WEBVTT

00:00:00.300 --> 00:00:01.250
A useful website

00:00:01.250 --> 00:00:02.200
monitor needs to

00:00:02.200 --> 00:00:03.692
remember what changed.

00:00:03.992 --> 00:00:05.368
We tested whether Grex

00:00:05.368 --> 00:00:07.361
could preserve fault identity,

00:00:07.361 --> 00:00:08.954
recognize resolution and

00:00:08.954 --> 00:00:10.403
produce useful reports

00:00:10.403 --> 00:00:12.504
across repeated observations.

00:00:13.467 --> 00:00:15.032
The fixture started clean,

00:00:15.032 --> 00:00:16.264
gained known faults,

00:00:16.264 --> 00:00:17.163
repeated them,

00:00:17.163 --> 00:00:17.896
then added an

00:00:17.896 --> 00:00:18.961
error burst before

00:00:18.961 --> 00:00:20.293
returning to normal.

00:00:20.593 --> 00:00:22.256
Expected answers stayed hidden

00:00:22.256 --> 00:00:23.303
from the workers.

00:00:24.258 --> 00:00:25.741
Grex computed fault facts

00:00:25.741 --> 00:00:27.696
from browser observations before

00:00:27.696 --> 00:00:28.909
writing a summary.

00:00:29.209 --> 00:00:30.889
A direct workflow summarized

00:00:30.889 --> 00:00:31.897
the same facts.

00:00:32.197 --> 00:00:34.157
Independent evaluation checked

00:00:34.157 --> 00:00:35.277
both the lifecycle

00:00:35.277 --> 00:00:36.677
and summary quality.

00:00:37.633 --> 00:00:39.083
4 faults were detected,

00:00:39.083 --> 00:00:40.215
retained and later

00:00:40.215 --> 00:00:41.488
resolved against the

00:00:41.488 --> 00:00:42.903
last accepted state.

00:00:43.203 --> 00:00:45.101
But 3 overflow attempts failed.

00:00:45.401 --> 00:00:46.937
Correct computed facts did

00:00:46.937 --> 00:00:48.405
not produce an acceptable

00:00:48.405 --> 00:00:49.540
complete report.

00:00:50.508 --> 00:00:52.036
All 3 production reports

00:00:52.036 --> 00:00:53.745
were eventually accepted,

00:00:53.745 --> 00:00:54.836
but only 2 arrived

00:00:54.836 --> 00:00:55.927
within 2 hours.

00:00:56.227 --> 00:00:57.859
Direct summaries were slightly

00:00:57.859 --> 00:00:59.128
preferred more often.

00:00:59.428 --> 00:01:01.248
Manual recovery and compressed

00:01:01.248 --> 00:01:02.799
timing limit the result.

00:01:03.758 --> 00:01:04.809
The study closed with

00:01:04.809 --> 00:01:05.977
its failure visible.

00:01:06.277 --> 00:01:07.568
Later code fixes did

00:01:07.568 --> 00:01:09.314
not retroactively pass the

00:01:09.314 --> 00:01:10.414
overflow stage,

00:01:10.414 --> 00:01:12.160
and clean production pages

00:01:12.160 --> 00:01:13.754
did not demonstrate real

00:01:13.754 --> 00:01:15.045
fault detection.