I think we can agree that for any "fair" process to be implementable, Rajlich must be "cooperative" to some sense. One principal leverage the ICGA had in this regard was PR, but as others have pointed out, Rajlich had no reason to cooperate with a "lynch" process (I use the term only by quotation). In the Rule #2 context, I'm not sure (even after Reul) the ICGA really considered that an accused person would be unresponsive toward the investigation (which then induces the need for a more robust framework for dispute resolution).syzygy wrote:I do see why a procedure in 2011 on whether to revise tournament results from 2006, if there must be one at all, should be "fair".
Vexatious has a proper meaning here, namely causing or tending to cause annoyance, frustration, or worry (in the legal setting this would typically be by abuse of process), or something brought up purely to cause annoyance. Asking the second time for an appeal (after the first rather clear response) might be warranted. A third is pushing it. For Rajlich (through his representative) to go beyond that is getting ridiculous, especially with the implied threats in the later emails.Rebel wrote:You may find that, I could say the same for your sense for fair procedures to be vexatious as well.
No, in this context it is not. If this was a criminal case, you could be correct. As the EC wrote in the "French team" case, regarding the attempt to invoke 6.3 of the ECHR: this is not a criminal case; this is a case in front of the competent organ of an international sports federation concerning the violation of its rules and of the sports law.Rebel wrote:An appeal is a fundamental right.
In contrast to your psychological evaluation, seven months later, he was still saying (as quoted in the Panel report regarding Rule #2 interpretation of "code"): Strelka contains Rybka code. "Isn't that all what matters?"Rebel wrote:I see human being horrified, who just became aware his program is cloned making an emotional public outburst, what's new? Then Hristo comes along and puts some sense back into him.
Turns out he was right.Rebel wrote:Osipov... the guy who wanted to commercialize Strelka and was pissed with Vas because didn't allow him?
And you believed him?
[And if you've gone over my posts from the Rybka forum at the time, you will find that I did not take Osipov too seriously (either in 2007 or after VR's Jan 2008 statement), it was only when ZW was more vocal in Aug 2008 that I became more suspicious -- for instance, see here my mocking of Osipov's "Fruitification" with move_order()].I guess I could re-write history as: Rajlich ... the guy who wanted to go commercial with "Rybka" already in 2003, but by June 2005 (having followed Fruit development closely for 15 months, as seen by CCC posts) was pissed because his engine still sucked, and didn't even bother to see if Letouzey would allow him to commercialise Fruit (under the Rybka name)?
"And you believed him?"BB+ wrote:How long should it take to construct a defense to the main claims (of evaluation similarities)
In other words, you have no real debate about the claims of evaluation similarities, and took 4-6 weeks to come to a psychoanalysis of the investigators?Rebel wrote:It took me (while scrutinizing your and Zach's document) 4-6 weeks to see something fundamental was wrong summarized in one word: tunnel-vision.
Rajlich would have the advantage of not having to conduct an outside investigation as you did.Rebel wrote:It took me about 3 months to conclude there is barely evidence Rybka started its life as Fruit.
I've also never been sure how you came to this "conclusion" (or what "Rybka started its life as Fruit" would mean); there seem to be many things still unaddressed (or one could say "conveniently ignored"). For instance, to give but one example (as close to "code copying" as possible), you have never responded to the last example in 6.3.2 in RYBKA_FRUIT where (as I say, just a simple example, no need to worry about its chess-playing effects) it is noted that Rybka copies a redundant use of "infinite || ponder" (not to mention that these [and other] variables are allocated in the same order as Fruit, or why the factoring of UCI parsing vis-a-vis time management in Rybka 1.0 follows the idiosyncrasy of Fruit, or why the iterative deepening code [Appendix A of RYBKA_FRUIT] differences are almost nil save numerological, etc., not to mention stuff that was found after RYBKA_FRUIT was written).But it should be stressed that the ICGA debate ended up being about the (higher-level) evaluation [feature] similarities, and whether these sufficed to break Rule #2, and whereas the above (partial) enumeration of low-level quirks might have required a point-by-point rebuttal, I don't think this is particularly true in the case at hand.
I am not sure which "2 documents" you mean here. If you mean ZW's and RYBKA_FRUIT, the first was made known to him already at the beginning of February, and the second was integrated into the open letter at the beginning of March. It is reasonable to think that he might already begin to familiarise himself with these at that point (though I guess his schedule was already "very full" by then). The principal "new" material (besides the Report itself) from the investigation was EVAL_COMP.Rebel wrote:In such a complex scenario it's only reasonable the accused gets a proper time to put up a defence and refute the 2 documents point by point. Which was denied to him.
But I also don't think the situation is that complex. Previously I noted that Rajlich had three main branches of argument to make (please add others if you can):
- (1) the contention that Rybka and Fruit have similar evaluations is erroneous (possibly because the studied Rybka versions are irrelevant to this discussion)
- (2) they are similar, but it is not out of the ordinary
- (3) they have extraordinary similarities but this is not prohibited by the rules
With (2), his main evidence would seem to be either the ponderhits, or the later analysis of Riis (as critiqued by Watkins). Note that Levy had already indicated in correspondence to Rajlich that code analysis should have priority over move-matching, but even with the latter the subsequent evidence seems to be rather against VR's claims, as codified in later work of Dailey, Hair, et. al. (and published as Move similarity analysis in chess programs here). Of course, Rajlich could go through EVAL_COMP point-by-point to dispute the "extraordinary" nature of the Fruit/Rybka similarities, but I think it would be easier just to take one or two examples from that text and highlight the difficulties and problems. In the Riis piece, Rajlich doesn't really dispute that the evals are quite similar, but rather lists: material imbalances, passed pawns, and tuning (also "a lot of little things") as being the main Rybka innovations, and also mentions decorrelation of terms. [I do love the pomposity of Rybka was the first engine to understand that major pieces are [relatively] more valuable in endgames -- err, one might argue that KAISSA could be said to already have "understood" this!].
So the dispute would be about (3), namely whether Rule #2 should include high-level copying (and to what extent), and it seems that the ICGA and Rajlich here have significantly different opinions. The result of the "poll" of Panel members (most of whom are ICGA programmers, unlike other lists of programmers that have been enumerated as holding an opposing view) indicated the general opinion that indeed Rule #2 should cover Rajlich's dependence on Fruit's evaluation features (and as FL has pointed out, it's just as much about what Fruit doesn't include as what it does).
