Operations Overview
Live · last updated 4s ago · East US 2
Availability
99.94%
↑ 0.02% from yesterday
All services up
Reliability
3events / 24h
→ steady
MTTR 8m 42s
Performance
423ms p95
↓ 12ms from yesterday
2 services watch
Load
12queued
→ stable
Oldest 4s
Resource
67%
↑ 4% from yesterday
3 services warm
Token / Cost
$
142M tok · $89
↑ 18% from yesterday
5 over budget
Agentic
28risk
↓ 3 from yesterday
0 above threshold
Security
8vulns · 2H · 6M
→ unchanged
4 services
Administration
4chg/hr
↑ 1 vs baseline
Last 7m ago
Cost ●
$4.2K/day
62% of monthly budget
Top: judge inference
Upstream ●
2degraded
OpenAI · 73% timeout · 4m
1 customer site silent
Service Wall · 23 services
action-gateway
Customers
47 tenants · 12 T1 · 28 T2 · 7 T3 (filtered)
Tier
Risk
acme-dental-group
Tenant Health Score
42
Red
Driven by
1. Availability impact — 4 services down (−28 pts)
2. Agentic risk elevated — Risk Score 72 on Sentinel PRS (−18 pts)
3. Security — 1 Critical vulnerability open (−12 pts)
2. Agentic risk elevated — Risk Score 72 on Sentinel PRS (−18 pts)
3. Security — 1 Critical vulnerability open (−12 pts)
Services this tenant uses
Active Overrides
Tier 2 sampling: 80% (default 50%)
@sre-lead
7d remaining
PagerDuty route: enterprise-emergency (default standard)
@sre-lead
30d remaining
Capacity Planning
Busy-hour analysis · Erlang B/C calculators · cost forecasting
Scope
Period
Days
M T W Th F
Percentile
Busy Hour (Time-Consistent)
Identified
14:00–15:00 ET
62% of days in window
P90 BH Utilization
76% across all services
↑ 4% from previous period
Utilization by hour-of-day (P90 across days)
00040812162024
Resource Utilization at Busy Hour
ResourceServiceBH P90TrendHeadroom
CPU
action-gateway
↑
22%
Memory
action-gateway
→
38%
DB Pool
action-gateway
↑
13%
Worker Pool
action-gateway
↑
29%
LLM Conc
sentinel-prs
↑
11%
Cost Busy Hour — last 30 days
Peak Hourly Burn
$42/hr
at 14:00 (Tue/Wed/Thu pattern)
P90 BH Cost Rate
$38/hr
Daily projection at BH: $912
Top contributing categories at BH
judge_inference$14/hr(37%)
telemetry_storage_traces$11/hr(29%)
tier1_classifier_inference$7/hr(18%)
Assets — CMDB
Configuration items required to run the services · Azure-heavy + non-Azure · running cost retained weekly / monthly / yearly · per-customer COGS
Provider
Class
Scope
Cost window
AssetClassProviderRegionSupportsServesCost · monthState
The CMDB is discovered, then reconciled — Azure Resource Graph inventory diffed against the declared
asset_registry. Orphan = running but undeclared (cost leak / shadow IT). Declared-missing = declared but not found. Per-customer cost reuses the §13.5.3 volume-proportional rule; dedicated assets attribute 100% to their tenant. Seam: inventory + reconciliation ride a separate OSCAR asset/discovery seam; this surface joins that with the monitoring read contract's cost/health telemetry (query class C) — the contract itself carries no inventory records.Problem Management
Top problem signatures · last 7 days · sorted by total impact
Correlations
Triaged groupings from the upstream application · root cause + symptoms · intake from product app, customer portal, and monitoring · read-only view into Linear
Severity
Intake source
Status
Scope
This is a read-only lens. Triage, deduplication, correlation, notification, and Linear creation happen in the upstream application — the console renders the result and deep-links to the issue it opened. The dedup ratio (raw signals → correlations) shows that intelligence at work. Seam: Correlations / Problems ride the upstream Ticket-Triage feed — a separate read contract from the OSCAR monitoring contract; this surface joins, it does not own.
Service Levels — SLO / SLA
Reliability, performance, engineering delivery, and AI cost commitments — tracked overall and per-customer
Linear
Workspace: airiam-ops · live
Filters
GitHub
Organization: airiam · live
Filter
Administration
AI Trust Services · Cost Health · Schema · Configuration · Audit
AI Trust Services Dashboard
Golden set persona-classes covered
14
Total labeled turns
2,847
Last refresh
18d ago
Inter-rater agreement
0.79
Disputed Judgments queue
23 pending ●
Spot-check progress this week
167/280 (60%)
Judge Calibration (7-day κ)
Faithfulness
κ = 0.71
Hallucination
κ = 0.68
Relevance
κ = 0.62
Tool Selection
κ = 0.58
Planning
κ = 0.45
Refusal Correctness
κ = 0.71
Cost Health
Annual target
$480,000
YTD spend
$187K
Projected
$445K (93%) ●
Top category (30d)
judge_inference $11K ↑28% ●
Vendor concentration
Largest 32% ●
Tenants outside COGS range
4
Schema Health
Current OCRM schema
v1.4.2
Active deprecations
0 ●
Consumers on min-version
12 / 12 current
Schema changes last 30d
7 additive, 0 breaking
Last config change
7m ago by @sre-lead
Decision Register · v10
DR-01 through DR-14 with status and scope. Replaces v9's "all ten resolved" framing.
Telemetry Sources · v10
Upstream signal sources for the active deployment.
Dependency Registry · v10.1
External dependency catalog. Empirical state from integration traffic; vendor status as context; operator overrides where applicable.
Vendor-reported state shown for context only; empirical traffic is authoritative. Disagreement between sources is surfaced as a signal.
Recent Administration Events
Settings
Display · Notifications · Saved Views · Account
Display
Token unit default
Used for all Token category charts; per-chart override available
Density
Spacing across cards and tables
Screenshare-safe mode
Anonymizes tenant names in the UI
Notifications
Email digest
Daily summary of incidents and metrics
Mobile push
2 devices registered · iOS, Android
Quiet hours
22:00–07:00 ET · Override for severity ≥ HIGH
Saved Views
My Watchlist
5 tenants · Created Mar 12, 2026
Tier 1 Critical Only
Auto-updates based on tier definitions
Agentic Drift Monitor
Filtered to drift signals across all agentic services
Account
greg.kovalcik@airiam.com
Role: SRE Lead · Scope: All tenants · Last sign-in: 14:00:23 EDT (web, East US 2)
Active sessions
2 sessions: 1 desktop, 1 mobile iOS