[Bob Ashlock]

Mr. P, glad to hear you found the problem. That early coil made for traditional Kettering inductive discharge does not play well with the Bosch CDI unit. It produces a weak yellow-appearing spark. Way too much inductance.. somewhere around 7 mH if I remember correctly!

Some things to share from my own experience:

(1) I have never seen an original black steel-can German CDI transformer that has randomly failed in normal operation. (They can be damaged by being connected as a conventional ignition coil.)
(2) Some aluminum-can replacements often have a short life and fail by arcing internally. Many of the shops I do business with report such failures, and I have some dead examples here. They draw current, but nothing comes out the high voltage tower. I was told by one shop that there was a recall. Unfortunately, date codes for the 'bad' ones are not known.
(3) Unpainted aluminum-can Brazil replacements primary inductance is around 400 uH. The original German ones are around 200 uH. This difference is not significant as far as normal operation goes.
(4) Recently one shop reports they are routinely buying the replacements again with no more problems.. newer ones with the problem apparently corrected. But without date code info it's hard to nail it down. These replacements are available from a wide variety of distributors and who knows whether they have good ones or bad ones on the shelf?
(5) Now here's another twist... recently I received a couple of CDI units to rebuild and the customer sent in the coils with them to check. At first glance, the coils appeared to be black originals, but then I realized they were imposters .... aluminum can ones that had been painted black, complete with proper sticker. He bought them that way! So, for a quick check for black imposters use a magnet and see if the can is aluminum. The original German ones are all in steel cans.
(6) So far, aluminum can ones always show the same stamped date code "908", but followed by an additional re-stamped "real" date code. For example, three I have here show "647", "653" and "988". (This was also discussed in the excellent Gerberding writeup.)