The trick will be nailing down the XML spec for this data. I'm not familiar with XBRL, but it seems like the key will be to get the spec "good enough" rather than "perfect." 

This seems like a no-brainer. We have to have XML to promote data interchange and interoperability, as well as making information accessible to all parties. Same is true for Web Services. These architectures work, are accepted, wide-spread and well known by most IT professionals. To invent something new for data transfer would just be crazy. 

While XML and Web Services are fine to a point, I am not convinced that they are ideal for recovery.gov. I'd much prefer to see recovery.gov embrace W3C Semantic web standards as explained here..

1) They require a significant degree of up-front agreement on schema that can inhibit the continual evolution of the data available. I'd far prefer data be collected and made available in a format that can freely evolve with the needs of a heterogeneous community of data sources.

2) XML-based Web Services provide fixed, limited ways to consume data. The only thing that people can do with data exposed via Web Services is what the people who created the services dreamed of. I'd much prefer an interface driven via a query technology such as SPARQL that would allow data *consumers* to choose precisely what information they are interested in.

3) XML based technologies do not easily adapt to Web-based mashups. For this purpose, I'd suggest that any technology approach chosen for data interchange and exposure should play easily in a JSON environment.

XBRL is interesting from a couple of perspectives

1) Already is use by the SEC and the FDIC for data collection

2) Enables the collection of more than just financial data; could be use to collect agreed upon metrics like # of jobs created.

As the original poster note - XBRL allow information modeling and expression of semantic meaning - supporting transparency and comparability

Getting to ARRA XML syntax would require development of the underlying semantics for ARRA, which XBRL will probably not fully satisfy. 

However, when building an ARRA terminology, XBRL could be used as a predefined set of definitions, taxonomy, concept map (DLG + Triples), and possibly thesaurus for the business relationship part of the terminology.