Why AM and Eurisko Appear to Work (1983) [pdf](aaai.org)

  EURISKO was given the rules of the game, and the constraints on the design process, and spent a great amount of time (roughly 500cpu hours on a Xerox Dolphin) managing a heuristically-guided evolution process. Fleets would fight, and the simulated battle would be analyzed (by some of EURISKO'S heuristics) to determine which design policies were winning, and--occasionally--what fortuitous circumstances could be abstracted into new design heuristics. An example of the former was when the Agility of ships gradually decreased, in favor of heavier and heavier Armor plating of the hulls. An example of the latter was when a purely defensive ship was included in an otherwise-awful fleet, and that fleet could never be fully defeated because that defensive ship, being very small, unarmored, and super agile, could not be hit by any of the weapons of the larger nearly-victorious fleet. The author culled through the runs of the program every 12 hours or so of machine time (i.e., each morning, after letting it run all night), weeding out heuristics he deemed invalid or undesirable, rewarding those he understood and liked, etc. Thus the final crediting of the win should be about 60/40% Lenat/EURISKO, though the significant point here is that neither Lenat nor EURISKO could have won alone. Most of the battles are tactically trivial, the contest being almost decided by the designs of the two fleets; that--and the thickness of the tulebooks--were the reason this appeared to be a valid domain for EUR1SKO.

  [...]

  Almost all the other entrants in the final tournament had fleets that consisted of about 20 ships, each with a huge spinal mount weapon, low armor, and fairly high agility, and a large number of secondary energy weapons (laser-type weapons). This contrasted with Eurisko's fleet, and such an enemy was rapidly decimated, at a loss of about a third of Eurisko's line ships, and no risk to its specialty ships.

  [...]

  The tournament directors were chagrined that a bizarre fleet-such as this one captured the day, and a similar fleet (though not so extreme) took second place. As a result, the rules for future years' TCS tournaments have been changed, to dramatically reduce the design singularities which EURISKO (work- ing with the author) found. 

  A second use of R7 in tile naval design task, one which also inspired a rules change, was in regard to the fuel tenders for the fleet. The constraints specified a minimum fractional tonnage which had to be held back, away from battle, in ships serving as fuel tenders. R7 caused us to consider using warships for that purpose, and indeed that proved a useful decision: whenever some front-line ships were moderately (but not totally) damaged, they traded places with the tenders in the rear lines. This maneuver was explicitly permitted in the rules, but no one had ever employed it except in desparation near the end of a nearly-stalemated battle, when little besides tenders were left intact. Due to the unintuitive and undesirable power of this design, the tournament directors altered the rules so that in 1982 and succeeding years the act of 'trading places' is not so instantaneous. The rules modifications introduced more new synergies (loopholes) than they eliminated, and one of those involved having a ship which, when damaged, fired on (and sunk) itself so as not to reduce the overall fleet agility. 

  R7: if f is an interesting function which takes a pair of A's as inputs,
      then define and study the coalesced function g(a) = ~f(a, a).

  My suspicion is that Lenat's TCS win tells us more about TCS than about EURISKO, that TCS is likely a game that's inherently vulnerable to the "find winning strategies by simulating a lots of games on a computer" meta-strategy. I've heard, for example, that battles are often tactically trivial, with the outcome of battles effectively determined by the composition of the two fleets (and fleet composition is what Lenat used EURISKO for). If that hypothesis is correct, though, it suggests it shouldn't be necessary to reimplement EURISKO specifically to get a program that's good at designing TCS fleets. If that turns out not to be the case, it would be evidence that there really is something special about EURISKO after all.