Emissions Data Lineage: The Question Every Carbon Number Should Be Able to Answer

Pick any row on a sustainability manager's Scope 1 spreadsheet. Ask a simple question: show me the exact document that number came from. Not the pivot table. Not the fuel export summary. The actual delivery docket, with the litres, the date, the site, and the supplier's letterhead.

This question can stall entire audit weeks. The number is real. The total is defensible. But the trail between the two is a stack of emails, a shared drive folder nobody has curated since 2024, and a workbook where the source of column G is "Sarah pulled it".

That trail has a name in data engineering. It's called lineage. And carbon accounting has almost none of it.

What data lineage means when the record is an emission

In a database team, lineage is the answer to "where did this field come from and what happened to it before it landed here". It's a chain. Extract from source, transform, load, aggregate, expose. Every step is logged. If a downstream number looks wrong, you walk back up the chain until you find where it went wrong.

Carbon accounting inherited none of this discipline. Emission figures still live in workbooks that get emailed around, cells that get pasted over, and totals that get rounded twice on the way to a board pack.

Lineage for a carbon record is the same idea, applied to a single emission line item. Which invoice. Which page. Which line. Which quantity. Which unit. Which factor, from which library, at which version. Which conversion rule. Which allocation logic. Which person or agent touched it. Which timestamp.

Get all of that persisted against every emission record and you have level-5 lineage. Miss any of it and you have less.

The five levels of lineage maturity

Not every workflow needs level 5 tomorrow. But every workflow should know where it currently sits, because the level dictates what kind of assurance opinion the auditor can give.

Level 1. Unknown. The number came from somewhere. Nobody can say where. This is more common than the industry admits. A single cell in a Scope 3 workbook that traces back to a 2023 email attachment somebody re-keyed by hand.

Level 2. Named source. The lineage says "SAP fuel export Q3" or "utility bills from AGL". A category, not a document. Fine for internal scorecards. Useless in an assurance walkthrough.

Level 3. File link. The record points to the actual CSV or Excel file on the shared drive. Better. But when the auditor picks a specific row to sample, you're now scrolling through a 40,000-line export to prove that one figure.

Level 4. Record link. The emission record points to the exact row inside the file. A specific transaction ID, a specific line number. Fast to walk through. But if the row itself was typed by someone reading a PDF, the trail dies at the row.

Level 5. Document link. The emission record points back to the supplier's original PDF, image, docket, or email. Page number. Field location. The extracted value alongside the pixel it came from. This is the level where assurance actually works.

Why level 5 is the assurance threshold

ASSA 5010 assurance engagements are walkthrough exercises. The auditor picks a sample of emission records and asks the team to demonstrate the full chain from source evidence to reported figure. Under limited assurance (which most Group 2 entities will face first under AASB S2), the sample is smaller but the walkthrough is still real. Under reasonable assurance, the sample expands and the tolerance for missing evidence shrinks.

If your lineage is level 2, the walkthrough becomes a scavenger hunt. Someone has to find the fuel export. Then find the underlying invoices the export was reconciled against. Then explain why the reconciliation matched. Then explain why 47 litres of AdBlue got booked under diesel. This is where a two-day audit week becomes a five-day audit week.

If your lineage is level 5, the walkthrough is one click. The record shows the invoice PDF, the extracted quantity, the factor version applied, the calculation, and the audit trail of who reviewed it. The walkthrough can move on.

In our view, evidence traceability is one of the control weaknesses most likely to come out of the first ASRS reporting cycle.

The audit-preparation week problem

Here is the pattern that hurts most: the audit team samples 40 emission records. The sustainability team then spends the next week finding the source documents for those 40 records. Not calculating anything new. Not revising numbers. Just chasing PDFs that were never linked to their emission entries in the first place.

Forty invoices, forty emails to procurement, forty file shares to open, forty scans to compare against the extracted quantity. That's the week. At professional rates, that is a significant amount of billable time spent on evidence retrieval alone. And most of the retrieval happens under stress, in the days before the sign-off deadline, which is when errors creep in.

The trigger for that week was a decision made 18 months earlier: to record emissions at level 2 lineage. Everything downstream is a compounding cost of that decision.

The ACCC greenwashing lens

The Australian Consumer Law does not care about your workbook. It cares about the public claim. If you say your Sydney warehouse cut Scope 2 by 22% year on year, that's a representation to consumers, and the ACCC can ask you to substantiate it.

Substantiation is a lineage exercise. Show the 2024 baseline records. Show the 2025 records. Show the invoices behind both. Show the factor library applied and the version at each period. Show the reconciliation of the meter changes if there was a tenancy switch.

If your lineage is level 2, the substantiation file is a narrative. Narratives do not survive infringement notices well. If your lineage is level 5, the substantiation file is a bundle of source documents with calculations attached. That bundle either supports the claim or it doesn't, and you can see which before you publish. Read our ACCC greenwashing penalties guide for the recent enforcement pattern.

The board-briefing lens

Boards ask variance questions. Emissions moved 8% quarter on quarter. Why. If the answer is "we're investigating" for two weeks, the CFO does not have a data problem. They have a lineage problem.

With level-5 lineage, the answer is a query. Group emissions by facility and by month. Sort the delta descending. The top of the list explains the 8%. Click through to the underlying documents to confirm. Answer the board in a paragraph.

Without it, the sustainability team goes back to the workbook, back to the fuel exports, back to the procurement team asking why the Perth site's diesel spend jumped, back to the fleet manager asking whether a new haul truck came online. Two weeks of forensics for a 90-second answer. Our writeup on variance analysis when the board asks why emissions moved walks through how the same lineage layer feeds board reporting.

What level-5 lineage actually persists

For every emission record, you want these fields captured automatically at ingestion, not typed by an analyst:

• The source document itself, stored, hashed, and retrievable
• The document type (invoice, docket, meter statement, email, CSV row) and the specific page or row the data was pulled from
• The extracted value, its unit, and the confidence score for the extraction
• The material or activity match, with the reasoning for why that match was chosen
• The emission factor applied, the library it came from (for example, NGA 2025 electricity factor for NSW), and the exact version of that factor at the time of calculation
• The calculation itself, expanded and reproducible
• The scope, category, and (for Scope 3) subcategory allocation
• The facility, business unit, and reporting period the record rolls up to
• Every state change with a timestamp and an actor: uploaded, extracted, matched, reviewed, approved, locked, restated

That last one matters more than most teams realise. Factors get restated. Baselines get restated. If the audit trail can show why a 2024 record now reads 2.7 kg CO2e instead of 2.6 (because the 2025 NGA factor edition applied to backfilled data), the restatement survives. If it can't, the restatement becomes an audit finding. We wrote about this in emission factor versioning and audit trail.

Where unit conversion sits inside lineage

Lineage isn't only about the document. It's also about every transformation between the document and the reported figure. Litres to gigajoules to tonnes CO2e is three conversions. If any of them are silent, the audit walkthrough finds it. See unit conversion errors in carbon accounting for the specific traps.

Level-5 lineage records the conversion factor at each step. Not "we converted litres to CO2e using the NGA factor". The specific fuel type (diesel B5 versus straight diesel), the specific factor (Scope 1 combustion versus Scope 3 well-to-tank), and the version. When the auditor asks whether well-to-tank was included, you can answer with a report field, not a memory.

What you get from Carbonly's ledger

With Carbonly, every emission record links to its source document for its life, and an auditor's sample comes out as an evidence pack rather than a week of digging. When we designed the ledger, level-5 lineage was the baseline, not a premium feature. Every emission record persists a link back to its source document, whether that document came in as a PDF invoice, a Word doc, a PowerPoint deck, an Excel export, a CSV feed, an RTF, a photo of a receipt, or a scanned image. Eight file formats, one lineage model.

The document arrives through the ingestion path you choose. OneDrive or SharePoint folder sync per project. Per-project email ingestion (each project gets its own inbox address). Manual drop. However it arrives, the emission record keeps its link back to that source document for its life. Our post on NGER audit preparation and evidence packs shows what this looks like on the auditor's side.

From there, the AI document engine extracts the fields. Five-tier material matching resolves the activity to the right factor. The factor version is pinned to the record at the moment of calculation. If DCCEEW updates the factor, the historical record still shows the version applied at the time; the new version applies to new records prospectively unless you choose to restate, with a recorded reason.

Several checks keep the lineage honest. Low-confidence matches go to a reviewer. Deterministic anomaly checks run every day at 6am and after each document batch and flag quantities that look out of pattern. The Data Health Agent flags gaps: scopes or months with no data, incomplete records and stale data.

The evidence pack export bundles that for an auditor sample: the source documents, the calculations with factor versions pinned, and the audit events behind each record. Not a week of digging.

For workflows where a consulting practice is running the engagement, the same lineage feeds their client-facing deliverable. Auditors ask consultants the same questions they ask internal teams. The lineage layer is what lets a practice scale from three assurance-ready clients to thirty without the file-hunting week multiplying.

The quantity-based advantage

There is a lineage argument buried inside the spend-based versus quantity-based methodology debate. A spend-based emission record traces to a general ledger line: dollars spent with a supplier, multiplied by an industry-average intensity factor. The lineage stops at the GL. There is no physical document to walk back to because the GL entry didn't require one.

A quantity-based record traces to the invoice or docket that reported the physical activity: litres of diesel, kWh of electricity, tonnes of concrete. The lineage goes all the way to the supplier's letterhead. When an auditor challenges a Scope 3 number, quantity-based lineage survives; spend-based lineage evaporates.

Many tools default to spend-based, in our view because extracting quantities from source documents at scale is hard. Construction is where that limit shows up first. Fuel dockets, concrete tickets and material invoices arrive in volumes nobody is going to key in by hand, in formats that change from supplier to supplier. That is the point where quantity-based reporting becomes practical rather than aspirational, and where lineage-per-record starts paying back the design decision.

Starting from level 2 or 3

Most Australian mid-market reporters are somewhere between level 2 and level 3 today. Getting to level 5 across the whole organisation in one quarter is not realistic. Getting there across one facility is.

Pick a single site. A warehouse, a plant, a depot, a project. Wire the source-document ingestion for that site: OneDrive or SharePoint folder sync for invoices, per-project email address for supplier PDFs. Run a month of emission records through the full lineage flow. Review the evidence pack export as if you were the auditor.

That month becomes the template. The controls it exposes (which supplier documents arrive late, which suppliers still email PDFs instead of CSVs, which fields are missing from the standard invoice) are the same controls you'll need for the other sites. Roll out to the next facility with a tuned process instead of a theoretical one.

The ASSA 5010 audit-preparation checks piece walks through the specific tests to run against that first month, and NGER audit errors and data quality covers the failure modes to watch for.

Where to start this week

Open your last quarterly emissions workbook. Pick five random rows. Ask, for each one, where the source document is. Time yourself.

If any of the five takes longer than sixty seconds to produce the original signed invoice or docket, you don't have a reporting problem yet. But you have a lineage problem, and the next assurance walkthrough will surface it.

Carbonly plans start at $100/month workspace fee with per-project pricing across Small, Medium, Large, and Enterprise tiers. If you want to see level-5 lineage running against your actual documents (not a demo dataset), send a folder of real invoices to hello@carbonly.ai and we'll show you the evidence pack that comes out the other side.