Every value on a MolGod SDS card carries a citation you can check yourself

Nonsensia Ltd (Company No. 15165427, 124–128 City Road, London EC1V 2NX) builds safety data sheets in which every stated value carries a citation the reader can look up — the company sells regulatory documentation only: SDS cards, certificates of analysis, labels, compliance work, LIMS and card audits. It sells no chemical substances of any kind.

There are three established ways to obtain a safety data sheet: download one from a database, write it in-house, or buy consultant hours. All three produce documents, and two of them produce judgement that no software should claim to replace. This article does not argue that you should drop any of them. It argues about one property that is easy to lose in all three, and that decides how expensive a document is to defend when it is questioned: whether every value printed on the sheet can be traced back to the named source it came from, by a reader who has never spoken to us.

Everything stated below about our own process was measured in the plugin source and in a rendered card, not estimated. Where we have not measured something, we say so rather than filling the gap with a plausible number.

How is a MolGod SDS card built?

A finished MolGod SDS card contains six fixed elements:

• 16 sections following the structure of REACH Annex II as amended by Regulation (EU) 2020/878;
• a physicochemical layer sourced from PubChem and the NIST WebBook, with citations attached to the values;
• the PubChem CID, InChIKey, molecular formula and molar mass, so the substance can be re-identified independently of the name printed on the sheet;
• a skeletal structure rendered from a MolBlock by RDKit (modules/m30-label-generator/tools/rdkit_skeletal_svg.py), so the geometry is computed rather than drawn by hand;
• supplier identity and logo written into section 1.3 from a single configuration point (MOL-GOD → Label Generator);
• a PDF composed by WeasyPrint.

A wrong company address is therefore a configuration error, not a typing error repeated across a hundred files.

How does a MolGod SDS card cite its sources?

Every data point on the card carries a citation, and 61 of the 71 registry entries carry a URL the reader can open without asking us. The registry lives in class-sds-citation-registry.php and holds 71 canonical entries. The institutions behind them are:

• ECHA / European Chemicals Agency — 16 occurrences;
• EPA — 9;
• PubChem — 3;
• OECD — 3;
• NIST — 2;
• IARC — 2;
• European Parliament, for legislative acts — 2;
• IPCS — 1.

Chicago style matters here for one reason: a citation in that form carries enough detail for a third party who has never contacted us to locate the source, and a DOI is included wherever a DOI exists (includes/class-chicago-citation.php). Each cited source also carries the date on which it was read.

The card exposes the state of each citation rather than hiding it:

• molgod-citation-doi-status — verified or unverified; when a DOI has not been confirmed, the card says so instead of presenting the reference as settled;
• molgod-citation-paywall — flags a source the reader will hit a paywall on, before they click;
• molgod-citation-weight — the weight assigned to the source, visible on the surface rather than buried in a pipeline.

We do not claim that every DOI in the set has been confirmed. Sources without a stable public URL — an ISO standard behind a paywall, for example — are omitted rather than cited blind, because a citation nobody can open is decoration rather than evidence. A document that marks its own weak points can be audited; one that presents every value with the same flat confidence cannot.

Which regulatory datasets stand behind the classification?

The reference data is already loaded on our side, and each Annex VI row carries the ATP and the CELEX identifier of the act it came from.

• ECHA CLP Annex VI, full table at ATP23 — 3855 records across 14 columns: index number, EC number, pictograms, signal word, H-codes, SCL / M-factor / ATE, Notes, ATP, CELEX.
• Specific concentration limits, M-factors and ATE values (clp_annex_vi_scl_mfactor.json) — 620 entries.
• Regulatory status seed (regulatory_status_seed.json) — 235 CAS numbers across seven inventories in five jurisdictions.
• ADR transport data (adr_dangerous_goods.json) — 39 CAS numbers with UN number, transport class and packing group.
• REACH Annex XVII (reach_annex_xvii.json) — restrictions on placing on the market.
• Citation registry — 71 canonical entries, 61 of them with a verifiable URL.

The regulatory status seed covers 235 CAS numbers across seven lists:

• SVHC — 224 CAS;
• REACH Annex XIV (authorisation) — 55 CAS, each with sunset date and authorisation-required flag;
• TSCA — 10 CAS;
• UK SVHC — 10 CAS;
• DSL toxic — 11 CAS;
• NDSL — 1 CAS;
• AICS — 12 CAS.

Our classification layer is assembled from these regulatory datasets rather than copied from an older sheet, so a correction is applied in one place — the dataset — instead of in every file that once quoted it.

How do you tell a draft SDS from a reviewed one?

A card with no reviewed_at value renders its PDF with a visible WERSJA ROBOCZA (working draft) watermark. The watermark disappears only when a person sets that field during review.

Draft status is the audit-relevant mechanism: the review status of an SDS card is a property of the PDF file itself, not of a spreadsheet kept next to it. There is no route to an unwatermarked card that skips the review. It is an interlock in the code, not a promise in the terms of service.

The draft watermark is deliberately unflattering to us: the automated output leaves our system visibly labelled as a draft, and there is no route through our pipeline that yields an unwatermarked card without review.

What is missing from a safety data sheet you download from a database?

An aggregator or a supplier library gives you a file. It does not give you the path back to the source of the values printed on it. During an inspection, or during a customer audit of your supply chain, the question is on what basis is this stated — not where did you download it. A file answers the second question and stays silent on the first. Nonsensia Ltd exists to close that gap in the document itself, not to argue that libraries are useless.

Harmonised classification or supplier self-classification: how do you tell which one your sheet uses?

Three different situations can produce the same H-code on paper:

• a harmonised entry in Annex VI to CLP, traceable to an index number, an ATP and a CELEX identifier;
• a supplier self-classification — a legitimate route, but one that carries different evidential weight and remains the supplier’s own call;
• a value carried forward from an earlier document without a stated source.

A downloaded PDF rarely distinguishes between the three, and neither does a flash point printed without a named measurement behind it. This is not a criticism of any particular provider — it is a property of the model. A file is a snapshot, and a snapshot does not carry its own provenance.

How can you check where each value on a safety data sheet came from?

Ask, value by value, which named source the number comes from and on what date that source was read. On a MolGod card the answer sits next to the value as a citation, and the registry behind it is described under how a MolGod SDS card cites its sources. On a downloaded file, the answer is whatever provenance its author chose to include.

Why does an SDS citation need an access date?

A regulatory source can change. A citation without a date tells you what a page said at an unspecified moment; a citation with a date is a check that somebody else can repeat and, if necessary, contradict. Printing the access date was an explicit decision taken on 2026-07-16, on the principle that the access date is part of the credibility of the reference rather than an ornament on it.

What happens to a downloaded SDS when CLP Annex VI is amended?

A sheet in your folder is a fixed artefact: when the underlying regulation moves, the file does not move with it, and nothing in the file tells you which revision it was written against. Two things make that readable on our side rather than assumed. Each Annex VI row carries the ATP and the CELEX identifier of the act it came from, so a row can be compared against the next revision of the act. And regulatory_status_seed.json carries a sunset date and an authorisation-required flag per entry — values that change on their own schedule, independent of any document.

How do you test an existing safety data sheet for traceability?

Take a sheet from your own files and ask, value by value, which named source each number comes from. The five-step procedure below applies the same test to ours, and prices are listed under how much a safety data sheet costs.

Why not write your safety data sheets in-house?

Building safety data sheets in-house is not a competence problem: a trained regulatory affairs specialist can write a defensible SDS without outside help. The in-house question is arithmetic and repeatability, and both are properties of the process, not of the person running it. Nonsensia Ltd supplies the assembly work; it does not supply the specialist’s judgement and does not claim to.

How many datasets does one safety data sheet require?

One safety data sheet is assembled by reconciling several independent datasets — six of them on our side, listed under which regulatory datasets stand behind the classification — before the first section is written.

A hundred products means the same lookups repeated a hundred times, and which lookups apply differs product by product: our transport data covers 38 CAS, regulatory status 235 CAS, specific concentration limits 620 entries. A substance may fall inside all of those sets, or inside none of them, and finding out which is itself part of the work.

How long does it take to write one safety data sheet in-house?

We do not publish an hours-per-card figure, because we have not measured your process and would be inventing the number. The figure that decides this question is yours to produce: time one card end to end, from the Annex VI lookup to a PDF you would hand to an inspector, then multiply by your catalogue. Do it on a substance nobody has touched before, not on the card your best person wrote last year. What we can put on the table is what we publish — see prices and turnaround — and you compare those against your own measurement, not against ours.

Why do in-house safety data sheets drift out of consistency?

Cost is the weaker argument against in-house SDS authoring; consistency drift across a catalogue is the stronger one. A card written in March by one person and a card written in September by another differ in layout, in which source was consulted, and in how much detail the physicochemical section carries. Each of them can be perfectly defensible on its own. As a set of one hundred, the inconsistency itself is visible before the content is — a set read as a set shows its own spread.

How does a MolGod card stay consistent across a hundred products?

Cards produced by one engine stay consistent with each other by construction rather than by discipline, through seven mechanisms:

• one 16-section layout, generated to the Annex II / 2020/878 structure rather than retyped from a previous file;
• supplier identity and logo resolved from a single configuration point, so a changed address is changed once and not one hundred times;
• molecular geometry computed by RDKit from a MolBlock, rather than drawn by hand and redrawn differently next quarter;
• the same identifiers on every card: PubChem CID, InChIKey, molecular formula, molar mass;
• the same citation registry behind every card — 71 canonical entries, 61 with an openable URL, in Chicago style with a DOI where a DOI exists;
• one PDF composition path through WeasyPrint;
• one mechanical draft rule: no reviewed_at, no unwatermarked card.

The mechanisms themselves are described in full under how a MolGod SDS card is built; what makes them a consistency argument is that none of them depends on anybody remembering to apply them.

What happens to your SDS catalogue when a new CLP ATP is published?

Manual authorship makes every update proportional to the catalogue. An amendment to Annex VI does not touch one product; it touches every product whose entry moved, and establishing which ones moved is itself a pass over the whole set. Authorisation deadlines shift on their own schedule, without anybody editing a document.

Three properties make that legible rather than assumed: the ATP and CELEX columns per Annex VI row, the sunset date and authorisation flag per regulatory status entry, and the access date printed with every citation. Where a DOI could not be confirmed, the card marks it unverified; where a source has no stable public URL, it is omitted rather than cited blind. Done by hand, an update is the same lookup work again, repeated for every affected product in the catalogue.

Which compliance decisions cannot be automated?

An engine assembles a safety data sheet; it does not decide. Five compliance decisions are not automatable, and Nonsensia Ltd does not attempt them:

• whether a substance or mixture is placed on the market at all, and in which jurisdictions;
• exposure scenarios and conditions of use specific to your customers, your processes and your packaging;
• your own assessment as a supplier, and the responsibility attached to it;
• the judgement call on any source the card flags as unverified or paywalled: whether it is good enough for your product;
• who signs, and on what basis.

Why not use a regulatory consultant instead of an SDS engine?

A regulatory consultant and a documentation engine answer different questions. The consultant is bought by the hour and paid for judgement; the engine is bought per card and paid for repeatability. Most chemical portfolios contain both kinds of work at the same time, and Nonsensia Ltd sells only the second kind. The useful question is not which model wins, but which part of your workload belongs to which.

What can a regulatory consultant do that SDS software cannot?

Some questions have no lookup answer, because the answer is an argument rather than a record:

• Non-standard mixtures. Bridging principles, additivity judgement calls and interpretation of missing endpoint data are reasoning tasks, not table joins.
• Disputes with an authority. Correspondence with a national enforcement body is advocacy; it needs a person who takes a position and defends it.
• Authorisation applications. Our regulatory status seed flags 55 CAS under REACH Annex XIV with sunset_date and authorisation_required. The flag tells you a decision is due. It does not make the decision.
• Registration strategy. Tonnage banding, consortium choices and data-sharing negotiations are commercial decisions with regulatory consequences.
• Paywalled normative texts. Our registry omits acts without a stable public URL — ISO standards, for example — instead of citing them blind. Reading and applying those texts is expert work by design.

When does a regulatory consultant stop being cost-effective?

The consulting model prices attention, and attention does not divide. A single difficult substance repays an hour of expert time many times over. One hundred routine substances repay it once; the remaining ninety-nine are charged again at the same rate for the same lookups.

Repetition across a catalogue means the same four operations on every substance:

• find the same Annex VI entry;
• assemble the same hazard statements;
• populate the same physicochemical section;
• format the same references in a consistent style.

Layout, structure rendering and the supplier block are bookkeeping, not judgement, and under an hourly model bookkeeping is charged at the same rate as judgement. Done by an engine, the hundredth card is assembled the same way as the first, at the price listed in the shop.

Which SDS tasks should be automated and which need a consultant?

Who reviews a MolGod SDS card before it is signed?

The Expert-Reviewed SDS at €249.99 goes through human review within 72 hours, and that review is the point at which a person takes a position on the content and sets reviewed_at. The mechanism that makes review visible in the file is described under how you tell a draft SDS from a reviewed one. We sell documentation, not a substitute for your own reviewer: an engine assembles, a person reviews, and the decision on whether the result is good enough for your product stays with you.

How long does it take to get a safety data sheet?

A working draft is generated automatically and sent by e-mail as soon as it is produced; a signed SDS card takes up to 72 hours of human review; a 100-product catalogue takes 7 days to first drafts and 30 days to the complete set.

• SDS Working Draft — generated automatically, e-mailed as soon as it is produced. Enough to start internal work, label layout and portfolio triage.
• Expert-Reviewed SDS — human review, 72 hours.
• Full LIMS deployment — 100 documentation records with hosting; drafts within 7 days, complete within 30 days.

Where consultant capacity is booked ahead, the queue — not the hourly rate — is what holds up sourcing, labelling and listing decisions that had nothing to do with the difficult question. The sequence that follows is simple: take drafts for the whole portfolio, review the ones that carry real risk, and spend consultant hours on the substances that survive that filter.

How much does a safety data sheet cost?

A safety data sheet costs €49.99 as an automatically generated working draft and €249.99 as a signed card reviewed by a person within 72 hours. Our published prices are these:

• SDS Working Draft — €49.99. Generated automatically and delivered by e-mail.
• Expert-Reviewed SDS — €249.99. Human review within 72 hours, no draft watermark.
• Full LIMS deployment — €2,999. 100 documentation records with hosting; first drafts within 7 days, complete set within 30 days.
• LIMS maintenance — €1,499 per year.

What does a consultant receive if an enforcement dispute arises?

If an enforcement question arrives, you will need a consultant. We are not arguing otherwise, and that is not a caveat appended to the sales case — it is part of it. The relevant difference is what that consultant receives on day one.

Which source was consulted and when is visible next to the value, together with its citation status — verified, unverified or paywalled. The registry behind those citations is described under how a MolGod SDS card cites its sources. Hours that would otherwise go into reconstructing where the numbers came from can go into the disputed question instead.

How do the four routes compare?

How can you verify an SDS provider before you pay?

You do not have to trust this article. The claims that decide whether you buy — the classification, the citations, the transport entry — can be tested by one person, at a desk, with a browser. The procedure below can be run by one person, at a desk, with a browser, and the only thing it needs from us is the card itself.

Run it on a molecule you know well, not on an easy one. A substance with a harmonised classification, a transport entry and an authorisation status will expose more than a simple solvent. The point of the exercise is to find the places where a document is weakest.

How do you check an SDS provider in five steps?

1. Take a CAS number from your own catalogue — preferably a difficult one.
2. Confirm the card describes the same substance, then compare its classification against the ECHA CLP Annex VI row for that substance.
3. Click a citation and confirm it lands on a real document from the institution named.
4. Check the UN number, transport class and packing group against ADR.
5. Find what the card marks as unverified or paywalled — and judge the document by whether it admits those gaps at all.

Step 1 — Which CAS number should you test an SDS provider on?

Pick a CAS number you ship, not one we suggest. Ideally choose something a customer, a carrier or an inspector has already questioned. That is the case where a wrong H-code or a missing UN number has a real cost.

• Prefer a substance with a harmonised entry in Annex VI over one classified only by supplier self-assessment.
• Prefer a substance that moves by road, so that Step 4 has something to compare.
• Write down, before you start, what your current supplier sheet says — hazard class, H-codes, signal word, UN number.

Step 2 — How do you check an SDS classification against ECHA CLP Annex VI?

Start with identity, not with hazard. Check the identifiers the card does carry — PubChem CID, InChIKey, molecular formula and molar mass — and confirm that the Annex VI row you are reading describes the same substance. An identifier check that fails makes every later comparison meaningless.

Then open the harmonised classification entry at ECHA and compare it against what the card states. The classification layer behind our sheets is the full Annex VI table at ATP23: 3855 records across 14 columns, with specific concentration limits and M-factors held in a separate set of 620 entries.

• H-codes and signal word — the full set, not a subset. Missing codes matter more than surplus text.
• Pictograms — as listed in the Annex VI entry.
• SCL / M-factor / ATE — where Annex VI carries them, they should agree.
• Notes and ATP — Notes change how a classification applies; the ATP tells you which revision the entry is from.
• Index number and EC number, where a sheet states them — an index number on a safety data sheet must match the ECHA CLP Annex VI entry character for character, including the check digit and the hyphenation, because matching H-codes against a wrong index number describes a different legal entry. Apply this test to every sheet you hold, ours included.

If your substance sits on a regulatory list, check that too. The regulatory status seed covers 235 CAS across seven inventories in five jurisdictions, with sunset dates and authorisation flags where they apply, and restrictions on placing on the market come from REACH Annex XVII. A sheet that is silent about an Annex XIV sunset date on a substance you are still importing is a sheet worth arguing with.

Step 3 — How do you verify that an SDS citation is real?

Pick any cited value and follow the reference. Our registry holds 71 canonical entries, 61 of which carry a URL you can open yourself, in Chicago style, with a DOI where a DOI exists and with the date the source was read. A citation should land on a document published by the institution named — not on a blog, a reseller page or a search results screen.

• Does the link resolve at all, or does it return 404?
• Does the target document actually contain the value quoted on the card?
• Is the institution on the page the institution in the citation?
• Where a DOI is present, does it resolve to the same work?

Step 4 — How do you verify the UN number and packing group against ADR?

A safety data sheet transport entry is correct only when the UN number, transport class and packing group all agree with ADR; a correct UN number with the wrong packing group still produces a wrong label and a wrong transport document. Our transport dataset covers 38 CAS numbers with those three fields.

If your substance sits outside that covered set, treat the absence of a transport entry as information rather than an omission to be filled in by guesswork. Ask us at /contact/ what the card can and cannot state for that CAS number.

Step 5 — What should a safety data sheet admit it has not verified?

Step 5 is the step most people skip: reading a safety data sheet for what it marks as uncertain rather than for what it asserts. Our pages carry explicit citation status markers — molgod-citation-doi-status with a verified or unverified value, molgod-citation-paywall and molgod-citation-weight. Where a DOI has not been confirmed, the card says unverified instead of presenting the reference as settled, and a source with no stable public URL is omitted rather than cited blind.

There is a second, mechanical admission built into the PDF. A card with no reviewed_at value renders its PDF with a visible WERSJA ROBOCZA (working draft) watermark. The watermark disappears only when a qualified scientific reviewer sets that field. It is enforced in the code, not promised in terms and conditions.

What does a finished MolGod SDS card contain?

The structure you will be checking is fixed, and it is the six-element list under how a MolGod SDS card is built. If you want to run the full procedure on our output rather than on a screenshot, an SDS Working Draft is €49.99 and is generated automatically — see the shop. The Expert-Reviewed SDS at €249.99 is the version that has passed human review within 72 hours and therefore carries no draft watermark.

How do you test a safety data sheet you already hold?

You can run the same procedure in the other direction, on a document you already own, using nothing but the public sources named above. When you do, the values that carry no traceable source at all are the category to work through first, because they show which values would have to be reconstructed if the sheet were challenged.

If you want a second pair of eyes on a sheet you already hold, write to us at /contact/ or pure@molgod.org, and we will tell you what we can check against named sources and on what terms.

Who is responsible for a safety data sheet under REACH?

Under REACH, the supplier placing a substance or mixture on the market is responsible for the safety data sheet; a document bought from a third party supports that assessment but does not transfer the obligation. A safety data sheet supplied by Nonsensia Ltd does not transfer your legal obligations as a supplier, importer or distributor.

As supplier you remain responsible for three things the card does not cover:

• your own assessment of the substance;
• the accuracy of the data you provide about your product;
• the decisions you take on the basis of the card.

What the traceable-citation model changes is the cost of defending a value: instead of reconstructing where a number came from, you open the reference printed next to it. We describe the process by which the document was built and expose every source it rests on. The outcome of any given inspection is not ours to promise, and we do not promise it.

What is the smallest sensible first step?

Take one substance from your catalogue — the difficult one, the one that has already been questioned — and run the five steps against a working draft at €49.99. That is a single, bounded test of the only claim that matters here: that every value on the card carries a citation you can check without asking us.

Prices and scope for draft, signed and full-catalogue work are on the shop page. Portfolios, edge cases and terms for reviewing sheets you already hold go to contact or pure@molgod.org.

Nonsensia Ltd, Company No. 15165427, 124–128 City Road, London EC1V 2NX, United Kingdom, sells regulatory documentation only: safety data sheets, certificates of analysis, labels, compliance work, LIMS and card audits. It sells no chemical substances.

Frequently asked questions