ApiaryActiveLive
Try: pause · settings · learn · wipe
← Community / Reading Room
KT
craft · 24 min read

Keep the Number and Its Unit Together

A catalog sheet for two portable display stands lists the width of one in inches, the transport length of the second in millimeters, and the mass of each…

A catalog sheet for two portable display stands lists the width of one in inches, the transport length of the second in millimeters, and the mass of each packed shipment in kilograms. The summary says the large stand is “900 inches wide, only 6 kg.” There are two visible numbers and a pair of plausible-looking units. Both numbers were in the source. Only one of them still refers to a true statement, and even that sentence has changed what the six kilograms describes.

This is the sort of mistake a fluent answer can make while getting every digit right. A specification is not a lucky association between a number and whichever noun is nearest. It is a claim about a particular object and property, under a stated condition, using an approved unit and degree of precision. If an extraction tool carries only the number forward, the rest of the claim can shift invisibly.

The examples below concern public documentation for invented, ordinary display products. They are not actual equipment recommendations, a substitute for design or fit review, or guidance for engineering safety decisions. A summary should never override a manufacturer's current applicable specifications or an authorized expert's interpretation.

AI disclosure: Drafted with AI assistance. Product sheets, numeric examples and review receipts are synthetic, not reports of actual products or testing.

Define the thing an auditor should check

A workable extraction keeps seven related facts together: source edition, product variant, measured property, value, unit, qualifier and condition. Not every sheet will supply all seven. That absence should be visible. “Unit not stated in reviewed source” is much more useful than a convincing guess. The reviewer can decide which facts are essential to report, and the writing tool can leave the claim unresolved instead of smoothing it over.

Consider “nominal display width: 24 in.” Nominal is a qualifier, display width is the property, and the unit is part of the value presentation. If the source attaches that phrase to the expanded state of stand S-1, a rigorous summary shouldn't use it for transport length on stand S-2. It also should not promise clearance in an enclosure of precisely that width. A nominal number has neither been observed on the reader's object nor established as a maximum.

For a short evidence card, copy a brief source fragment and identify its table heading, row and applicable notes. Store the interpretation separately from the fragment. The card may say, “S-1 extended display width, nominal 24 in., visible in Sheet A revision 4, width row, column Extended.” That is a claim a person can challenge. “S-1 measures 24” is not useful enough to support comparison or action.

If you need to aggregate many fields, don't force public presentation to mirror the evidence table. The audit record can keep dense details that readers do not need in every sentence. A public summary can show the most important dimensions in clear prose and indicate limits. The key constraint is that a shortened sentence does not assert a broader or different fact than the reviewed record.

Follow a number to its heading, not to the nearest word

Public equipment documents often have wide tables, merged headings, footnotes and repeated row labels. A copy into plain text can preserve all the words in the wrong order. “900 | Packed | 6” might look well structured when it was actually part of three distinct columns joined during capture. Open the actual source view rather than asking a model to reconstruct a missing table from familiarity.

A common grouping has variants across the top and conditions nested underneath. For example, Width might have a separate entry under Base, Extended and In carton for each of two models. A small vertical rule on paper can be the only visible clue. If the document is rendered as plain text or OCR, record the layout consequence explicitly. The parser's ability to return a rectangular array says nothing about whether its cells followed the intended grouping.

When only the row title contains the unit, attach that unit before passing the value around. If a note says “unless noted, lengths in mm,” check whether the exact cell explicitly uses another unit. Some cells might say in. beside a number. Do not resolve a conflict by choosing whichever symbol occurs most often. A useful output reads “cannot determine from extracted text; original table needed” rather than fabricating a majority interpretation.

A copied table can also move a note away from the relevant row. A qualifier like “packed with optional tray” might apply to a specific mass for one variant. If the cropped view omits it, that numeric fact should remain unverified. For authorized teaching fixtures, make a test that deliberately removes the footer containing the scope. The extractor should show an unresolved condition instead of reusing the same answer in both full and cropped cases.

Reviewers can record “source found, association not confirmed” as a distinct outcome. If the check fails because formatting hid the unit, that is different from a clear association being summarized incorrectly. The first calls for better source capture or review. The second calls for changes in how the interpretation is assembled or tested. A one-word error category like numeric won't reliably route the work.

Use a worksheet that keeps the evidence claim-sized

Build a compact register for one public sheet with only a few sample fields. Here is a fictional set:

Candidate observationSource interpretationImportant limit
24 in.S-1 nominal display width under ExtendedNot its packed width
900 mmS-2 packed length of outer shipping cartonNot its display width
6 kgS-2 packed mass for standard kitNot bare frame mass
About 18 in.S-2 useful display opening in catalog illustrationNo exact tolerance established

These values cannot be swapped simply because a common adjective describes both models. Write a provisional summary from the register, not directly from the text stream. “The S-1 sheet lists a nominal extended width of 24 inches. The S-2 sheet lists a standard packed mass of 6 kg and a carton length of 900 mm; neither value is the bare-frame size.” The negative sentence might not be necessary in an ordinary catalog, but it's useful for verification because it identifies the most likely cross-association.

Give each record a claim ID supporting review and updates. If a document changes its revision, the same claim ID can refer to a new candidate that must be checked. A stable ID does not make content immutable. Save what the reviewed edition actually said or link a permitted snapshot so an older summary can be reconstructed. Never let a claim continue to appear approved merely because a mutable source anchor still resolves.

Use categorical review statuses that distinguish copied, interpreted, not comparable, unresolved and unsupported. A valid property claim can be not comparable to another model because one number describes a packed version and another a bare version. A sentence asserting one is lighter then needs the same basis for both. If that evidence is not present, publish the two legitimate statements separately or leave the comparison out. The system's job is to make that gap easy to notice.

A note explaining the next check can stay short. “Need unpacked mass for S-2 before making a like-for-like comparison” is better than “please verify.” It gives the next person a defined question and avoids training the writer to ask the supplier to confirm everything again. The underlying reading principles remain useful even if a future workflow stops using a language model entirely.

Preserve the direction of comparisons and limits

Three is not necessarily three. In a statement “up to three sample trays,” three is the stated maximum, not exactly three, not at least three and not a capacity instruction for every use. “At least three trays in this bundle” is the opposite direction. “Approx. three in the illustration” is not a purchasing quantity. Keep the logical role next to the number when assessing model summaries.

Imagine a fictional panel bracket described in an ordinary display catalog as supported in ranges “from 10 to 20 mm” for a specific decorative insert configuration. The entry tells you something about the scope written there. It does not make every model with a 14 mm measurement interchangeable, and it does not establish measured performance of a user's install. If the exact inclusion of endpoints or fit matter, refer the reader to the authoritative documentation and an appropriate reviewer. This article is about checking a claim, not choosing parts for use.

A comparison such as “S-3 is smaller” needs a property, comparator and state. If the same variant family has a shorter box and a wider display, “smaller” is incomplete. Write precise comparisons only when the source supports both sides in equivalent conditions. “The documented shipping cartons differ in listed length” is a narrower fact. Keep superlatives such as smallest, lightest and most compact out unless you have an authorized defined comparison group and the evidence actually supports it.

Test changes in degree separately from numerical replacement. One model takes “nominal 24 inches” and output says “exactly 24 inches.” Another preserves “nominal” but joins it to the wrong property. A third omits a known range and prints its middle value as if the source reported it. A strong test catches all three failures and gives a different explanation for each. It should also accept a faithful paraphrase that expresses the same meaning in natural words.

Format values without pretending to convert them

The NIST SI guide's chapter on expressing quantities discusses writing values with units. Unicode CLDR's numbers specification describes locale-sensitive number representations. A symbol, spacing convention or decimal presentation can legitimately change when formatting a value. Neither source tells you which unit the catalog author actually used, whether your system should convert units, or how to infer the original document's precision.

A workflow can store literal source text and a structured value when the interpretation is authorized and verified. Keep those layers apart. Seeing 1,250 can suggest a grouping separator in one context and a decimal in another. Confirm the document's representation convention before digit normalization when it affects interpretation. If the source's context is mixed or unknown, repeat the notation as clearly sourced or hold numeric conversion until a qualified reviewer settles it.

A format rule can preserve the same quantity in the same unit. A conversion changes the unit and usually the printed value. That can be useful for a deliberately approved comparison, but it brings other decisions about rounding, measurement precision, reporting limits and which output is authoritative. The default in this article is no conversion. A writer should not swap inches for millimeters because the surrounding paragraph happens to feature more metric values. If an authorized system converts, keep a clearly labeled calculated derivative and record the source value and unit.

Typography can influence interpretation too. A line break between a quantity and its unit may make the unit look like part of a different row. A portable format can use predictable layout without making the value inaccessible to readers using assistive technology. This guide is not a tutorial on table rendering. The narrow point is that a visual reviewer can catch a broken association in the final prose even if the data were correct upstream.

Ask what the source measured before discussing precision

An approved document might report a whole number because it is a category size, a nominal marketing length, a rounded typical value, or a specification with its own defined basis. A summary should not inject extra digits merely to look technical. Likewise it should not blindly round a value when the last digit affects the documented distinction. Ask which precision is part of the source and which belongs to a later presentation choice.

An invented stand description lists “about 5.5 kg unpacked” and “6 kg shipping weight.” A familiar-looking rewrite might say “5.500 kg” to match decimal display across the whole catalog. That implies a degree of precision the source doesn't claim. Another equal-width column might round both to 6 kg. If used without labels, that could obscure the source distinction between actual states. Prefer wording that carries approximate and packed separate from the numbers.

A figure can also be measured under one configuration. A source says “depth with handle folded” or “width before accessory fitted.” Those conditions shape the number as much as a numerical qualifier. Do not drop them to save a line. A reviewer who sees a puzzling discrepancy between source photos and summary may first need to confirm the condition, not ask a model to choose the higher number by guessing a safety margin.

Distinguish the claim from any recommendation a reader might derive. A catalog description tells what the document says about the named item. It does not determine whether a user can safely fit it through a doorway, carry it comfortably, or place it on their stand. If a high-consequence measurement is important, the reader needs approved confirmation or a proper measurement process. An accessible summary can help them find the relevant source without assuming their context.

Use a regression puzzle with a matching number in the wrong place

Suppose a safe, synthetic fixture includes two variants:

FieldModel AModel B
Display width600 mm24 in.
Packed width24 in.590 mm
Packed mass6 kg6 kg
Bare frame massNot statedApprox. 5.5 kg

A generated comparison says both display widths are 24 inches because it sees the statement under model B and a value of 24 inches elsewhere in model A's column. A digit-and-unit check alone passes. A claim check ties each figure to its property and group. If no authorized conversion is included, the clean answer lists both displayed widths in their original units without calling them equal or different by an inferred calculation.

Introduce the same mass situation. The model's opening states that both frames weigh 6 kg. This one passes all unit checks and purports to compare the same component, but the source refers to packed mass. An improved summary says both standard packages are listed at that mass, subject to the document's actual conditions. It can mention bare mass for model B if useful while noting no bare mass is stated for A in the fixture. That absence is a source fact, not evidence that model A weighs nothing.

Give reviewers one output with a clearly correct relation spelled differently. “The manufacturer lists six kilograms for the packaged standard version of both models” preserves the relationship if that is what the source establishes. A rigid exact text match could reject it for spelling the number in words. The validator should test meaning in its verified scope. Use direct quotation only when needed; a reliably checked paraphrase makes a public summary easier to read.

Close the exercise by changing one heading from packed to assembled. The numerical data may stay identical. A well-designed workflow should re-evaluate dependent sentences even though the digits did not change. This reveals whether the source association is real or merely a decorative note appended to the output.

Capture what the model was authorized to see

Good auditing is easier when you know exactly which source went into the prompt. Name the approved edition, variant subset and table excerpt, including material headings and notes. If a source is public, that does not mean any context you append about a customer's situation should be shared with a separate tool. Use synthetic or authorized context and do not request a model to resolve a real purchasing or safety decision from an incomplete extract.

An instruction can say: “Extract only the supported source claims. For each, keep the model, property, observed value, unit, qualifier, condition and source reference. Do not infer conversions, equivalence, fit, contents or overall recommendations. Mark missing context rather than choosing a likely unit.” That makes refusal to invent part of the work. A structured result helps but doesn't guarantee a useful audit. If the wrong table was retrieved, an elegant schema can still contain wrong facts.

The NIST Generative AI Profile describes confidently incorrect output and oversight as relevant risks. A mixed-unit spec summary can be confidently wrong in a way that's invisible until somebody traces a particular noun. The checks here are small responses to that problem: keep approved input available, compare each significant output against it and retain plausible alternative interpretations as unresolved rather than automatically selecting one.

Never ask a model to validate a measurement by citing itself. Use the actual source or a duly authorized later clarification. A second model trained on similar conventions can repeat the same extrapolation. Separate the aid that proposes a reading from the records and people allowed to confirm it. Two confident assertions are not independent evidence if both came from the same incomplete layout or assumed conversion.

Tell the difference between two equally plausible source editions

An equipment catalog can publish a broad family overview and a narrower technical card. One document's name might look authoritative without establishing its scope. The family overview can be useful for orientation; the applicable sheet for a named revision can be necessary to support a precise measurement. Don't choose whichever contains the number the summary already proposed. Establish edition, model mapping and audience before reconciling the two.

Consider an invented sheet that describes a folding display frame introduced in a prior season. It lists a 600 mm width without naming the exact notch configuration. The later guide for a specific retail variant lists two widths under two arrangement labels, and the source receipt doesn't show which one the retailer sells. The replacement answer is not necessarily the newer number. The supported conclusion may be “the family document lists 600 mm; the specific frame's applicable arrangement has not been verified.” Send the mapping question to the catalog owner or qualified reviewer.

Save claim disagreements without declaring the source is wrong. A question about packed width may have an answer in one edition and no answer in another. A source could use inches with a rounded display number while a different source uses millimeters with its own rounded value. You cannot declare exact equivalence merely because a plausible conversion looks close. Ask whether the two documents are about the same physical object, same configuration, and same property, and whether the stated precision supports any comparison people need.

Keep the as-of date and original page context. Whether a guide is publicly reachable now is not necessarily evidence that it was the document used in the past. Likewise a current detail page can contain display-oriented summaries that omit material conditions. If the answer needs exact provenance, refer to the actual authorized edition. Provide a clear note when the result is based on a saved example rather than the live product record.

A useful reconciliation decision separates “this document selected for this claim” from “all prior documents are now invalid.” A support page can appropriately continue to point to an older configuration while current retail materials cover the new one. If a model combined them, removing the conclusory sentence might be better than forcibly choosing which guide wins across the entire product family.

Treat pictures and diagrams as useful, limited evidence

A diagram with a double-ended arrow can supply the direction a number measures; a photograph can help a reader identify which drawer a label refers to. Neither is a license to infer a quantity by measuring pixels when the document doesn't establish scale. A catalog diagram might be schematic. Even an accurately scaled drawing could represent a different revision or accessory configuration. Keep numeric assertions tied to published values and their stated scope.

For a fictional display panel, the document has a shaded outline with 900 mm across an external carton drawing and two shorter values across internal packing diagrams. Crop away the drawing title and the longest line might look like display span. A multimodal tool can plausibly describe it without recognizing that it is looking at packaging. Resolve the diagram caption, arrows and associated unit first. Check that the visible callout can be identified by a second reviewer without relying on which reading the model gave first.

An illustration could show an open tray while its text only reports carton width. A caption “representative configuration, accessories not included” matters to any claim about contents. Don't use the image to override the quantity stated in the text or to conclude it applies to both optional and base kits. A narrow summary might include “the illustration depicts a stand; stated shipping length is for the standard kit.” If the visual context cannot determine the ambiguity, leave it open and label the assistance limitation.

Preserve the useful part of diagram references with as little sensitive or unnecessary data as practical. A public link and page reference may be enough if the authorized source is stable. A local fixture can use invented geometric shapes, labels and callouts without reproducing a manufacturer's complete protected material. An editor auditing a short public summary does not need to distribute every diagram that helped them understand it.

Build a synthetic hold case where the missing unit is the point

Suppose you're given two example tables after a text export. One has rows reading S-1 package length 900 and S-2 package length 910. The exported table lost a general unit note. A different source, received in the same batch, uses inches for external display widths. You can't decide from those neighboring tables whether the package values are millimeters. A human reviewer may judge some interpretations highly improbable for handheld displays. Plausibility is still different from reliable source attribution.

A correct extraction can note the exact raw values, unavailable units, named property from the excerpt and missing source note. It should not propose a comparative measurement as if the units were known to be shared. Perhaps the units are consistent across the two cells, but the allowed reporting convention in this workflow requires evidence. A useful next action is reopen the authorized source version and identify the scope of its header or unit note. If you only have the export, describe the numbers without using them in a released measurement comparison.

Change the fixture in one way: supply a header identifying millimeters for both package-length values, while the external display-width document stays unchanged. Now the package claims can be stated and compared within their documented scope. This shows the model that an abstention is not the preferred final answer regardless of evidence. An auditor wants it to wait when evidence is missing and report supported facts when the necessary condition is clear.

Make the missing cases realistic without rewarding timidity. For a second fixture, include a clear qualifier and uncommon but readily legible unit symbol. A system trained only to never risk a unit error might abstain on all less familiar settings. The evaluator should count missed supported facts separately from invented facts. A brittle overly hesitant summary is a different concern from an overconfident one. Good review notices which actual evidence could have made the answer useful.

Keep comparisons and promotional interpretations away from raw extraction

Marketing and support can use ordinary measurements for different purposes. A highlighted claim “portable” might be appropriate in approved product branding, while a factual summary of a packed mass should not derive that word as an objective grade about a specific buyer's mobility needs. A writing model's habit of turning quantities into conclusions can sound helpful while creating expectations the underlying information never established.

Try a neutral comparison vocabulary. “The document lists a smaller packed length for model A in the same units and packaging basis” is supported if the source rows actually establish that relationship. “Fits in any car” needs a different evidence set and user context. “Designed for easy transport” might describe stated intent if confirmed, but it doesn't prove ease for a particular person. The safer workflow isolates these remarks rather than asking the auditor to grade a whole paragraph as right or wrong.

A source's capacity statement can be constrained by another condition. “Holds up to six sample inserts in the pictured layout” describes a particular count and arrangement. A summary that says “accommodates six of any display item” makes a broader fit claim. A three-part check can ask which kind of objects, in what arrangement, and whether the count is maximum or exact. The correct answer can feel slightly less impressive and still be more useful for shopping.

One editor practice is to highlight extra adjectives in the model output and ask what they assert. Compact could merely be an article's heading or present an unsupported comparison. Accurate can be a meaningful precision claim where context matters. Official can imply endorsement by an authority when the only source was a reseller description. Don't strip every descriptive word. Keep those materially changing the factual promise visible for review.

Treat the written result and the evidence receipt as different surfaces

A reviewer needs more detail than a reader who wants a useful short summary. The reader may need two plain sentences and a reference to the applicable guide. The audit may hold raw source wording, unstructured OCR excerpts, reviewer initials, qualifier check and unresolved context. Keep the layers related without dumping your entire workflow into a customer-facing note. An audit reference can be readable by the people with authorized access and still unnecessary to expose publicly.

A well-made receipt records what was actually verified: standard package, exact variant and document edition, approved unit, approved context, no conversion and any missing bare-frame property. If later someone proposes converting or comparing, that is a new operation with its own policy and assumptions. Do not let a checked numeral become implicit approval for a derived claim. Treat the proposal as a different interpretation, not merely a restyled fact.

Use reason codes sparingly. For small teams, missing unit, unsupported qualifier, wrong property, wrong variant, mixed basis and outdated source are enough to start. The usefulness is in routing cases. A source-missing condition needs a retrieval step. A correctly interpreted row using the wrong output comparison needs editorial repair. A technically consequential interpretation unavailable to an ordinary reviewer should go to the appropriate source owner or qualified professional. No generic AI rewrite solves all three.

Check revision propagation when output leaves the draft space. A plain-text export might drop a source note a web view keeps. A product card could split a metric unit onto a line that disappears in a narrow crop. For this article, the focus is correctly attaching the unit and condition in the summary, not designing complex visual tables. But observing the reader's actual surface can reveal that a technically approved sentence isn't what the product showed.

Give the reviewer an independent route to the source

A friendly authoring workflow can unintentionally hide the evidence. Suppose the standardized brief shows the model's extracted row beside its final paragraph, but opening the original requires a separate permission request. A reviewer may compare the two convenient views and approve them as if they were independent. Both can repeat an incorrect extraction of the same header. Plan the relevant original access under the organization's rules before asking a person to verify content they cannot inspect.

Create a closeout fixture with two independent references to the same claim. A permitted view of the original establishes the variant, stated value and table basis. The evidence register establishes how the writer interpreted it. If the interpretation differs, ask which part moved. The task is not to decide whether the register is generally reliable, but whether this exact reviewed claim can be located in the source. A stable link can help, as long as its destination and snapshot authority are not themselves assumed.

Try an intentionally misdirected reference. The draft points to the correct PDF and page, but adjacent rows describe folded and expanded width. A generic link check passes. A person who opens the page without the intended row context may think the number is supported somewhere and move on. A receipt should state the property, heading, and enough context to make the wrong row apparent. Concise wording is fine; vague reading is not.

Build a handoff for reduced source access too. An editor might be allowed to view publicly released specifications, while a product owner holds an internally approved clarification that cannot be circulated. Do not replace the document with an unsupported assertion. Have an authorized person confirm what exact claim can be shared and record its approved basis under appropriate access controls. If nobody can provide sufficient evidence for public output, keep the claim out. Respecting that limit makes the summary less dependent on a specific employee's inbox.

Finally, distinguish observed source date from verification time. The first is about the material you reviewed, the second about when it was read and compared. A month-old verified paragraph may be appropriate for a dated archived sheet and stale for a constantly updated model page. State what the current task demands. This lets a colleague schedule a narrow recheck when one relevant edition changes rather than reinterpret every inch and kilogram each time the catalog refreshes.

Audit the next revision before trusting a reusable summary

Build your first pilot around a small source with several clear claims and a few deliberately difficult ones. Keep a witness of each intended failure: wrong variant, same digit with changed unit, source rounding turned into exactness, packed-versus-unpacked comparison, and ambiguous extraction due to lost caption. Run both the initial extraction and the release check. A pilot that evaluates only one normal table might never encounter the exact type of drift you need to prevent.

Try holding the prompt and output review constant while changing the source's packing definition. Suppose one edition weighed the standard kit with an extra insert and the current edition explicitly excludes it. Even equal listed values do not make the experimental result equivalent across conditions. Change the wording that names the condition, update dependent statements, and preserve the previous documentation when authorized. Auditors should be able to understand why an identical-valued metric remained worth rechecking.

Measure outcomes in ways the sample actually supports. You can count seeded errors found and supported claims retained across ten synthetic fixtures. That does not estimate the model's accuracy on thousands of real equipment sheets. A complete trace showing which important errors were caught may justify trying the workflow on another authorized family. Haven't checked a particular unit, locale or table form? Say so. Silence is not evidence that the tool has it covered.

Keep the evidence and deadline aligned. A handoff can note that one comparison remains blocked until its compatible source arrives and that already verified single-property statements can proceed only under normal review. That lets coworkers continue independent work without turning an incomplete comparison into a promise.

Finally, give the next person a short rule to keep: when a figure moves out of its original context, move its claim identity along with it. Name the object, property, unit, qualifying language, condition and evidence. If the point of the summary is to help people compare, demand compatible bases instead of making the number sound more certain. A precise sentence that explains what cannot yet be compared earns more trust than an impressive chart built from two true numbers and a false relationship.

Frequently asked
What is Keep the Number and Its Unit Together about?
A catalog sheet for two portable display stands lists the width of one in inches, the transport length of the second in millimeters, and the mass of each…
What should you know about define the thing an auditor should check?
A workable extraction keeps seven related facts together: source edition, product variant, measured property, value, unit, qualifier and condition. Not every sheet will supply all seven. That absence should be visible. “Unit not stated in reviewed source” is much more useful than a convincing guess. The reviewer can…
What should you know about follow a number to its heading, not to the nearest word?
Public equipment documents often have wide tables, merged headings, footnotes and repeated row labels. A copy into plain text can preserve all the words in the wrong order. “900 | Packed | 6” might look well structured when it was actually part of three distinct columns joined during capture. Open the actual source…
What should you know about use a worksheet that keeps the evidence claim-sized?
Build a compact register for one public sheet with only a few sample fields. Here is a fictional set:
What should you know about preserve the direction of comparisons and limits?
Three is not necessarily three. In a statement “up to three sample trays,” three is the stated maximum, not exactly three, not at least three and not a capacity instruction for every use. “At least three trays in this bundle” is the opposite direction. “Approx. three in the illustration” is not a purchasing quantity.…
References & sources
  1. Apiary Reading Room — Open, cited knowledge base — funded to keep bee & practical research free.
From the Apiary Reading Room. Opinion & editorial — not financial advice. We don't overclaim.
More from the Reading Room