Why Verdandi extracts from legislation directly
Most compliance intelligence tools sit one or two layers away from the primary source. Verdandi is built on direct extraction from official EU publications. This article explains why that architectural decision was the only viable one for a regulatory domain where the law is still being written.
The question we kept asking about every alternative
When we were deciding how to build Verdandi, the most direct question was not which AI model to use or how to structure the retrieval pipeline. It was: where does the knowledge come from?
There are several plausible answers to that question. You could build on consultant knowledge, capturing the interpretation of experienced practitioners and making it queryable. You could build on analyst curation, having specialists read official publications and translate them into structured, digestible outputs. You could build on model training, using a language model trained on a large corpus of regulatory text to answer questions about what the law requires.
Each of these approaches has a real product behind it somewhere. Each of them has a structural problem when applied to EU sustainability regulation specifically. Understanding those problems is the clearest way to explain why Verdandi is built the way it is.
The problem with consultant knowledge
The case for building on consultant knowledge is obvious. Experienced sustainability compliance consultants have spent years developing an understanding of how EU sustainability regulation applies in practice. They know which provisions are genuinely contested, where enforcement authorities have signalled their approach, and what adequate compliance tends to look like across different business profiles and sectors.
The problem is that this knowledge exists at a point in time. EU sustainability regulation is not settled law that experienced professionals have finished interpreting. It is a body of regulation that is still being written, where the Commission issues new guidance, EFRAG publishes Q&As, consultation papers describe standards that will become binding within eighteen months, and national implementing measures add a further layer of variation.
A knowledge base built on consultant expertise reflects the state of that expertise when it was captured. If the Commission published a FAQ last month that changes how materiality assessment under CSRD is understood for companies without direct EU operations, a consultant knowledge base does not automatically update. The answer you get reflects what the consultant knew at the time they were interviewed or their content was written, not the current state of the official sources.
This is not a criticism of consultants. It is a structural observation about the mismatch between how expert knowledge is captured and how fast the regulatory environment is moving. For businesses making compliance programme decisions in 2026 against obligations that will be enforced in 2027 and 2028, the gap between what an expert knew eighteen months ago and what the official sources say today can be material.
The problem with analyst curation
Analyst curation is a step closer to the source. A specialist reads official publications as they are released, interprets them, and makes the interpretation available in the form of alerts, summaries, or structured workflow tasks. This is the model most enterprise regulatory intelligence platforms use, and it addresses some of the currency problem by keeping a human close to the source.
The gap it introduces is a different one: every curation step adds a layer of interpretation between the user and the source. The analyst has read the document and decided what it means for a compliance professional. That decision may be correct, well-reasoned, and useful. But the user receiving the curated output is relying on the analyst’s reading, not on the document itself.
In a stable regulatory domain where the interpretation is well established, that reliance may be low-risk. In EU sustainability regulation, where a Commission guidance document can shift the interpretive framework in a way that changes what adequate compliance looks like, the distinction between “what the analyst thought this meant” and “what the document says” can have real consequences.
The further problem with analyst curation is scale. EU sustainability regulation spans CSRD, CSDDD, EUDR, CBAM, and the EU Taxonomy across adopted law, Commission proposals, EFRAG and Commission guidance, consultations, and CJEU and General Court case law. Curating that corpus continuously, across all five regulatory instruments and all five source streams, is not feasible at the depth that compliance use requires. Platforms that attempt it either narrow their scope significantly, relying on their own judgment about which developments matter, or surface materials in a cursory way that gives a signal that something has happened but does not tell the user what it means for their specific situation.
The problem with training-based AI
The third alternative is the most visible one in 2026: train a language model on regulatory text and let users ask questions. The appeal is clear. The user experience is familiar, the outputs read as authoritative, and the cost of building and scaling the product is relatively low.
The structural problem is that a model trained on regulatory text knows what EU sustainability regulation said up to its training cutoff. EU sustainability regulation has continued to develop past any plausible training cutoff. More importantly, the model has no way of knowing which version of a regulation its answer is based on, whether a provision it describes has since been amended, or whether the guidance it is drawing on has since been superseded by a Commission FAQ that takes a different position.
For EU sustainability regulation, this failure mode is particularly sharp. The EUDR application dates have moved twice since the regulation was adopted. The CSDDD transposition timetable is still developing across member states. EFRAG guidance on ESRS application continues to be published. A model trained even eighteen months ago is describing a regulatory landscape that has materially changed. And because the model produces its output with equal fluency regardless of whether the underlying information is current, the user has no way to tell from the output alone when it is outdated.
The deeper problem is accountability. A compliance professional acting on an answer about their regulatory obligations needs to be able to show, if asked, what that answer is based on. “A language model said so” is not an answer that satisfies an auditor, a board, or an enforcement authority. The standard that applies is the same one that has always applied: here is the source, here is the provision, here is my reading of it.
Why direct extraction from official sources is the only viable foundation
Verdandi is built on direct extraction from official EU publications because it is the only approach that does not introduce a layer of interpretation or currency risk between the user and the source.
The system retrieves documents directly from official EU publications, not from a knowledge base built by practitioners, not from analyst summaries, and not from model training weights. Every document carries its provenance: where it came from, when it was published, and what its legal status is. The system knows what it has and what it does not have, because its knowledge is defined by what it has retrieved, not by what patterns exist elsewhere.
This matters most in the five-stream structure that Verdandi uses. Adopted law, Commission proposals, EFRAG and Commission guidance, consultation papers, and CJEU and General Court case law are not the same kind of thing, and treating them as interchangeable is an error that has real compliance consequences. Acting on a proposal as though it were binding, or missing a guidance document that changes how auditors expect a standard to be applied, are both failures that direct extraction from a properly structured source library is designed to prevent.
When the Commission published a clarification on EUDR due diligence statements, it is available in Verdandi as a guidance document with its own status, distinct from the adopted regulation it interprets. When EFRAG publishes a Q&A on materiality assessment under CSRD, it is indexed alongside the ESRS standard it addresses, not blended into a general model of what CSRD requires. A user querying Verdandi gets an answer grounded in a specific document, with the document visible and verifiable, not an averaged reading of everything that has ever been said about the topic.
What this means in practice
The practical implication is that Verdandi’s answers are checkable. A user who receives an answer to a question about CSDDD stakeholder engagement obligations can see which document the answer comes from, open that document, and read the passage themselves. The answer is not an assertion. It is a starting point for informed professional judgment, with the source attached.
This is the standard that compliance use requires, for the same reason that a compliance decision needs to be documentable. The answer a compliance professional reaches needs to be defensible, and defensibility requires a traceable path back to the authoritative source.
For businesses operating at the intersection of multiple EU sustainability instruments, that traceability is more than a procedural requirement. It is the mechanism that allows them to distinguish what is currently binding from what is proposed, what the adopted text says from how guidance has developed the interpretation, and what has been in force since adoption from what has changed since their compliance programme was last reviewed.
None of those distinctions are reliably available through a knowledge base, a curated alert service, or a trained model. They are available through a system that goes to the source directly and keeps the source visible throughout.
Verdandi monitors EU sustainability regulation continuously and delivers personalised impact analysis anchored to verified official sources. Ask questions directly against the source documents, and see what a development means for your specific business. Start for free.
For the architecture behind this approach and why it produces auditable outputs, see why deterministic RAG beats generative AI for research. For an overview of what Verdandi monitors and who it is for, see introducing Verdandi.
Subscribe for news updates.
Corlytics combines pre-legislative monitoring with enforcement analytics. Forseti is built around source-anchored interrogation of the full EU regulatory lifecycle. This article maps the difference honestly, identifies which firm types get more value from each, and explains why CELEX-cited RAG is a different product category from NLP-classified alert feeds.