The call infrastructure playbook

What this is. This is the field guide to the layer underneath the phone call. That layer decides whether your number rings as a name or as SPAM LIKELY, before a word is spoken. It is the exact counterpart to the deliverability guide: same shape of problem, different channel. It covers what the scoring engines watch, and what they refuse to publish. Why a brand new number can arrive already flagged. How number reputation gets built and lost. What to do when you get labeled. And which pieces of the standard advice do not survive contact with the primary sources. Read it before you buy a dialer, because most of the decisions that matter here are made before the first dial.

01Three engines, not one carrier#

Most people picture a single gatekeeper deciding whether their call gets labeled. There isn't one. There are three private analytics companies, they are wired to different carriers, and they disagree with each other constantly.

T-Mobile uses First Orion. AT&T uses Hiya. Verizon uses TNS. The mapping matters because reputation does not travel. Each engine keeps its own score, runs its own registration portal, and reaches its own verdict about you. A number can be perfectly clean on one network and labeled on another, and neither engine will tell the other. When somebody says "my number is clean," the only honest reply is: clean according to whom.

How far apart they are is not a matter of opinion. In 2023 the caller-identity firm Numeracle ran a test and filed the results with the FCC. They took a hundred numbers from a freshly allocated block in the 930 area code. None had ever been assigned to a customer. None had ever placed a single call. Then they checked how the engines saw them.

Hiya labeled thirty-two of them as spam. TNS labeled twenty-four as potential spam. First Orion labeled none. Of Hiya's thirty-two and TNS's twenty-four, only ten were the same numbers.

Sit with that. Those numbers had no call volume, no duration profile, no answer rate, no complaints, no history of any kind. Roughly a quarter to a third were condemned anyway, and the two engines that condemned them mostly disagreed about which ones. That is the environment you are operating in. And it reframes the whole problem. You are not starting from zero. You are starting from wherever the engines happen to have you. And you do not get to see the file.

Numeracle's filing made a second point worth keeping in mind. The engines sell remediation for the labels they apply themselves. The filing cites one charging over forty thousand dollars a year to register and remediate a hundred numbers. That is not a reason to dismiss them. It is a reason to read their advice knowing who profits from the problem.

02What they really watch#

Nobody publishes a threshold. That is the single most important fact in this guide, and almost every article you will read violates it.

The FCC comes closest to an official list. In a 2023 order it encouraged providers to consider "common indicia of illegal calls including, but not limited to: call duration; call completion ratios; large bursts of calls in a short time frame; neighbor spoofing patterns; and sequential dialing patterns." Five signals. Zero numbers. And the FCC declined to set numbers on purpose, reasoning that each campaign has unique qualities better handled case by case.

The most candid account from an engine came under oath. First Orion testified to a congressional subcommittee. They described labeling calls on "a sudden burst of incoming calls from a previously unknown number, a high failure to ring through rate, very short duration after the call is answered." Alongside consumer complaints. And, notably, Do Not Call violation data from the FTC. Then they said the line that should govern how you think about all of this:

Hiya went furthest in September 2025 when it launched a reputation product that grades four factors from A to D. Maturity: are your numbers well established, and do you avoid rotating them. Connection: do recipients choose to answer. Engagement: do they stay on the line once they do. Sentiment: do they refrain from complaining and blocking. Their developer documentation adds a useful detail. Maturity gets computed from traffic even before you register, and it is accurate immediately. The other three stay empty until enough calls accumulate, sometimes for weeks.

TNS, for its part, publishes nothing. Its product sheets are marketing. Any claim about what TNS weighs, beyond the unallocated and invalid numbers it says it screens, is not sourceable to TNS.

So the honest picture is roughly a dozen known signals, three engines that demonstrably disagree, and no published threshold anywhere. Which brings us to the numbers you have certainly been told.

03The numbers that are not real#

You will read that you should stay under a hundred and fifty dials per number per day. You will read that calls under fifty seconds look suspicious. That you should let it ring thirty seconds. That recovery from a flag takes two to four weeks. That one spam label costs you sixty to eighty percent of your answer rate.

None of those figures traces to a carrier, an engine, or a regulator. They come from vendor blog posts and content farms, unattributed. Take the most widely repeated one, the hundred-and-fifty-dial ceiling. It appears in a dialer vendor's 2025 post, described as "industry best practice," with nothing cited. It is absent from the same vendor's own best-practice post two years earlier. A number that materializes in marketing content without a source is a marketing artifact, not a discovered threshold.

This matters more than it sounds. Build your operation around a fabricated ceiling and you get false confidence. You stay under a hundred and fifty dials and assume you are safe. Meanwhile the thing that flagged you was a burst pattern. Or a run of short calls. Or a complaint from the very first person you called.

04Maturity beats rotation#

Here is where the standard advice inverts, and it is the most actionable finding in this guide.

The folk wisdom says rotate. Spread the load across many numbers, retire them when they get hot, keep fresh ones coming. It sounds like the domain rotation that works in email, and the analogy is why people believe it.

The analogy is wrong, and Hiya says so directly. Their first graded factor is maturity, defined as whether you use established numbers without rotating. Their guidance says a number with little or no calling history is itself a red flag. It suggests a business changing numbers to evade labeling. Rotation is not a defense against the scoring engine. To the scoring engine, rotation is the behavior of somebody with something to hide.

And the Numeracle result explains why fresh numbers cannot save you anyway. A new number is not a clean slate. A quarter to a third of never-used numbers were already labeled before they placed a call. Rotating into a fresh number is rolling the dice, not resetting the board.

There is a second half to Hiya's guidance that makes this practical rather than despairing. A number becomes mature either by carrying higher volume, or by being seen making calls across multiple days and weeks even at very low volume. Time substitutes for volume. That means a small, stable, patiently-aged set of numbers is a legitimate strategy, and it is available to a team of any size.

So the build is: buy few, register them, warm them slowly, dial them consistently, and keep them for a long time. The instinct to churn numbers is the instinct to look like a spammer.

05Consistency, and why it is the one rule with teeth#

Of everything in the folklore, the consistency rule is the one with real support behind it.

Two of the FCC's five named indicia are about pattern rather than quantity: "large bursts of calls in a short time frame" and "sequential dialing patterns." First Orion's sworn description leads with "a sudden burst of incoming calls from a previously unknown number." None of those is a volume ceiling. They are all descriptions of shape. What matters is not how much you dial. It is how erratically.

Which means the practical instruction is not a number, it is a discipline. Dial roughly the same quantity every day. Twenty on Monday and a hundred on Tuesday is the exact silhouette the engines are built to catch. It is worse than a hundred every day would have been. Ramp a new number gradually rather than switching it on at full volume. Spread the day's calls across the day instead of firing them in a block. Keep weekends and holidays in mind, because a pattern that stops dead and restarts is also a pattern.

This is the same discipline the deliverability guide teaches for email volume, for the same reason. Automated reputation systems look for the signature of automation. Steady human-scale behavior is the thing that does not look like a machine.

06The other half of the score is the conversation#

Two of Hiya's four factors have nothing to do with your infrastructure. Engagement asks whether people stay on the line after answering. Sentiment asks whether they refrain from complaining and blocking. First Orion names "very short duration after the call is answered" as a labeling signal.

Read that as an operational fact rather than a compliance note. A rep who gets hung up on repeatedly is damaging the number, not just the day. Every call that dies in the first few seconds contributes to a duration profile that looks like a robocall. Every prospect annoyed enough to press the block button feeds a sentiment score.

This is where infrastructure stops being separate from the calling itself. You can register everything correctly, age your numbers patiently, and hold your volume flat. And still get labeled, because the person dialing opens badly and gets hung up on forty times a day. The cold calling guide covers the opener in detail. That is why it belongs in the same conversation as DNS records and attestation. The two are the same system.

The same logic applies to your list. Calls to disconnected and invalid numbers are named in FCC material as a signal. Bad data does not just waste dials, it degrades the number placing them. Data quality is a deliverability input on the phone exactly as it is in the inbox.

07Expect it, then remediate#

You will get flagged. Treat that as a cost of operating rather than a failure, because a quarter of numbers that have never placed a call are flagged already.

What matters is that labels are contestable. Every engine runs a registration and dispute process. The whole point of Section 01 is that you must do this three times. A clean result at one engine tells you nothing about the other two.

Two honest gaps, which this guide will not paper over. First, the exact submission paths still need documenting properly. The fields each engine requires, and the realistic turnaround. And the widely-quoted two-to-four-week recovery figure has no authoritative source behind it. Second, registration is not immunity. Numbers that have been registered still get flagged again when their calling pattern drifts from what was registered, or when enough people report them. Registration buys you a hearing, not a shield.

The operational habit that follows: check your numbers' status on a schedule rather than waiting for a rep to notice that nobody is picking up. By the time answer rate visibly drops, you have been burning dials against a label for days.

08What is still open#

An honest guide says where it stops. Four things in this area need work before anyone should act on them with confidence.

The remediation runbook. The step-by-step for each engine and each carrier, with the actual forms and current turnaround times. This is the most useful missing piece and it needs verifying against live portals rather than blog summaries.

Attestation. STIR/SHAKEN signs calls at three levels, A, B and C, and A-attestation is becoming the gate for positive branding rather than a defense against labeling. What must a business do to get A-attestation on its outbound numbers? And what does B or C cost in practice? Both need establishing from FCC and carrier sources.

The consent question. Some experienced operators hold that no purely cold call is fully legitimate. Their reasoning: carriers expect prior consent before a call counts as wanted. That conflates two different things, what the law requires and what a private scoring engine prefers, and they are not the same. Three things need establishing properly. The TCPA position on manually-dialed B2B calls. The state statutes that go further. And the current status of the consent rule struck down in 2025. This is the one section of this guide that should be reviewed by a lawyer before it goes to a client. Nothing here is legal advice.

Your own data. Close3x sits on connect-rate data that could settle the local-presence question directly, which no public source currently can. A controlled test of local presence against a matched control would be worth more than every vendor claim on the subject combined.