EBA validation rules, checked in your browser

An EBA validator checks a supervisory filing against the validation rules the European Banking Authority publishes with its reporting framework, such as a total that must equal the sum of its parts. Line4 Check does this for xBRL-CSV and XBRL-XML filings, in your web browser, and is free to use. EBA reporting is the first framework in Line4 Check, a regulatory filing checker where you choose your country, then your regulator and return.

Your national authority, the one that supervises your firm. Every EU Member State, and Iceland, Liechtenstein and Norway, uses the EBA reporting framework, and each national authority collects the returns of the firms it supervises. In a few countries one body supervises and another collects the returns, and investment firms can have a different authority from banks. The Line4 Check page lists the authority for each country, with sources.

No. Your file never leaves your computer. Your browser downloads the Line4 engine (WebAssembly) and EBA's rules from line4.ai, then opens and checks the file on your own computer. Nothing is uploaded, and nothing from the file is recorded. If you ask for the report by email, only the report you see is sent to Line4, not the file.

For each check: the return checked, the EBA framework version, the file format (xBRL-CSV or XBRL-XML), the kind of result (passed, warnings only, blocking errors, nothing could be checked, or a return not covered), the address of the page it ran on without anything after ? or #, and the time. Never the file or any part of it, its name, the entity or its LEI, the reporting date, any value or message, or which or how many checks failed, and no name, email, cookie or IP address with the record. Visits are counted by Vercel Web Analytics without cookies.

Filings for EBA reporting framework 4.2, 42 modules. Financial statements (FINREP): financial reporting (FINREP 9) and financial reporting, solo data points (FINREP data points); Capital, liquidity and large exposures (COREP): own funds and capital requirements (COREP own funds), leverage ratio (COREP LR), liquidity coverage ratio (LCR), large exposures (COREP LE), net stable funding ratio (NSFR), additional liquidity monitoring metrics (ALMM) and market risk (FRTB); Funding, interest rate risk and other bank returns: asset encumbrance (AE), funding plans (FP), interest rate risk in the banking book (IRRBB), environmental, social and governance risks (ESG), global systemically important institution indicators (G-SII), financial conglomerates (FC) and intermediate EU parent undertakings (IPU); Resolution, MREL and TLAC: resolution planning: organisation and liabilities (RESOL1), resolution planning: critical functions and services (RESOL2), loss-absorbing capacity (MREL and TLAC), decisions on loss-absorbing capacity (MREL decisions) and bail-in impracticability notifications (Notifications); Investment firms (IFR): investment firms, class 2 (IFR class 2), investment firms, class 3 (IFR class 3), investment firm groups: group capital test (IFR group capital test) and investment firms: threshold monitoring (IFR threshold monitoring); Remuneration: remuneration benchmarking, credit institutions (REM benchmarking), remuneration benchmarking, investment firms (REM benchmarking (IF)), diversity benchmarking (REM diversity), gender pay gap, credit institutions (REM gender pay gap), gender pay gap, investment firms (REM gender pay gap (IF)), high earners, credit institutions (REM high earners), high earners, investment firms (REM high earners (IF)), approved higher bonus ratios, by country (REM higher ratios) and approved higher bonus ratios, by institution (REM higher ratios); Supervisory benchmarking of internal models (SBP): benchmarking of credit risk models (SBP credit risk), benchmarking of IFRS 9 models (SBP IFRS 9), benchmarking of market risk models (SBP market risk) and benchmarking of initial market valuation (SBP IMV); Payment fraud and crypto-assets (PSD2 and MiCA): payment fraud (PSD2 fraud reporting), SEPA instant payments (SEPA_IPR) (SEPA instant payments), asset-referenced and e-money tokens (MiCA) and own funds of token issuers (MiCA own funds). Each can be an xBRL-CSV report package or an XBRL-XML instance: drop the package .zip (or all the CSV files from its reports folder, including the filing indicators and parameters files, FilingIndicators.csv and parameters.csv), or the .xbrl file (or a .zip holding just that file). Both formats run the same checks. EBA itself accepts only xBRL-CSV for reference dates from March 2026; XBRL-XML is the older format. EBA's Pillar 3 disclosure modules have no active checks (EBA has switched them all off), so there is nothing to run there. Instant payments (SEPA IPR) are checked too: EBA switched their 248 checks back on from 17 March 2026. Filings for the earlier releases 4.0 and 4.1 can be checked too, against their own release's rules: EBA 4.0 (COREP own funds, COREP leverage ratio, Investment firms (class 2) and MiCA token reporting (MICA_ITS)) and EBA 4.1 (ESG risks, MiCA token reporting and MiCA own funds). So can filings for the later releases, from the first reference dates EBA gives them: EBA 4.2.1 (FINREP solo data points); EBA 4.3 (Third-country branches: core reporting, Third-country branches: head undertaking, Third-country branches: liquidity and AMLA risk assessment); EBA 4.4 phase 1 (Resolution reporting (RESOL1), Resolution reporting (RESOL2), MREL decisions, FINREP under IFRS 18, FINREP under IFRS 18, solo data points, AMLA eligibility, AMLA risk assessment and Third-country branches: liquidity).

Yes. Line4 Check runs all 2,404 of EBA's active NSFR checks, across the NSFR templates (C 80.00 to C 84.00), in your browser, on an xBRL-CSV package or an XBRL-XML instance. By Line4's count every one of them is a warning in EBA's 4.2 package: a failure does not block the return, but your supervisor can ask you to explain it. Try EBA's own NSFR sample from the list above.

Yes, all 3,601 of EBA's active IRRBB checks in reporting framework 4.2, across the IRRBB templates (J 01.00 to J 11.02): the worst-case EVE and NII, the breakdowns of sensitivity, repricing cash flows and parameters. None of them blocks a filing in EBA's 4.2 package; all are warnings.

Yes: 2,500 of EBA's 2,527 active resolution, MREL and TLAC checks (98%) across five modules, the two resolution planning reports, on organisation and liabilities and on critical functions and services (RESOL1 and RESOL2, templates Z 01.01 to Z 17.00 and Z 07.01.1 to Z 09.04), MREL and TLAC (M 01.00 to M 07.00), MREL decisions and bail-in impracticability notifications. The checks not run yet are mostly checks on text cells of open tables. It runs EBA's rules, not the Single Resolution Board's own data quality checks.

Yes. The ALMM maturity ladder and the other additional liquidity monitoring metrics (C 66.01 to C 71.00): all 11,864 of EBA's active ALMM checks. Asset encumbrance (F 32.01 to F 36.02): all 2,114 of EBA's active asset encumbrance checks. COREP large exposures (C 26.00 to C 29.00): 89 of EBA's 89 active checks.

Yes: all 2,399 of EBA's active PSD2 fraud reporting checks, across the fraud reporting templates (Y 01.01 to Y 08.01). This is the file a national authority sends the EBA each half-year, adding up the fraud data of its payment service providers under Article 96(6) PSD2 and the EBA Guidelines on fraud reporting (EBA/GL/2018/05). By Line4's count one EBA rule can block it, the one that says no credit transfer total may be negative; the rest are warnings.

Yes. Funding plans (P 01.01 to P 05.00), which the largest banks in each Member State report by 15 March under the EBA’s funding plan guidelines (EBA/GL/2019/05): all 1,358 of EBA's active funding plan checks. By Line4's count all of them are warnings in EBA's 4.2 package.

Yes: 306 of EBA's 311 active benchmarking checks (98%) across the four benchmarking modules of reporting framework 4.2, credit risk (C 101.00 to C 105.03), IFRS 9 (C 111.00 to C 118.00), market risk (C 107.01 to C 120.06) and initial market valuation (C 106.00 to C 106.02).

The rules of the release each file was built for, read from the file itself: its report.json entry point (…/fws/tcb/4.3/mod/tcb_core.json) or its XBRL-XML schemaRef. Never two releases on one file. Line4 Check then compares the file's reference date (refPeriod in parameters.csv) with the dates EBA applies that version of the module to, from EBA's validation rules list of 12 October 2026. If the date is outside them, the report says so and names the version EBA applies on that date: EBA would refuse the file (filing rule 1.5, the entry point must suit the module and the reference date). A file that names no entry point, such as loose CSV tables, is checked against the version EBA applies on its reference date, or 4.2. EBA's releases are modular: each one re-publishes only some modules, and every other module keeps its earlier version. So a FINREP filing for 31 December 2026 runs FINREP's 4.2 rules, although EBA calls that period framework 4.3, while a resolution filing (RESOL1) for the same date runs 4.4's. For 4.2, the rules EBA publishes with the 4.2 taxonomy (42 modules). For 4.2.1, 4.3 and 4.4 phase 1: EBA 4.2.1 (FINREP solo data points); EBA 4.3 (Third-country branches: core reporting, Third-country branches: head undertaking, Third-country branches: liquidity and AMLA risk assessment); EBA 4.4 phase 1 (Resolution reporting (RESOL1), Resolution reporting (RESOL2), MREL decisions, FINREP under IFRS 18, FINREP under IFRS 18, solo data points, AMLA eligibility, AMLA risk assessment and Third-country branches: liquidity), 25,454 of their 25,496 active checks. For 4.0 and 4.1: EBA 4.0 (COREP own funds, COREP leverage ratio, Investment firms (class 2) and MiCA token reporting (MICA_ITS)) and EBA 4.1 (ESG risks, MiCA token reporting and MiCA own funds), 43,599 of their 43,961 active checks; filings of FINREP, LCR and the other returns at that time used their earlier versions, which are not covered. Always with EBA's own severity for each rule (blocking or not) and without the rules EBA has switched off, as its list of validation rules of 12 October 2026 stands: for 4.2, those in its January 2026 severity package and 357 more rule codes in the 42 modules that it switched off in its quarterly updates of 17 March, 10 June and 14 September 2026. EBA asks authorities not to validate filings against rules it has switched off. A file that names a release Line4 Check has no rules for, such as one before 4.0, is not checked, and the page says so.

121,859 of the 122,161 checks EBA has active in these 42 modules (99%), every one of them in 29 modules; the list above gives each module. EBA writes many rules out once per row or column, and each written-out check counts once; checks EBA has switched off (by its list of 12 October 2026) are not counted. The few not run yet are mostly COREP checks that compare one sheet of an open table with another (one exposure class with another, for example), sums over entities or countries whose EBA formula also counts the total itself (mostly financial conglomerates; EBA has switched off the FINREP ones), checks whose EBA formula also takes in cells of other templates (mostly in the 4.0 and 4.1 releases), and a handful that add up a cell that may be empty or divide by a cell. Checks on text and coded cells of open tables, yes or no answers, dates and text patterns are run. Three FINREP checks are left out on purpose: their EBA formula also takes in cells of other templates, where the reference processor can fail a correct filing. 9 more are left out until a test can show them failing: on EBA's samples, no single change to a value makes their result change.

No. It is a free check to catch problems before you submit. Your competent authority's own validation decides whether a filing is accepted, and it also runs XBRL checks and filing rules that Line4 Check does not run, or runs only in part.

EBA marks every validation rule as blocking or non-blocking. Errors are failed blocking checks, which normally stop a filing being accepted. Warnings are failed non-blocking checks, which your authority may still ask you to explain.

Yes, 15 of them: the ones the open-source reference processor also checks on xBRL-CSV packages. They cover how the package’s report, parameters and filing indicator files are written (report.json, parameters.csv and FilingIndicators.csv), the decimals for each kind of number, the reference date, the entity, currencies, facts reported twice with different values, nil and empty values, and how each value is written (EBA XBRL Filing Rules 5.9, rules 1.4, 1.5, 1.6, 1.6.1, 1.9, 2.8, 2.10, 2.16, 2.18, 2.19 and 2.23, and 4 of EBA's extra rules for xBRL-CSV). Some are checked in part. Not checked yet: a filing indicator for every template, facts in templates marked as not reported, coded values from the wrong list and the format of key values such as dates. Where EBA's current filing rules and the reference processor differ, Line4 Check follows EBA: amounts may be accurate to ten thousands (decimals -4) for filings submitted since 1 April 2025. Files built for EBA 4.0 are checked against filing rules 5.5, the version EBA published with that release, which has no separate rules for xBRL-CSV: a decimals suffix or a special value such as #none is reported under rule 1.9 (a valid xBRL-CSV report), and a missing or empty key column is not reported. Files built for 4.1 follow version 5.7, which reads the same as 5.9 for every rule checked here; in 4.1 ESG, amounts may have decimals down to -6 (millions), as EBA's accuracy table allows for ESG.

It reads the instance in one pass, in the browser, and matches each fact to EBA's datapoint by its concept and dimensions, exactly as EBA's own xBRL-CSV metadata defines them, so the same checks run on both formats. Filing indicators, which say which templates are filed, come from the instance’s filing indicator element (find:fIndicators). It then checks 27 of EBA's filing rules for XBRL-XML (rules 1.4, 1.5, 1.6, 1.6.1, 1.6.2, 1.9, 2.1, 2.2, 2.4, 2.7, 2.9, 2.10, 2.11, 2.13, 2.14, 2.15, 2.16, 2.17, 2.18, 2.19, 2.21, 2.22, 2.23, 3.1, 3.2, 3.4 and 3.5), the ones the open-source reference processor also reports for EBA: encoding, one link to the taxonomy (schemaRef), filing indicators, contexts (one subject, one reference date, instants only, no segment), duplicate facts, precision, decimals, nil facts, units and currencies, and namespace prefixes, some in part. Facts that are no datapoint of the module are reported, not silently dropped. Duplicated contexts are not flagged: EBA's own samples repeat one, and the reference processor does not report them for EBA. As for xBRL-CSV, amounts may have decimals -4, as EBA's current filing rules allow.

EBA fills its sample filings with dummy values to show the format, not to pass the rules, so thousands of failures on the sample are expected. A real filing usually fails far fewer.

Every check Line4 Check runs on EBA’s FINREP 9 and COREP own funds samples (50,536 checks) was compared one by one with the open-source reference processor, running the same EBA rule list of 12 October 2026: all with the same result, down to the number that fail (9,449 and 9,366). So were 1,809 checks on EBA’s leverage ratio and liquidity coverage samples and 6,162 on edited copies of those four samples, with templates left unfiled, cells left out, values set exactly on the rounding boundary and conditions switched on and off. For the 37 modules added on 10 November, every check Line4 Check runs was compared with the same processor on EBA’s own sample for each module (64,557 results) and on 113 edited copies (169,219 results, with sums cleared and broken, values on the rounding boundary, signs flipped, cells left out and templates unfiled): all the same. The filing-rule checks were compared with the same processor on 166 edited packages and broken files: the same verdict and rule codes on 160. The other 6 differ for known reasons: EBA’s current rules allow decimals of -4 for amounts (two files), EBA allows a fact repeated with exactly the same value, an empty last line, and two kinds of value Line4 Check does not check yet. XBRL-XML was tested the same way: EBA’s xBRL-CSV sample of every module, converted to XBRL-XML, gives the same result as the xBRL-CSV package on every one of its checks (121,543 in all); on EBA’s own XBRL-XML sample of every module and on edited copies (filing indicators switched off, nil facts, precision instead of decimals), 78,457 checks were compared with the reference processor, all with the same result. The XBRL-XML filing rules were compared on 99 instances: the same verdict and rule codes on 93. The other 6 differ for known reasons: decimals -4 for amounts (two files), a relative link to the taxonomy (schemaRef) the processor cannot open, a one-currency check the processor also applies to the LCR tables EBA asks for in each currency, and two files where it adds messages that follow from the same error. For EBA 4.0 and 4.1, every check Line4 Check runs was compared with the same processor, on that release’s own packages, on EBA’s own sample of each module in both formats and on three edited copies of each (35 files): 217,800 results, all the same, each evaluated the same number of times. Checks whose EBA formula also takes in cells of other templates are run only where the processor was seen not to take those cells in. Their filing rules were compared on 37 edited 4.0 packages (the same verdict and rule codes on 34; the other 3: decimals -4, and repeated facts that EBA allows) and 19 XBRL-XML files (the same on 15; decimals -4, extra messages the processor adds to a nil fact, and two files too large for the processor in our memory limit). For EBA 4.2.1, 4.3 and 4.4 phase 1, every check Line4 Check runs was compared with the same processor on EBA’s own sample of each of the 13 modules (4,872 results; for FINREP solo data points 4.2.1, a sample of its checks on EBA’s xBRL-CSV and XBRL-XML files) and on an edited copy of each, with one value set so a failing check clears, or a passing one fails (4,786 results): all the same. Each was evaluated the same number of times, except where EBA’s sample reports the same fact in two templates, which the processor evaluates once per copy; on copies with each fact reported once (1,912 results), the counts are equal too. On 10 November, 570 checks were added: checks on text and coded cells of open tables, yes or no answers, dates and text patterns, and instant payments (SEPA IPR). Every added check gives the same result as the processor on EBA’s own samples (instant payments in both formats), each evaluated the same number of times, and on 40 edited copies where one fact was changed for each added check so that its result changes: 10,951 results, all the same. The browser version gives the same result as Line4’s command-line engine, check for check, on 92,947 test runs (255.6 million check results), in both formats, including all 51 of EBA’s official xBRL-CSV sample packages, its XBRL-XML samples and 919 deliberately broken or edited files. It is a prototype, not certified software.

On our test MacBook in Chrome, EBA's FINREP sample (14,148 facts, 20,441 checks) is checked in 10 to 25 milliseconds. The first check also downloads EBA's FINREP rules (593 KB) and prepares them in about 60 milliseconds; after that they are kept for the next file. Firefox takes a few times longer, still well under a second. XBRL-XML takes longer to read: EBA's own FINREP sample as XBRL-XML (13 MB) is read and checked in about a fifth of a second in Chrome (about half a second for the first XBRL-XML file, which also downloads and starts the XML reader), and a 230 MB instance in about 4 seconds, in the background so the page stays responsive; Firefox takes several times as long. Every result shows how long the check took on your own computer.

Yes. There is no account, no upload and no limit on how many filings you check (each package can be up to 300 MB). After a result you can leave your email address to get the report by email, if you want it. For premium support or bespoke validation (your own national rules, other returns or formats), contact us at team@line4.ai.