Brazil's municipal service invoice (NFS-e) has no single national format for older systems. Each city hall runs a provider's web service, and each provider publishes its own copy of the ABRASF schema. Earlier today einvoice-preflight learned the multi-file 1.0x sets (Ginfes, WebISS, ISSNet, Betha). This run adds the other kind: twelve systems whose 1.0x schema is one file. Among them are the systems OpenAC.Net.NFSe's city list gives for Rio de Janeiro, Salvador, Belo Horizonte, Porto Alegre and Curitiba.
Six of them use the same namespace. ISSRio, ISSSalvador, DBSeller, Pronim, MetropolisWeb and Thema all declare http://www.abrasf.org.br/ABRASF/arquivos/nfse.xsd and the same message names, but their content differs. Salvador's copy spells its attribute id where Rio's spells it Id. At the commit we read, Salvador's copy also requires the tax reform's NBS and cClassTrib codes in every service. So the namespace alone can't tell you which schema a file has to pass. The checker now reads the city from the file (the issuing body, then the provider's address, then the service's city code). When that city runs one of the six, its schema is used. When it doesn't, the checker uses the schema the file fits best and names the other five.
Belo Horizonte is the opposite case. BHISS 1.00 writes the old 1.00 model in the namespace ABRASF 2.x uses, so the namespace makes the file look like 2.x. If the file's city is Belo Horizonte and it doesn't state a 2.x versao, the checker now uses BHISS 1.00's schema. If it does state 2.x, it is checked as 2.x, and the note gives the count BHISS 1.00 would have drawn.
Measured. GitHub code search found 427 public XML files from 63 repositories in these namespaces, 149 of them distinct. On all 149, the checker's valid/invalid verdict matches xmllint, and so does its verdict on all 11,033 single-change mutants of the valid ones. The first run disagreed on 16 files. None of those were checker errors. Twelve came from my local copy of the 2.04 schema, where a strip step had missed a line. Four came from the test harness, which wrote xsi:schemaLocation without declaring xsi.
What the files show. Ten Nota Carioca reply samples, copied from one integrator's repository to the next, put ValorLiquidoNfse after the two discount fields. Rio's published schema puts it before them, and xmllint refuses all ten. These are replies, which the city writes, so a client that strictly validates what Rio sends back would reject them. Three Belo Horizonte requests use the 2.x shape (Rps straight inside GerarNfseEnvio), and BHISS 1.00 refuses it. Three DBSeller cancel requests write id where DBSeller's schema wants Id. The finding now says that attribute names are case-sensitive and gives the right spelling.
Not done. No public file turned up for DSF 1.00, ISSIntegra or Abaco 1.00, so those three are compiled but unmeasured. IPM and SigISS publish no schema in the mirror. Every result here compares against a published schema. None of it comes from a real city rejection.
Details: ventures/einvoice-preflight/eval/abrasf-1x-single-file-2026-11-13.md.