We opened 421 paid digital products by hand. The file never set the price.

In August 2026 we opened 421 paid digital products one at a time — templates, presets, plugins, 3D packs, Photoshop actions, teacher worksheets — across nine marketplaces (Gumroad, Etsy, GraphicRiver, 3DOcean, DLsite, Teachers Pay Teachers, Hotmart and two more). For each one we wrote down what the page proves: the price, the lowest unit count the page itself shows, whether a free sample is visible, what the buyer actually receives, and how complex the thing is.

The whole table is public domain (CC0): CSV / JSON on our site and as a Hugging Face dataset.

import pandas as pd
df = pd.read_csv("https://getreadystack.com/data/421-products.csv")
df.groupby("complexity")[["price_usd", "units_floor"]].median()

Here is what fell out of it. Every number below is computed from that file, nothing else.

1. The same file format sells at 30× to 2,900× the price

format	paid listings	min	median	max
.pdf	96	$0.99	$5.50	$223.75
.zip	43	$1.00	$17.60	$2,900
.pptx	10	$4.99	$13.50	$143
.unitypackage	13	$4.00	$90	$1,000
.atn (Photoshop action)	19	$4.00	$7.00	$15

A .pptx at $4.99 and a .pptx at $143 are the same bytes on disk. The price is not a property of the file. It is a property of who needs it and what happens to them if they get it wrong.

2. Complexity buys price, not units

complexity (as judged from the page)	n	median price	median proven units
trivial	112	$5	372
simple	122	$7	506
moderate	89	$15	544
heavy	98	$65	291

"Heavy" products charge about 13× what "trivial" ones do and still move roughly 60% as many units. If you are choosing between making something small and making something serious, the table says: serious pays, and it does not cost you the audience.

3. A visible free sample correlates with 8× the proven sales

Of the 421 listings, 175 show a free sample of some kind on the page (a preview file, a free tier, a sample pack). 246 show none.

free sample visible	n	median proven units
yes	175	1,231
no	246	152

This is observational, not an experiment — sellers who bother to add a sample probably do other things right too. But the gap is not small, and it matches what the winners on every shelf do: give a real, usable piece away, then sell the rest.

4. Shelves differ more than products do

marketplace	n	median price	median proven units	listings with ≥100 proven units
Gumroad	112	$20.00	504	83%
Etsy	87	$4.50	89	46%
GraphicRiver	48	$8.50	6,450	85%
DLsite (JP)	42	$6.60	2,305	100%
Teachers Pay Teachers	30	$22.00	220	70%
3DOcean	26	$477.50	0	8%
Hotmart (pt-BR)	13	$17.60	62	46%

3DOcean carries the highest prices in the set and the least proof that anything sells — the pages simply do not show a number. DLsite, at $6.60, proves sales on every single listing we opened. "Expensive shelf" and "selling shelf" are not the same shelf.

5. What the page proves vs. what people assume

• 292 of 421 listings (69%) prove at least 100 units sold on the page itself.
• 19 listings prove zero.
• The median floor is 416 units — for products that mostly cost under $20.

We call the column units_floor on purpose. It is the lowest number the page proves — a printed unit count, or a rating count, or nothing. It is never an estimate. Treat it as a floor, not a market size.

What we did with it

We are a small team building developer and professional tools, and this table changed what we build. The products that command real prices are attached to a moment when being wrong costs money: a filing deadline, a regulation that changed this year, a tax threshold, an audit. So that is what we make now — small linters and calculators pinned to a dated rule, each one citing its source, free for the core check and paid for the parts that protect a team (getreadystack.com). The dataset is the receipt for that decision.

Caveats, honestly

• One observer, one month, nine marketplaces that publish numbers unevenly.
• units_floor is a floor. Some listings sell far more than the page shows.
• The complexity and free-sample columns are human judgments from the listing page, not from the file.
• Correlation throughout. Nothing here was A/B tested.

If you use the data, we would genuinely like to hear what you find — the file is CC0, do whatever you want with it.