How we judge the site.
1,165 pair pages are prebuilt against a budget of 1,240, each chosen by the page score, as of .
The measures
What each measures, the target, and today's value.
| Measure | Counts | Target | Today |
|---|---|---|---|
| Price accuracy | Audited prices that match the provider's page | 99% or more | from the weekly audit (D3) |
| Time to correct | From a wrong price to its fix | under 48 hours | from the weekly audit (D3) |
| Wrong price in a first place | Count per month | 0 | from the weekly audit (D3) |
| Coverage | Visits that land on a prebuilt page | 90% or more | waiting for the first signals file |
| Precision | Pair pages with a visit in 90 days | rising monthly | waiting for the first signals file |
| Regret | Unbuilt pairs beating the weakest page | under 5% of budget | waiting for the first signals file |
| Time to page | New offers, first seen to listed | under 10 days | same day; no offer first seen in 90 days |
| Funnel yield | Scouted providers that get listed | reported | from the funnel's table (D2) |
| Time awaiting approval | End of shadow to a decision | under 7 days | from the funnel's table (D2) |
| Approval agreement | Decisions a rule would match | 95% before automating | from the funnel's table (D2) |
Price accuracy
- Counts
- Audited prices that match the provider's page
- Target
- 99% or more
- Today
- from the weekly audit (D3)
Time to correct
- Counts
- From a wrong price to its fix
- Target
- under 48 hours
- Today
- from the weekly audit (D3)
Wrong price in a first place
- Counts
- Count per month
- Target
- 0
- Today
- from the weekly audit (D3)
Coverage
- Counts
- Visits that land on a prebuilt page
- Target
- 90% or more
- Today
- waiting for the first signals file
Precision
- Counts
- Pair pages with a visit in 90 days
- Target
- rising monthly
- Today
- waiting for the first signals file
Regret
- Counts
- Unbuilt pairs beating the weakest page
- Target
- under 5% of budget
- Today
- waiting for the first signals file
Time to page
- Counts
- New offers, first seen to listed
- Target
- under 10 days
- Today
- same day; no offer first seen in 90 days
Funnel yield
- Counts
- Scouted providers that get listed
- Target
- reported
- Today
- from the funnel's table (D2)
Time awaiting approval
- Counts
- End of shadow to a decision
- Target
- under 7 days
- Today
- from the funnel's table (D2)
Approval agreement
- Counts
- Decisions a rule would match
- Target
- 95% before automating
- Today
- from the funnel's table (D2)
Pair pages, by family
Pages built of the pairs that could have one, the budget, and which rule kept each page.
| Family | Pages | Budget | Kept by |
|---|---|---|---|
| /vps/vs | 325 of 325 | 400 | under budget 325 |
| /vps/compare | 400 of 1,116 | 400 | tenure 400 |
| /ai/compare | 250 of 636 | 250 | tenure 250 |
| /ai/vs | 20 of 3,268 | 20 | tenure 10, floor 10 |
| /gpu/compare | 75 of 811 | 75 | tenure 75 |
| /platforms/compare | 75 of 318 | 75 | tenure 75 |
| /platforms/vs | 20 of 36 | 20 | tenure 10, floor 10 |
/vps/vs
- Pages
- 325 of 325
- Budget
- 400
- Kept by
- under budget 325
/vps/compare
- Pages
- 400 of 1,116
- Budget
- 400
- Kept by
- tenure 400
/ai/compare
- Pages
- 250 of 636
- Budget
- 250
- Kept by
- tenure 250
/ai/vs
- Pages
- 20 of 3,268
- Budget
- 20
- Kept by
- tenure 10, floor 10
/gpu/compare
- Pages
- 75 of 811
- Budget
- 75
- Kept by
- tenure 75
/platforms/compare
- Pages
- 75 of 318
- Budget
- 75
- Kept by
- tenure 75
/platforms/vs
- Pages
- 20 of 36
- Budget
- 20
- Kept by
- tenure 10, floor 10
Capacity
Read weekly, with when to act.
| Measure | Today | Act when |
|---|---|---|
| Files per version | counted by every build; npm run capacity prints it | over 16,000; the build fails at 18,000 |
| Pair pages | 1,165 prebuilt of a budget of 1,240, in 7 families | a family over its budget fails the build |
| Requests a day | npm run capacity, from Cloudflare | the Worker's own pass 60,000 of 100,000 |
| Server errors | npm run capacity, from Cloudflare | over 0.5% of requests on any day |
| Largest paint, 75th percentile | npm run capacity, from PostHog | a top-ten page over 2.5 s for two weeks |
| Responsiveness, 75th percentile | npm run capacity, from PostHog | a top-ten page over 200 ms |
| Database | npm run capacity, from Neon | storage over 0.4 GB, or compute over 80 hours a month |
Files per version
- Today
- counted by every build; npm run capacity prints it
- Act when
- over 16,000; the build fails at 18,000
Pair pages
- Today
- 1,165 prebuilt of a budget of 1,240, in 7 families
- Act when
- a family over its budget fails the build
Requests a day
- Today
- npm run capacity, from Cloudflare
- Act when
- the Worker's own pass 60,000 of 100,000
Server errors
- Today
- npm run capacity, from Cloudflare
- Act when
- over 0.5% of requests on any day
Largest paint, 75th percentile
- Today
- npm run capacity, from PostHog
- Act when
- a top-ten page over 2.5 s for two weeks
Responsiveness, 75th percentile
- Today
- npm run capacity, from PostHog
- Act when
- a top-ten page over 200 ms
Database
- Today
- npm run capacity, from Neon
- Act when
- storage over 0.4 GB, or compute over 80 hours a month
How any one page came to be built: npm run why -- /vps/vs/contabo-vs-hetzner. See also how Kuraberu works.
How we count
The measures and their targets are section 8 of the design paper for discovery and ranking. A number this page prints is computed when the site is built, from the same files the pages are built from. A number another step publishes names the step: the weekly price audit, the provider funnel, or the weekly capacity report (npm run capacity), which reads Cloudflare, PostHog and Neon.
Coverage, precision and regret need the weekly signals file, the visits and clicks per page counted from our analytics. No section has one yet.
The page score chooses which pair pages are built: merit, how new a thing is, and, once there are enough of them, visits and clicks. A page stays at least 90 days once built, and while a family has fewer candidates than its budget, every candidate has a page. Until the daily refresh records which pages it built, the pages built before the score keep their place.