Timber is deeply, deeply flawed

[geebee-1015]

The supposed purpose of Timber is to prevent Ancestry customers from being mislead by a phenomenon known as "excess sharing".  This is believed to occur as the result of having multiple distant shared ancestors who have passed on enough DNA to give the appearance that the common ancestry must be more recent.  Timber "adjusts" the shared DNA to what Ancestry believes it should be to reflect the actual degree of relatedness.

However, I know for a fact that time and time again, Timber is simply wrong.  That is, rather than the adjusted cM being more indicative of the true relationship, it is less so -- that is, it makes the relationship appear more distant than it is.

Here's one example:

"GK" is a 2nd cousin who shares 265 cM with me in 15 segments, with a longest shared segment of 73 cM.  This amount of sharing is, of course, protected from adjustment by Timber because it's greater than 90.0 cM -- the limit above which Timber is prohibited from making any adjustments.

My daughter, of course, is a 2nd cousin once removed to GK.  Average sharing for this relationship is 122 cM -- which would also be safe from adjustment by Timber.  However, the range of sharing is 14-353 cM, and the amount of shared DNA with GK that my daughter inherited from me was 87 cM in 6 segments, with a longest shared segment of 26 cM.  Timber went to work on this match because it was no longer inhibited from doing so -- by 3 cM! -- and adjusted the sharing amount downward by 40 cM.  So what Ancestry reports for this match is 47 cM.  (https://dnapainter.com/tools/sharedcmv4)

Ancestry suggests that there is only a 21% chance that the relationship between GK and my daughter is that of 2nd cousin once removed.  There's about a 78% chance that this would be 3rd cousin or more distant.  But the thing is, what is the actual source of the 40 cM that Timber chopped off?  Me, the 2nd cousin.  It is clearly not "excess sharing".

I did a review of all of the matches of both my wife and myself that are "Timber-safe" for the two of us due to being greater than 90.0 cM.  Virtually all of these matches -- many of whom are 2nd cousins -- have had a downward adjustment for our daughter, with the exception of those which are above 90.0 cM for her.  In some cases, the adjustments are fairly small; but others are by 20, 30, or more cM.

[JFWinstone]

Looking through my list this seems to be mostly impacting my 2nd cousin once removed - 3rd cousin once removed matches the most, on average they are being timbered by 40cM. The smallest was 10cM and the largest 60cM but most were around the 40-45cM mark.

[geebee-1015]

So, for the first time in weeks I got a match with sharing greater than 90.0 cM -- so therefore safe from Timber's axe.  This match is on my maternal grandmother's maternal grandmother's side.  We share 103 cM in 6 segments and are predicted to be 2nd-3rd cousins; our longest shared segment is 42 cM, so the sum of the remaining 5 segments must be 61 cM.

But once again Timber "sees" excess sharing in the match between this person and my daughter.  Instead of using the unweighted amount of 62 cM, it's adjusted to only 49 cM.  This is in 2 segments, again with a longest shared segment of 42 cM.  That means the either Timber is finding excess sharing in both segments, or only in the smaller of the two segments -- which would make its adjusted size just 7 cM.  Of course this is possible, though the average size of my smaller segments would be 12.2 cM, not 7 cM.

This doesn't mean the two segments I passed on to my daughter aren't 42 cM and 7 cM, of course.  But the plain truth is that if none of my sharing with this match is "excess", then there is no reason to suppose any of my daughter's sharing is excess.  Of course, without a chromosome browser I can't tell how well my daughter's smaller segment compares with any of my smaller segments.

Ancestry, however, can tell, and I'm convinced that if they cared to look they would see lots and lots of instances in which Timber's decision to "trim" sharing is questionable at best.

EDIT:

Based on my shared matches with AS, it seems highly likely that he is a descendant of my grandmother's grandmother's parents -- that is, my 2nd great grandparents.  I found a person with the same name as my match in my mother's home town, and if it's the same person we're within about two years of being the same age.  My guess would be that this match is indeed a 3rd cousin to me, and a 3rd cousin once removed to my daughter -- not 4th-6th cousins as predicted by Ancestry on the basis of the Timberized sharing amount.

After a bit of digging, it indeed appears that AS is related to me in the way I first guessed -- we're 3rd cousins.  As a side note, AS reportedly has 1% Indigenous Americas - North, which is a common ancestry among my relatives in this part of my tree; along with 1% Indigenous Americas - Bolivia & Peru.  It's my contention that Ancestry's various Indigenous Americas categories are not nearly as separate and distinct as they think they are.  

Currently Ancestry says I have 2% Indigenous Americas - North, but other times they have labeled some or all of my IA as Indigenous Americas - Mexico.  In fact, right now they're identifying the daughter of one of my full sisters as having 1% Indigenous Americas - Mexico, with no Indigenous Americas - North; and they're identifying the daughter of another of my full sisters as having 1% Indigenous Americas - Ecuador, also with no Indigenous Americas - North.  In both cases, the probable immediate source of this ancestry is their mothers, whose ancestry is 100% the same as mine, even though we obviously only share about half of our DNA.

AS also has 1% Eastern European Roma, another ancestry appearing in this part of my tree.

[geebee-1015]

Here's a trio -- mother, son, and daughter -- that both my daughter and I share with.

With LW, the mother, I share 193 cM in 7 segments, with a longest shared segment of 54 cM.  My daughter shares 140 cM in 5 segments with LW, again with a longest shared segment of 54 cM.  So LW is a "Timber-safe" match for both of us.  LW is my 2nd cousin once removed -- or to be more precise, she's my mother's 2nd cousin.  That makes her my daughter's 2nd cousin twice removed.

LG, the daughter, and I share 158 cM in 8 segments; LG's sharing with my daughter is 112 cM in 6 segments.    So once again, LG's sharing with both of us is safe from "adjustment" by Timber.

MG, the son, shares 136 cM with me in 6 segments, with a longest shared segment of 47 cM.  But, while MG and my daughter share a significant amount at 83 cM in 4 segments, that's pre-Timber.  Since the total is a "whopping" 7 cM less than the "Timber-safe" level of 90.0 cM, Timber is free to "see" any amount as "excess sharing".  In this case, the amount chosen is 48 cM!  Of course, this is ridiculous given my sharing with MG, his sister LG, and their mother LW.  The sharing amounts are consistent for the relationships involved.

Since my daughter inherited the 83 cM she shares with MG directly from me, and he inherited it directly from his mother, it cannot suddenly be "excess sharing" -- no matter what Ancestry's algorithm may "think".  So if there appears to be any excess sharing here on the part of other matches, that is purely coincidental to our own matching.

And I'll state this once again.  Results like this are not rare, but can be found with many of my daughter's matches the instant sharing falls below 90.0 cM.  Basically, if Timber can "chop", it will -- and sometimes by large amounts even when it makes no real sense in terms of the actual relationships involved.

[geebee-1015]

One of the things I've been able to quantify now is that out of 99 matches that are "Timber safe" for either my wife or me because of being greater than 90.0 cM, 53 0f these are not Timber safe for our daughter, since they are less than 90.0 cM.  That isn't surprising in itself, since intuitively you might expect her sharing to average about half of each parent's sharing -- so for matches in which a parent shares less than 180.0 cM, she would often share less than 90.0 cM.

But, because of this threshold not being met in these cases, Timber is free to adjust the sharing amount even though any DNA deemed not to be "excess" for one parent or the other is not going to suddenly be "excess" for our daughter.  Yet in some cases, not only are there reductions, they are also very large.  

For example -- as I may have mentioned in a previous post -- there is a 2nd cousin GK with whom I share 261 cM in 15 segments, with a longest shared segment of 73 cM.  My daughter shares 87 cM with GK, none of which is at all likely to be "excess".  Despite that, Timber chops a full 40 cM off this amount, reducing it to only 47 cM.

Or, there's BH.  BH is my 1st cousin twice removed.  She's the granddaughter of my 1st cousin DS, with whom I share 1,048 cM in 24 segments, with a longest shared segment of 111 cM.  My daughter shares 489 cM with DS, who of course is her 1st cousin once removed.  She shares 81 cM with her 2nd cousin once removed, BH, but that's before reduction by Timber.  Timber chops off 35 cM of imagined excess sharing between my daughter and BH, reducing their reported sharing to only 46 cM.

In fact, of the 53 matches which were Timber-safe for one parent or the other but not our daughter, 24 had reductions of 20 cM or more.

In addition to the above, there were many instances in which neither the sharing of the connecting parent nor our daughter was "Timber safe" but the amount of Timber reduction was inconsistent.  For example, 17 cM of my wife's sharing with MP -- her 3rd cousin -- was deemed by Timber to be "excess".  But this amount grew to 28 cM in the case of our daughter, who would be a 3rd cousin once removed to MP.  Why did 11 cM more of the matching suddenly become "excess" with the addition of one more generation?  The unweighted sharing was nearly identical -- 65 cM in 5 segments for my wife, and 64 cM in 5 segments for our daughter, yet post-Timber sharing was reported as 48 cM and 36 cM respectively.

Timber was also the likely reason that 325 connecting parental matches showed sharing of less than 20 cM even though our daughter's sharing was greater.  This includes two matches in which no sharing was shown for either parent -- meaning either that these were false matches or that the connecting parent's sharing was reduction to below the 8.0 cM threshold for being reported as a match.

[geebee-1015]

Here's the kind of thing I see with great regularity:

• CW versus my wife.  29 cM in a single segment -- no adjustment by Timber
  
• CW versus my daughter.  14 cM in a single segment, after adjustment of -14 cM by Timber, or half the total.  So Timber "decided" that in the daughter, 14 cM of the inherited segment is "excess", even though it isn't in the mother.
  

What's especially annoying here is that 14 cM doesn't show up as a shared match, but 28 cM does.  Yet in terms of the predicted relationship, there's actually very little difference.  In either case, the top prediction is for 3rd cousin 1x removed.  So here Timber seems to serve no real purpose, as is actually pretty typical.

[geebee-1015]

Here are a few more examples of matches that are "Timber-safe" for me (due to being at least 90.0 cM unweighted), but not for my daughter (due to being less than 90.0 cM unweighted).  For her, Timber reduced each match by at least 30 cM.

The first three of these have in common that we all descend from the same pair of most recent common ancestors, who were 2nd cousins to each other.

• MG and I are 3rd cousins, sharing 136 cM in 6 segments.  The longest shared segment is 47 cM.  Ancestry reports MG's sharing with my daughter as only 35 cM, but this is after a reduction by Timber of 48 cM (from 83 cM unweighted).  I also happen to share with LW, who is MG's mother.  My sharing with her is 193 cM in 7 segments, with a longest shared segment of 54 cM.  LW's sharing with my daughter is 140 cM in 5 segments, again with a longest shared segment of 54.  Given my sharing with MG, as well as my daughter's sharing with MG's mother, it's clear that Timber has no legitimate reason for ANY reduction between MG and my daughter -- let alone a reduction of 48 cM!
  
• KG is also my 3rd cousin, sharing 210 cM in 13 segments.  The longest shared segment is 35 cM.  KG is a 1st cousin once removed to LW, and a 2nd cousin to MG.  My daughter's reported sharing with KG is 44 cM after Timber, but but that's a 44 cM reduction (from 88 cM unweighted).  
  
• GK is my 2nd cousin.  We share 261 cM with me in 15 segments, with a longest shared segment of 73 cM.  However, GK's reported sharing with my daughter is only 40 cM after Timber, which represents a reduction of 40 cM (from 87 cM unweighted).  I share 148 cM in 8 segments with GK's niece LS, who is his brother's daughter).  My longest shared segment with LS is 54 cM.  My daughter shares only 42 cM unweighted with LS, but Timber actually reduces even this amount by 31 cM to just 11 cM -- in other words, even below the threshold for reporting matches as shared.  Like me, GK is a 2nd cousin once removed to LW and a 3rd cousin to MG and KG.  (Although both have the same last initial, they don't have the same surname.)
  

The remaining three are all on my father's side.

• BH shares 237 cM in 12 segments with me, with a longest shared segment of 70 cM.  She's the granddaughter of one of my 1st cousins, making her my 1st cousin twice removed.  My daughter shares 81 cM unweighted with BH -- her 2nd cousin once removed -- but Timber reduces this by 35 cM to 46 cM.
  
• AD shares 108 cM with me in 4 segments, with a longest shared segment of 51 cM.  She shares 77 cM with my daughter -- unweighted -- but this is reduced by 31 cM to 46 cM by Timber.  AD is my 3rd cousin once removed, and my daughter's 3rd cousin twice removed.
  
• KC -- who is AD's sister -- shares 131 cM with me in 7 segments, with a longest shared segment of 51 cM.  Sharing between KC and my daughter is 54 cM in 3 segments after adjustment by Timber from 84 cM unweighted, so a reduction of 30 cM.  Yet the longest shared segment alone is still reported as 51 cM.
  

One of the problems of Timber is that it actually can't tell if any sharing represents excess sharing, only that it lies in a region in which some matches appear to show excess sharing.  That's one reason for the 90.0 cM restriction on Timber, but it doesn't mean that relatives who lack this amount of sharing should be "fair game".  

It does at least prevent Timber from reducing sharing between relatives as close as parents and their offspring, or between siblings -- which Timber would certainly do if no restriction existed.  It also ignores the reality of half or double relationships, however.

For example, I have numerous half 2nd cousins.  These have exactly the same number of steps from our shared ancestor as full 2nd cousins do, but on average only about half as much DNA.  While some of these half 2nd cousins still share more than 90.0 cM, others do not (and nearly always show sharing reductions by Timber).

[geebee-1015]

Here's another illustration of what a crock Timber is.  My daughter just received a new match "KR" for whom Ancestry reports sharing of 22 cM in a single segment.  The two have ten shared matches, all paternal.  However, while Ancestry reports my daughter's match as being on her "Paternal Side", here's what I see when I attempt a direct comparison between myself and KR:  [KR] is either not a DNA match or has not taken a DNA test.

The same thing is said if I do a direct comparison between KR and my daughter's mother.  This, of course, makes no sense if this is a real match.  However, let's take a closer look at the ten shared matches between KR and my daughter:

1. GL & my daughter: 1,058 cM in 31 segments, longest shared segment 84 cM; GL & me: 2,026 cM in 53 segments, longest shared segment 119 cM.  GL is my niece, so my daughter's paternal 1st cousin.
   
2. VM & my daughter: 428 cM in 12 segments, longest shared segment 53 cM; VM and me: 831 cM in 27 segments, longest shared segment 116 cM.  VM is my paternal 1st cousin, so my daughter's paternal 1st cousin once removed.
   
3. TC & my daughter: 62 cM after Timber adjustment (85 cM before adjustment) in 4 segments, longest shared segment 48 cM; TC & me: 42 cM after Timber adjustment (89 cM before adjustment) in 5 segments, longest shared segment 49 cM.  TC is my paternal 3rd cousin, so my daughter's paternal 3rd cousin once removed.
   
4. TCT & my daughter: 34 cM after Timber adjustment (49 cM before adjustment) in 3 segments, longest shared segment 31 cM; TCT & me: 16 cM Timber adjustment (60 cM before adjustment) in 2 segments, longest shared segment 49 cM.  TCT is TC's daughter, so my 3rd cousin once removed and my daughter's 4th cousin.
   
5. NV & my daughter: 23 cM after Timber adjustment (28 cM before adjustment) in 2 segments, longest shared segment 25 cM; NV & me: 39 cM after Timber adjustment (55 cM before adjustment) in 4 segments, longest shared segment 24 cM.  See item 6.
   
6. CV & my daughter: 22 cM after Timber adjustment (38 cM before adjustment) in 2 segments, longest shared segment 25 cM; CV & me: 30 cM after Timber adjustment (46 cM before adjustment) in 3 segments, longest shared segment 25 cM.  I believe NV and CV are siblings, but I am not certain of their relationship with my daughter of me, although they are marked as paternal matches.
   
7. SM & my daughter: 21 cM after Timber adjustment (33 cM before adjustment) in a single segment; SM & me: 9 cM after Timber adjustment (42 cM before adjustment) in 2 segments, longest shared segment 34 cM.  Relationship with SM is unclear, but she is marked as "Paternal Side".
   
8. JS & my daughter: 21 cM after Timber adjustment (32 cM before adjustment) in a single segment; JS & me: 19 cM after Timber adjustment (51 cM before adjustment) in 2 segments, longest shared segment 34 cM.  Relationship with JS is unclear, but she is marked as "Paternal Side" for me and as "Uncertain" for my daughter.  This is nonsense.  Possibly the algorithm is confused by my daughter supposedly sharing more DNA with JS than I do.  Of course, this is only because Timber "chopped" much more aggressively for me than for my daughter -- which makes no sense whatsoever.
   
9. LVW & my daughter: 21 cM after Timber adjustment (28 cM before adjustment); LVW & me: no match.  Also no match between LVW and my wife.  Despite that, the match is marked as "Paternal Side".  It is likely that this simply means that I share more than 28 cM with LVW before any adjustment by Timber, and less than 8.0 cM after adjustment.  Frankly, such matches should be ignored because they lack credibility.  If Ancestry wants to report matches for customers with two tested parents, then they should show the connecting parent's unweighted sharing even if the Timber-adjusted sharing is otherwise below the reporting threshold.
   
10. FC & my daughter: 20 cM after Timber adjustment (28 cM before adjustment) in a single segment; FC & me: 11 cM after Timber adjustment (30 cM before adjustment) in a single segment.
    

Rather than Timber proving useful here, Timber actually has the opposite effect.  Several of these matches are clearly made to seem more distant than they really are, particularly in the case of my 3rd cousin TC and his daughter.  I happen to know that TC and I share a set of 2nd great grandparents, and I have many other confirmed 3rd cousins who share significant amounts of DNA with me.  Yet because TC shared a single cM less than the 90.0 cM "Timber safe" threshold, Timber was free to make a 47 cM reduction!

TC's sister JB, however, does not have this problem.  She and I share 107 cM in 5 segments, with a longest shared segment of 53 cM -- pretty close to TC's longest shared segment with me of 49 cM.  I realize siblings can inherit different amounts of shared DNA, but the actual difference is not as extreme as the post-Timber difference:  107 cM for one sibling and 89 cM for the other, as opposed to 107 cM and 45 cM.

And this is not even getting into the fact that Timber reduced my sharing with TC much more than it did my daughter's.

[geebee-1015]

Here's another new match at Ancestry: EJ.  She actually tested previously at 23andMe, so I knew what her results were over there.  There's a bit of a difference -- 23andMe says we share 151 cM in 9 DNA segments, with a longest shared segment of 34 cM; Ancestry says it's 158 cM in 10 segments, but agrees on the longest shared segment of 34 cM.

Where it gets more interesting is that 23andMe reports sharing between EJ and my daughter as 67 cM in 3 segments -- also with a longest shared segment of 34 cM -- while Ancestry says sharing is only 52 cM in 3 segments, though again the longest shared segment is 34 cM.  The difference, of course, is that Timber has chopped off 15 cM of sharing, because unweighted sharing is shown as 67 cM.

The problem is that in both cases, my daughter actually inherited a full 67 cM from me of my 158 (or 151, whoever you believe) of sharing between EJ and me.  The lower number reflects what Timber "thinks" sharing actually should be when corrected for "extra" shared DNA from more distant common ancestors.  And without a doubt, EJ and I share more than just our more recent common ancestors.  Of course, we most likely inherited every bit of our shared DNA from those most recent common ancestors -- not by any other path.

In this case, the common ancestors are a pair of my great grandparents -- who would be EJ's 2nd great grandparents, and my daughter's 2nd great grandparents (making them 3rd cousins).

I also share with EJ's father at 23andMe, RB.  RB and I share 207 cM in 11 segments, with a longest shared segment of 37 cM.  Two of my siblings share even more DNA with RB than I do, while my other three siblings share a bit less.  My daughter shares only 70 cM with RB, so I have no doubt that if he tested at Ancestry, Timber would reduce this amount.

One problem with Timber is that it's a rather dull axe, and often used when there is really no reason why it should be.  And when I say "often", I think that's actually most of the time it's used.  This may not be true for everyone, but it definitely seems true for me and my daughter -- but also for her mother.

But the issue isn't simply whether Timber sometimes works.  Maybe it does ... sometimes.  The issue is, it's very clear that sometimes Timber actually makes matches seem more distant than they really are -- which is the opposite of what Ancestry says it's supposed to do.  I think it's arrogant that they continue to use it, although I would feel better about their using it if they did the following:

1. Allow customers to sort by unweighted sharing instead of just by Timberized-sharing, or simply make unweighted sharing the default.
   
2. Use the unweighted sharing number for determining whether a match should be on shared match lists.  (20.0 cM)
   
3. Use the unweighted sharing number for determining whether to report a match at all.  (0.8 cM)
   

It's ridiculous that someone with two tested parents will sometimes not see a tested parent in their shared match list, or -- even worse -- direct comparisons between a match and both parents will say "[match name] is either not a DNA match or has not taken a DNA test."  If taken literally for people you know have taken a DNA test, it would tell you the match is impossible.  A child can only share DNA with someone when at least one parent does.

Despite this, my daughter has a number of matches with whom -- according to Ancestry -- neither her mother nor me share any DNA.  This is true even for some matches greater than 20.0 cM who also have multiple shared matches on one parent's side -- just not the parent on that side (or on the other side).

[geebee-1015]

I did a little playing around with some of the tools found at https://dnapainter.com/tools/sharedcmv4.  I was again comparing my daughter and her 3rd cousin EJ.  

When I input Ancestry's Timber-adjusted sharing amount of 52 cM,  the probability of a 3rd cousin relationship was shown as 18%.  However, there was also a probability of an even closer relationship of 8%, so the combined probability for 3rd cousin or less was 26%.  The probability of a more distant relationship with this level of sharing was 75%.  (The numbers don't all add up exactly due to rounding.)

Using instead the unweighted probability of 67 cM, the probability of 3rd cousin changed to 25%.  There was also a 17% probability of an even closer relationship.  This changed the probability of 3rd cousin or closer to 42% -- meaning that the unweighted sharing amount was a much better fit for the actual, known relationship.  The probability of a more distant relationship that 3rd cousin was given as 58%, so even the unweighted sharing was a bit on the low side for the actual relationship.

Using my sharing amount was EJ of 158 cM, the probability of a relationship of 2nd cousin once removed -- the actual relationship -- was shown to be 53%.  There was also a 27% probability of an even closer relationship, such as 2nd cousin.  But the probability of a more distant relationship than 2nd cousin once removed was only about 20%.

[geebee-1015]

I have a total of 83 "Timber safe" matches at Ancestry, meaning matches that Timber can't mess with because unweighted sharing is at least 90.0 cM.  In most cases, I know exactly what the relationship is -- there are two nieces, five 1st cousins, three half 1st cousins, several 1st cousins once removed, 2nd cousins, 2nd cousins once removed, and numerous 3rd cousins.  There are a few for whom I haven't yet been able to determine an exact relationship.

Now, the rationale for these matches being "Timber safe" is that the amount of DNA shared is greater than what might be expected with "excess sharing" of more distant origin, at the 5th great grandparent level or beyond.  If the relationship is comparatively close -- say, great grandparents or 2nd great grandparents -- it's much less likely that the shared DNA is due to anything but the closeness of the relationship.

But it only takes until the 22nd of these matches to find one not Timber safe for my daughter.  "GK" is one of my 2nd cousins, sharing 261 cM with me in 15 segments, with a longest shared segment of 73 cM.  If my daughter had inherited half, she share about 130 cM and it would be safe for her, too.  Instead, she inherited 87 cM unweighted, in 6 segments.  The longest shared segment is 26 cM.  Yet Timber still declared that somehow 40 cM of this had suddenly become "excess sharing", and chopped the match down to 47 cM.

From this point, it was basically hit and miss.  That is, some matches were still at least 90.0 cM for my daughter and remained "safe"; others were less and invariably were reduced even further by Timber.  For example, my 23rd match -- BH, a 1st cousin twice removed -- shares 237 cM with me in 12 segments, with a longest shared segment of 70 cM.  BH is the granddaughter of one of my 1st cousins, DS, who shares  1,048 cM with me in 34 segments, with a longest shared segment of 111 cM.  DS also shares 489 cM with my daughter in 19 segments, with a longest shared segment of 91 cM.  However, my daughter's unweighted sharing with BH (her 2nd cousin once removed) is only 81 cM, making it fair game for reduction by Timber.  Ancestry says the match is "really" only 46 cM, a 35 cM reduction.  The longest shared segment by itself is larger than this, at 50 cM.  That's even before adding up the other 3 shared segments.

And it gets worse.  My 80th match is DTS, who shares 93 cM with me in 5 segments, with a longest shared segment of 35 cM.  DTS is one of my 3rd cousins.  My daughter's unweighted sharing with DTS is a perfectly reasonable 51 cM in 3 segments, with a longest shared segment of 32 cM.  But Timber chops this match down by 33 cM, to 18 cM.  That means that it not only is no longer a "safe" match, it isn't even sufficient to make the threshold to be in a shared match list!

That isn't the worst one, though.  Even before my match with DTS, there's LS.  She admittedly is a somewhat more distant match.  She's a 4th cousin once removed, although she has this relationship with me twice -- once through each of my maternal grandfather's maternal grandmother's parents, who were 2nd cousins to each other.   It's possible that my daughter shares absolutely nothing with LS.  That is what Ancestry shows, but in reality all it's possible to know for sure is that she and LS share less than 8.0 cM after Timber.  Once Ancestry drops a match, it's dropped -- there's no way to know if it actually does exist.

In fact, my daughter has some matches that Ancestry confidently asserts to be maternal or paternal, and then on direct comparison neither parent shares anything with that person!  This is even true for a few matches with whom my daughter has multiple shared matches, making me think these matches might actually be real.  What seems likely to be happening is that Timber simply adjusted the parental match to below the 8.0 cM threshold, so that it disappeared.

This is why I think that, at the very least, the thresholds should only be applied to the unweighted sharing amounts.  So a match showing unweighted sharing of 51 cM, for example, wouldn't be excluded from shared match lists simply because Timber reduced it below 20.0 cM.  Ancestry should also show matches greater than 8 cM unweighted, even when Timber has reduced them to below that -- even to below 6 cM.  To apply the thresholds after Timber has done its work places a confidence in Timber that is demonstrably unjustified.

I'm not even saying that Timber never does what it's supposed to.  But it's unfair for Ancestry to use Timber in a way that makes some sharing amounts seem less "misleading" for some customers at the cost of making the matches more "misleading" for other customers.  The fact is, what Timber is doing with its "adjustments" is changing the actual sharing amount between two matches.  Even if that sharing is perhaps not simply the result of recent common ancestry, it's still DNA that was passed on directly from parent to child except in cases of related parents.  If it turns out that there are more distant related relatives, all the DNA from them still has to go through the "chain" leading to the child.

So, for example, I know that my father's maternal grandparents were related to each other.  They were 2nd cousins twice removed.  So it turns out that my most recent common ancestors with both of them are a couple who are simultaneously my 4th great grandparents and my 6th great grandparents.  This does not mean I have huge amounts of "excess sharing" with my matches who share the same ancestry.  It might mean I have a slightly increased chance of sharing any DNA at all with them.