Operations
Demo mailbox, 8 messages, 11 work items. Organization is in collaborative mode.
Autonomous resolution
0%
§114 — of items the sender was authorized to receive
Human touch rate
73%
Items needing a person before they go out
Not-found rate
36%
Answered honestly rather than invented
Unauthorized disclosures
0
§114 — the one metric that must stay at zero
Running on the heuristic classifier
No model credentials are configured, so intent confidence is capped below every autonomy threshold in §38. The autonomous resolution rate above reflects that cap, not the system's ceiling. Set ANTHROPIC_API_KEY to see the real number.
By intent
| Intent | Policy | Threshold | Items | Sent |
|---|---|---|---|---|
| SHIPMENT_STATUS | autonomous | 0.95 | 6 | 0 |
| POD_REQUEST | autonomous | 0.98 | 2 | 0 |
| INVOICE_REQUEST | collaborative | 0.99 | 1 | 0 |
| RECORD_UPDATE_REQUEST | human_required | 1 | 1 | 0 |
| RATE_REQUEST | collaborative | 0.99 | 1 | 0 |
Autonomy readiness
§70 — the system can recommend enabling autonomy for an intent. It never enables it.
SHIPMENT_STATUS
Not yet — keep collecting evidence
- ○300+ processed requests
- ○95%+ human acceptance
- ✓0 unauthorized disclosures
- ✓98%+ shipment matching
Mailbox
Volume, and who it came from. System senders are counted separately from unknown ones on purpose - bounce notices and no-reply blasts would otherwise pile into a number people read as a security signal.
8
Received
0
Sent
7
Verified sender
0
Unverified sender
1
Unknown sender
0
System sender
0
Internal sender
Task handling
11
Open
0
Done
0
Agent replies sent
0
User replies sent
0
No reply sent
8
Awaiting review
“No reply sent” is deliberate silence, not a failure to answer - an email whose whole content is “thanks” is a completed conversation, and replying to it is how two inbox agents end up thanking each other indefinitely.
Tasks by status (11)
Tasks by detected intent
Age and resolution are separate columns because they describe different populations: age is the backlog, resolution is the finished work. Blending them makes a growing backlog look like fast service. Max sits beside average because a six-hour average with a thirty-day maximum is a six-hour service plus somebody’s forgotten claim.
| Detected intent | Count | % | Max age | Avg age | Max resolution | Avg resolution |
|---|---|---|---|---|---|---|
| Shipment status | 6 | 54.5% | 0s | 0s | 0s | 0s |
| POD request | 2 | 18.2% | 0s | 0s | 0s | 0s |
| Invoice request | 1 | 9.1% | 0s | 0s | 0s | 0s |
| Request for record update | 1 | 9.1% | 0s | 0s | 0s | 0s |
| Rate request | 1 | 9.1% | 0s | 0s | 0s | 0s |