Consider an organisation that needs a sufficiently large starship fleet
(by number of hulls) that it decides to buy and/or build sufficient
shipyard capacity to do the build in-house over a number of years.

For some concrete discussion numbers, let's say 4k vessels of the same
class at a base price of GCr 1 (or milieu equivalent) apiece.  For this
discussion, mission, commercial vs military vs paramilitary, basing, etc
are irrelevant.  Prototypes have been built and issues rumbled in
testing have been resolved.  This is full-scale production.

Will hull #4000 cost the same to build as hull #1?  More?  Less?

Keep in mind, the _reason_ for vertical integration was to capture any
benefits from large-scale production.

 From my comparatively-untutored viewpoint, it would seem the answer to
the above question would be "less".  But how much?

Possibly out of my own ignorance, I'd lean towards the reasonably well
known "learning/experience curve" (T.P. Wright, 1936, etc - see
https://web.archive.org/web/20120830021941/http://cost.jsc.nasa.gov/learn.html
for an example + calculator)) - namely, the observation that, ceteris
fnordibus, unit prices fall by a certain, compounded, fraction with each
doubling of overall volume.  For instance, with a 20% learning curve,
the 4th unit of a run (2 doublings beyond the first) would cost 0.80 *
0.80 = 64% of the first unit's cost. Likewise, under a 10% curve, the
4th unit would cost 0.9 * 0.9 = 81% of the first unit's cost.

Obvious question is obvious - assuming a learning-curve model is
least-worst, what approximate learning curve rate would apply to serial
starship production?

Does progress down such a curve stall out at some level?  What things
would act to reset curve progress?

Second question - if this is a less-than-useful model, what would you
suggest as a more useful model?

--