Privacy in the Cloud   

Murray Cantor on "Next Gen Lean Software"    

• We need to try to get past the debates between the various flavors of project management and their overly prescriptive aspects.
• While the principles in Don Reinertsen's book on "second-generation lean" are applicable, there are 175 principles in his book, and that's way too many. The book is also hard to read if you don't have operations research knowledge.
• It makes no sense to talk about "six sigma" in software development. Six sigma is about the "consistency of artifacts" when you make thousands or millions of units of the same thing. When each product is unique, it doesn't apply. That's even true in, for example, the construction business: you can build identical chain stores all over the place, but the Sydney Opera or the Boston Big Dig are unique projects where Six Sigma doesn't apply either. And those examples demonstrate that it's not just software that suffers huge budget and schedule overruns. Murray doesn't believe software is as uniquely bad as some claim.
• In an Agile method, you don't necessarily know in advance what your final system will be, but you know what artifacts you need to produce: user stories, code, tests, documentation, etc. This is an artifact-centric process.
• You should let the team decide what activities are needed to produce those delivered artifacts. The method doesn't matter if the results are there.
• The lifecycle can be defined in terms of the state transitions of the artifacts.
• If you size a team to keep everyone close to 100% busy, you can apply queuing theory and demonstrate that you will have long queues of things waiting to be done. To minimize the queues (hence the schedule) you need more resources, hence less than 100% utilization and therefore higher costs. This is just math! On the other hand, the team itself should be able to reallocate work as needed to minimize the wait times.
• These principles apply as soon as you have about 8-10 people in a team.
• Outsourcing often creates problems because of the attempt to define everything in advance in a contract. If you absolutely have to do it, like some government organizations that have money but are under orders not to hire more people, then there should be people with "gray badges" in the team -- client representatives who live with the developers even though they are not part of their organization.
• Deterministic predictions of release dates and budgets are doomed to fail. We need to move the conversation to a probabilistic view: there's x% probability that the software can be delivered by date D. Prediction markets help do this, because team members know best whether something can be done. A prediction market allows them to express their understanding without directly confronting a manager who often made clear what the only acceptable answer was -- Murray calls this "asking your people to lie to you."

If you would like to have access to Murray's wisdom, contact the Cutter Consortium directly or through c�b�.

Many of us who work with foreign colleagues, clients and suppliers, social network members, etc., rely on Google Translate to understand fragments of foreign documents or to generate a (hopefully understandable) reply in a foreign language. The tool (smartphone app or Web site) can be very accurate at times, and strangely incorrect at others. It helps to understand how it works.

The translation process is not based on rules, but on matching patterns in the input text to fragments observed in millions of documents for which a human-generated translation exists, and then applying statistical methods to choose the "best" translation.

One of the sources used by Google is the huge library of United Nations documents, which are all published in six languages (Arabic, simplified Mandarin Chinese, British English, French, Russian, Spanish). As a result, direct translation between those languages is the most reliable. Translation between other languages is poorer if there aren't as many pairs of documents to examine for possible translations, or if the translation needs to happen in two phases, using English as an intermediate language.

"Google Search will be your next brain."

-- Title of an article by Steven Levy in BackChannel

(tweeted by Vince Kellen, CIO of the University of Kentucky)