Engine eviction acceleration based on strange move time frequencies

Hi all

It seems yesterday:
en.lichess.org/tournament/WVyJqKSJ

Just from the move times, there was a min of 0.5 seconds per move and a max of 3 seconds per move (except very 1st move which shows delays of up to 10 seconds).

The problem I have with this as a human being is basically:

1) I like to premove to get the 0 seconds delay
2) In scary or complex positions, I am pretty sure I start panic thinking and can spend even more than 10 seconds on a move in a bullet game

So my question is - when there is this artificial move frequency of an apparently successful bullet player, perhaps on move frequency alone they can be suspended from the tournament after say 30 games if they are also crushing everyone.

I just think in particular the max 3 seconds a move is particularly telling - it betrays the engines 'no fear', 'no confusion', 'no panic' state and rather simply just a complete moron who has made the setting of max 3 seconds per move.

It seems a shame Ferrari, had to lose to him about 8 times - I had better luck even winning 4 out of about 9 encounters. One of those is where I ran it out of time.

Anyway - perhaps move frequency heuristics can be applied to new players as a compromise.

I am glad this morning, the engine has been identified, and Ferrari becomes the winner now.

It was great fun overall - but that is my question about it - could an eviction have taken place earlier based on move time frequency ?!

If this is the case, could it be useful to bear in mind for future bullet tournaments in particular ?!

BTW i am talking about this account:

en.lichess.org/@/AjedrezClasico

which got marked this morning as using engine assistance.

Sorry more complex:

en.lichess.org/9OiRzqMp/black#8

There was a move here played at 0 seconds in the opening stage. Possibly an engine opening book was being used in the first few moves.

Okay it is really complex. The opening moves need to be discarded as well.

That kind of detection, unfortunately, leads to a scary amount of false positives for now.
On the other hand, we are willing to implement something similar as soon as both sensitivity (the positives that are correctly identified as such) and specificity (the negatives that are correctly identified as such) are both acceptable.
Until then, we shall not trust an automatic detection and can, at best, use those parameters to help human moderation.

FBD

Can anyone please explain what it means to use computer assistance? Isn't impossible to win a bullet game with c.a?

I've lost bullet games to people who I felt were definitely using computer assistance. They would consistently find difficult tactical resources within 1 sec in complex positions and yet have a rating of 1600. Their move times were suspicious. And the average centipawn loss would show 0 blunders, 1-2 mistake and a few inaccuracies. These observations led me to believe 1. they were occasionally using an engine to maintain their ratings at a certain level. 2. The engine wasn't at full strength eg not the latest or strength turned down. 3. They had figured out a way to programmatically hook an engine directly into lichess.

You can program a script to auto play the moves
then make it so it makes a move in 3 or fewer seconds. these may not be perfect moves however almost everyone will make a mistake in a bullet game which will lead to a win if you dont, which is what this engine then does

@3



pgn of that game with time info, elapsed_time/remaining_time format as move comment.

[Event "Monthly Bullet Arena"]
[Site "lichess.org/9OiRzqMp"]
[Date "2016.06.26"]
[Round "?"]
[White "ferrari9182"]
[Black "AjedrezClasico"]
[Result "0-1"]
[WhiteElo "2461"]
[BlackElo "2520"]
[PlyCount "32"]
[Variant "Standard"]
[TimeControl "60+0"]
[ECO "C25"]
[Opening "Vienna Game: Vienna Gambit, Hamppe-Allgaier Gambit"]
[Termination "Normal"]
[Annotator "lichess.org"]

1.e4 {2.0s/58.0s} e5 {2.0s/58.0s} 2.Nc3 {0.0s/58.0s} Nc6 {0.5s/57.5s} 3.f4 {1.0s/57.0s} exf4 {1.0s/56.5s} 4.Nf3 {0.0s/57.0s} g5 {0.1s/56.4s} 5.h4 {0.0s/57.0s} g4 {0.5s/55.9s} 6.Ng5 {0.0s/57.0s} h6 {1.0s/54.9s} 7.Nxf7 {1.0s/56.0s} Kxf7 {1.0s/53.9s} 8.Bc4+ {0.0s/56.0s} d5 {1.5s/52.4s} 9.Nxd5 {1.0s/55.0s} Na5 {1.5s/50.9s} 10.Bb3 {10.0s/45.0s} Nxb3 {1.0s/49.9s} 11.axb3 {0.0s/45.0s} Nf6 {3.0s/46.9s} 12.Nxf4 {2.0s/43.0s} Nxe4 {1.0s/45.9s} 13.Qe2 {3.0s/40.0s} Qe8 {1.5s/44.4s} 14.O-O {2.0s/38.0s} Kg7 {2.0s/42.4s} 15.d3 {8.0s/30.0s} Bc5+ {1.5s/40.9s} 16.Be3 {4.0s/26.0s} Ng3 {2.0s/38.9s} 0-1

So..., are you saying that the evidence of cheating is based in time-controlls only?

@6

That [3] is definitely a more serious one. Using an engine at 1+0 is just crazy, someone must be using a system called "computer playing without human assistance".