Extraction can turn a 16-section PDF into structured fields in seconds, but the fields are only as good as the reading. A transposed CAS digit, a density read in the wrong unit or a concentration range collapsed to one number will flow quietly into inventory totals and Tier II reports. This guide walks through a field-by-field review using an illustrative composite SDS, gives you a checklist for the judgement calls, and shows how reviewed data moves into inventory and reporting in NeoTech's platform.
Where extraction goes wrong, and why it matters
A current SDS follows a standard 16-section format: Section 1 identifies the product and supplier, Section 2 covers hazard identification (classification, signal word, hazard statements, pictograms), Section 3 lists the composition and ingredients, Section 9 gives physical and chemical properties, and Section 16 carries other information, including the date of preparation or last revision. Extraction tools read those sections, but supplier layouts vary, and the common errors are predictable:
- Units: relative density or specific gravity stored as if it were pounds per gallon
- CAS numbers: digits transposed, or a CAS number attached to the wrong ingredient
- Ranges: "10–30%" stored as a single value, or ">99%" stored as a maximum
- Line breaks: a second hazard class on a continuation line dropped
- Dates: 03/04/2025 read as March 4 when the supplier meant 3 April, or the reverse
- Missing items: emergency phone, product code or trade-secret ingredients left blank
These matter because EPCRA reporting draws on the same data. A Tier II report asks for chemical name and CAS number, whether the chemical is an extremely hazardous substance, mixture components, physical and health hazard categories and the maximum amount on site, and that maximum often comes from container counts converted to pounds with the SDS density.
Source29 CFR 1910.1200: OSHA Hazard Communication StandardHow review works in NeoTech
In NeoTech's AI SDS extraction workflow, nothing extracted reaches the approved library until a person has reviewed it. The controls below are the ones the platform provides.
Treat every pre-filled value as a proposal until you have compared it with the PDF. The example below shows what that comparison looks like.
Illustrative example: reviewing a composite acetone SDS
| Field | Source SDS text (composite) | Extracted value | Reviewer action |
|---|---|---|---|
| Product identifier (Sec. 1) | Example Acetone Solvent, product code EX-100 | Example Acetone Solvent | Add product code EX-100 so the record matches container labels |
| Supplier (Sec. 1) | Example Chemical Co.; emergency phone (555) 010-0199 | Example Chemical Co.; emergency phone blank | Add the emergency phone from Section 1 |
| Revision date (Sec. 16) | Revision date: 03/04/2025 | 2025-03-04 | Check whether the supplier writes day/month or month/day before accepting |
| Substance or mixture (Sec. 3) | 3.1 Substances | Mixture | Change to pure (substance) |
| CAS number (Sec. 3) | Acetone, CAS 67-64-1 | 67-46-1 | Correct the transposed digits to 67-64-1 |
| Concentration (Sec. 3) | >99% | Max 99% | Enter 99% as the minimum, not the maximum |
| Physical hazard (Sec. 2) | Flam. Liq. 2 — H225 Highly flammable liquid and vapour | Flammable liquid, Category 2 | No change; matches the source |
| Health hazards (Sec. 2) | Eye Irrit. 2A — H319; STOT SE 3 — H336 (next line) | Eye irritation, Category 2A | Add STOT single exposure, Category 3 |
| Signal word (Sec. 2) | Signal word: Danger | Danger | No change |
| Density (Sec. 9) | Relative density: 0.79 (water = 1) | 0.79 lb/gal | Correct the unit; 0.79 × 8.34 lb/gal ≈ 6.59 lb/gal |
8.34 lb/gal is the approximate weight of a gallon of water, used to convert relative density to pounds per gallon. Record the basis of any conversion you make.
The density row shows why unit checks come first. Say a site holds 4 drums × 55 gal = 220 gal of this product. At 6.59 lb/gal that is 220 × 6.59 = 1,449.8 lb; with the unit error, 220 × 0.79 = 173.8 lb, about one-eighth of the real amount, and the error would carry into every threshold comparison that uses it.
Mixtures and ranges: decide, record, stay consistent
EPA's hazardous chemical reporting rule sets out how mixtures are counted. For an extremely hazardous substance (EHS), you add the EHS present as a component of every mixture to all other quantities of it at the facility at any one time; the component quantity is the concentration multiplied by the mixture weight. You may report a component or the mixture as a whole, but for each mixture the choice should be consistent between Section 311 SDS reporting and Section 312 inventory reporting where possible. A hazardous chemical in a mixture need not be counted at 1% concentration or less (0.1% or less if it is a carcinogen).
Ranges make the arithmetic a judgement call, which is why extraction should keep both ends of the range rather than one number.
If an ingredient is withheld as a trade secret, extraction cannot fill in what the SDS does not show. Mark it as trade secret and ask the supplier what you need for reporting.
Source40 CFR part 370: Hazardous chemical reporting (eCFR)Human-review checklist
Run these checks with the PDF open next to the pre-filled form. Anything you cannot confirm should stay in review or go back to the supplier, not be approved on trust.
Units and mixtures
- Density or specific gravity carries the unit the SDS states
- Any conversion to pounds is recorded with its basis (for example, gallons × lb/gal from Section 9)
- Concentration ranges are kept as a minimum and a maximum
- "Greater than" and "less than" values are entered as the correct bound
- Substance vs mixture matches Section 3
- Components at or below 1% (0.1% for carcinogens) are handled by your documented rule
Source traceability
- Manufacturer, product name and product code match the PDF and the container label
- Revision date matches Section 16, with the date format confirmed
- Language and page count match the document you meant to upload
- The original PDF stays on file with the record
- Reviewer name and review date are recorded in your review log
Missing fields
- Every Section 3 ingredient appears with a CAS number or is marked trade secret
- Emergency phone and manufacturer address are filled in from Section 1
- Every Section 2 hazard class is present, including ones on continuation lines
- Fields left blank are blank because the SDS is silent, and that is noted
Uncertain extraction
- Scanned, rotated or low-quality pages are compared line by line
- Values from tables and footnotes are checked by hand
- Fields you could not confirm are queried with the supplier before approval
- Looser validation is acknowledged only when you are sure the record is correct, and the reason is written down
For an audit trail of library activity, the Workspace SDS Library's "Download report" exports an activity CSV you can filter by date range, facility, user or event ID. Keep your own review log for the field-level reasoning behind each correction.
Handing reviewed data to inventory
Once a record is approved, it sits on the Library tab and inventory rows can link to it. In NeoTech's chemical inventory software (Product Inventory), that link is made by manufacturer and product name:
- Open the inventory batch for the facility and reporting year.
- For a single product, use Guided Add and "Find SDS" with the manufacturer and product name.
- For a bulk import, upload the CSV or XLSX, then use "Find All" to match an SDS to every row and see which rows are matched and unmatched.
- Resolve unmatched rows, usually a product name spelled differently from the approved SDS record, before you rely on the list.
- Save the batch; nothing from an import is recorded until you click Save.
If you correct an SDS record after rows are linked, recheck any quantities you converted with the old values.
Handing inventory to reporting
The reporting step is where an unreviewed extraction error would finally surface. NeoTech's Tier II reporting software works from the linked inventory: "Calculate 302 | 311" or "Calculate 312 · Tier II" works out each decision from the inventory and the reviewed SDS data: converting units with the SDS density, totaling each chemical across products and locations, and counting EHS components inside mixtures, so a wrong density or concentration flows straight into the result. Run it again after any SDS or inventory correction.
Reports are generated per facility and reporting year: 302 as PDF, 311 as PDF or ZIP, and 312 Tier II as PDF or T2S. The Report Center shows each file's status, generation timeline and SHA-256 checksum. A named person certifies the batch, which locks it. You then either file with your state yourself and upload the confirmation receipt, or use NeoTech's filing service, which submits the Tier II to the SERC, LEPC and fire department. Either way, the owner or operator certifies the report.
Federally, the Tier II report is due by March 1 each year and covers the previous calendar year, going to the SERC, the LEPC and the local fire department. States may route submissions through one portal, so confirm the method and recipients with your state.
SourceEPA: EPCRA Sections 311-312Regulatory statements in this article were checked against the official sources linked above on .