Charcoal HQ

ArtOfCode

12:11 AM

there you go, @Undo

yellow and orange align to left-hand Y axis, blue to right-hand

Undo

No chance we're actually taking a rolling average, right?

ArtOfCode

yellow and orange are the previous graph (TTD, with and without FIRE), blue is percent with deletion logs

@Undo where?

I'm not taking a rolling for these graphs, they're just daily averages

Undo

@ArtOfCode Yellow line

ArtOfCode

Nope, daily average

Undo

Here's what I'm seeing: TTD gradually increasing while deletion log coverage is lower than normal, then gradually returning to normal once deletion log coverage is normal

ArtOfCode

12:13 AM

AVG(TIMESTAMPDIFF(SECOND, posts.created_at, deletion_logs.created_at))

Undo

Although that could also be an artifact of Smokey's DL system w/ cause going the other way

ArtOfCode

@Undo yep, I was seeing that as well. Glad we're on the same page.

So now the question is why did DL coverage drop?

Undo

Was that the timeframe where we were having issues with Thomas' instance networking?

ArtOfCode

Could have been, I guess

Undo

Could explain the Smokey DLs dropping off

Find it suspicious that it gradually rises to the exact level as time w/ FIRE

ArtOfCode

12:17 AM

There were a few issues around then, yeah

Undo

still a weird way for it to manifest itself

@ArtOfCode Is blue % with deletion logs overall, or % with deletion logs from Smokey?

ArtOfCode

@Undo overall

Undo

If Smokey logs drop to low levels, that'd explain why it lines up well with the FIRE logs

ArtOfCode

That graph might have to be a job for tomorrow, but I wouldn't be surprised if you're right

Which would explain basically the whole thing away

Undo

I can throw it together

ArtOfCode

12:20 AM

If @Andy can generate a graph with a long-period rolling average instead of an overall trendline, with any luck it'll start dropping down again after that

Undo

No discernible difference in counts without a graph

Wed, 01 Feb 2017 => 108,
Thu, 02 Feb 2017 => 197,
Fri, 03 Feb 2017 => 130,
Sat, 04 Feb 2017 => 161,
Sun, 05 Feb 2017 => 70,
Mon, 06 Feb 2017 => 152,
Tue, 07 Feb 2017 => 160,
Wed, 08 Feb 2017 => 191,
Thu, 09 Feb 2017 => 144,
Fri, 10 Feb 2017 => 119,
Sat, 11 Feb 2017 => 122,
Sun, 12 Feb 2017 => 81,
Mon, 13 Feb 2017 => 156,
Tue, 14 Feb 2017 => 198,
Wed, 15 Feb 2017 => 167,
Thu, 16 Feb 2017 => 183,
Fri, 17 Feb 2017 => 151,
Sat, 18 Feb 2017 => 147,
Sun, 19 Feb 2017 => 85,
Mon, 20 Feb 2017 => 160,
Tue, 21 Feb 2017 => 173,

There is something weird around Apr 11, though

ArtOfCode

welcome to your irregularly scheduled Amateur Data Analytics with Undo and Art, @JF

@Undo yeah, and dropping away again afterwards

Undo

Explained by our initial not-recording of API key id

~~... waaait~~

Yep, started recording it on Apr 24

which is where the yellow line and orange line diverge, naturally

ArtOfCode

@Undo and I just checked this, exploder works fine now

Undo

awesome

ArtOfCode

12:27 AM

For a test, use the last 100 posts and this script:

let posts = store.posts;
results.push(['Post ID', 'TTD']);

for (let i = 0; i < posts.length; i++) {
    let post = posts[i];
    let created = Date.parse(post.created_at);
    let deleted = Date.parse(post.deleted_at);
    let diff = deleted - created;
    diff = isNaN(diff) ? null : diff / 1000;
    results.push([post.id, diff]);
}

Undo

cool

Transcript for