Thank you Jonathan. To address your points in order:
1. We are not able to measure the rate of change in the number of profiles without sources, because we are not able to get an accurate count of same without extremely laborious, essentially manual, review. Kay's BioCheck app helps, but the process is still time-consuming and produces statistical estimates, not actual counts. That is the essence of the issue.
2. I agree the more profiles identified the better, and that we can't correct profiles unless we identify them. The number identified with the {{Unsourced}} template rises and falls independently of the actual (unknown) number of profiles without sources. As long as that holds true, the number of profiles identified has no utility for monitoring progress toward reducing the problem. The remedy for this is to identify all profiles without sources. That's a task for a bot.
3. The issue has nothing to do with Suggestions, except to the extent Suggestions can help identify profiles without sources. See point #2, above.
4. My main analysis of this issue did not filter for presence or absence of birth or death dates in any way. In particular, the study of 160 profiles, that yielded the estimate of 4.9 to 8.1 million profiles, used "a sample taken from all Open profiles" regardless of whether the date fields contained data. Your criticism about "selection of profiles from the group consisting of missing birth and death dates" has no foundation in fact.
5. I rejected as sources "links to trees or profiles behind paywalls," not records behind paywalls. Such a thing put the profile in my "unsourced" bin only if no other valid source was provided. Selection criteria defining a valid source have long been discussed on G2G and need to be established in order to identify all profiles without sources (as I maintain WikiTree must do). In fact, each member now adding or removing an {{Unsourced}} template applies his or her own subjective criteria, which may be more or less stringent than what I used.
6. Difficulty in detecting sources within free text does not justify not attempting it. The BioCheck app shows that it can be done with some success. Once in place, a detection algorithm can be expected to improve over time.
7. Your statement beginning "if the numbers decrease significantly" invites untested or meaningless comparisons. I do not mean to diminish in any way the efforts of those involved in the Source-a-Thon. 38,000 profiles removed from the total is a very good thing in which the participants can and should take pride. Is it "significant?" Compared to the estimated total of several million profiles without sources, I would say No. If comparing to newly created profiles without sources, no one knows that quantity, so who can assess the significance? If comparing to newly identified {{Unsourced}} profiles, please note that about 22,000 profiles were "identified" in the month preceding the Source-a-Thon. See point #2, above. As to whether we are "directionally correct," that depends on whether your "If" applies, and no one can judge that with any level of objective accuracy. That is the essence of the issue, which brings us full circle back to point #1, above.
Edited to correct typo.