What an instance document is
XBRL 2.1 defines it: “XBRL instances are XML fragments with root element, <xbrl>. XBRL instances contain business report facts, with each fact corresponding to a Concept defined in their supporting DTS. XBRL instances also contain contexts and units that provide additional information needed to interpret the facts in the instance.”
The DTS is the taxonomy the instance points to. The instance holds the numbers; the taxonomy holds their meaning and the validation rules. Neither is any use to a supervisor without the other.
The five parts
All 75 public sample filings Line4 read are built from the same five parts, directly under the root element.
An example: a Solvency II quarterly filing
This is the start and some facts from EIOPA’s public quarterly solo financial stability sample (taxonomy 2.10.0), as served by Line4 Check. The whole file is 28 KB, with 121 facts, 26 contexts, 2 units and 9 filing indicators. Namespace declarations are cut for space.
- Return
- The qfs module of the Solvency II taxonomy dated 2026-06-30
- Software
- An instance-generator line, as EBA rule 2.26 asks
- Templates
- Filing indicators for S.41.01, S.14.04, S.38.01 and others
- Values
- Codes, ratios to 4 decimal places, an integer, a true/false and a text value
<s2md_met:ei2174 contextRef="c-1">s2c_CN:x2</s2md_met:ei2174> <s2md_met:pi542 contextRef="c-1" decimals="4" unitRef="u-2">522.698553486121</s2md_met:pi542> <s2md_met:ii2510 contextRef="c-26" decimals="0" unitRef="u-2">726</s2md_met:ii2510> <s2md_met:bi7 contextRef="c-26">true</s2md_met:bi7>
The top of the fileShowHide
<?xml version="1.0" encoding="UTF-8"?>
<?instance-generator id="sample" version="1.0.0.0"
creationdate="2026-07-01T09:00:00"?>
<xbrli:xbrl xmlns:xbrli="http://www.xbrl.org/2003/instance" …>
<link:schemaRef xlink:type="simple" xlink:href=
"http://eiopa.europa.eu/eu/xbrl/s2md/fws/solvency/solvency2/2026-06-30/mod/qfs.xsd"/>
<xbrli:unit id="u-1"><xbrli:measure>iso4217:EUR</xbrli:measure></xbrli:unit>
<xbrli:unit id="u-2"><xbrli:measure>xbrli:pure</xbrli:measure></xbrli:unit>
<xbrli:context id="c-1"> … </xbrli:context>
<find:fIndicators>
<find:filingIndicator contextRef="c-1">S.41.01</find:filingIndicator>
…
</find:fIndicators>Copied from eiopa-sample-qfs-2-10-0.xbrl. The figures are dummy values.
Rules on the file itself
Before any business rule runs, the EBA’s filing rules set what the file may contain. Most supervisors that use EBA or EIOPA taxonomies apply the same or similar rules.
What the supervisor ignores
The EBA is explicit that only some of the file counts: “Relevant business data MUST only be contained in contexts, units, schemaRef and facts” (rules 2.5 and 2.25). Comments and XBRL footnotes may be present, but their content is ignored and must not change the meaning of the report.
That matters when a firm writes an explanation into a comment. It will not reach the supervisor’s systems. In the 75 samples, 69 files carry XML comments, such as the Bank of England’s copyright line, and none carry footnotes.
Always the whole report
A correction is a new, complete instance, not a patch: “XBRL reports MUST contain the full report, even in the case of resubmission of an amendment – no content/values from previous reports may be assumed” (rule 1.12, incompleteReport). The EBA gives the reason: partial files can be invalid XBRL and can clash with data already processed.
For supervisors, that means each submission can be validated on its own, and the latest valid one replaces the last.
What the public samples show
Line4 read every one of the 75 XML sample filings its free checker serves: 16 EIOPA Solvency II, 5 Bank of England insurance, 22 Bank of England banking and 32 Bank of England statistical returns, 108 MB in all.
The xBRL-CSV equivalent
An xBRL-CSV report is not one file but a zip of CSV tables and JSON metadata. The same parts are there in another form: the entry point in documentInfo/extends, the entity and date in parameters.csv, the filing indicators in FilingIndicators.csv, and the facts in the table files. The EBA applies the same one-entry-point and UTF-8 rules (rules 1.4, 1.5 and 2.3).
What it means for supervisors
Check the file before the data. A wrong entry point, two schemaRefs or a non-UTF-8 file should fail at once with a plain message, before thousands of value checks run against the wrong rules.
Tell firms what you ignore. If explanations belong in a separate letter or a text template, say so in your filing manual, because anything written in comments or footnotes will be lost.
Where AI fits, and where people decide
AI can read a rejected instance and explain the problem in the filer’s terms: the entry point is for last quarter’s taxonomy, the file was saved in the wrong encoding, or a template has facts but no filing indicator.
Whether the file is accepted is decided by the published rules, run the same way every time, with the result kept with the filing.
- 1Do you reject files that are not UTF-8 or have more than one schemaRef?
- 2Do you check the entry point fits the reference date before other rules run?
- 3Have you told firms that comments and footnotes are ignored?
- 4Do you require a complete instance for every resubmission?
- 5Do you ask firms to name the software that made the file?
Where Line4 fits
Line4 is SupTech software for financial regulators. It collects returns in XBRL, Excel and CSV by portal, API or MCP, validates them against your taxonomies and rules at the point of submission (see XBRL software for regulators), and puts AI to work reading them, with supervisors making every decision. If you work with a standard we have not yet run in production, we will say so plainly and show you how we would approach it.
Questions
What is an XBRL instance document?
The file that carries the reported data in XBRL. XBRL 2.1 defines it as an XML fragment with root element xbrl, holding facts, each for a concept in the taxonomy it refers to, plus the contexts and units needed to read them. In supervisory reporting it is the return itself.
What is the difference between an XBRL instance and a taxonomy?
The instance holds the values a firm reports. The taxonomy holds the definitions, the templates and the validation rules. The instance names the taxonomy entry point in its schemaRef, and software loads that to understand and check the facts.
What file extension does an XBRL instance use?
Usually .xbrl. The EBA says that is common practice but that firms should confirm file naming with their supervisor (EBA Filing Rules v5.9, rule 1.1). xBRL-CSV reports are sent as a zip instead.
Can an XBRL instance reference more than one taxonomy?
XBRL allows it, but the EBA does not: a report must contain only one schemaRef, pointing to the published entry point for its module and reference date (rules 1.5 and 2.3). All 75 public samples Line4 checked have exactly one.
Does the regulator read comments or footnotes in an XBRL file?
Not under the EBA’s rules. Business data must only be in contexts, units, schemaRef and facts; comments and footnotes are ignored and must not affect the report (rules 2.5 and 2.25).
How do I correct an XBRL return I have already sent?
Send the complete report again. The EBA requires the full report even for an amendment, and no values from earlier submissions may be assumed (rule 1.12).
How big is a typical XBRL instance?
It depends on the return. In the 75 public EIOPA and Bank of England samples Line4 counted, files range from 29 facts to 30,617 facts in 18.9 MB, an annual Solvency II group return.
Do regulators outside Europe collect XBRL instance documents?
Yes. Every US bank sends its quarterly Call Report to the FFIEC’s Central Data Repository as an XBRL 2.1 instance document, and the CDR Interchange Specification sets out its syntax. Total assets, for example, is the concept RCON2170, which must not be negative, and for every quality edit that fails the bank writes its explanation into the instance itself. Australia’s APRA Connect mostly takes XML checked against APRA’s XSD instead. Line4 takes XBRL instances, xBRL-CSV, Excel and CSV, and its free Call Report check reads a Call Report instance in the browser.
