PRIIPs · post-publication

The published-KID checklist: ten checks for documents you have already shipped

Article references are to Regulation (EU) 1286/2014 and Delegated Regulation (EU) 2017/653 as amended by 2021/2268. Frequencies are measured, not estimated: our scans cover 5,800+ published documents from 59 manufacturers in eight languages, with every counted finding verified by hand. Last updated 25 September 2026.

What a Key Information Document must contain at drafting time is not a mystery: the regulation itself is the checklist, and every production platform encodes it. The half nobody encodes is what happens afterwards. A KID that was correct on the day it was generated can still be wrong on the day it is read: the review date lapses, a sibling language version moves on without it, a distributor keeps serving last year's file. Each document is internally consistent, so no internal control catches it.

These are the ten checks we run on published estates. They are orderable: each one assumes you still trust the output of the one before it.

01What is actually served, against what you believe is published

List every product you have live, then fetch the KID for each from the place an investor would: the fund page, the document portal, the listing venue. Dead links, superseded versions and gaps between the product list and the document list all count. Everything downstream depends on this inventory being real rather than assumed.

BASISArt. 5(1) and Art. 13-14, Reg. 1286/2014: the KID is published on the manufacturer's website and provided before the retail investor is bound.

IN THE WILDIndustry bodies have assured regulators in writing that investors will always find the latest version. Our corpus is, to our knowledge, the first attempt to actually measure that assumption.

02The review date, on every document, against the twelve-month clock

Every KID carries a date of production or last revision, and the information must be reviewed at least every twelve months. A document past the interval is not automatically defective: a review that finds nothing to change requires no revision. But a shelf where old dates cluster is a shelf where no reader can tell whether the review happened at all, and that is the question a supervisor asks first.

BASISArt. 10(1), Reg. 1286/2014; Art. 15, Del. Reg. 2017/653.

IN THE WILDIn our first population study, one document in ten carried no date newer than twelve months. The oldest live document we have found was produced in 2020 and is now past six years without a visible review.

03The template era: nothing on the pre-2023 layout

Delegated Regulation 2021/2268 applies since 1 January 2023. It replaced the "reduction in yield" cost presentation with the annual cost impact, changed the scenario methodology and reworded prescribed labels. A document still carrying RIY language was produced on the abolished template and never regenerated: the signature of a back-book nobody re-ran.

BASISDel. Reg. 2021/2268, applicable from 1 January 2023.

IN THE WILDOne major European issuer serves 18.5% of sampled live products on the abolished template. Sometimes the update lands part-way: we have seen documents where the footnote carries the post-2023 wording while the table row above it still carries the old label.

04The prescribed title and section headings, in their literal wording

The template in Annex I is not a suggestion: the title and the section headings are prescribed word for word, in each language, and the language versions of the annex are the reference, not your translator. Rewritten headings are a distinct defect class from missing sections, and they concentrate where documents are produced manually or translated from another language version.

BASISArt. 8(2)-(3), Reg. 1286/2014; Annex I, Del. Reg. 2017/653.

IN THE WILDWe have found the prescribed "required by law" statement rewritten across an issuer's entire programme in two languages, and full heading sets rewritten shelf-wide: a template decision, repeated in every document it produced.

05All four performance scenarios, stress included

The performance section names four scenarios: stress, unfavourable, moderate and favourable. Check presence, not just the table shape: layouts differ, and a row can be absent while the table still looks complete. Where a scenario row is missing, it is usually missing from every document of the product line, because the gap sits in the template, not the document.

BASISAnnexes IV and V, Del. Reg. 2017/653.

IN THE WILDWe have measured shelves where one scenario row is absent from an entire product line while every other manufacturer in the same market prints it.

06Both cost tables, with their entry and exit rows

Two tables are prescribed: costs over time, and the composition of costs with its one-off, ongoing and incidental rows. Shelves lose whole tables to template choices, and individual rows, entry and exit costs above all, to layout decisions. A cost section that reads smoothly can still be structurally incomplete.

BASISArt. 8(3)(f), Reg. 1286/2014; Annex VII, Del. Reg. 2017/653.

IN THE WILDEntire shelves missing the composition-of-costs table, and entire shelves missing the entry/exit rows: in both cases every document of the manufacturer showed the same gap, because the defect lived in the template.

07The arithmetic: components against totals, impact against figures

Do the cost components sum to the stated total? Does the annual cost impact reconcile with the amounts shown? This is the defect class hardest to defend in front of a supervisor, because it requires no legal interpretation: the numbers either agree or they do not. It is also where automated checking earns its keep, one document at a time.

BASISAnnexes VI and VII, Del. Reg. 2017/653.

IN THE WILDThe ESAs found inconsistent cost tables in 25% of fund KIDs they examined. In our own corpus we have found documents whose unconditional cost components sum to more than the stated total.

08Language versions of the same product, against each other

A translated KID is the same document in another language, not a cousin. Same risk class, same costs, same holding period, same production date. Divergence has a mechanism: one language version gets regenerated and its siblings do not, so each document is correct on its own and the inconsistency exists only between them. No single-document control can see it.

BASISArt. 7, Reg. 1286/2014.

IN THE WILDOf twenty products we held in more than one language version, fourteen disagreed with themselves in at least one value: same ISIN, same worked example, a different total cost depending on the language the investor reads.

09The copies your distributors serve, against your current version

The person advising on or selling the product provides the KID, and most retail investors meet the document on a platform, not on your website. A distributor caching last year's file serves a document you have already superseded, under your name. The same applies to the data feed behind the document: the EPT your distributors consume should say what the PDF says.

BASISArt. 13-14, Reg. 1286/2014.

IN THE WILDA leading index manager's public EPT feed reported management costs of 0.19% while its current KIDs said 0.14%, consistently, in four languages. Four weeks after our note, the feed was regenerated.

10Whether a machine can read what you published

This is the forward-looking check. The Retail Investment Strategy brings a five-value dashboard and points towards a machine-readable KID; supervisors already parse published documents at scale. A PDF whose text layer is broken, words glued together, values rendered as graphics, passes every visual review and fails every automated one, including the ones your regulator runs.

BASISForward-looking: the RIS rewrite and the European Single Access Point filings from 2028. What RIS changes and when.

IN THE WILDWe have measured shelves where the text layer of every document collapses under standard extraction: the dates and headings are printed for the human eye and invisible to every parser, ours included until we taught it otherwise.

The pattern behind all ten: defects on published shelves are not random. They are template decisions, so they repeat across every document a producer generates, and they are invisible from inside the production system that made them. That is why the same platform can produce one manufacturer's clean shelf and another's defective one, and why the fix is usually one template change, not a thousand documents.

Findings referenced above are aggregate and anonymised. We notify manufacturers before publishing anything that could identify them, with a correction window; verified fixes are reported on the outcomes section of the main page. This page is not legal advice.

Run the first five checks on your own shelf, free

Send your ISIN list or fund range and we draw a sample of five published documents across your languages and vintages. Full per-document findings within 48 hours, no obligation. Methodology: the full report.

Request the free check

or write to hello@kidcheck.eu