Performance
and split tests.

What the channel is actually doing, measured so the comparisons hold: equal age, medians rather than averages, and one variable under test at a time.

Account @augustusvancePosts 4Comparing at V1Updated 2026-08-28 17:16
Daily · 60 seconds

Log the numbers.

Open TikTok, read each post's counters, type them here. Every save is a timestamped snapshot, not an overwrite — that is what lets the board compare posts at equal age instead of comparing a post from today against one from last week.

PostViewsLikesCommentsSharesSavesLink
p002blink
p003ears
p004grid
p001pauselock

Blanks are skipped, so you can log views only and fill the rest from the TikTok Studio CSV later.

Leaderboard · measured at V1

Every post, same age.

Latest is the raw counter at whatever age each post happens to be — useful, but not a fair comparison. The column beside it is: views interpolated to V1, the oldest checkpoint most posts have reached, so nothing is rewarded for simply being older. A post at 3× the account median is flagged an outlier: that is the one to clone, not the one with the most raw views.

Posts4
Measurable0
Median V1
Total views139
Outliers0
PostFormatSkinAgeLatestV1× medianEngCommentAvg watch
p002episodes_blink_002 blinkheat0.6h 9 —× —%—% —s
p003episodes_ears_002 earsheat0.6h 25 —× —%—% —s
p004episodes_grid_001 gridheat0.6h 13 —× —%—% —s
p001episodes_geo_002 pauselockheat0.8h 92 —× —%—% —s
Split tests · one at a time

What we are actually testing.

Only one test runs at a time and everything in its hold row stays frozen while it does. Running four variables across twenty posts on a lottery-shaped account does not give you four answers, it gives you noise wearing four hats.

activeE1

Which mechanic travels furthest on a cold account?

pauselock1/5
blink1/5
ears1/5
grid1/5

Next post should be pauselock · 16 posts to fill this test.

Hold fixed skin=heat · cover_variant=0 · post_hour=evening

Decision rule Kill the bottom two mechanics, then run E2 on the survivors.

queuedE2

Which skin gets picked up in a cold feed?

heat4/5
glitch0/5
sticker0/5

Waiting — 11 posts of runway once it opens. Do not vary this yet.

Hold fixed format=winner of E1 · cover_variant=0

Decision rule Sticker reads brightest in-feed; if it does not beat heat by 1.5x, stop rendering three skins.

queuedE3

Does the cover variant change the pickup rate?

04/5
10/5
20/5

Waiting — 11 posts of runway once it opens. Do not vary this yet.

Hold fixed format=winner of E1 · skin=winner of E2

Decision rule If flat, drop to one cover variant and stop generating three.

queuedE4

Does the hook shape matter more than the mechanic?

number+you4/5
number0/5
you0/5
plain0/5

Waiting — 16 posts of runway once it opens. Do not vary this yet.

Hold fixed format=winner of E1 · skin=winner of E2

Decision rule Rewrite every live script to the winning hook shape.

queuedE5

Is the 64s band actually right, or does longer hold?

<65s0/5
65-72s2/5
72-80s2/5
>80s0/5

Waiting — 16 posts of runway once it opens. Do not vary this yet.

Hold fixed format=winner of E1

Decision rule Retune the lint runtime band to the winner. Never go under 60s (Creator Rewards floor).

Read-out · medians, never means

What each variable is worth.

Each bar is the median V1 for posts carrying that value. A verdict of signal needs five posts in the thinnest arm and a 1.5× gap; lean is a hint you may not act on yet; insufficient means the numbers are real but the conclusion is not.

Nothing measurable yet. This fills in once posts have views logged.

Ingest · three doors

Getting numbers in.

1 · The form above

Fastest daily loop. Saves a timestamped snapshot through the same queue the Generate buttons use, so nothing needs to be installed on your phone.

2 · Browser snapshot

Signed into TikTok on the desktop, open your profile, scroll to the bottom and run pull_stats.js. It captures views for every tile at once and copies JSON to the clipboard.

python3 ~/quizforge/tracker.py ingest -

3 · Studio CSV

TikTok Studio → Analytics → Content → Download data. This is the only door that carries average watch time, completion and follows, which are the numbers that actually explain a flop.

python3 ~/quizforge/tracker.py csv ~/Downloads/export.csv