It could be a bug. Can you reboot your computer and try to reproduce your rollouts? On my two computers (running a slightly different version of XG), I get more reasonable results.
While I wouldn't necessarily consider a slightly negative MWC to be a bug (many people would though; see explanation below), the reported normalized equity for 21/14 is way off (based on 3.28% wins [no gammons] and 96.72% losses). (Note: I don't consider -1.53% MWC to be a "slightly negative" MWC.)
For 21/17 18/15, 3.30% wins (no gammons) corresponds to -0.934 normalized equity (or if 3.30% is a rounded number and it's actually 3.295% wins, -0.9341 normalized equity). Note that the cube is frozen at 1 due to the Crawford rule. The difference between -0.934 and -0.941 could be explained by statistical noise in the rollout (note that XGR+, XGR++, and your rollout are all rollouts which are adjusted at the end by the sum of the variance reduction effects).
For 21/14, 3.28% wins (no gammons) corresponds to -0.9344 normalized equity (if it's actually 3.275% wins, -0.9345 normalized equity). The difference between -0.9344 (-0.9345) and -1.169 is huge. My guess is that it's due to a bug that occurred during your rollout.
Note: If the correct normalized equity is -1.000 (0% MWC) and a rollout says normalized equity = -1.001 (i.e. a slightly negative MWC) due to noise in the rollout, I wouldn't complain, though I would advise Xavier to put some "sanity" checks into XG to prohibit negative MWCs in the final rollout result (actually, I think he already added various sanity checks to the XG code at some point in time).
On my two computers running XG 2.19.211.pre-release (using seed 1, the same seed that you used) I get these results:
           
 
       
 | | White is Player 1
score: 6
pip: 26 | 7 point match
Crawford | pip: 164
score: 0
Blue is Player 2 |
|
| XGID=-DEBcbAa----------bb-bc---:0:0:-1:43:6:0:1:7:10 |
| Blue to play 43 |
| 1. | Rollout1 | 21/17 18/15 | eq: -0.9339 |
| Player:
Opponent: | 3.30% (G:0.00% B:0.00%)
96.70% (G:95.31% B:0.45%) | Conf.: ±0.0010 (-0.9350...-0.9329) - [83.8%]
Duration: 30.5 seconds |
|
| 2. | Rollout1 | 21/14 | eq: -0.9347 (-0.0008) |
| Player:
Opponent: | 3.26% (G:0.00% B:0.00%)
96.74% (G:95.28% B:0.33%) | Conf.: ±0.0012 (-0.9359...-0.9336) - [16.2%]
Duration: 28.5 seconds |
|
| |
1 1296 Games rolled with Variance Reduction.
Moves: 3-ply, cube decisions: XG Roller
|
eXtreme Gammon Version: 2.19.211.pre-release, MET: Kazaross XG2
In the above rollouts, -0.9339 normalized equity corresponds to 0.597% MWC. -0.9447 normalized equity corresponds to 0.590% MWC.
XGR+:
| | White is Player 1
score: 6
pip: 26 | 7 point match
Crawford | pip: 164
score: 0
Blue is Player 2 |
|
| XGID=-DEBcbAa----------bb-bc---:0:0:-1:43:6:0:1:7:10 |
| Blue to play 43 |
| 1. | XG Roller+ | 21/14 | eq: -0.9366 |
| Player:
Opponent: | 3.17% (G:0.00% B:0.00%)
96.83% (G:95.30% B:0.53%) | |
|
| 2. | XG Roller+ | 21/17 18/15 | eq: -0.9379 (-0.0013) |
| Player:
Opponent: | 3.10% (G:0.00% B:0.00%)
96.90% (G:95.39% B:0.53%) | |
|
eXtreme Gammon Version: 2.19.211.pre-release, MET: Kazaross XG2
XGR++:
| | White is Player 1
score: 6
pip: 26 | 7 point match
Crawford | pip: 164
score: 0
Blue is Player 2 |
|
| XGID=-DEBcbAa----------bb-bc---:0:0:-1:43:6:0:1:7:10 |
| Blue to play 43 |
| 1. | XG Roller++ | 21/17 18/15 | eq: -0.9361 |
| Player:
Opponent: | 3.19% (G:0.00% B:0.00%)
96.81% (G:95.21% B:0.50%) | |
|
| 2. | XG Roller++ | 21/14 | eq: -0.9362 (-0.0001) |
| Player:
Opponent: | 3.19% (G:0.00% B:0.00%)
96.81% (G:95.24% B:0.48%) | |
|
eXtreme Gammon Version: 2.19.211.pre-release, MET: Kazaross XG2