[lycier]

lycier, on 2016-August-29, 16:43, said:

In the dynamic bidding sequences, Gibs do not know when to add value, when to devaluate.
On the Gib TP evaluation world, there is only addition, no substraction.
So Gibs often only make many one-sided description, this is a kind of misrepresenting description, too many rediculous story will occur.

I think this is a biggest problem.

See my Gib hand.

---

The result of this hand is 7♠S=, how about 7♠ contract ? It seems there is no problem.
However, on the contrary, here there is a serious problem on it. Now I would make a dialogue for Gibs.

- Gib N says " Hi, pd, I have 5+♥ with 11-21hcp and 12-22TPs".

- Gib S replies " Hi, pd, I respond 1♠ and promise 4+♠ with 6+TPs ".

- Gib N says " see my jump 3♥ to show 6+♥,3-♠,17-20TPs "

- Gib S replies " see my rebid-4♣ over 3N, even Gib CC says" it's a new suit--- 4+♠,rebiddable ♣,9+TPs", this is a slammish bid over 3N in fact."

- Gib N says " I see, now I would tell you that I have exact 3-card ♠ and 6+♥ with 17-20TPs".

- Gib S would think there is 20TPs in total in its hand, so Gib S would start to make addition TP evaluation : 17-20TPs + 20TPs=37-40 TPs = grand slam contract !

This is a Gib's evaluation calculation logic. So a rediculous story would occur.

Now it is time to see how rediculous story will occur.
Hand-1

---

Hand-2

---

[Stephen Tu]

This is just a repeat of a theme you introduced 2 TP threads ago. GIB leaps to slam instead of making a lower bid (RKC) to check if not off too many aces for the level it is contemplating.

If your wish is that "GIB shouldn't use TP totals to make jumps to slam in preference to checking keycards", then I completely agree with you. On these sort of auctions it should be checking keycards, and then using simulations (not TP totals) to then decide whether 7 is a good bet or not (and maybe follow up with king asks or side suit asks instead of just jumping to 7 after the initial RKC respose). Leaps to slam should be pretty low on the priority list and defined so that they rarely come up, mainly they should be defined if humans employ them and marked as absolute signoffs.

If your wish is that TP should be entirely abandoned (and replace with what??? HCP only?), then I don't agree. Distributional strength matters. But on most of these auctions, at some point simulations should kick in for it to evaluate how high it should go, not simply combining estimated TP totals. And between RKC and simulations it should be able to figure these hands out. I think one problem is you use basic bots a lot for your testing that have simulation disabled in a bunch of spots, although this is useful for finding bugs in how the book bids are defined and prioritized.

But bad rule priorities just kind of screw everything up. On this hand, on my older PC GIB on these hands it usually makes bad bid of 6nt after 1h-1s-3h by rule, but sometimes simulate 4c is better and choose that. But if force it to bid 4c, after 4s it does the right thing, bids RKC, then does the right thing afterwards. So at some point this was broken, apparently, by subsequent fixes to BBO GIB (they have made many changes since last release for commercial GIB). But it demonstrates that with proper rules (book bid 4nt over 4s, correct priority over 7s), things work fine, and that the problem is more with rule definition/priorities rather than "TP superstition".

The problem is you never state exactly what you want, other than a general whining against the use of TP.