I just noticed that ICU 2.6 replaces the previous 943_P14A table with one with a lower year in its name, but which has later dates internally, and differs only by now agreeing with Microsoft on the 5 or 6 codepage characters that previously mapped to different Unicode codepoints.
<<ibm-943_P14A-2000-1999.ucm.txt>>
It sounds like problems with these characters in Unicode software against Japanese codepage databases have been surfacing across the industry. We've seen them at Siebel, and Oracle has also seen problems and introduced the JA16SJISTILDE charset in Oracle 9.2 which makes their one problem 2U:1C character Wave Dash conform to Microsoft.
I'd like to find any available information about the change in 943_P14A. We'd be particularly interested in plans for DB2 to offer this charset.