Transportation across the Pacific is scarily cheap; it's about a thousand dollars for a container that holds about twenty tons, so five cents per shirt.

This is because it's done with a big boat, and boats are amazingly efficient; five thousand containers use about a fifty-megawatt engine for about two weeks, that's twenty kilowatt-weeks (a couple of tons) of fuel to take each container to China or back. You're adding the price of about a cup of super-cheap marine fuel per shirt per direction.

Remember that shipping wheat from Egypt to Rome was cost-effective with sailing ships two thousand years ago!

And human industry also emits significantly more SO2 than volcanoes; you don't get a Pinatubo every decade, and China alone emits two Pinatubos of SO2 annually.

The laser-tracking protocol is defined to run at 25 pulses a second; pulling them back and forward by tiny amounts, to take advantage of the electronics in the orbiter that are designed to measure tiny time differences in order to do the LIDAR altimetry, is a really nifty classic NASA hack.

But the press release did not make a good job of pointing out that NASA were working under that restriction. Obviously if you were trying to do laser communication you'd do something else; ESA have done 50Mbit/second laser communication from low-Earth orbit to geostationary and from geostationary back to Earth, with their Artemis satellite.

Yes, an SoC is a significantly bigger job than a pure CPU core. But Intel hasn't been producing pure CPU cores for a long time; an Ivy Bridge has a large GPU, a collection of video accelerators, two DDR3 controllers, a PCIe 3.0 interface, and quite a fancy power-management microcontroller. The die is less than 50% occupied by CPU cores.

Similarly, I built a two-socket Opteron machine out of ebay parts a few years ago; companies are really bad at realising the value in the components in equipment that they're getting rid of, so quad-core dual-socket Opteron processors that sold for four figures new were two for $99. A current similar effect is happening with the slower-speed Infiniband interconnects from decommissioned supercomputers; DDR Infiniband (20gbit/second and much lower latency than 10GbaseT ethernet) cards are pocket-money cheap on ebay.

I have a quad 6168 in my shed; I do very parallel number-theory jobs, and it's really very good at those.

But for my work each core offers less performance than one hyperthread of a current i7; my 48-core machine is comparable to about 20 cores of Ivy Bridge. For the cases where I'm not taking full advantage of the 6168's amazing memory and interconnect bandwidth (and I don't have enough of those to keep the machine busy), I'd be getting the same performance for half the price with five i7/3770K boxes; they'd take more space and a bit less electricity. I'm hoping the 6168 will keep working for several more years, but I can't see why its successor wouldn't be a pile of haswell+1 machines.

And how does your yield compare to an equal area being farmed competently by people working for Archers-Daniel-Midland?

I have a garden, it's probably the most expensive luxury item I own; it cost me about £40,000 on the price of the house compared to getting a similar-sized apartment with no garden in a similar area, say about £200 a month on the mortgage. It produces enough tomatoes for my purposes, and they are tasty tomatoes; but £200 a month buys from Tesco enough tomatoes for a hundred people's purposes.

If I remember rightly, India has a reasonable number of large cities; what is it that stops people with irredeemably awful prospects in Bihar from getting on a train and becoming people with irredeemably awful but less walled-in prospects in Delhi?

People with irredeemably awful prospects in Gansu have been getting on a train and turning into people with quite decent prospects in Shenzhen by the dozens of millions for the last twenty years.

If you want satellite images, you can write a cheque to DigitalGlobe without much difficulty, and spend money on training local imagery-analysts to be good at producing the information that your planners and developers need.

If you want your own satellite that you can task, you can write a cheque to Astrium who will build you one, and another cheque to get Starsem to launch it from Guyana.

You don't need to develop your own rocket to get the satellite goals; and the technology involved in actually building rockets has only one spin-off use on the ground. Admittedly, nuclear deterrence is a pretty awesome spin-off use if you've got Pakistan on your western border.

Alternatively, you've taken some smart programmers who could be working on designing better interfaces for web sites connected with the poverty-alleviation programme, and you've got them designing interfaces for the ISRO miscellaneous small item procurement internal website. You're taking people who could be working on complicated investment strategies allowing Grameen bank to do more good with its resources, and you're using their time to design low-thrust trajectories for Lunar injection. You're taking fluid-dynamicists who could be working on the feedstocks for the big dams in Kashmir, to get a few more megawatts out of the turbines with the same volume of water, and getting them to work on the manifolds for the turbines pumping high-pressure hydrazine.

The money doesn't go away, but it's no more use than if you were spending that much to train people to be chess grandmasters or marathon runners or professors of analytic number theory - experts in fields of self-referential inutility.