Systems and records
International dataset measurement units: validating pounds vs kilograms
By SourceX Editorial · Updated
9 sections
Short answer
To validate a shipment that mixes pounds and kilograms, keep the recorded value and unit untouched, record where the unit came from, convert once with the exact factor (1 lb = 0.45359237 kg) into a separate field, and compare with the specification using the interval implied by the source's rounding. Ambiguous units such as ton or cwt, and gross versus net weights, get a reviewer flag instead of a guess.
Countries and languages
International dataset
Key takeaways
- The pound-to-kilogram factor is exact; most errors come from unit provenance, ambiguous tokens, weight basis and rounding.
- Store the original text, a coded unit, its provenance and the weight basis before converting anything.
- Compare with limits using the interval implied by the source precision, and allow an indeterminate result.
- Ton, hundredweight and unqualified pound each have more than one reading and must be resolved from evidence.
- Keep jurisdiction, locale and capture date with the value so dual statements are not mistaken for errors.
How do you validate a shipment task that mixes pounds and kilograms?
Keep the value exactly as recorded, state where its unit came from, convert once with the exact factor into a separate normalized field, and compare against the specification with a tolerance that allows for how the source value was rounded. A check that passes or fails a shipment without those four steps is testing the converter, not the shipment.
The exact factor is not in dispute. NIST Handbook 44 gives one avoirdupois pound as 0.453 592 37 kilogram, and the Unicode CLDR unit data uses the same constant. What goes wrong in international dataset measurement units is everything around the factor: a unit inferred from a country instead of read from the record, a ton that could mean three different masses, a gross weight compared with a net limit, or a rounded figure treated as exact.
This guide gives a unit provenance and conversion verification task you can attach to any record that carries a weight, with a worked example, an exception and the cases that need a domain reviewer.
Where did the unit in a shipment record actually come from?
From one of five places, and the place determines how much you can trust it. An explicit unit code in its own field is strong evidence. A unit inferred from the shipper's country or the software's locale setting is weak evidence and should never be silently promoted to fact.
Locale data shows why. CLDR's measurement data marks the United States and Liberia as using the US system and Great Britain the UK system, with metric as the world default, and its own comments say that element is being replaced by finer preferences. A record from a British warehouse may hold kilograms for freight and pounds for a person's weight. Country is a prior, not a unit.
| Provenance level | Example in a record | Confidence | What to store |
|---|---|---|---|
| Explicit coded field | A unit column holding KGM or LBR from UN/ECE Recommendation 20 | High | The code, the code list and its version |
| Unit in the same text as the value | A packing list line reading 1,100 LB | High if unambiguous | The original string and the token parsed as the unit |
| Regulatory default for the field | A US export filing's shipping weight, defined in kilograms | High for that field only | The rule and the field it governs |
| System default | A WMS weight column with a site-level unit setting | Medium; settings change | The setting value and the date it was read |
| Locale or country inference | A carrier in Texas, so pounds is assumed | Low | The inference and a flag that a reviewer must confirm it |
Which unit assumption could reverse the answer?
The ambiguous words, more often than the pound itself. Ton and hundredweight each name more than one mass, pound can mean three different units outside freight, and weight fields differ on whether packaging is included. Any of these can move a value across a limit, so the dataset must say which reading it used and why.
The factors below come from UCUM and CLDR definitions, rounded to four decimals for display; keep the exact definition in the data.
| Token in the record | Possible readings | Mass in kilograms | How to resolve |
|---|---|---|---|
| ton | Short ton (2,000 lb), long ton (2,240 lb), or a misspelled metric tonne | 907.1847, 1016.0469 or 1000 | Read the document's origin and any spelled-out form; flag if neither settles it |
| cwt | Short hundredweight (100 lb) or long hundredweight (112 lb) | 45.3592 or 50.8023 | Use the trade lane and contract; never default silently |
| lb | Avoirdupois, troy or apothecary pound | Avoirdupois is 0.45359237; the others differ | Freight is avoirdupois; precious metals and old pharmacy records may not be |
| lb used for force | Pound-force in a load rating | Not a mass; CLDR converts it through gravity | Keep it out of mass comparisons |
| weight | Gross, net, tare or chargeable weight | Depends on the basis | Store the basis as its own field before comparing |
| 1.100 | One point one, or one thousand one hundred, depending on decimal separator | Differs by a factor of 1,000 | Parse with the source locale and keep the original string |
What should the unit provenance and conversion verification task include?
Ten fields per measured value, so that any reviewer can recompute the result without the supplier's systems. The task input is one record with a measured value and one specification it must meet; the expected output is a filled row and a result of pass, fail or indeterminate.
At a glance
What should the unit provenance and conversion verification task include
Use UN/ECE Recommendation 20 codes for trade interchange, where KGM is the kilogram and LBR the pound, and add a UCUM code when the data will feed scientific or health systems, where the avoirdupois pound is written [lb_av].
| Field | Content | Rule | Example (illustrative) |
|---|---|---|---|
| source_text | The value and unit exactly as they appear | Never edited after capture | 1,100 LB |
| source_value | The parsed number | Parsed with the record's locale | 1100 |
| source_unit_code | Rec 20 code for the source unit | Blank if the unit is unknown, never guessed | LBR |
| unit_provenance | One of the five provenance levels | Locale inference sets review_needed | unit in same text |
| weight_basis | Gross, net, tare or chargeable | Required for any weight | gross |
| source_precision | Smallest step the source value can resolve | From the system or document, or inferred and flagged | 10 lb |
| factor | Conversion factor and its reference | Exact defined value, not a rounded one | 0.45359237 per NIST HB 44 Appendix B |
| normalized_value | Converted value in the target unit, with interval | Computed in one place; source fields untouched | 498.95 kg (496.68 to 501.22) |
| spec_reference | Limit, its unit, basis and source document | Must state gross or net | 500 kg gross per load plan |
| result | Pass, fail or indeterminate, plus reviewer flag | Indeterminate when the interval straddles the limit | indeterminate; review_needed |
Which checks should the validator run on every converted value?
Eight checks catch most unit faults before a person reviews anything. Each is mechanical, so a buyer can rerun them on delivery.
Run them three times: when records are extracted, after every transformation that touches a numeric column, and on the final package. Unit faults are often introduced by a well-meant cleanup step, such as a spreadsheet that reformats numbers or a script that normalizes a column in place. Record each run's date and counts of pass, fail and indeterminate so the buyer can see whether a later step changed anything.
- Round trip. Converting the normalized value back to the source unit reproduces source_value within the stated precision.
- Factor identity. The factor matches the defined constant for the unit code; a factor such as 0.45 or 0.4536 fails.
- Single conversion. No value has been converted twice, which shows up as a normalized value about 0.45 or 2.2 times the expected size.
- Basis match. The record's weight basis equals the specification's basis, or the comparison is marked indeterminate.
- Interval comparison. The comparison uses the interval implied by source precision, not the point value alone.
- Ambiguous token. Any ton, cwt or unqualified pound in a non-freight context forces a reviewer flag.
- Locale parse. Values parsed with a decimal comma or a thousands point are re-parsed both ways, and a large difference raises a flag.
- Provenance floor. Records whose unit came only from locale inference cannot carry a pass without reviewer confirmation.
Worked example (illustrative): a pallet at 1,100 lb against a 500 kg gross limit
The figures in this example are invented to show the task. A packing list from a US shipper records a pallet as 1,100 LB gross. The receiving site's load plan, written in a metric jurisdiction, caps each pallet at 500 kg gross. The shipper's scale reports to the nearest 10 lb.
Point conversion gives 1,100 times 0.45359237, or 498.95 kg, which looks like a pass. But a reading of 1,100 at 10 lb resolution means a true weight between 1,095 and 1,105 lb, which is 496.68 to 501.22 kg. The interval straddles the limit, so the correct result is indeterminate with a reviewer flag, and the verified outcome comes from the receiving site's own scale record, if one exists.
The exception. A second record in the same illustrative set, from a UK supplier, reads 1.1 ton. Read as long tons it is about 1,117.65 kg; read as short tons it is about 997.90 kg; read as tonnes it is 1,100 kg. The document has no spelled-out unit and no code. The task output is source_unit_code blank, result indeterminate and a note naming the three readings, which is the behavior a model trained on this data should learn.
A generic training example would have converted both records and reported two passes. The verification task reports one indeterminate and one unresolved unit, which is what a careful logistics analyst would say.
What language, date and jurisdiction context must travel with the unit?
The context that explains why the record used that unit: the document language and locale, the jurisdiction whose rules applied to the field, and the date the value was captured. Without it, a reviewer cannot tell a deliberate dual statement from an error.
Two rules show how jurisdiction shapes the record. In the EU, Directive 80/181/EEC allows supplementary indications in non-SI units only alongside the SI indication, which must predominate. In the US, the Fair Packaging and Labeling Act regulations require net quantity statements in both SI metric and inch-pound units and point to NIST Handbook 133 for conversions. A product record from either regime may legitimately hold two values for the same quantity; the dataset should keep both, mark which one is the primary declaration, and not average them.
Weight is only the most common case. Dimensions recorded in inches against a specification in centimeters, volumes in cubic feet against cubic meters, and temperatures for cold-chain goods in Fahrenheit against Celsius all need the same treatment. Dimensional or chargeable weight adds a further trap, because the divisor a carrier applies depends on its own rules and on the unit system of the dimensions, so store the divisor and its source with the result rather than recomputing it later.
Store the original document text with a language tag, the locale used for parsing numbers and dates, the jurisdiction, and the capture date in ISO 8601 form. Translation of surrounding text must not touch numbers or unit tokens.
Which records need a localization or domain reviewer before release?
Any record where the unit or the basis was inferred rather than read, any value that converts to within its own rounding interval of a limit, and any record from a trade where non-avoirdupois units are normal. These are a small share of most freight datasets but carry most of the risk.
Narrow the scope rather than guess when a whole source system has no unit field and no documented default. A smaller dataset with provenance on every value is more useful to a buyer than a larger one in which a fifth of the weights could be off by a factor of 2.2. Regulated filings, hazardous goods records and anything tied to customs value also need counsel or a trade compliance reviewer to confirm that the records can be shared at all.
Make the review itself part of the record. For each flagged value, store the reviewer's decision (confirmed unit, corrected unit, or excluded), the evidence used, the reviewer's role and the date. A buyer can then tell a confirmed value from one that simply passed the mechanical checks, and the same decisions become examples of how an expert resolves an ambiguous unit.
Preparing unit-consistent international records for licensing
Unit provenance is one of the first things an AI buyer's engineers test, because a silent conversion error poisons every downstream comparison. Records that already carry the ten fields above, or can be rebuilt with them from source documents, move through review faster.
SourceX helps companies license selected business records as AI datasets, running rights review, contracting, preparation and delivery. It works with U.S., Canadian and EU companies that had at least 50 full-time employees at their peak, not counting contractors; companies in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia can be considered when at least 90% of the offered data is in English, measured on the records actually offered. Checking whether a company qualifies starts with a brief assessment and needs no uploaded records.
Frequently asked questions
What is the exact conversion factor from pounds to kilograms?
One avoirdupois pound is exactly 0.45359237 kilogram, the value NIST Handbook 44 gives and CLDR uses. Store this full value as the factor and round only for display, so a validator can recompute every normalized value exactly.
Should a dataset store weights in the original unit or converted to kilograms?
Both. Keep the original value, unit text and unit code unchanged, and add the converted value in a separate field with the factor and its source. Overwriting the original removes the evidence a reviewer needs to catch a wrong unit.
How should a dataset record a unit it cannot determine?
Leave the unit code blank, keep the original text, set the provenance to unknown or inferred, and mark any comparison indeterminate. A guessed unit that later proves wrong is harder to find than a blank one.
Which unit codes should an international dataset use?
UN/ECE Recommendation 20 codes, such as KGM for kilogram and LBR for pound, are designed for international trade data. UCUM codes such as [lb_av] suit scientific and health systems. Name the code list in the schema so a buyer knows which one each field uses.
Why is a ton ambiguous in shipping records?
A short ton is 2,000 pounds, a long ton is 2,240 pounds and a metric tonne is 1,000 kilograms. Unless the document spells out the unit or uses a code, the record cannot be compared with a limit and should be flagged for review.
Sources
- NIST Handbook 44, Appendix B, gives one avoirdupois pound as exactly 0.453 592 37 kilogram. Source
- Unicode CLDR unit data defines lb_to_kg as 0.45359237 and converts the ton as 2,000 pounds, the tonne as 1,000 kilograms and pound-force through gravity. Source
- CLDR measurement data marks the US and Liberia as using the US system and Great Britain the UK system, with metric as the default, and notes the element will be replaced by finer preferences. Source
- UCUM defines the avoirdupois pound [lb_av] as 7,000 grains, separate codes for troy and apothecary pounds, short and long hundredweights of 100 and 112 pounds, and short and long tons of 20 of each. Source
- UN/ECE Recommendation 20, as published in UNECE's current Rec 20 vocabulary, assigns the code KGM to the kilogram and LBR to the pound. Source
- The US Foreign Trade Regulations define shipping weight as the total weight of a shipment in kilograms, including goods and packaging. Source
- FTC regulations under the Fair Packaging and Labeling Act require net quantity statements in both SI metric and inch-pound units and direct conversions to NIST Handbook 133. Source
- Directive 80/181/EEC allows supplementary indications in non-SI units only alongside an SI indication, which must predominate. Source
See if your company qualifies
A short company assessment. No data uploads are needed.