How every figure is sourced, dated, and computed
StartupPerks exists to be trusted. Every value, eligibility rule, and score on this site traces to one place, the provider's own official page, carries a confidence tier, and is stamped with when we last checked it. This page is the full account, so you never have to take a number on faith.
What a confidence dot means
Every figure carries one of four tiers. The dot's color is the same on every surface, so a claim's KIND is always legible at a glance. Color is never used alone: a text label always accompanies it. A jobs site we also built applies the same idea to pay, and AITraining.jobs' pay methodology explains its four labels, from a rate read live from a job feed down to an estimate.
Confirmed against the provider's own official program page.
Stated by the provider, not re-confirmed this cycle.
From a public directory or announcement.
Derived by StartupPerks (the value score); shown with a leading ~.
Editorial as-of vs live retrieval
A comparison site fails the moment it presents a stale figure as current. We keep two clocks and never let one stand in for the other.
The curated, code-owned catalog (program identity, categories, this copy, the source registry) is dated as of Nov 14, 2026. It changes only when the code is edited, and it is never shown next to a live number.
Each program's value and eligibility carry their own fetched_at from the last time we re-checked the provider's page, and read as verified X ago. Right now the catalog shows:
Of the 2,903 programs, 2,882 have had their terms checked on the provider's own pages, by the automatic re-read or by our research, and 2,881 of them within the last 60 days. A program page shows its own last check; one not yet checked at its source says so rather than borrowing another program's date.
A verified value whose re-check has not run within its cadence is visibly downgraded (its dot turns amber, then red) rather than shown as confidently live. A stalled pipeline surfaces as stale, never as silence.
When a provider's own site shows a program has closed, or never offered it, the program leaves the catalog and its page says so, with the evidence and the date we read it. They are listed in startup programs that ended.
Every read is kept with its date. When a read no longer states the figure an earlier read stated, or finds a program has stopped or resumed taking applications, that change is listed on the program's page under Terms history and in what changed in startup programs. A reworded page is not a change.
The comparison guides, program briefs and articles quote each term word for word from the provider's page. A scheduled check re-reads those pages around the clock, with no rewriting: a quote is either still on its page or it is not. A pick whose every quote was found again says so with the day ("still stated on"), and a quote a page no longer states is re-read by a person and corrected. Each program's own value works the same way: a read keeps the words on the provider's page that state it, the check finds them again, and the program counts as checked on that day; only a program whose words disappear is read again in full.
How each figure on a program is built
Three of these are read straight from the provider's own page. One, the value score, is the only figure we derive, and we show exactly how.
Benefit value
VerifiedThe headline worth of a program ("up to $100K in credits", "90% off for a year"), taken verbatim from the provider's own official program page. It is compared on the same basis across every program, and every value on the site links straight to the page that states it.
Source: The provider's official program page
Eligibility
VerifiedWho actually qualifies: startup stage, funding or backing required, company-age and funding-raised caps, and geography, read directly from the provider's terms. We never infer eligibility, and a rule the provider does not state is shown as unknown, never as open.
Source: The provider's official program page
How to apply
VerifiedThe exact apply path and any referral requirement, linked to the provider's own application page. A referral link never changes the value or the eligibility rule.
Source: The provider's official program page
Value score & ranking
ComputedThe one number we compute ourselves. We weigh a program's benefit value, how broadly startups qualify, and how well its terms are confirmed, then map that onto a 0 to 100 scale so headline offers compare on one basis. It is a transparent proxy for how much a program offers, not a verdict on whether it fits you: a high score means a large, widely-open offer, not a guaranteed match.
Source: StartupPerks re-verification
The source registry
Every fact on the site names one of these sources. A Primary mark means it is load-bearing: the provider's own page, and the re-verification that keeps it current.
The source of record
The benefit value, eligibility terms (stage, funding, caps, geography), and how to apply, exactly as the provider states them. Every figure on the site names the specific page it came from.
Each provider (AWS, Google, Mercury, Neon, ...)
How we discover programs
Community-maintained lists of startup credits and deals, used to discover programs and cross-check that our catalog is complete. Never used as a value source: every figure is re-verified against the provider's own page.
Open-source community (GitHub)
New product and startup deals as they launch, used to spot programs to add. Value and eligibility are always confirmed on the provider's page before a program goes live.
Product Hunt
How we keep it fresh
A scheduled job reads program pages every three hours on a rolling schedule (the programs people search most at least every two weeks), updates each program's value and availability from the words its page states, and keeps every read with its date, so every surface shows the true age of the data and every change is listed. Each read keeps the words that state the value, and a second job re-reads those pages and every page the guides quote, hourly, confirming each quote is still there with no AI; only a program whose quote disappears is read by the model again.
StartupPerks
Neutrality, and the caveats we never hide
StartupPerks takes every value and eligibility rule from the provider's own page, and a partner never changes a value, an eligibility rule or a source. We show a value only when we have a real source for it; when we do not, we show an honest gap, never an invented number. AITraining.jobs, Founden, AIData.Broker and GenMagic are run by the same company as StartupPerks: their listings are free, ranked by the same rules as every other program, and labeled "Same owner as StartupPerks" wherever they appear.
Authorship and expertise
Every value and eligibility rule on StartupPerks is drawn from the provider's own startup-program page and names it. We publish what the source says, cite where it came from, and stamp when we retrieved it.