[uva72uva72]

{comments}

---

Match points, ACBL robot individual.

I previously reported a hand where playing the ♣K vice the ♣A when declarer was known to have both cards derailed the robot defense. The attached hand shows a huge difference in result stemming from declarer's choice of apparently meaningless spot cards.

At 7 tables, declarer reached the identical contract via an identical auction, received the ♥7 lead, covered with the ♥8 from dummy and won East's ♥10 with his/her ♥Ace. Each of the 7 declarers then went after ♦s via double finesse. When the ♦10 lost to the ♦Queen, declarer won the ♠return in hand and led to the ♦Jack, losing to the ♦King. Here is where declarer's spot card play worked its magic. Those 5 declarers who thoughtlessly led the ♦4 at trick 2 went down like dogs when East returned another spade. One declarer made a promising start when he/she led the ♦8 at trick 2, but fell from grace by leading the ♦6 for the second finesse at trick 4; back came a ♠. All 6 of these declarers (myself included) finished down 1 and received the 58.9% score they deserved. One clever declarer, though, led the ♦ 8 at trick 2 and then found the Zia-esque lead of the ♦4 at trick 4, cleverly concealing ♦6. East bit, hook, line and sinker. No matter that West had, as at every table, echoed in ♦. East was bedazzled by the missing ♦6 and returned a ♣ instead of a spade, resulting in +600 NS and a richly deserved 100% score for declarer's legerdemain.

Note that playing the ♥King vice the ♥Ace at trick one does no good. One declarer not included in the group of 7 discussed above (but presumably having read my earlier posting about playing the King when you're known to have both Ace and King) did play ♥King at trick one, but then undid the effort by leading the ♦4 at trick 2 and went down like the rest of us.

[1eyedjack]

GIB run simulations to determine its next move, at each turn.

It is a bit more complicated than that after trick 2 when using advanced robots, but I believe that sims still feature.

The sim process involves generating a set of random possibilities based on (1) initial conditions at decision time and (2) a random number "seed". That set is (particularly in the early stages of play) a small subset of the total population of possibilities.

In robot tourneys the robots use the same random number seed for a specified set of initial conditions, and therefore reliably produce identical actions in those circumstances. The only exception to this is in "Instant" robot tourneys, where that remains the case for all tables other than your own.

However the initial conditions have to be precisely the same for the random generator to produce the same set of sample situations from which it derives its play.

I speculate that a club switch by East is indicated on a large-ish minority of hands. Occasionally, if the sample size is small enough, the sample might erroneously show this to be the majority action, but only on a minority of sets of initial conditions. Those initial conditions are a function of the order of play of spot cards, and indeed choice of touching cards/honours.

I don't know how difficult it would be to get the simulator to disregard such variations, but my guess is that it would be hellish difficult, not least because on some hands the order of spot cards (and more so of play of touching honours) does logically have a real impact on the indicated line of play.