[gta02-core] gta02-core prototypes: distribution and finances

[Werner Almesberger]

These are my ideas for how we can split the gta02-core prototypes
we'll make. There are two issues: cost and risk. The risk is that
we may end up with anything between a glorious pile of junk to
perfect devices, depending on design and manufacturing performance.
So we need a reasonably fair solution for sharing that risk.

First of all, here's roughly what all this will cost:

- the only sponsorship we have is Openmoko donating the components.
  PCBs and SMT cost, any components we add, import fees, etc., will
  have to come out of our own pockets.

  (If we can find a way to cut costs, that will be great, but for
  now we have to assume that we'll be on our own.)

- my rough estimate is that, if we make 20 boards, each board will
  cost about EUR 200 for the PCB, EUR 200 for SMT, and EUR 100 for
  miscelleaneous costs (components, etc.). If we make fewer than
  20 boards, the total cost will decrease but the per-board cost
  will increase.

- so the final cost per device should be something between EUR 500
  and EUR 1000. To this, add shipping plus any taxes your customs
  will claim.

- I would make all the payments for the production costs, and then
  ship the devices against reimbursement for the respective share
  of the cost. (That would be by SWIFT transfer.)

Since the number of component sets may be smaller than the number
of devices people want, I was thinking of the following rationing
scheme:

- I'll start with a show of hands of who wants to buy a prototype.
  That's just indicative so that I can plan ahead, not a firm
  commitment.

- active contributors to the gta02-core project get priority and
  can "pre-order" up to two devices each.

- after that, anyone else's "pre-orders" will be considered, again
  up to two devices each. If we exceed the 20 units, the first
  pre-orders by time of arrival get registered, and all the rest
  goes into a queue.

- if we have less than 20 pre-orders, the rest will get allocated
  to those who want more than 2 units, by repeatedly applying the
  above rules while incrementing the per-person quota by one, until
  we reach 20 or run out of pre-orders.

- I'll use this information to order the GE865 modules (which have
  a bit of a lead time), and to obtain a more precise cost estimate.

- once we're approaching the point where we'll order the PCBs and
  I'll have a better cost estimate, I'll ask for confirmations. If
  you've pre-ordered but have changed your mind, you can drop all or
  part of your pre-order. If you haven't pre-ordered but now want a
  prototype, or you want to increase your order, your extension goes
  to the back of the queue and I'll apply the above rules on the
  available devices.

  You can of course also communicate any changes before the
  confirmation deadline.

- once the orders have been confirmed, I'll announce the results and
  you're expected to stick to your decision. I'll accept payments as
  soon as the various orders have been placed and confirmed, i.e.,
  once I'm confident things will run to completion.

When the first board comes out of SMT, we'll proceed as follows:

- try to make them it do something, such as draw power and not smoke.
  Maybe even talk to JTAG. In general, test as much as possible to
  verify that there are no SMT errors, such a badly soldered parts,
  misplaced or misoriented parts, or wrong parts. The emphasis here
  is on "possible", not on "much". The SMT line can't wait forever
  while we're performing a meticulous autopsy.

- if the first board looks really bad, it goes to the scrap pile. If
  it looks so-so, it goes to the "problem" pile. If it miraculously
  works great, we put it on the "we've defeated Murphy" pile. Then
  we'll make any necessary adjustments and try the next one.

- repeat until things look good or we run out of boards. Then SMT the
  rest - if any - quickly and do the triage as above. (Even when the
  SMT process looks good, there may still be flaws. So once in a
  while, there will be a bad board.)

In the end, we'll have up to three piles: scrap, problem, and good.
Note that this classification is only the result of an initial
inspection. A "good" board may have hidden serious flaws and a "scrap"
board may be perfect after a small rework. But I think it's at least
indicative enough that we can and should use this information for the
further allocation.

The allocation would then go as follows:

- each order of two or more boards gets a board from the best pile
  ("good" if there are any, "problem" of none were good, let's not
  dwell on the third possibility.)

- if there are more candidates for a board from the best pile than
  boards, a lottery deemed fair by yours truly will decide.

- if there are boards left in the best pile, one goes to each of
  the remaining orders. Again, with lottery if necessary.

- the remaining boards are distributed to the unfulfilled orders
  randomly. If we have more boards than orders, we'll keep the worst
  ones back.

- once the allocation is complete, the units can be shipped, along
  with information on how the testing went.

Does this sound reasonable ?

- Werner