The 100–500 scale is a good common language and a bad contractual clause. This essay gathers what the literature and the industry have already documented — and the reason ISO 19650 replaced it with the level of information need (EN 17412-1): granularity per element and purpose, and documentation with a place in the contract. A natural continuation of “what is LOD”.
In one sentence: LOD condenses the information need of an element into a single number — and that is exactly what fails: the number does not say for which use the information is needed, it is not validated by machine and it has no place for documentation. The European standard's answer is called LOIN (Level of Information Need, EN 17412-1, adopted by ISO 19650): a specification per element and per purpose, in three components — geometry, alphanumeric, documentation.
Two lineages set out from the same problem and arrived at different places: the American one pushed the scale (AIA → BIMForum); the European one abandoned it (PAS 1192-2 → ISO 19650).
“In contrast to the LOD specification proposed by BIMforum and other industry guidance, the LOIN standard deliberately refrains from defining explicit levels.” Tomczak, v. Berlo, Krijnen, Borrmann & Bolpagni (2022), DOI 10.1088/1755-1315/1101/9/092024 — the European standard refuses explicit levels on purpose
“LOD 300” applies to an element, but information needs do not live per element: they live per element × purpose. The same door needs fine geometry for coordination, complete properties for asset management, and almost nothing for a massing study. The scale forces choosing an average number — and an average number means over-modelling in some uses and under-information in others. The literature is clear:
“These different levels are not always compliant, they are not yet standardized, and they are subject to interpretation. […] It does not allows to identify relevant information according to product requirements and consequently to BIM uses.” Tolmer, Castaing, Diab & Morand (2017), Visualization in Engineering 5:21
There is also a question the scale hides entirely: which objects exist in the model, at all — is the facade one element or wall + insulation + cladding? Tolmer et al. call it abstraction and propose making it explicit (borrowing from the CityGML lineage, Biljecki et al. 2014); LOIN answers it in practice by specifying per object and purpose.
A “LOD 300” requirement can be checked as a label — is 300 written there? — but not as a state:
“It can be verified whether the value is 200, 300 or 400, but it cannot be automatically validated if the state would not change since the computer is only aware of the current location.” Tomczak et al. (2022), same review — LOD is a semantic label, not an acceptance criterion that can be automated
It is the reason buildingSMART created the IDS (Information Delivery Specification): requirements in machine-readable form — “every IfcDoor must have property X with value Y” — validatable against IFC without opinion. A LOD will never be that; a LOIN with explicit properties, will.
The 100–500 scale describes the model; but much of what is exchanged in a project is not model — it is drawings, calculation reports, certificates, manuals. In LOD, documentation has no place in the contract: it is either ignored or appended outside the logic of the levels. In LOIN it is one of the three components, assessed per element and purpose: the fire door for asset management needs a certificate and a maintenance manual — and that is written into the requirement, beside the geometry and the properties. It is the difference between scaling the documentation with the use or improvising it at the margin.
LOD is not useless: it is the lingua franca that a client and a supplier recognise in a kick-off meeting, and BIMForum keeps it up to date. The mistake is promoting it to a clause. A serious tender document does not ask “which LOD?”; it asks, per element and per purpose, which geometry, which properties and which documentation — and accepts that the answer be granular: it requires complete documentation where there is asset management, and symbolic geometry where that suffices. The 100–500 scale does not even have anywhere to write that. LOIN does — it was designed for it: EN 17412-1 describes geometry in five aspects (detail, dimensionality, location, appearance, parametric behaviour), the alphanumeric as properties, and documentation as a component in its own right. And with buildingSMART's IDS, what LOIN specifies can be checked by machine against IFC.
The figure already published here says it all: in 1,059 documents from 705 architecture competitions (2010–2026), read in full, zero cite LOD (meta-concurso [measured]). The Portuguese market never procured with LOD — and the NP EN ISO 19650 (published in May 2025) already speak of level of information need, not of LOD. When the vocabulary reaches the tender documents — the 2027 pilots and the 2030 obligation will bring it — it can arrive through the standards, not through the import of an American habit already superseded in Europe. The EIR text this collective publishes in the meta-concurso follows that line: LOIN per exchange, no generic LODs [measured].
One number per element does not say for which purpose the information is needed, is not validated by machine and has no place for documentation. It works for conversing; it fails as a specification.
No. It remains the common language the market recognises (the BIMForum spec remains up to date, 2025 edition). The reasonable position: LOD to converse, LOIN to procure, IDS to verify.
Level of information need (EN 17412-1): a specification per element × purpose with three components — geometry (5 aspects), alphanumeric and documentation. The novelty is granularity: complete documentation and symbolic geometry can be required for the same element, as the use demands.
Per element×purpose pair, in three columns: geometry required, properties required, documentation required. More verbose than “LOD 300”, but enforceable, measurable and without over-modelling out of fear.
With neither, for now (verification as of 2026-09-03): 0/1059 competition documents cite LOD. The NP EN ISO 19650 (May 2025) already speak LOIN — when the vocabulary arrives, it arrives settled in favour of LOIN.
The BEP99 audits BIM execution plans clause by clause and says what is missing — including where a tender document asks for generic LOD instead of an information need per use. It runs in the browser.
Open BEP99The whole corpus is published in the open: 1301 works — 503 harvested from OpenAlex on
03/09/2026 with 21 registered queries, plus 798 that the updates of 10/09
brought with the evolved queries grown from the corpus itself (see below) —, with DOI, year, venue,
citations, open access, the queries that caught each work and the thematic classification of the
lists above — file lod-corpus.csv (at
xyzbim.eu/pesquisa/lod-corpus.csv, machine-readable link in the Dataset markup of this
page).
Two different numbers live on this page and they should not be swapped: 1301 is our
corpus (the works the queries returned); 299 is the sample of publications
analysed by Abualdenien & Borrmann (2022) in their study — their number, cited
as such. The 1301 works are today all classified: 80 in list A (critique
and conceptual treatment of LOD), 180 in B (CityGML lineage), 162 in C
(institutional successors — LOIN, IDS, MVD, PDT, automatic verification of requirements),
8 in D (the Portuguese line) and 871 marked ∅ — screened and outside the
lists. The queries are broad by design — recall before precision — and the corpus contains
acknowledged false positives (316 labelled as such, from the genetics of LOD scores to
effort modelling across all the sciences and to the environmental impact assessment of everything); whoever wants the clean list filters by the
classification.
Two layers, with the precedence declared in the file. The first is the review of the A–D lists
on this page: 15 works read closely (three in full text), classified by hand.
The second, added on 10/09/2026, is the systematic screening of the remaining 1286: each
abstract read (1009 works) or, when OpenAlex has no abstract, the title (277), against the
published criteria of the lists — work assisted by language models, with the
confidence and the reading basis recorded per line in the columns classificacao_entrada,
leitura_registada and classificacao_origem. The review's labels
always win; the screening is abstract reading, not full text — the A and B candidates of
high confidence are the reading list for the continuation of this review. The re-screening also corrected
the internal audit itself: nine entries in the lists (Biljecki 2013, 2016 and the 2017 thesis, Boersma 2019, Kumar & Labetski 2019, the 2013 predecessor by Tolmer et al.,
Oliveira et al. 2022, Osman et al. 2021, Pereira & Azenha 2026) were already in the corpus
since the harvest — invisible for not having a DOI recorded in the review — and a tenth,
Bolpagni & Ciribini 2016, only entered in the update of 10/09, caught by one of the
evolved queries written for the purpose. The EN 17412-1/ISO 7817-1 standards, assumed to be outside
OpenAlex, are indexed.
Licence: OpenAlex metadata (CC0); the classification and screening layer is our work, published under CC BY 4.0 with the attribution “XYZ BIM, corpus LOD, 2026-09”. The suggested citation accompanies the CSV.