequality, group and attribution. Do not translate until you can restate the source neutrally. Use five boundary tests for strict versus non-strict expressions. Record mistakes without rewriting the source.

Progress is not measured by how many comparative words one can recite. It is measured by whether another reader can recover the same relation and apply it to an unfamiliar numerical case. At the end of the session, select one error that would have changed a real decision, and describe the exact check that would have caught it before release.

Practice 2: Week two: numerical and logical precision

Construct contrasting pairs of absolute differences, relative increases, percentages, percentage points, ranking and equalities. Have a second reviewer independently compute the differences. Practise turning the correct interpretation into natural target prose without adding reasons, predictions or approval.

Practice 3: Week three: context and register

Take comparisons from instructions, research, customer communication, contracts, subtitles and everyday conversation. Observe which fields are explicit and which must be recovered from context. Translate twice for different target audiences; preserve the relation while changing formality, length and genre.

Practice 4: Week four: blind verification

Give a reviewer the target text without the source and ask them to reconstruct a comparison record. Reconcile it with the source record. Log each mismatch by category and severity, not just by whether a word ‘sounds off’. Retest after revision, and save a concise decision for future similar expressions.

10. Thirty-two questions readers ask about translating comparisons

1. Can I translate ‘more than’ with a single equivalent word?

Sometimes, but only after identifying its complement, baseline and threshold. A single word may not cover an idiomatic degree construction in another language.

2. Is ‘not more than’ always the same as ‘less than’?

No. The former commonly includes equality with the threshold; the latter normally excludes it.

3. How can I check ‘no less than’?

Turn it into an explicit minimum, test equality, and separately consider whether the expression adds rhetorical emphasis.

4. Does ‘as much as’ always mean equality?

No. In different constructions it can express equality, a possible maximum, emphasis or a concessive relation.

5. How should ‘rather than’ be translated?

First decide whether the source expresses selection, contrast or correction. It need not claim that one option has a higher measured value.

6. What is the difference between percentage and percentage points?

Percentage points subtract two percentage levels; relative percent change divides the difference by a chosen starting level.

7. What does ‘fifty percent higher’ compare against?

The reference value following ‘than’ or supplied by context. Without a baseline the arithmetic is incomplete.

8. Why is a ratio’s order important?

A:B = 3:2 describes a different ordering from B:A = 3:2; switching labels without inverting the ratio changes meaning.

9. Is higher always better?

No. Higher can describe a longer delay, greater expense, more output or stronger performance. The evaluated dimension matters.

10. Does ‘the best’ require a reference group?

For precise reporting, yes: best among which alternatives and on which criterion? Context may supply the group but the translation must not expand it.

11. Can I translate ‘lowest tested’ as ‘lowest worldwide’?

No. This silently enlarges the comparison set beyond the source evidence.

12. Can an average become a typical individual result?

Not automatically. Averages, medians, percentiles and individual outcomes are different statistical claims.

13. Should I calculate during translation?

Use arithmetic as a verification aid when values are given. Do not invent missing quantities or editorially correct a source without authorisation.

14. Can a comparison be correct with different wording?

Yes. Different syntax may be natural in the target language provided the relational truth conditions and communicative function remain intact.

15. What if the source is mathematically wrong?

Flag the discrepancy with evidence; ask the responsible author or client. Do not quietly replace source claims with new claims in a translation.

16. What if the source is ambiguous?

Retain ambiguity where possible, query the author when consequential, and document any authorised interpretation.

17. Should rates be compared before units are harmonised?

No. Check whether numerator, denominator, population and period actually refer to compatible quantities.

18. Can ‘up to’ promise the maximum?

No. Its ordinary upper-bound use is compatible with values below the maximum.

19. What is an equality test?

Imagine that both compared values are identical. If a candidate translation has different truth conditions from the source, it likely mishandles strictness.

20. What is a reversal test?

Swap the compared entities and check whether the inverse relation is expressed correctly.

21. What is a boundary test?

Try values below, at and above the stated threshold to see whether the candidate text allows the same cases.

22. What is baseline drift?

A shift in the comparator, such as replacing a past value with a competing product or a forecast with a measured result.

23. Is a preference a performance comparison?

Not necessarily. People can prefer an option for many reasons; preference does not by itself report a measured advantage.

24. Can machine translation make these mistakes?

Yes. Evaluate outputs against explicit source relations, including numbers, negation, groups and attribution rather than relying on fluency.

25. Is this guide a substitute for qualified review?

No. It is a diagnostic method; high-stakes material needs the appropriate specialist and bilingual verification.

26. How do I document an issue?

Record the source sentence, candidate target, actual comparator, failed test, decision, evidence and reviewer.

27. When should I stop revising?

When the required quality standard and risk-based acceptance criteria are met, including any mandatory subject-matter review.

28. Does this chapter teach comparative grammar?

It touches grammar only to protect a cross-language meaning relation. The complete grammar-learning owner remains separate.

29. What if a term is unfamiliar?

Resolve the source meaning and term first, using relevant domain references. Word knowledge is not itself evidence of an accurate comparison.

30. Does a comparative always involve numbers?

No. Preferences, evaluations, rankings, similarity and contrast can be qualitative, but they still require clear parties and standards.

31. Can translations reduce the length of comparisons?

Yes when the medium requires it, provided omission does not delete a baseline, qualifier, direction, scope or material distinction.

32. What should I learn next?

Return to the translation root for the complete workflow. Follow the preceding specificity chapter if referents are unstable, Vocabulary for lexical gaps, and How English Works for monolingual grammar or usage.

11. Revision rubric: a small scorecard with serious consequences

Review area	Question	Suggested action
Source relation	Can the reviewer identify who or what is compared, and on what dimension?	Pass, revise, or query the source; document any material decision.
Baseline integrity	Is the comparator the same control, prior value, competitor, forecast or threshold?	Pass, revise, or query the source; document any material decision.
Logical integrity	Do inequality direction, strictness, negation and equality behave identically?	Pass, revise, or query the source; document any material decision.
Magnitude integrity	Are unit differences, ratios, percentages, percentage points, ranks and approximations kept distinct?	Pass, revise, or query the source; document any material decision.
Population integrity	Are group, denominator, period, sample and exclusions preserved?	Pass, revise, or query the source; document any material decision.
Evidence integrity	Does the target preserve observation versus estimate, preference, claim and hypothetical?	Pass, revise, or query the source; document any material decision.
Reader action	Would the same plausible reader take the same materially relevant decision from both versions?	Pass, revise, or query the source; document any material decision.
Target naturalness	Is the message idiomatic, readable and appropriate without a change in meaning?	Pass, revise, or query the source; document any material decision.

This scorecard is a practical editorial aid, not an official certification scale. An error affecting a legal limit, payment, access, clinical communication or public reporting deserves greater attention than a harmless stylistic difference. The severity of the consequence must be considered alongside error type; a fluent sentence may still be unacceptable if it reverses the reader’s decision.

12. Transfer and next routes

There is no need to translate one fixed grammatical form everywhere. A skilled translator can reformulate a comparison with a clause, a preposition, a paired expression, an equation, an explicit baseline or a short explanation as target conventions require. The freedom is on the surface of the sentence. The discipline is beneath it: keep the same parties, the same measured property, the same direction, the same threshold, the same reference group and the same evidential status.

Before release, review the target independently for naturalness and then compare it with the source for accuracy. This two-sided evaluation avoids two familiar traps: a literal but unusable translation and a smooth translation that distorts the claim. Whenever a key value determines eligibility, safety, payment, timing or accountability, do not waive review because a sentence reads beautifully.

This chapter’s contribution to the larger architecture is the comparative record: subject, comparator, dimension, direction, magnitude, denominator and evidence. It does not claim ownership of reference identity, lexical acquisition, English-only grammar, domain regulatory advice or all quantitative localisation. Those layers can be linked when relevant without taking over their scope.

Continue with the Master Art of Translation root for the whole system of source reading, terminology, equivalence, workflow and bilingual revision. If it is unclear which entity a comparison refers to, use Definiteness and Specificity Fidelity. If the problem is knowledge of comparative words or numerical vocabulary, enter the protected Vocabulary Learning Hub. If the issue is using English comparative constructions correctly when no translation is happening, use the protected How English Works. These are next routes, not duplicate owners.

13. References and standards for careful terminology

Cambridge Dictionary, English Grammar Today: Comparison: adjectives. Used for authoritative distinctions between comparative and superlative constructions and the role of ‘than’. This is a language-description source, not a universal translation template.

Cambridge Dictionary, English Grammar Today: Rather. Used to distinguish degree use from the alternative/preference function of ‘rather than’.

W3C Multidimensional Quality Metrics community terminology. Its distinction between accuracy, terminology and fluency informs the separation between a meaning error and a well-formed target sentence.

W3C MQM issue-type taxonomy. Used as a reference for classifying accuracy-related translation problems; the scorecard here is the article’s own teaching model, not an official MQM certification.

Reference scope note: grammatical descriptions of English help identify source-control forms; they do not establish direct equivalence for every language pair. Numerical cases are deliberately hypothetical. Follow the current official standards, source documents, authorised terminology and professional review required by the actual translation commission.