WEBVTT

1
00:00:00.000 --> 00:00:06.433
This overview explains why method rankings
require a common requested artifact and lifecycle.

2
00:00:06.433 --> 00:00:13.986
A fast scalar count and a complete truth vector answer different
requests, so their bars cannot establish one shared winner.

3
00:00:13.986 --> 00:00:14.000
A fast scalar count and a complete truth vector answer different
requests, so their bars cannot establish one shared winner.

4
00:00:14.300 --> 00:00:18.433
A point evaluation and a witness search can both use A or B, but

5
00:00:18.433 --> 00:00:20.633
they deliver different objects.

6
00:00:20.633 --> 00:00:24.733
The point query fixes the input and asks for one bit.

7
00:00:24.733 --> 00:00:30.400
The witness search asks the method to find an input making the formula true.

8
00:00:30.400 --> 00:00:33.467
A full vector supplies four outputs.

9
00:00:33.467 --> 00:00:36.767
Counting all satisfying assignments gives three.

10
00:00:36.767 --> 00:00:40.367
Counting distinct selected A values gives two.

11
00:00:40.367 --> 00:00:44.267
These are three more tasks, and a time from one should not be inserted into

12
00:00:44.267 --> 00:00:46.900
a ranking for another.

13
00:00:46.900 --> 00:00:51.200
Organize evidence by requested artifact and eligible method.

14
00:00:51.200 --> 00:00:56.400
A comparison is meaningful only after the methods agree on the delivered object, input

15
00:00:56.400 --> 00:00:59.400
universe, lifetime, and boundary.

16
00:00:59.400 --> 00:01:04.733
That grid is more informative than one universal winner label.

17
00:01:04.733 --> 00:01:04.749
A ranking needs a common job.

18
00:01:04.749 --> 00:01:07.933
A ranking needs a common job.

19
00:01:07.933 --> 00:01:15.367
Start with the answer required, then the workload,
lifecycle, comparator and timed stages.

20
00:01:15.367 --> 00:01:19.633
A single podium often hides these conditions.

21
00:01:19.633 --> 00:01:22.700
The grid separates one-bit evaluation,

22
00:01:22.700 --> 00:01:29.200
full truth output, scalar count, and distinct
projected count into different rows.

23
00:01:29.200 --> 00:01:35.700
Its columns separate cold construction, warm resident queries, and reload.

24
00:01:35.700 --> 00:01:40.733
The same engine can occupy different positions across this grid.

25
00:01:40.733 --> 00:01:44.433
The retained studies already include a CM kernel gain,

26
00:01:44.433 --> 00:01:51.233
an EPFL parity result, slower public wrappers and strong counting controls.

27
00:01:51.233 --> 00:01:56.367
A scalar count may avoid constructing the truth vector entirely.

28
00:01:56.367 --> 00:02:03.333
That is a legitimate advantage for a counting
request, not proof that it won a full-output task.

29
00:02:03.333 --> 00:02:09.400
Preserve negative cases, refusals and unpromoted routes in the grid.

30
00:02:09.400 --> 00:02:14.391
A blank cell means the comparison has not been established.

31
00:02:14.391 --> 00:02:14.400
A blank cell means the comparison has not been established.

32
00:02:14.700 --> 00:02:19.233
Unknown means there is no matching retained evidence for that cell.

33
00:02:19.233 --> 00:02:22.733
It does not mean the method is slow or incorrect.

34
00:02:22.733 --> 00:02:26.533
The result remains an open comparison.

35
00:02:26.533 --> 00:02:31.767
Refused means the method or API did not accept this input or contract.

36
00:02:31.767 --> 00:02:34.967
That may be an expected eligibility boundary.

37
00:02:34.967 --> 00:02:40.233
It is different from completing the correct job and losing on elapsed time.

38
00:02:40.233 --> 00:02:44.833
A measured loss means the method completed the same task correctly but had worse measured

39
00:02:44.833 --> 00:02:47.333
cost within the declared scope.

40
00:02:47.333 --> 00:02:51.667
Keeping unknown, refused, and measured loss separate prevents missing data

41
00:02:51.667 --> 00:02:55.600
from becoming a fabricated ranking.

42
00:02:55.600 --> 00:02:55.603
Match the deliverable, workload, and lifecycle before ranking methods.

43
00:02:55.603 --> 00:03:02.067
Match the deliverable, workload, and lifecycle before ranking methods.

44
00:03:02.067 --> 00:03:07.833
Keep negative results and unestablished comparisons visible.