Roughly one in four live products meets a bar most enterprise buyers already enforce.
Every founder says the product is production-grade. We measure whether it is — from the outside, deterministically, on 5,317 launched services, and without the company’s cooperation. That last part is why it works for diligence: a target has no way to prepare for it.
Security scores across every live web product we measure. The shape is not a bell curve — it is two populations. One mass sits at 20–45: no Content-Security-Policy, no HSTS, defaults left as shipped. A second sits at 80–100, where the team did the unglamorous work. Very little sits between them.
n = 5,086 live web products, Security frame, measured from HTTP headers and transport. Red bars fall at or below the median of 45.
The gap is binary because the work is binary. A team either wired up headers, consent and a real 404, or it did not — and that decision predicts far more than the demo does.
Six categories, from developer tools to consumer productivity, and the median Security score is 45 in all of them. The 90th percentile is 100 in all of them too. The spread lives inside each category, not between them.
Amber dot: category median. Green dot: 90th percentile. Bar: the range between them.
For a fund this is the useful finding. “It’s an AI agents company” predicts nothing about engineering quality. The specific team does — which is exactly the variable diligence is supposed to resolve, and the one a pitch cannot settle.
Point the API at a list of domains and apply the bar your later-stage co-investors will apply anyway. Most of a pipeline does not clear it.
Filters run against the live catalog: Security ≥ 70, then Privacy ≥ 50. One call, no vendor contact.
Roughly one in four live products meets a bar most enterprise buyers already enforce.
Add consent and a reachable policy and the field halves again.
Nothing is requested from the company. The measurement is of the public surface it already ships.
Security against Performance for one category. If polish predicted rigour these points would form a line; they form a cloud. The product that loads instantly in the demo is as likely as any other to be shipping without headers.
34 AI & Agents products, real scores. Hover any point for the name. Green: clears a Security 70 policy. Red: does not.
This is the chart to open in an IC meeting. It puts the target somewhere specific rather than somewhere flattering, and the evidence behind every point is a public page anyone in the room can open.
Every service is re-measured on a schedule, so engineering quality becomes a series rather than a snapshot. A board deck claims the team hardened the platform last quarter; the series says whether it did.
Across companies re-measured so far, over a 100-day window.
The rest held flat or slipped — which is itself the signal you want on a portfolio call.
Re-measurement accrues weekly. Biggest movers so far: Mindwalk +60 · Salestrics +20 · Legit.Show +20.
We would rather show a small honest series than a smooth invented one. The history is 27 companies deep today and widens every week; a licence includes it from the day it exists for the names you care about.
Per-frame scores are open and quotable — that is how the measurement travels. What a fund pays for is the shape you cannot assemble from single lookups.
| Field | Access | What it is for |
|---|---|---|
| Per-frame scores | public | Screen one company, cite a number in a memo, link the evidence page |
| Category benchmarks | public | Place a target against its market, as in the charts above |
| Overall score | licensed | One number to rank a pipeline or a portfolio by |
| Score time series | licensed | Post-investment monitoring, regression alerts, quarterly reporting |
| Bulk index | licensed | Join 5,317 services against your own CRM or pipeline |
| AI attention | licensed | Which answer engines read a product, and how often — an early discovery signal |
Open any product page and read the frames and the evidence behind them. If the shape is useful for your pipeline, we will run your current watchlist and send back the screen.
Ask for a pipeline screenOr start free: the query API · how the seven frames are measured · the published reports