Counting Scale by Weight: How to Count 5,000 Bolts Without Counting
Count ten. Weigh the rest.
That is the whole method. A counting scale by weight does not count anything. It weighs a small handful of pieces you have counted by hand, works out what one piece weighs, and divides the weight of the full bin by that figure. Five thousand bolts become one weighing and one division — a minute’s work instead of most of a lunch hour. Below: how the arithmetic works, how big the sample must be, where the errors really come from, when counting mode must not be trusted, and how the count reaches a phone without re-typing.
How does a counting scale by weight actually work?
Two weighings and one division.
First, the sample. You count out a known number of pieces — say ten M10 x 40 hex bolts — and put them on the scale. The scale divides the sample weight by the sample count and stores the result as the average piece weight. An M10 x 40 mild-steel bolt weighs about 35 g, so ten weigh about 350 g and the stored piece weight is 35 g.
Second, the bulk. The whole bin goes on the platform. The scale takes the net weight — 175 kg, say — and divides it by the stored piece weight: 175,000 g ÷ 35 g = 5,000 pieces.
It is the formula a government inspector uses on a package sold by count. NIST Handbook 133 (2026 edition), the US manual for checking packaged goods, has inspectors weigh a counted sample, work out what one piece should weigh, and convert the package’s weight error into a count error. The international scale standard OIML R 76 files “counting” under non-weighing modes — the instrument is calculating, not measuring — and requires that the scale can always return to plain weighing.
The scale measures weight. The count is arithmetic on two weighings, and every error in it traces back to one of those weighings or to the assumption that every piece weighs the same. On an HSCo scale the counting mode is part of the base firmware: place the counted sample, key in its count, and the display shows weight, stored piece weight and total count together.
Why counting by hand fails at 5,000
Because the human eye was never built for it.
Vision researchers have measured it. Up to about four items, people take in the number at a glance. Beyond four, the brain counts one item at a time, at roughly a quarter to a third of a second each, and the researchers call that process slow, effortful and error-prone (Trick and Pylyshyn, Psychological Review, 1994; Vetter, Butterworth and Bahrami, PLoS ONE, 2008). Accuracy falls further when attention is divided — the normal condition of a stores counter.
Five thousand bolts at 0.3 seconds each is 25 minutes of unbroken counting. One interruption and the heap starts again. Then the number is written on a slip and typed in an hour later — the second error. A study in Management Science (2008) examined about 370,000 inventory records across 37 stores and found 65 % of them inaccurate; field experiments across 62 stores (Production and Operations Management, 2013) found inaccuracy to be the normal state.
India runs on pieces. Its fastener trade alone is worth roughly ₹95,000 crore a year by one market estimate, and every bag, kit and challan in it carries a number somebody counted.
How many pieces should the sample be?
More than you think — and the answer is regulated, not guessed.
When a US inspector verifies a package labelled by count, NIST Handbook 133 sets the sample at 10 % of the labelled count or at least 50 scale divisions, whichever is larger — on a scale reading to 1 g, a sample weighing at least 50 g.
The one place US law licenses a scale’s counting feature for trade is the pharmacy prescription scale, and even there NIST Handbook 44 sets the floor at at least ten pieces, each at least three scale divisions and the sample at least thirty, with the scale required to refuse anything smaller. It binds pharmacy scales, not bolt counters — a floor, not a target.
Why more is better is basic measurement science. The international guide to measurement uncertainty (JCGM 100:2008, the “GUM”) says the uncertainty of an average falls with the square root of the number of readings: a 100-piece sample gives a piece weight about three times surer than a 10-piece one. The factory rule of thumb: ten is a floor, a hundred is comfortable, and the sample must weigh at least 50 divisions of the scale you sample on — 250 g on a platform reading to 5 g, 5 g on a bench scale reading to 0.1 g.
Where the error really comes from
From the sample weighing, not the bulk weighing.
NIST Handbook 133 says of its own count-by-weight procedure: “the precision of this procedure is only ± 1 %.” That follows from its sample rule — 50 divisions round to half a division either way, and half of 50 is 1 %. The worked example below, for 5,000 M10 x 40 bolts at about 35 g each, is arithmetic rather than a sourced statistic — check it against your own scale.
| How the piece weight was taken | Rounding error in the piece weight | Error in a lot of 5,000 |
|---|---|---|
| 10 bolts on a platform reading to 5 g | ± 0.71 % | about ± 36 pieces |
| 100 bolts on the same 5 g platform | ± 0.07 % | about ± 4 pieces |
| 10 bolts on a bench scale reading to 0.1 g | ± 0.014 % | under 1 piece |
| The 175 kg bulk weighing itself, on a 300 kg class III platform | — | about ± 2 pieces |
The bulk weighing barely matters: NIST Handbook 44’s tolerance table lets the heavy platform be out by 75 g on 175 kg — about two bolts. The ten-bolt sample on that same platform is worth thirty-six. If the pieces themselves differ by 1 %, the average of ten carries a further ± 16 pieces of scatter; the average of a hundred, about ± 5.
Now set that against the law. India’s Legal Metrology (Packaged Commodities) Rules allow a package declared by number to be out by 2 % of the declared quantity — 100 pieces on 5,000. Internationally the bar is tighter: OIML R 87 allows 1 % above 50 items, and the US pharmacy table — the tightest any regulator writes down — 0.5 %. A ten-bolt sample taken on the platform eats a third of India’s allowance before the bin is lifted.
Three rules follow:
- Take the sample on the finest scale you have — a bench scale reading to 0.1 g for the piece weight, the heavy platform for the bulk.
- Sample more pieces than the legal floor — ten if you must, a hundred when the lot is valuable.
- Never mix lots.
When counting mode fails — and what it is not
Counting by weight assumes every piece weighs the same. Break that assumption and the count is wrong with complete confidence.
Suppose a ten-piece sample accidentally holds seven M10 x 40 bolts at about 35 g and three M8 x 30 bolts at about 16 g. The stored piece weight comes out at about 29 g, and a true 5,000 is displayed as about 6,000 — a 17 % error no scale can detect, because the scale did exactly what it was asked. Plating, head style or shank length shifts it just as surely.
NIST Handbook 133 orders inspectors to count by hand when the piece weight “varies excessively”, and adds a test: take the piece weight from two separately counted samples and proceed only if they agree within one scale division. NIST Handbook 44 makes the user responsible for the stored piece weight, “especially critical” when the product comes from different lots. Adopt both as habits: two samples agreeing within a division before any lot is counted, and a fresh piece weight for every new lot.
The second thing counting mode is not: a legal-for-trade result. Under India’s Legal Metrology Act the thing verified and stamped is the weight; the count is a calculation on it. The US requires most counting scales to carry the words “the counting feature is not legal for trade”, and an Indian inspector checking a package sold by number finds the error by actual counting. For stock, kitting and dispatch checks that is exactly what you want. For a sale by number, the packer remains responsible for the declared count within the 2 % rule. An HSCo counting scale is verified and stamped for its weighing like any other trade scale; the count it shows is arithmetic on that verified weight.
What changes on Monday morning
- Goods inward is checked while the truck is still there. Weigh the crate, read the count, compare with the challan. A short delivery shows in a minute.
- Kitting and dispatch stop being a counting job. A bag of 500 becomes a target weight. An HSCo counting platform scale — 100 kg to 1,000 kg platforms reading to 5 g or 10 g, with unit-weight memory, a running total and a serial output for a printer or computer — adds up the bags as they go.
- The piece weight lives with the part number, not in someone’s head. Each part gets its own stored piece weight, refreshed when the lot changes.
- The record is never re-typed. With a Bluetooth-enabled scale built around your workflow, the weight, the piece weight and the count can travel to a phone together, and the app is designed to log each count as one record — part number, piece weight, sample size, net weight, count and time — for export. The IoT connected weighing scales page lists the connection options; an existing scale can be given a voice with a Bluetooth module rather than replaced.
How to start: one part number, thirty days
- Pick one part number — the one your stores counts most often, or the one most often short.
- Take its piece weight on the finest scale you have, with a sample of at least ten pieces weighing at least 50 divisions. Take a second sample; proceed only if the two agree within a division.
- Count every lot of that part by weight for thirty days. Once a week, hand-count one lot and compare.
- Record every count with its piece weight, sample size, net weight and time.
- At day thirty, compare by-weight counts, hand counts and the stock record. Roll out only if the numbers justify it.
Hindustan Scale Company has made weighing scales in Mumbai since 1955 and builds the counting firmware, the Bluetooth hardware and the phone app itself — one supplier answers for the whole chain.
Frequently asked questions
How accurate is a counting scale by weight?
As accurate as the sample you gave it. Government inspectors who count by weight accept about ± 1 %; a ten-piece sample taken on a coarse platform can be worse, and a hundred-piece sample on a fine bench scale can be better than one piece in a thousand. The bulk weighing itself contributes almost nothing.
How many pieces should I put in the sample?
At least ten, and the sample must weigh at least 50 divisions of the scale you take it on (250 g on a scale reading to 5 g). A hundred pieces gives a piece weight about three times surer than ten. Take two samples and proceed only if they agree within one division.
Can I sell bolts by number using the scale’s count?
The scale’s weighing is what is verified and stamped; its count is a calculation, and Indian inspectors check packages sold by number by counting them. You may use the scale to fill the packs, but as the packer you remain responsible for the declared count within the 2 % permitted by the Packaged Commodities Rules. Use it for stock, kitting and dispatch checks without hesitation.
Why does the count go wrong when a new lot arrives?
Because the stored piece weight belongs to the old lot. A different supplier, plating, head style or length shifts the piece weight, and the scale multiplies that shift across every piece. Take a fresh piece weight for every new lot and keep it filed against the part number.
Can my existing platform scale count pieces and send the count to a phone?
A scale with a serial output can be fitted with a Bluetooth module so its readings reach a phone. Counting mode with a stored average piece weight is part of HSCo’s base firmware; the app is designed to log each count (part number, piece weight, sample size, net weight, count, time) for export, set up to your workflow.
What is average piece weight?
The weight of one piece, worked out by weighing a counted sample and dividing by the number of pieces in it. The scale stores it and divides every later net weight by it to produce the count. It is the single number the whole method depends on.
Next step
Tell us which part number your stores counts most often, and we will propose a thirty-day count-by-weight pilot for it — sample scale, platform and phone record included.
Hindustan Scale Company (HSCo) · Mumbai · est. 1955 · www.hindustanscale.com · hsc@hindustanscale.com
Sources
- NIST Handbook 133 (2026 edition) — Checking the Net Contents of Packaged Goods, Chapter 4, clause 4.2 “Packages Labeled by Count”: sample = greater of 10 % of the count or 50 scale divisions; audit precision ± 1 %; hand-count when piece weight varies excessively; two-sample agreement within one division.
- NIST Handbook 44 (2026 edition) — Specifications, Tolerances, and Other Technical Requirements for Weighing and Measuring Devices, Section 2.20 Scales: S.1.2.3, S.6.6, N.1.10, Table S.6.3.a note 13, Table 6, Table T.N.3.10 and UR.3.13 (prescription-scale counting rules, quoted by analogy; “counting feature is not legal for trade” marking).
- OIML R 76-1:2006 — Non-automatic weighing instruments, clause 4.20 “Modes of operation”: counting as a non-weighing mode.
- OIML R 87:2016 — Quantity of product in prepackages, Table 1: tolerable deficiency for packages sold by count.
- Legal Metrology (Packaged Commodities) Rules 2011 (as amended) — First Schedule Table II (2 % by number), rule 2(e), rule 19(6)(c), verification of commodities packed by number.
- The Legal Metrology Act 2009, section 24(1) — verification of any weight or measure used in a transaction.
- JCGM 100:2008 — Guide to the expression of uncertainty in measurement (GUM), clause 4.2.3: uncertainty of a mean falls with the square root of the number of observations.
- Trick, L. M. and Pylyshyn, Z. W. — “Why are small and large numbers enumerated differently?”, Psychological Review 101(1), 1994; Vetter, P., Butterworth, B. and Bahrami, B. — “Modulating attentional load affects numerosity estimation”, PLoS ONE 3(9), 2008.
- DeHoratius, N. and Raman, A. — “Inventory Record Inaccuracy: An Empirical Analysis”, Management Science 54(4), 2008.
- Hardgrave, B. C., Aloysius, J. A. and Goyal, S. — “RFID-Enabled Visibility and Retail Inventory Record Inaccuracy: Experiments in the Field”, Production and Operations Management 22(4), 2013.
- Indian fastener manufacturer’s weight chart for MS hex bolts and nuts (IS 1363/67), kg per 100 pieces — approximate piece weights used in the worked example.
- IMARC Group — India Fasteners Market report page (2026): indicative market estimate, single commercial source.
- Hindustan Scale Company — Counting Platform Scale (HSC-CPS) product page; IoT Connected Weighing Scales page; Bluetooth module product page.