Caller Reputation Monitor

Caller ID Reputation Monitor

Keep the user's outbound dialing numbers (DIDs) out of carrier spam labels: register them (prevention), monitor them daily (detection), rest and remediate the ones that degrade (recovery). Built for SmrtPhone users doing REI cold calling, but the concepts apply to any dialer.

The core insight: carriers never tell you a number is flagged. The truest signal you can get for free is your OWN answer rate cratering on a specific number. This skill reads that straight from the user's SmrtPhone call log. No Telnyx account, no port, no monthly fee for the default setup.

What is in this skill

• references/quick-start.md - the 15-minute setup guide. START HERE for a new user.
• references/registration-and-remediation.md - why numbers get flagged, the free registration walkthrough (Free Caller Registry), and the runbook for clearing a number that is flagged right now.
• references/reputation-workflow.md - the full design: signal layers, fusion rules, the number lifecycle state machine, data model.
• references/telnyx-api-contract.md - verified Telnyx endpoints and fields (only needed for the optional carrier-grade upgrade).
• references/methodology.md - the deep research behind the whole system (carrier analytics engines, STIR/SHAKEN, healthy call-behavior targets, sources).
• scripts/ - the complete working monitor. Stdlib-only Python except the two Playwright helpers.

How the monitor works (know this before helping)

Two signal layers, fused as worst-of:

L1 healthy targets (SmrtPhone's own published bands): ASR >= 30%, ALOC >= 30 seconds, short calls (under 6s) <= 15% of outbound. A breach makes a number at least Watch, never Flagged on L1 alone. Below 15 dials in the window the sample is too small and L1 abstains (protects warm-up numbers).

L2 mapping (if enabled): spam_category non-null (a real label) = Flagged; spam_risk high = Flagged; medium = Watch; low = Clean.

Trend escalation: 3 non-Clean reads in a row = Flagged even if no single read is critical.

Lifecycle per number: NEW -> WARMING (ramp 30 -> 75 -> 125 dials/day over 3 weeks) -> ACTIVE (cap 75/day) -> WATCH (cap 37/day, stays dialable so it can clear itself) -> RESTING (benched 30 days) -> RECOVERED back to ACTIVE, or RETIRED after 3 consecutive surviving flags. All tunable in config/thresholds.json (inline notes explain every knob).

Outputs per run: output/caller_reputation.html (dashboard), output/active_pool.json (recommended dial pool with per-number caps, for loading the dialer), optional Slack digest via SLACK_WEBHOOK_URL.

Helping a user set up (first session)

Follow references/quick-start.md. The sequence:

1. Copy the scripts/ folder somewhere permanent on their machine (it writes state next to itself). Python 3.10+ required; Playwright only for the two helper scripts.
2. python smrtphone_login.py - headed browser, they log into SmrtPhone once, session saved locally.
3. cp config/numbers.example.json config/numbers.json, set business_name, then python pull_smrtphone_numbers.py --commit to auto-fill their real numbers.
4. Backdate added in numbers.json for numbers that have already been dialing for weeks, otherwise the monitor treats them as brand new and puts them in warm-up.
5. python monitor.py (dry run, no writes) to sanity check, then python monitor.py --commit for the first live scan.
6. Open output/caller_reputation.html with them and read the results.
7. Register every number at freecallerregistry.com (free, one form, biggest prevention lever). Walk them through references/registration-and-remediation.md.
8. Optional: schedule the daily run (Task Scheduler command in quick-start; use Register-ScheduledTask, never schtasks /TR on a path with spaces, and disable the battery-kill defaults on laptops).

Interpreting results for a user

• Clean + ACTIVE: dial normally, respect the cap in active_pool.json.
• Watch: early warning. Keep dialing at the reduced cap. Check list quality (wrong-number rate) and daily volume on that number first; those are the usual root causes.
• Flagged / RESTING: pull it from the dialer pool now. Re-submit it at freecallerregistry.com (free, unlimited resubmits) and follow the remediation runbook. It rests 30 days; resting alone does not clear a carrier flag, the resubmission + fixed root cause does.
• Low sample (abstain): under 15 dials in the window, no judgment. Normal for new or lightly used numbers.
• A number keeps re-flagging (3 strikes): retire it. But remediation beats churn: a fresh number needs a 2-3 week warm-up and recycled numbers can inherit a prior owner's spam history, so do not replace numbers casually.

Key coaching points backed by references/methodology.md: three private analytics engines label calls (Hiya for AT&T, First Orion for T-Mobile, TNS for Verizon) and they do NOT share scores; STIR/SHAKEN attestation is authentication, not immunity; steady volume with no spikes and no dead days is what healthy traffic looks like; even 5-10 consumer complaints can tip a low-volume number.

The optional Telnyx upgrade (L2)

Only mention if the user specifically wants the carrier's literal label or automated remediation. It requires porting their numbers into their own Telnyx account, Level 2 verification, and $100/month flat. Setup is telnyx_setup.py (register -> loa -> upload-loa -> enable -> status -> associate) with the enterprise identity from config/enterprise.example.json. Details in quick-start section 7 and references/telnyx-api-contract.md. For most users L1 is the product; skip this.

Troubleshooting

• "Session expired" or no L1 data: re-run python smrtphone_login.py. The monitor idles cleanly (no crash) when the session is missing.
• Short-call % reads ~0 on SmrtPhone: known limitation, the SmrtPhone log's duration field does not capture sub-6s connects. ASR and ALOC carry the detection; this never false-flags.
• Run "succeeded" but nothing changed: the failure mode to check is a silent no-op. Verify state/numbers_state.json has today's last_checked and the dashboard file is fresh. The monitor self-checks this and alerts to Slack if configured.
• Everything shows WARMING: added dates in numbers.json are today. Backdate them for established numbers.

Safety rails

• Never put API keys or webhooks in code or config that gets shared; they belong in a local .env next to the scripts (SLACK_WEBHOOK_URL, and for L2 only: TELNYX_API_KEY, TELNYX_ENTERPRISE_ID).
• smrtphone_state.json is a live login session. Treat it like a password: never commit, upload, or share it.
• The monitor is read-only against SmrtPhone. The only billable actions anywhere are the opt-in Telnyx L2 calls, and fresh L2 rechecks only fire on numbers a free layer already degraded, capped per run.
• This system exists to help legitimate businesses keep compliant outbound calling healthy. Pair it with real list hygiene and DNC compliance; it is not a way to evade labels earned by bad traffic, and remediation fails if the root cause is not fixed.