INDUSTRY FIELD GUIDE
Wired telecom service incident data for AI dispatch review
A dispatcher has to distinguish a single service complaint from evidence of a wider incident.
NAICS 517111 · Wired Telecommunications Carriers
A dispatcher has to distinguish a single service complaint from evidence of a wider incident. Scale lists telecommunications among its public enterprise data partnership interests. The FCC’s outage reporting work shows that some network information has reporting and confidentiality considerations. Start with a simulated dispatch task and get authorization for the particular data use. Scale Data Partnerships · FCC outage reporting and confidentiality order
Make the escalation boundary visible
Create fictitious service tickets using generic service areas. Add one delayed network observation, a maintenance notice and a separate confirmation from a simulated operator. Record which facts were available when the dispatcher decided whether to open a shared incident. Hold out a pattern in which several tickets arrive together but belong to separate issues. Test whether the assistant asks for evidence before merging the tickets or escalating the case.
Use the timeline and evidence labels to test a routing or documentation assistant. Keep network control, restoration actions and formal regulatory reporting outside the task.
Use no real addresses or network topology
Exclude precise infrastructure locations, subscriber records, network diagrams, access credentials and detailed incident information. Get security and data-owner authorization for any actual export and keep the first sample fully fabricated.
Send one invented incident timeline and its routing rubric. Describe a dataset
Original Fiund exercise
Several complaints need a shared-incident link before they are merged
An isolated dispatch tabletop uses generic service areas, delayed observations and a fictional escalation rubric. It contains no subscriber, topology, network control or restoration task.
| Record | Source | What it records |
|---|---|---|
| WD1 | 09:00 ticket snapshot | Tickets T1 and T2 from practice area A report unavailable service. T3 from area B reports a different symptom. No shared incident confirmed. |
| WD2 | 09:05 received observation | Operator's synthetic observation made at 08:58 links T1 and T2 to incident I-A. It explicitly excludes T3; earlier maintenance notice concerns B on another day. |
| WD3 | Practice routing rule | Before operator linkage, keep tickets separate and request review. After verified linkage, group only the named tickets and notify the incident lead. |
Question for the model
What grouping is supported at the 09:00 information cutoff versus after the 09:05 receipt, and does the maintenance notice justify adding T3?
Show the reference answer and checks
At the 09:00 receipt cutoff, WD1 lacks a verified shared-incident link, so WD3 calls for separate tickets and review. Once WD2 arrives at 09:05, it supports grouping T1 and T2 under I-A and the stated lead notification. Its earlier observation time does not make it available at 09:00. T3 is explicitly excluded, and an unrelated-day maintenance notice supplies no contrary link.
- Using a delayed observation before its receipt-time information cutoff.
- Merging all nearby complaints or applying a maintenance notice from another day.
Fields to preserve
- Ticket, area and symptom
- Preserves the scope of each report.
- Observation time and receipt time
- Separates when a fact occurred from when dispatch could know it.
- Verified incident membership
- Defines the precise grouping.
- Routing-rule version
- Limits the assistant to documented escalation.
Test another case
Use several same-time area-A tickets that an operator explicitly links to two independent incidents. Test whether the model follows membership evidence rather than arrival clustering.
Use fabricated timelines only. Exclude addresses, subscribers, network diagrams, credentials, infrastructure locations and detailed real incidents.
Discuss a contribution
Describe the sources you control, the task they could support and what has already been checked. Use the existing-data assessment for an archive or the collection brief for a new recording project. The contributor agreement defines Fiund’s licensing authority and compensation. A guide or fictional exercise does not establish an accepted collection or a sale.