How to read a chemical certificate of analysis

The answer, up front

A certificate of analysis is a statement of test results for a batch, issued by whoever tested it — usually the producer’s own quality control department. It is a first-party document. It is not a third-party verification, not an accredited test report, and not a warranty that the material will do what you need.

We could not find a general standard, Indian or international, that defines what a certificate of analysis must contain. There are authoritative documents that come close for particular sectors — the World Health Organization publishes a model certificate of analysis; USP General Chapter <1080> and the IPEC guide cover pharmaceutical excipients; ISO/IEC 17025:2017 clause 7.8 governs what a laboratory’s test report must contain. None of them binds a bulk chemical producer. So the content of the COA you receive is a matter of what you asked for in the purchase order, and almost nothing else.

That is the practical point of this article. A COA can be entirely truthful and still tell you nothing, in four specific ways: the result has no method behind it, the number is on a basis you did not expect, the sample it describes was never representative of what arrived, or the word “complies” is standing where a number should be. Each has a clause that prevents it.

What follows is a buyer’s reading guide, not legal advice. Where it names a statutory duty it says whose duty it is, because that is the part most often got backwards.

Four documents that are not the same thing

The single most common failure is accepting the wrong document. All four arrive as a PDF with a company letterhead and a table of numbers.

A document with no batch number, no date of manufacture and no method references is not doing a COA’s job, whatever the heading says — and the useful response is to ask for the missing fields rather than argue about the label. Producers are candid about this in their own documents. Tronox’s caustic soda liquid data sheet states “This document does not constitute a specification” and that nothing in it “shall be deemed a representation or warranty of any kind”. Tata Chemicals’ dense soda ash data sheet (file dated 29 July 2024) heads its value column “TCL Assured Specs”, cites compliance to IS 251:2024, references no test method for any parameter, and states that “No warranties, expressed or implied, are made” and that the parameters “can be changed without prior intimation”. That is a specification sheet doing exactly what a specification sheet should do. It is not a COA and was never offered as one.

For the authoritative contrast: USP <1080> describes a COA as relating specifically to the results of testing a representative sample drawn from that batch, and the IPEC Certificate of Analysis Guide for Pharmaceutical Excipients, Version 2.1 (2024), section 3, recommends reporting actual data and observations rather than non-specific “passes” or “conforms” statements — unless the test is qualitative, or the compendial criterion is itself the acceptance criterion.

The one industry that solved this, and what it tells you

Metals got there first. EN 10204:2004 gives inspection documents names that say who stands behind the numbers, and the ladder is worth knowing because we know of no equivalent vocabulary in chemicals for naming who stands behind a number:

• Type 2.1, declaration of compliance with the order — the manufacturer declares conformity, with no test results at all.
• Type 2.2, test report — the manufacturer declares conformity and supplies results based on non-specific inspection: results from material made by the same process to the same specification, not necessarily the material actually supplied. It often is the delivered material; what is missing is any assurance that it was.
• Type 3.1, inspection certificate — results of specific inspection on the products actually supplied, validated by the manufacturer’s authorised inspection representative, independent of the manufacturing department.
• Type 3.2 — the same, additionally validated by the purchaser’s authorised inspection representative or an inspector designated by official regulations.

Only type 3.2 brings in anyone outside the manufacturer — and even then it may be your own inspector rather than an independent third party. A chemical COA in Indian trade sits somewhere between a 2.2 and a 3.1, and the document itself usually does not tell you which — whether the figures came from the material in your tanker or from a representative production sample, and whether the person who signed is independent of the plant that made it. Those are both fair questions to ask in an enquiry, and the answers are more informative than the numbers.

EN 10204 covers metallic products and has no application to chemicals by its own scope, though the 2.2 / 3.1 / 3.2 vocabulary does get borrowed in contracts for other materials. It is cited here as the analogy, because it names a distinction chemical trade leaves unnamed. The type definitions above are taken from two independent third-party reproductions that agree with each other; we have not read the standard itself, which is paywalled.

What a usable COA states

There is no mandatory list, so this one is assembled from the two most authoritative sources available — the WHO model certificate of analysis, published as Annex 4 to WHO Technical Report Series No. 1010 (2018), and the reporting requirements ISO/IEC 17025:2017 places on a laboratory test report — and reduced to what is checkable on a bulk chemical delivery. WHO’s document states that it is intended to serve as an example and not to be prescriptive, so read the items below as what a good certificate provides for, not as what anyone is obliged to print.

1. The product, its grade and the specification it is being judged against, with the edition year of any standard named.
2. A batch or lot number that matches the number on the packages. WHO’s model ties the two together deliberately, providing for the batch number “used by the original manufacturer and repacker or trader” to appear on the certificate — so that it can be matched to the label. Where a repacker has assigned a number of its own, that is the number you will be matching, and you need to be told so.
3. The date of manufacture, and the quantity the batch represents. Without the quantity you cannot tell whether the certificate covers your 20 tonnes or a 500-tonne production campaign.
4. The name and address of the original manufacturer. WHO’s model provides for it, plus the repacker’s or trader’s where they supplied the sample.
5. Each test by name, with four things against it: the method reference, the acceptance criterion, the numerical result, and the conformity statement. WHO’s model sets out all four as columns, and calls for results “in numerical form, whenever applicable”.
6. The basis of every number — dry or as received, 100% basis or as supplied, w/w or w/v.
7. The units, including for trace parameters, and the quantitation limit behind any result written as “less than”.
8. Which tests were not performed on this batch. The IPEC guide is explicit: where a specification test was not run on the batch, that should be indicated on the COA. USP <1080> — a general information chapter, which carries no mandatory requirement unless a monograph or a chapter below <1000> specifically references it — says that reduced-frequency testing should be clearly stated on the certificate and each affected test indicated.
9. Whether any result came from an outside laboratory, and which.
10. Who sampled, and how. This is the field most often absent and most often decisive — see below.
11. The name, position and date of the person authorising it. ISO/IEC 17025 requires a test report to identify the person authorising it; WHO’s model provides for the name, contact details, date and signature of the head of the laboratory or another authorised person. A certificate with no identifiable signatory names nobody who can be asked about it.

A usable COA also carries a unique certificate number and “page 1 of n” on every page, which sounds like clerical detail until a two-page certificate arrives as one page and the page that is missing is the one with the trace metals on it.

ISO/IEC 17025:2017 clause 7.8 applies to a laboratory’s test report, not to a producer’s COA. The standard does not require anything of a producer’s certificate. It is borrowed here because it is the best available checklist, not because it governs.

“Complies” is not a result

A column of “Complies”, “Conforms” or “Pass” against each parameter is the most common defect in a chemical COA, and both authoritative sources say to report the number instead. The IPEC guide recommends actual data and observations rather than non-specific statements, except where the test is qualitative or where the compendial criterion is itself the acceptance criterion. USP <1080> recommends the same.

The reason is not pedantry. A number tells you where in the range the material sits, and that is operational information. Caustic lye certified at 48.2% and caustic lye certified at 49.8% both comply with a 48–50% specification, and they differ by 16 kg of sodium hydroxide per tonne of solution — about 3.3% more NaOH from the stronger consignment. If you are dosing to a setpoint, the difference shows up as a consumption variance that nobody can explain, because the certificates for both consignments read “Complies”.

There is a second reason, and it is the stronger one. A conformity statement is a conclusion, and a conclusion requires a rule. ISO/IEC 17025:2017 clause 7.8.6 makes this explicit for a laboratory: before issuing a statement of conformity, the laboratory must have documented a decision rule, and the statement must identify which results it applies to, which specification is met or not met, and which decision rule was applied — the last of those being excused only where the decision rule is inherent in the specification or standard requested. The duty to document and apply a rule is not excused. Where you are handed a bare “complies” against a specification that does not itself set a decision rule, nobody has told you what rule produced it.

Why a result exactly on the limit is a coin toss

This is the part of a COA that buyers almost never interrogate, and it decides borderline consignments.

ILAC G8:09/2019, the international guidance on decision rules, describes simple acceptance — the rule where the acceptance limit is the specification limit, with no guard band. It also calls that rule “shared risk”, because where a measurement result is exactly on the tolerance limit the probability of being outside it “may be as high as 50%” — assuming, as the guidance says, a symmetric normal distribution of the measurements. The Eurachem/CITAC guide Use of uncertainty information in compliance assessment, 2nd edition (2021), puts the same case at a 50% probability of an incorrect decision, and adds the qualification that matters: where the limit falls inside the expanded uncertainty interval of the result, it says that without further information the result cannot be used to decide compliance — and the further information is precisely what a decision rule supplies.

Deciding compliance therefore needs three things, not one: the specification limit, a measurement uncertainty, and a decision rule. A chemical COA almost never carries an uncertainty figure, and that is normal rather than a defect — WHO recorded in 2018 that pharmacopoeias and regulators did not require expanded uncertainty to be reported for compliance testing. But it means that a result sitting on your limit cannot be resolved from the certificate alone, and that is worth knowing before you reject a tanker on it.

Rounding is a convention, and the conventions do not agree

Three different approaches to the same borderline result are all respectable, and they do not always give the same answer — which is why the approach belongs in the contract rather than in the analyst’s head.

• The Indian convention is round half to even. IS 2:2022, “Rules for rounding off numerical values”, provides at clause 4.2.3 that where the figure after the last retained place is 5 alone, or 5 followed only by zeros, the retained figure is increased by one if it is odd and left unchanged if it is even — zero counting as even — so that rounding up and down balance out over many values. Clause 4.4 requires the rounded value to be obtained from the most precise value available in one step only, not by successive rounding, and clause 4.5 requires rounding in one direction only where safety limits have to be respected.
• The pharmacopoeial convention is round half up, against the limit’s own decimal places. USP General Notices 7.20 rounds the result to the number of decimal places in the limit expression, considering only one digit beyond it, and increases the preceding digit where that digit is 5 or higher. Its own worked example: against an assay limit of 98.0% minimum, a result of 97.95% rounds to 98.0% and conforms, while 97.92% rounds to 97.9% and does not. USP adds two rules that matter as much: acceptance criteria are fixed numbers and are not rounded, and numbers should not be rounded until the final calculation is complete.
• The third approach does not round at all. Compare the observed value with the limit as it stands, and 97.95% against a 98.0% minimum simply fails.

On that 97.95% example the first two conventions happen to agree — the retained digit 9 is odd, so round-half-to-even also takes it up to 98.0% — and the third approach is the one that differs. The two rounding conventions part company where the retained digit is even: applying each rule as written, a result of 97.85% against a 97.9% minimum rounds to 97.8% and fails under round-half-to-even, and to 97.9% and conforms under the pharmacopoeial rule. The same number, the same limit, two defensible answers.

ASTM E29 has long formalised the underlying choice — between an “absolute method”, where the observed value is compared with the limit without being rounded at all, and a “rounding method”, where it is rounded first — and makes the governing point: an unqualified limit cannot be regarded as carrying a definite operational meaning, so the choice should be specified in the document that sets the limit. If your specification says “98.0% min” and nothing else, it has not told your supplier whether 97.95% passes.

The IS 2 clauses above are in IS 2:2022, the current Second Revision, which supersedes IS 2:1960. The ASTM position is a different matter: the current edition is ASTM E29-22, and the absolute-method and rounding-method wording quoted here is from the superseded E29-90, which we read. We have not read E29-22, so treat the ASTM point as the long-standing structure of that practice rather than as a quotation from the current text, and read the current edition before writing a clause on it.

The method defines the number

“Purity 99%” and “moisture 0.3%” are not facts about a material until you know how they were measured. Three illustrations, each from a primary source.

Water is not volatile matter. USP <731> defines loss on drying as the volatile matter of any kind driven off under the specified conditions. USP <921> determines water, by Karl Fischer titration among other methods. They measure different quantities, and the pharmacopoeia does not let the analyst choose: the method is specified in the individual monograph. A material carrying a residual solvent will usually read higher by loss on drying than by Karl Fischer, and neither number is wrong. It can also run the other way: a hydrate whose water of crystallisation is not released at the drying temperature specified reads lower by loss on drying than by Karl Fischer. The point is not which is higher but that they are different measurements, so the method has to be named.

Some Indian standards handle method disagreement by naming a winner. IS 251:1998, the then-current soda ash specification, designated the barium sulphate gravimetric method as the referee method for sulphate with the turbidimetric method as the alternate, and Nessler cylinder comparison as the referee for iron with bipyridyl colorimetry as the alternate. That structure — a designated referee method and a permitted alternate for routine work — is the thing to borrow, and your contract is where you say that the referee method governs, because the standard designates it without saying so. If your contract does not name the referee method, a dispute begins with an argument about which laboratory was right rather than about the material.

A titration result is not an assay result. IS 252:1991 required sodium hydroxide content to be computed as total alkalinity minus the carbonate contribution — NaOH = B − (A × 40/53), where B is the total alkalinity expressed as NaOH and A is the carbonate determined separately as Na₂CO₃, because carbonate titrates as alkalinity too. The factor converts the carbonate into the sodium hydroxide it titrates as: one mole of Na₂CO₃ takes up two protons and so titrates like two moles of NaOH, giving 2 × 40.00 / 105.99 = 0.755 — which the standard writes as the factor 40/53, that is 80/106. A certificate reporting “total alkalinity as NaOH” and a certificate reporting “NaOH content” are reporting different numbers for the same drum, and the gap is the carbonate converted at that 0.755.

Edition warning: the referee-method and alkalinity-arithmetic examples above are read in IS 251:1998 and IS 252:1991, both superseded. The current editions are IS 251:2024 and IS 252:2013. The structural points are transferable; the clause numbers and values are not. Check the current edition before writing either into a contract.

“Not detected” is three different numbers

A result written “<0.001%” or “ND” is uninterpretable without the method and a stated limit, because at least three distinct quantities get written that way. Eurachem’s Fitness for Purpose of Analytical Methods, 3rd edition (2025), section 5.3.1, distinguishes the critical value (the result indicating an analyte level significantly different from zero), the limit of detection (the lowest concentration detectable at a specified level of confidence) and the limit of quantification (the lowest level at which performance is acceptable for the application) — and records that “reporting limit” is a synonym of the limit of quantification, not of detection. ICH Q2(R2) adds the usual estimation approaches: a signal-to-noise ratio of about 3:1 for the detection limit and at least 10:1 for the quantitation limit, or 3.3σ/S and 10σ/S from the standard deviation of the response and the slope of the calibration.

So a heavy-metals line reading “not detected” may mean “below 1 ppm” or “below 50 ppm”, and only the laboratory knows which. If a trace limit matters to you, specify the method and require the quantitation limit to be printed on the certificate beside the result.

The sampling question, which is the one that actually decides disputes

Everything above concerns the quality of a number. This section concerns whether the number has anything to do with your consignment, and it is where most of the real risk sits.

Before any of it: drawing a sample from a tanker or a bulk container of a corrosive, oxidising or flammable chemical is a hazardous operation, and none of what follows is a procedure for doing it. Top hatch access is work at height; the hatch may be under pressure or vacuum; the vapour space is an inhalation hazard, and in a nitrogen-blanketed or otherwise inerted tank it is an asphyxiation hazard rather than merely a toxic one; on a flammable load, sampling equipment lowered into a vapour space is a recognised ignition source, so bonding and earthing of the vehicle and the equipment, and exclusion of ignition sources, belong in the procedure. Splash protection and eyewash provision are not optional. ISO 3165:1976 exists specifically for safety in sampling chemical products for industrial use. Whether an atmosphere needs testing, and what controls apply, is a competent person’s call — agree who draws the sample, under whose written procedure and with what protection, before the vehicle arrives.

The standards are unusually blunt about it. ISO/IEC 17025:2017 requires a test report to carry a statement that the results relate only to the items tested, and requires that where the laboratory was not responsible for the sampling, the report state that the results apply to the sample as received. In other words, the most authoritative document in the field refuses to let a report speak for a consignment the laboratory did not sample. A COA is in the same position, usually without saying so.

A one-litre sample of 48% caustic lye weighs about 1.5 kg, and on a 20-tonne load it speaks for some thirteen thousand times its own mass. Whether it represents that mass is a matter of design, not of good faith.

What representative sampling actually requires

The vocabulary exists, and it is worth using. ISO 6206:1979, Chemical products for industrial use — Sampling — Vocabulary (confirmed 2023), defines the terms the rest of this section uses — increment, bulk sample, blended bulk sample, laboratory sample, reference sample, lot, batch, consignment, sampling unit and sampling plan — and it is in scope for industrial chemicals, which is more than can be said for the standard everybody reaches for first. Write your sampling clause in those terms and both sides know what the words mean.

For tank geometry the fullest treatment is ISO 3170 (current edition ISO 3170:2025; the definitions below are read in the superseded 2004 edition, whose text is available as a public preview). It defines a representative sample as one having physical or chemical characteristics identical to the volumetric average, and distinguishes a spot sample taken at one location and time, a running sample accumulated while passing through the total liquid height in both directions, and a composite sample made by combining spot samples in defined proportions. Upper, middle and lower spot samples sit at one-sixth, one-half and five-sixths of the liquid depth below the surface. Sampling begins when the contents are at rest. And a manual spot sample taken from a pipeline, the standard warns, may not be representative of the bulk.

Those positions are the standard’s, for a hydrocarbon in a deep fixed tank. How many increments to take, and whether those positions are the right ones for your product and your vessel, is a judgement rather than a table lookup. It depends on the product, the vessel geometry and the parameter at risk, and it belongs to a competent person — a stratifying or settling product in a shallow road tanker can need more sampling than depth alone would suggest. What a buyer can usefully do is require the number and position of increments to be agreed in writing beforehand, rather than discovering afterwards that the certificate rests on one dip from the top hatch.

ISO 3170 covers hydrocarbon liquids — petroleum products, crude oils and intermediates — and does not apply to caustic lye, acids, hypochlorite or any other non-hydrocarbon chemical. It is cited here for the tank-geometry detail it adds: borrow the structure, not the authority. For industrial chemicals the standards in scope are ISO 6206:1979 for the vocabulary, ISO 3165:1976 for safety in sampling, ISO 8213:1986 for technique on particulate solids, and product-specific methods where they exist — ISO 1552:1976 for liquid chlorine, for instance. What we did not find is a general ISO standard prescribing a sampling plan for bulk non-hydrocarbon liquid chemicals, which is why the plan has to come from your contract.

For packed solids, Indian product standards tend to carry their own sampling annex rather than cross-refer to a general standard, and the pattern is worth describing because it is what a well-drafted contract clause looks like. IS 252:1991 Annex B defined the lot as all containers in one consignment from a single manufacturing batch, treated separate batches in one consignment as separate lots, and scaled the number of containers to open by lot size — 3 containers from a lot of 3 to 50, 4 from 51 to 200, 5 from 201 to 400, 6 from 401 to 650, 7 from 651 to 1,000 — selected systematically, every r-th container where r is the integer part of N divided by n. It then required the portions to be combined into a composite and split three ways: one set to the purchaser, one to the supplier, and one referee sample bearing both parties’ names, kept at a location they agree, for use in a dispute.

And then the part that is easy to miss: the conformity criteria were not uniform. For containers, iron had to be met by every individual sample, while the other characteristics were judged on the composite. A composite average can conform while an individual container does not. If you have a parameter where one bad container ruins a batch, the contract has to say that it is judged individually. And for bulk vehicles the same annex took the stricter line: a tank car or wagon conformed only if its own individual sample met every requirement, with no compositing across vehicles — which is the rule a multi-tanker contract should borrow.

Edition warning, and it cuts against us as much as anyone: IS 252:1991 is superseded by IS 252:2013, which we have not read. So treat the annex above purely as a drafting template — the shape of a clause that works — and make no assumption about what the 2013 edition says. Worth knowing: BIS operates a facility for viewing Indian Standards under compulsory certification free of charge, and several of the chemical standards in this article fall under Quality Control Orders, so the current text may be readable without buying it. For solids generally, ISO 8213:1986 (confirmed 2023) gives sampling technique for particulate chemical products up to 100 mm particle size, but it supplies technique rather than a sampling plan — it directs the user to a previously established plan.

And what a “batch” is, when the plant never stops

For a continuously produced bulk liquid there is no natural batch, so a batch has to be defined. The authoritative construction is the WHO one: a batch is a defined quantity processed in a single process or series of processes, and in the case of continuous manufacture the batch must correspond to a defined fraction of the production. The EU GMP glossary carries the same idea in the alternative form — all the units manufactured in a given period of time.

So for a continuously produced liquid, require the COA to state what the batch number represents: the time window or the tonnage, and the quantity covered. Without that, “Batch No. 2614” is an arbitrary label, and a certificate against it is a certificate against nothing in particular.

What the COA is not, and what sits elsewhere

Four things routinely get asked to do a COA’s job.

• The safety data sheet does not carry quality data. An SDS or an international chemical safety card is substance-level hazard information — hazards, first aid, exposure limits, physical constants — with no batch content whatsoever. It is the right document for handling the material and the wrong one for accepting it.
• The package declarations are not analysis. The Legal Metrology (Packaged Commodities) Rules, 2011 govern identity and quantity — manufacturer, packer or importer, the generic name of the commodity, net quantity, month and year of manufacture, and consumer-complaint contact details. They are a labelling regime, and they were never quality data. For bulk industrial supply much of it does not apply at all. The exemption from the retail-package chapter, as substituted by G.S.R. 629(E) of 23 June 2017 with effect from 1 January 2018, has two limbs that matter here: packages containing more than 25 kg or 25 litres — with cement, fertiliser and agricultural farm produce exempt only in bags above 50 kg, so such a package between 25 and 50 kg is not exempt — and, separately, packaged commodities meant for industrial or institutional consumers. Those consumer definitions extend to buying from an importer or a wholesale dealer as well as direct from the manufacturer, so a trader’s sale can sit inside that limb — but only where the package carries the declaration “not for retail sale”, and the institutional limb additionally requires use not for commercial or trade purposes. These Rules are amended often; this describes the position as at October 2026, and it is worth checking rather than assuming.
• ISO 9001 certification is not accreditation. Certification of a management system says nothing about the technical validity of any test result and confers no accredited status on any certificate.
• An accredited laboratory does not make an accredited certificate. Accreditation attaches to parameters within a defined scope. NABL’s own policy — NABL 133, Policy for Use of NABL Symbol and / or Claim of Accreditation by Accredited Conformity Assessment Bodies, Issue 09 of 13 December 2019 with Amendment 03 of 3 September 2024 — requires the NABL symbol to be used for parameters inside the accredited scope, prohibits its use and any claim of accreditation for parameters outside it, requires non-accredited parameters appearing in a report to be clearly identified and segregated, and states that a separate report or certificate should be issued for non-accredited parameters. WHO’s model makes the same point from the other side, listing a reference to ISO/IEC 17025 accreditation as content to include only where the laboratory holds it and the accreditation relates to the specific analysis. The buyer’s question is not “is the laboratory accredited” but “is this parameter, by this method, on its published scope”.

A COA is not a warranty of fitness — and what the law supplies instead is a condition, on conditions

This is where a buyer’s assumptions are most expensive, so it is worth stating precisely. Under the Sale of Goods Act, 1930, section 16 opens by providing that, subject to the Act and to any other law, there is no implied warranty or condition as to the quality or fitness for any particular purpose of goods supplied under a contract of sale. It then sets out the exceptions, and both of them are conditional:

• Fitness for a particular purpose arises where the buyer makes the purpose known to the seller in circumstances showing reliance on the seller’s skill or judgment, and the seller deals in goods of that description — with an exception for a specified article sold under a patent or trade name.
• Merchantable quality arises where goods are bought by description from a dealer in goods of that description, but not as regards defects which an examination the buyer actually made ought to have revealed.

Section 15 adds the one most useful to a chemical buyer: on a sale by description there is an implied condition that the goods correspond with the description, and on a sale by both sample and description it is not enough that the bulk matches the sample — it must also match the description.

The word “condition” is doing real work there. Under section 12 of the same Act, breach of a condition gives rise to a right to treat the contract as repudiated — to reject the goods — while breach of a warranty gives a claim for damages but no right to reject. So what sections 15 and 16 supply, where their conditions are met, is stronger than a warranty, not weaker.

And then the qualification that governs all of it. These are default terms. Section 62 provides that a right, duty or liability arising under a contract of sale by implication of law may be negatived or varied by express agreement, by the course of dealing between the parties, or by usage. So the contract is where this is decided — which is why reading the other side’s terms of sale matters as much as reading the certificate.

The practical consequence is a drafting one. A COA number creates no fitness warranty, and matching a retained sample does not discharge the description. So state the purpose in the purchase order in writing, and make the grade, the standard and its edition part of the contractual description of the goods — because the description is the thing the law will hold the seller to.

Where a statutory duty exists, whose duty it is

Two regimes get regularly inverted in procurement discussion, and both inversions put an obligation on the wrong party.

Food-grade material. Under the Food Safety and Standards Act, 2006, “food” means any substance intended for human consumption, a food additive is a substance not normally consumed as a food by itself but added for a technological purpose, and a food business is any undertaking handling food at any stage — which includes distribution and sale. The duty to ensure compliance sits on the food business operator at all stages of production, processing, import, distribution and sale within the businesses under his control, and section 31(1) provides that no person shall commence or carry on a food business except under a licence, with registration rather than licensing provided for petty manufacturers, petty retailers, hawkers and tiny food business operators. Section 26(4) goes further: an operator selling food to a vendor must give a written guarantee as to the nature and quality of the food, and the bill, cash memo or invoice itself counts as that guarantee even where the prescribed form is not used.

Read the attribution carefully, because it runs in both directions. Those are duties on whoever sells the food-grade material — and separately on you, if you are using it in food, because that makes you a food business operator with your own section 26(1) and section 31 obligations for the business under your control. The seller’s section 26(4) guarantee is a defence of limited scope, not a transfer of your duty. What it is not is a reason to treat the seller’s statutory position as your diligence exercise, or your own position as discharged by an invoice. On content, the Food Safety and Standards (Contaminants, Toxins and Residues) Regulations, 2011, as amended, set entry-specific metal limits rather than one blanket figure — including an entry for chemicals not otherwise specified used as ingredients or in the preparation or processing of food — so a food duty needs the applicable entry and limits written into the specification, and a technical grade that happens to assay well is not a substitute.

Pharmaceutical use. The revised Schedule M to the Drugs Rules, 1945, notified by G.S.R. 922(E) of 28 December 2023, places the obligations on the licensed drug manufacturer: starting materials are to be purchased only from approved suppliers and, where possible, directly from the producer; each consignment’s containers are to be checked at least for integrity of package and seal and matched against the order, delivery note and supplier labels; each batch in a delivery is treated separately for sampling, testing and release; there must be procedures to ensure the identity of the contents of each container; and only materials released by quality control and within shelf-life may be used. A supplier’s COA does not discharge any of that. The supplier’s job is to be an approvable vendor furnishing a certificate the manufacturer can check.

We are a bulk supplier and trader, not a manufacturer, and nothing here states or implies that Velora holds any particular licence or registration. Where an application brings a licensing requirement — food, pharmaceutical, or a scheduled substance — tell us the application and we will confirm in writing whether we can supply against it. Require the same of any supplier, and do not accept a grade name as the answer.

What a trader can and cannot give you

Velora buys from producers and sells on, so it is worth being plain about what that means for documentation, because the honest answer is more useful than a reassuring one.

Where the producer issues a batch certificate and we supply in the original unopened packaging, we pass that certificate through with the batch number that is on your packages. Two honest qualifications. Where a producer issues a certificate covering a production campaign rather than your consignment, we will tell you what the certificate actually covers before you order. And where we break bulk or repack, the number on the packages is ours rather than the producer’s, and we will say so rather than let you discover it on arrival. We will name the producer where our supply arrangements permit it, and tell you where they do not.

We do not operate a laboratory. Where a parameter justifies it we can arrange testing at an independent laboratory, or at one you nominate, on a jointly drawn and sealed sample — and you should receive that report alongside the producer’s certificate, not instead of it. The IPEC guide provides that an additional COA is issued where a distributor has additional testing performed, and the companion practice of furnishing the original manufacturer’s certificate as well sits in the IPEC Good Distribution Practices Guide for Pharmaceutical Excipients, section 7.5. That is the practice we work to where we have had our own testing done.

What we do not give is a warranty that the material will perform in your process, and our terms say so — nor can any pass-through turn a producer’s first-party statement into third-party verification. But read that alongside the point made above under A COA is not a warranty of fitness, because it runs the other way: where a buyer makes the purpose known to a seller who deals in goods of that description, in circumstances showing reliance on the seller’s judgment, section 16(1) can impose a fitness condition by operation of law — subject to the patent or trade-name proviso, and subject to section 62, which lets express terms negative or vary it. In other words the contract decides, and ours says what it says. So tell us the purpose and let us answer it in writing, rather than assume either that a warranty exists or that none can. What is also worth asking any trader for: the batch and its date of manufacture, and the storage history of anything that has been held in stock.

WHO’s model certificate adds a condition worth writing into a purchase order for material that has been in storage or transit: it provides for the expected shipping, packaging and storage conditions to be stated, with a statement that deviations from them would invalidate the certificate. A certificate issued at the plant gate says nothing about what six weeks in a yard did to the material.

The editions, because this is where citations go stale

A standard cited without an edition, or in a superseded edition, is a contract clause that will not do the work you want it to. These are the editions we could verify as current at the date of this article; verify before you cite, because several of these turned over recently.

Three cautions on that table, and they apply to us as much as to anyone. First, we verified the edition designations from BIS, ISO, ILAC and standards-reseller records, but we have not read the texts of IS 251:2024, IS 252:2013, IS 266:2025 or IS 11673 (Part 1):2019 — so no value, clause or method in this article is attributed to those editions, and you should read the current text before writing a clause on it. Several of them sit under Quality Control Orders, and BIS operates a free-of-cost viewing facility for standards under compulsory certification. Second, BS 5309-4, still cited in Indian tender documents as the sampling standard for chemical solids, was withdrawn in 2013 with no replacement edition. Third, BIS has an Indian adoption of the 2005 edition as IS/ISO/IEC 17025:2005, but we could not establish whether the 2017 edition has been adopted; do not write “IS/ISO/IEC 17025:2017” without checking.

The short version

A certificate of analysis is worth what the purchase order made it worth. Five clauses do most of the work: name the standard and its edition; require numerical results with the method, the basis and the acceptance criterion against each one; name the referee method and the governing laboratory; agree the sampling protocol, the lot definition and a sealed referee sample before the material moves; and state the decision rule and rounding convention that apply at the limit. None of them costs anything at the enquiry stage, and each of them removes an argument you would otherwise have after the tanker has discharged.