Dataset Viewer
Auto-converted to Parquet Duplicate
slug
stringclasses
9 values
name
stringclasses
9 values
url
stringclasses
9 values
answering
bool
1 class
measures
stringclasses
9 values
method
stringclasses
9 values
source_endpoint
stringclasses
9 values
headline_label
stringclasses
9 values
headline_value
int64
14
721k
detail_1
stringclasses
9 values
detail_2
stringclasses
9 values
detail_3
stringclasses
8 values
as_of
stringclasses
9 values
stale
bool
1 class
error
float64
fetched_at
stringclasses
9 values
skillworks
SkillWorks
https://skillworks.thecompound.tech
true
Every Claude Code skill, subagent, plugin and marketplace repository.
Each listing's files are read from source and scored 0–100 on four weighted components. A listing that does not parse into something Claude Code could load is counted as broken, not quietly dropped from the denominator.
https://skillworks.thecompound.tech/api/stats
listings read from source
720,952
that do not load: 63,762
share failing: 8.8%
repositories: 29,174
2026-09-03T07:46:15.519+00:00
false
null
2026-09-04T14:42:56.555Z
stillshipping
StillShipping
https://stillshipping.thecompound.tech
true
Whether an agent tool is still being worked on.
Commit, release and issue activity pulled from the GitHub API nightly and turned into one of three verdicts. The verdict is recomputed every night rather than recorded once, so a project that goes quiet changes its own row without anybody editing it.
https://stillshipping.thecompound.tech/api/stats
tools tracked
340
maintained: 295
slowing: 21
dead: 24
2026-09-04T06:20:52.557+00:00
null
null
2026-09-04T14:42:56.568Z
tooldrift
ToolDrift
https://tooldrift.thecompound.tech
true
What the AI coding tools changed overnight — pricing, models, rankings.
Vendor changelogs, pricing pages and store rankings are fetched on a schedule and diffed against the previous capture. A price is recorded from the page that states it, and a page that has moved is followed and the redirect published rather than silently followed.
https://tooldrift.thecompound.tech/api/stats
tools watched nightly
36
models tracked: 428
pricing pages checked: 33
ranking rows: 1,459
2026-09-04T09:51:14.279+00:00
false
null
2026-09-04T14:42:56.571Z
kitgrade
KitGrade
https://kitgrade.thecompound.tech
true
SaaS starter kits, graded on what is actually in the box.
Kits are graded from measured facts rather than from their own landing pages, and the hands-on count is published separately from the documented count — a kit somebody installed and ran is a different claim from a kit somebody read about.
https://kitgrade.thecompound.tech/api/stats
starter kits graded
36
installed and run: 16
open source: 22
archived since listing: 2
2026-09-04T05:41:48.01+00:00
null
null
2026-09-04T14:42:56.574Z
storeready
StoreReady
https://storeready.thecompound.tech
true
Which AI app builders produce something the App Store will actually accept.
Every verdict is attached to numbered evidence — a policy clause, a rejection thread, a shipped binary. A builder nobody has evidence for is published as unknown rather than given the benefit of the doubt.
https://storeready.thecompound.tech/api/health
app builders assessed
14
pieces of evidence: 47
ship: 4
ship with caveats: 8
2026-09-03T08:50:16.47898+00:00
false
null
2026-09-04T14:42:56.575Z
blockdex
BlockDex
https://blockdex.thecompound.tech
true
Every component, block and theme inside every public shadcn registry.
Each registry is crawled and its items are read individually, so an item that disappears between crawls is counted as removed rather than forgotten. Removals over the last thirty days are published beside the total.
https://blockdex.thecompound.tech/api/stats
registry items indexed
75,550
registries crawled: 976
removed in 30 days: 11,076
with source available: 22,719
2026-09-04
false
null
2026-09-04T14:42:56.577Z
rulestack
RuleStack
https://rulestack.thecompound.tech
true
How teams actually write agent config — AGENTS.md, CLAUDE.md and the rest.
Config files are read out of public repositories and measured: length, whether they carry runnable commands, whether they carry code. The format shares are computed from the files found, not from a survey of what people say they use.
https://rulestack.thecompound.tech/api/stats
agent config files read
7,109
formats compared: 8
repositories: 1,729
most common format: cursor-rules
2026-09-01
null
null
2026-09-04T14:42:56.579Z
citerank
CiteRank
https://citerank.thecompound.tech
true
Whether answer engines cite a site when a buyer asks about its category.
A question is put to a live answer engine through a real browser and the answer it returns is recorded — one row per question per engine per run. Nothing is inferred from search rank; a citation counts when the engine actually produced it.
https://citerank.thecompound.tech/api/stats
answers recorded
186
engines answering: 5
free checks run: 21
null
2026-09-03
false
null
2026-09-04T14:42:56.580Z
doc-extract-bench
doc-extract-bench
https://github.com/kyisaiah47/doc-extract-bench
true
Document-extraction APIs, scored on pinned public datasets.
Every vendor is run against the same pinned subsets and EVERY RAW RESPONSE IS COMMITTED, so the published accuracy can be recomputed offline by anyone with no API keys and no account. It is the only one of the nine whose numbers a stranger can fully re-derive.
https://raw.githubusercontent.com/kyisaiah47/doc-extract-bench/main/results/latest.json
documents scored
1,210
vendors compared: 3
pinned datasets: 5
extraction failures: 18
2026-07-10T13:37:49.489Z
null
null
2026-09-04T14:42:56.582Z

Toolproof: the nine indexes and what each one currently measures

One row per index under the Toolproof masthead: what it measures, the method behind it, the public endpoint its figures come from, and the headline figure that endpoint returned at the moment of the cut.

Rows in this cut 9
One row is one index
Cut 2026-09-04
Refreshed Monthly, on the first of the month
Measured by Toolproof
Method https://toolproof.thecompound.tech/methodology
Licence Creative Commons Attribution 4.0 International
Publisher Compound Labs
Also on Hugging Face · Zenodo

How it is measured

Read from https://toolproof.thecompound.tech/api/index.json, which builds itself by calling each index's own public stats endpoint at request time. An index that did not answer is kept in the file with ok=false and the reason it gave, because a consumer has to be able to tell a missing index from an index that measured nothing.

What a citer needs to know

  • The headline figures move between cuts. A citation of one of them should carry the cut date, which is in the filename and in as_of.
  • as_of is the run that produced the figure. fetched_at is when this export read it. Those are different dates and both matter.

Files

File Format Size
toolproof-indexes-2026-09-04.csv CSV 0.00 MB
toolproof-indexes-2026-09-04.json JSON 0.01 MB

Every cut is also a GitHub release with the same files attached, so a citation can pin the exact edition it quoted.

Schema

Column Type Empty in sample
slug string 0.0%
name string 0.0%
url string 0.0%
answering boolean 0.0%
measures string 0.0%
method string 0.0%
source_endpoint string 0.0%
headline_label string 0.0%
headline_value number 0.0%
detail_1 string 0.0%
detail_2 string 0.0%
detail_3 string 11.1%
as_of string 0.0%
stale boolean 44.4%
error null 100.0%
fetched_at string 0.0%

Cite it

Compound Labs (2026). Toolproof: the nine indexes and what each one currently measures. Toolproof, https://toolproof.thecompound.tech. Cut of 2026-09-04. Creative Commons Attribution 4.0 International (CC BY 4.0). https://github.com/kyisaiah47/compound-datasets/tree/main/toolproof-indexes
@dataset{compound_toolproof_indexes_2026,
  title     = {Toolproof: the nine indexes and what each one currently measures},
  author    = {{Compound Labs}},
  year      = {2026},
  publisher = {Compound Labs},
  url       = {https://github.com/kyisaiah47/compound-datasets/tree/main/toolproof-indexes},
  note      = {Cut of 2026-09-04. Measured by Toolproof, https://toolproof.thecompound.tech},
  license   = {CC-BY-4.0}
}

The whole collection has a DOI, 10.5281/zenodo.22844096, at https://doi.org/10.5281/zenodo.22844096. It resolves to the newest Zenodo version. To pin the exact deposit this cut belongs to, cite 10.5281/zenodo.22844097 instead.

Attribution is the licence condition, and it is the only one. Quote a figure with the publisher, the product that measured it and the cut date, and there is nothing else to ask.

Downloads last month
50