Deliverability in NimbusOS is not a page you check after a campaign underperforms. It is a live brain that scores every inbox, every domain, and the entire fleet; pauses campaigns that cross thresholds; and alerts humans when intervention is needed. This article covers the Nimbus Deliverability Score (NDS), the health models, the alert system, and how the brain operates in automated and supervised modes.
The Nimbus Deliverability Score (NDS)
NDS is a 0-to-100 composite score. One number that rolls up the state of an inbox, a domain, or the whole fleet. Most competitor tools surface separate metrics for reputation, bounce rate, complaint rate, and authentication. NimbusOS combines them into one.
NDS components:
- Inbox placement rate. Percentage of sends that land in the primary inbox (not promotions, not spam).
- Reputation signal. ISP-reported reputation from Gmail Postmaster, Microsoft SNDS, Yahoo FBL.
- Authentication health. SPF, DKIM, DMARC alignment pass rate.
- List hygiene. Bounce rate, complaint rate, unsubscribe rate.
- Engagement quality. Open rate and reply rate trends.
Each component is weighted. The default weights favor authentication and reputation signals because those are the hardest to fix after the fact.
NDS bands:
- 90 to 100: Excellent. Full cold outreach clearance.
- 75 to 89: Good. Cold outreach allowed with standard throttles.
- 60 to 74: Moderate. Nurture only. Cold outreach gate blocks.
- 40 to 59: Poor. Warmup only. Fleet or inbox needs attention.
- 0 to 39: Critical. Immediate intervention required.
The Three Health Models
NDS rolls up from three underlying models.
InboxHealthScore
Per inbox daily snapshot. Captures the state of a single EmailAccountHealth on a given date. Fields:
overall_scorereputation_scoreauthentication_scoreengagement_scorelist_hygiene_scoredaily_send_countdaily_bounce_countdaily_complaint_count
One row per inbox per day. Time series queries read from this table. The per-inbox page shows the trend over the last 30 days.
DomainHealthScore
Per domain daily snapshot. Aggregates across all inboxes on the domain. Fields similar to InboxHealthScore but at the domain level.
A domain with 5 inboxes might have one inbox at poor health and four at excellent. The DomainHealthScore is the weighted average, and the per-domain issue is whether the one poor inbox is poisoning the domain reputation.
NimbusDeliverabilityScore
Fleet level daily snapshot. The single NDS number that describes the whole sending fleet. Written daily by compute_nimbus_deliverability_score Celery task.
The fleet NDS is what the campaign resume gate reads. A fleet NDS over 80 is the signal that the fleet can safely launch new cold outreach campaigns.
The Deliverability Brain
Six Celery tasks run on schedule and keep the deliverability view live.
- inbox_health_score_daily (daily 3 AM UTC). Computes
InboxHealthScorefor every active inbox. - domain_health_score_daily (daily 3:30 AM UTC). Computes
DomainHealthScore. - compute_nimbus_deliverability_score (daily 4 AM UTC). Computes fleet NDS.
- sync_isp_monitoring (every 6 hours). Pulls Gmail Postmaster, Microsoft SNDS, Yahoo FBL.
- evaluate_deliverability_alerts (every 15 minutes). Checks thresholds, fires alerts.
- check_domain_rotation (daily). Evaluates whether any domain needs rotation.
The 15 minute alert check is the fastest-moving loop. A campaign that suddenly spikes in bounces triggers an alert within 15 minutes.
Auto-Pause Thresholds
Default auto-pause thresholds on active campaigns:
- Bounce rate over 2 percent in a 24 hour window: cold outreach pauses.
- Complaint rate over 0.1 percent in a 24 hour window: cold outreach pauses.
- Unsubscribe rate over 2 percent in a 24 hour window: cold outreach pauses.
- Inbox NDS dropping by more than 10 points in 24 hours: inbox pulled from rotation.
Thresholds are campaign-type aware. Nurture has looser thresholds. Re-engagement has its own profile.
Pauses are not destructive. The campaign moves to paused status, pending sends stop, and the human is notified. Resume is manual after the human confirms the cause is addressed.
The Alert System
DeliverabilityAlert rows capture every alert fired. Fields:
severity: info, warning, criticalalert_type: bounce_spike, complaint_spike, reputation_drop, blacklist_hit, authentication_failure, inbox_pulled, campaign_pausedsubject_ref_typeandsubject_ref_id: what the alert is about (inbox, domain, campaign)triggered_atresolved_atsummaryrecommendation
Alerts route to:
- In-platform alerts feed
- Email to workspace alert recipients
- Slack webhook if configured
- Webhook endpoint if configured
Critical alerts cannot be dismissed without marking the underlying issue resolved. This prevents silent ignoring of serious reputation events.
The Campaign Resume Gate
Before a cold outreach campaign launches, the resume gate runs. The gate has three hard conditions:
- At least 20 inboxes at stage 3 or higher.
- At least 25 percent of the fleet at stage 3 or higher.
- 7 day fleet bounce rate under 2 percent.
All three must be true. Failure on any one blocks the launch with a specific reason.
The gate is stage-based, not day-based. An inbox that has been active for 30 days with only 100 warmup sends is still at stage 2 and does not count. This prevents the common failure of "my fleet is old enough, why am I blocked".
See Email Warmup for the full stage engine.
Domain Rotation
When a domain's reputation is falling, the brain can rotate to a backup domain. Rotation is automatic in autopilot mode and supervised in manual mode.
Rotation logic:
- If
DomainHealthScoredrops below 50 and does not recover within 7 days, flag for rotation. - If any inbox on the domain hits a blacklist and does not clear within 24 hours, flag for rotation.
- If two inboxes on the domain pause within 24 hours, flag for rotation.
In supervised mode, flagging fires an alert. In autopilot mode, the brain moves active campaigns to a healthy backup domain and retires the unhealthy one.
DomainRotationLog records every rotation event with the reason and the replacement domain.
Blacklist Monitoring
The brain checks every sending domain against 20+ public DNSBLs (Spamhaus ZEN, SORBS, Barracuda, UCEPROTECT, and others) every 6 hours. A listing fires a critical alert.
Some DNSBLs are more reliable than others. UCEPROTECT level 2 and 3 have high false positive rates and fire informational alerts rather than critical. Spamhaus ZEN is authoritative and fires critical.
The blacklist alert includes the listing URL and a suggested delist procedure.
ISP Monitoring Integration
The brain reads three ISP sources.
Gmail Postmaster Tools. Domain reputation, IP reputation, authentication, delivery errors. Requires domain verification through Postmaster and API credentials.
Microsoft SNDS. Per IP reputation, complaint rates, trap hits. Requires IP whitelisting at Microsoft.
Yahoo Feedback Loop. Complaint events from Yahoo recipients.
Not every workspace has all three configured. The brain uses whatever is available. Missing sources downgrade the confidence on the ISP reputation component but do not prevent NDS computation.
See ISP Monitoring for detailed setup.
Operating Modes
The brain has three operating modes.
Supervised. Default. Brain evaluates and alerts, but every action requires human approval. Appropriate for conservative senders and newer workspaces.
Autopilot. Brain executes approved action types directly. Human sees a log and can reverse. Appropriate for mature workspaces with good data quality.
Manual. Brain alerts only; no action proposals. Appropriate for very unusual sending patterns that do not fit the brain's training.
Mode is per campaign. You can have a mix: tier A campaigns in supervised and generic nurture in autopilot.
Platform-Wide Intelligence
The brain reads platform-wide signals via PlatformDeliverabilityIntelligence. This captures aggregate patterns across every fleet on the platform, anonymized.
Examples of signals the brain uses:
- "This week, Gmail is filtering Sendgrid-origin mail more aggressively than normal."
- "Microsoft recently updated spam scoring on first-time-sender domains."
- "Major Outlook outage between 2 and 4 PM UTC caused widespread delivery delays."
These signals feed back into per-workspace scoring so a brief platform-wide event does not tank your NDS unfairly.
Working with the Deliverability Dashboard
The dashboard at /deliverability has four tabs.
Overview. Fleet NDS, top inboxes needing attention, recent alerts, trend over 30 days.
Inboxes. Per inbox NDS with stage, send volume, bounce rate, complaint rate. Sortable and filterable.
Domains. Per domain NDS with inbox count, aggregate health, rotation history.
Alerts. Full alert log with severity filter, resolved status, and resolution notes.
The Alerts tab is where you spend most of your time during an incident.
Troubleshooting
"NDS dropped but I cannot see why"
Click into the NDS detail. It shows per-component scores and which component dropped most. Usually one component explains most of the delta.
"Alert keeps firing even though I fixed the issue"
Alerts re-evaluate every 15 minutes against live data. A bounce rate alert persists until the 24 hour window rolls past the offending period. Wait or manually resolve.
"Brain paused my campaign and I disagree"
The pause action is reversible. Resume from the campaign page. If the pause was wrong, check the alert detail for the specific metric the brain read; you may find the metric was spiking briefly due to another cause.
"NDS at 65 but my campaign launched"
NDS bands are guidance; the cold outreach gate uses specific stage and bounce rules, not NDS directly. A fleet with NDS 65 could still pass the gate if the 20 inbox and 25 percent rules are met and bounce rate is fine.
Frequently Asked Questions
How is NDS different from Gmail Postmaster's reputation score?
Postmaster reports Gmail's view of your domain. NDS is a cross-ISP aggregate that includes Postmaster, SNDS, Yahoo FBL, and NimbusOS-internal signals. NDS is generally more conservative.
Can I turn off auto-pause?
Yes, per workspace. Not recommended. Auto-pause is the fastest intervention the brain can take. Turning it off means reputation damage proceeds until a human intervenes, which could be hours.
Does the brain see my email content?
For deliverability purposes, yes. Spam forensics reads subject and body to find trigger words. The data stays inside your workspace and is not shared.
Why does my fleet NDS lag my perceived deliverability?
NDS is a 7 day rolling measure by default. Sudden improvements take a week to reflect. You can switch to 24 hour rolling in Account Settings for a more responsive (but noisier) signal.
What to Read Next
Useful next pages after this one: SPF, DKIM and DMARC for the authentication that feeds the score, Bounce Intelligence for the bounce handling subsystem, and ISP Monitoring for the per-ISP dashboards.