The evidence
standard.
The rulebook the research agent runs under. It is written down so that when a lead is wrong you can tell us which rule failed — and so that a claim on this site can be checked rather than believed.
Rubric version
v4
Signal window
120 days
Intent half-life
30 days
No evidence, no lead.
Before scoring, before enrichment, before anything else: a company with no dated, sourced signal is dropped. Not downgraded — dropped. This is the single biggest reason Antiqore returns fewer rows than a list vendor, and the reason those rows survive scrutiny.
Confirmed
Full weightA named source states it directly.
A funding announcement in a publication, a role on the company's own careers page, a statement in a filing or a transcript. The claim and the page that makes it are the same thing.
Signal
Reduced weightA source implies it, and the inference is one step.
A tool detected on the marketing site implies the stack. A competitor named in a job description implies the incumbent. The source is real; the conclusion is ours, and it is labelled as ours.
Hypothesis
Cannot lift a tierA pattern suggests it, and nothing states it.
Headcount growth inferred from a careers page over time. Kept because it is sometimes the thing that matters, shown because you should be able to discount it. It can never lift a company to tier A on its own.
Every signal is stored append-only with the date it was observed, so the record of what we knew and when we knew it cannot be quietly rewritten later.
Score the three things separately.
A single blended number is unfalsifiable. Three numbers can each be wrong in a specific way, which means each can be corrected.
Fit
65% of the composite
- Industry and vertical against your defined segments
- Employee band and revenue band
- Region, and whether you can actually serve it
- Regulatory drivers that make you relevant
- Tools already in place that you sit beside or replace
- Anti-signals: what rules a company out no matter what else is true
Intent
35% of the composite
- Driven by the strongest single signal, not the count of them
- Freshness decayed on a 30-day half-life
- Signals older than 120 days leave the window entirely
- A hiring post from last week outweighs three from spring
Confidence
Caps the tier
- How many independent sources agree
- The evidence tier of each one
- How recently each was observed
- Whether the contact's address is published rather than derived
- Tier A needs confidence of at least 0.5 — a great story on one thin source stays at B
How the tier falls out
composite = fit × 0.65 + intent × 0.35
Tier A needs a composite of 75 or more and confidence of at least 0.5. Tier B needs 55. The band — hot, warm, cold — is read off intent alone, because that is what the question “call them now?” actually asks.
Say what an address really is.
“Verified” is the most abused word in this category. We do not use it. Every address carries a number from 0 to 100 and a reason for that number.
- 95–100
- Published
- The address appears on a page we can link to — a team page, a filing, a conference listing, a public profile.
- 60–80
- Documented format
- We observed the company's address pattern across several known addresses and applied it. The sample count is shown with the score.
- Below 60
- Weak
- The pattern is thin or inconsistent. Shown so you can decide, never presented as a working address.
- Never
- Guessed
- We do not invent an address and label it verified. For senior people we give you the professional profile instead, because that is the channel that works.
Consent is a state, not a checkbox.
The same standard applies to the people who land on your site. What we collect depends on what they allowed, and withdrawal deletes rather than flags.
Before consent
Only what is needed to make the page work and to record the consent decision itself. No memory recall, no personalisation, no marketing signals.
Global Privacy Control
Honoured automatically, without asking again. A browser that sends the signal has marketing categories forced off.
Withdrawal
Purges that visitor's events and memories. The consent record itself is kept, because we must be able to show what was allowed and when.
Erasure
A visitor can be forgotten entirely on request, through an endpoint you can wire to your own privacy page.
Session replay
Input fields are masked at capture, not at playback — the keystrokes never reach the server. Retention is a number you set.
Research subjects
Business contact data from public professional pages, used for business communication. Suppression is permanent and applies across every programme.
If a rule here conflicts with what the product actually does, the rule is wrong and we want to know. Tell us and we will fix one of the two.
Tell us