Background listening: why the system stays quiet
A false positive is more expensive than a missed signal: it burns a relationship with a prospect and erodes trust in the system. That’s why RevSign OS operates silently by default and only proposes an action once a signal clears three independent checks, in this order: the intent score, the conflict filter, and the no-repeat window.
Intent threshold: the rule
The intent score is computed from technology-stack category deltas (a tool adopted or removed within a category, with a fixed weight per category and a 0.5 multiplier for removals), capped at 100. That score is compared against sequence-intensity bands, not a single binary cutoff:
| Score | Sequence intensity | Touches |
|---|---|---|
| 0–49 | Low priority | 2 (e.g. email + LinkedIn) |
| 50–79 | Medium priority | 3 (email + LinkedIn + call) |
| 80–100 | High impact | 3, higher density |
The band is selected as the highest one the score still satisfies —it isn’t three isolated cases, it’s an ordered lookup over a configurable band table.
The conflict filter
Before being proposed, every signal is checked against a company-domain exclusion list. If the domain has an active entry on that list, the signal is marked suppressed and discarded without reasoning, without asking —it’s never presented to the human operator as a decision to make.
Classifier agents: functional fit, not declared title
A contact isn’t prioritized by the title they declare on their profile. It’s evaluated by functional fit to a buyer persona, with a three-state output contract: clear fit, plausible fit, or rejected. Contact details (email, phone) aren’t revealed until that fit is clear or plausible —a late-reveal pattern that avoids spending enrichment budget on contacts that won’t qualify.
| Evidence type | Examples | How it’s used |
|---|---|---|
| Declarative | LinkedIn title, press release, self-report | Input to the classifier, never the output |
| Business event (with confidence) | Funding round, layoffs, product launch, leadership change | Classified with an explicit confidence level: high, medium or low |
Noise isolation: the no-repeat window
A signal of the same type for the same domain within a 60-day window from the last detection is not re-inserted —it’s discarded outright. This window is a single constant applied across two layers of the system to make sure no prospect gets the same sequence twice for the same reason in a short span.
What “operational silence” means in practice
| Scenario | Signal emitted? | Why |
|---|---|---|
| Score 85, clean domain, no prior signal in 60 days | Yes, high-impact band | Clears all three checks |
| Score 90, domain on active exclusion list | No | The conflict filter suppresses it before intensity scoring |
| Score 60, same company had a same-type signal 20 days ago | No | Inside the 60-day window, discarded |
Frequently asked questions
- What happens when a signal repeats domain and type inside the cooldown window? It’s discarded instead of inserted: the system never fires the same signal twice for the same domain within 60 days of the last detection.
- Does a high score guarantee an action? No. The score sets the outreach intensity band (2 or 3 touches), but the signal has to clear the conflict filter and the cooldown check first.
$ curl revsign.os/docs/validation-pipeline
→ {
title: "Validation Pipeline: How a False Positive Gets Filtered",
answers: "How is a false positive filtered out of a B2B intent signal?",
schema: [TechArticle, FAQPage]
}