WEBVTT

1
00:00:00.000 --> 00:00:08.867
This overview explains how D8 carried a frozen rewrite package
to Linux while separating output agreement from profitability.

2
00:00:08.867 --> 00:00:17.359
Correct outputs and rule incidence can reproduce even when
overhead-inclusive timing reverses direction on another host.

3
00:00:17.359 --> 00:00:17.367
Correct outputs and rule incidence can reproduce even when
overhead-inclusive timing reverses direction on another host.

4
00:00:17.667 --> 00:00:23.300
Use the verified distributive identity with the same variable meanings on both hosts.

5
00:00:23.300 --> 00:00:26.400
Keep the source and candidate identities fixed.

6
00:00:26.400 --> 00:00:32.767
Exact matching and equal outputs can transfer even when the performance result does not.

7
00:00:32.767 --> 00:00:38.833
On fictional host A, matching costs two and execution costs four, totaling six.

8
00:00:38.833 --> 00:00:41.100
The matching control costs eight.

9
00:00:41.100 --> 00:00:45.300
Under this declared boundary, the candidate is cheaper.

10
00:00:45.300 --> 00:00:50.400
On fictional host B, matching costs five while execution remains four.

11
00:00:50.400 --> 00:00:53.833
The total is nine against eight, reversing the result without

12
00:00:53.833 --> 00:00:56.200
changing the mathematical answer.

13
00:00:56.200 --> 00:01:02.893
These teaching values explain the mechanism and are not the retained D8 timings.

14
00:01:02.893 --> 00:01:02.900
These teaching values explain the mechanism and are not the retained D8 timings.

15
00:01:02.900 --> 00:01:08.433
D8 carries a frozen one-pass rewrite package to Linux.

16
00:01:08.433 --> 00:01:16.233
Exact outputs and rule incidence can reproduce
even if overhead-inclusive timing changes direction.

17
00:01:16.233 --> 00:01:19.667
They are separate result columns.

18
00:01:19.667 --> 00:01:26.333
Use a conceptual rewrite with matching,
execution and delivery on one timeline.

19
00:01:26.333 --> 00:01:32.733
In this illustration, the same transformation
and admitted cases run on the second host.

20
00:01:32.733 --> 00:01:39.733
Both hosts have matching outputs and application
counts, while their total charged times differ.

21
00:01:39.733 --> 00:01:46.833
This shows how exact engineering success can
coexist with an unprofitable whole request.

22
00:01:46.833 --> 00:01:51.867
The retained D8 comparison reversed
the Windows profitability direction on Linux,

23
00:01:51.867 --> 00:01:56.633
and unconditional one-pass rewriting was not promoted.

24
00:01:56.633 --> 00:02:05.533
That is a result of the frozen experiment, not a reason to hide
the second machine or retune the criterion after seeing it.

25
00:02:05.833 --> 00:02:10.200
Exact behavior and runtime transfer are separate checks.

26
00:02:10.200 --> 00:02:14.367
A system can preserve every decision across hosts while losing its

27
00:02:14.367 --> 00:02:18.500
cost advantage because overhead changed.

28
00:02:18.500 --> 00:02:22.967
If a threshold is edited after seeing the unfavorable result, the new rule

29
00:02:22.967 --> 00:02:24.767
is a new candidate.

30
00:02:24.767 --> 00:02:27.500
The old result should remain visible.

31
00:02:27.500 --> 00:02:32.000
Editing a parameter cannot turn the already observed frozen test into

32
00:02:32.000 --> 00:02:35.700
an unseen successful confirmation.

33
00:02:35.700 --> 00:02:39.200
The revision can enter a new development and validation process

34
00:02:39.200 --> 00:02:41.700
with appropriate fresh evidence.

35
00:02:41.700 --> 00:02:46.933
Keep its identity and scope distinct from the original host-transfer result, including any

36
00:02:46.933 --> 00:02:51.600
retained regressions and non-promotion decision.

37
00:02:51.600 --> 00:02:51.611
Report exactness and profitability as separate gates.

38
00:02:51.611 --> 00:02:56.500
Report exactness and profitability as separate gates.

39
00:02:56.500 --> 00:03:04.467
Preserve the Linux loss and use the admitted control
unless a justified successor meets the conditions.