Passport Summary: an AI-generated narrative overview of an asset's maintenance history where every factual claim is backed by citations to specific signed events in the passport.
Key Takeaways
- 1Summaries provide quick executive overviews of maintenance history
- 2Every statement cites specific event IDs from the passport
- 3Citations are clickable links to the source credentials
- 4Invalid citations cause the summary to be rejected and regenerated
- 5Summaries are included in passport packs for verifiers
The problem: information overload
A well-maintained asset might have hundreds of maintenance events over its lifetime. For a verifier evaluating the asset, reading through every oil change, inspection, and repair is impractical. They need a quick overview to understand the maintenance story.
Traditional approaches to this problem fall short:
Manual summaries
Asset owners write their own summaries, which may be biased, incomplete, or simply fabricated. No way to verify claims.
Simple AI summaries
Standard AI summarization can hallucinate details, invent events that didn't happen, or misrepresent the maintenance history.
No summary at all
Verifiers must read raw event lists, which is time-consuming and makes it easy to miss important patterns.
The solution: traceable summaries
UMP passport summaries solve this with a simple but powerful constraint: every factual claim must cite its source. The AI generates the narrative, but every statement links back to specific signed events.
AI analyzes events
The AI model receives all signed events for the asset and analyzes the maintenance history to identify key patterns and highlights.
Summary with citations generated
The AI generates a narrative summary where each claim includes citations in the format [event-id]. Multiple claims can cite the same event.
Citation validation
The system validates every citation: Does this event ID exist? Does the claim match what the event actually says? Invalid citations fail validation.
Regeneration if needed
If any citation is invalid, the summary is rejected and regenerated. The AI is forced to produce accurate, traceable claims.
Clickable citations
In the final summary, citations become clickable links. Readers can click any claim to see the source event credential.
Every claim is cited
AI assists. Proof comes from signed events.
The AI generates the narrative, but every factual claim traces back to a cryptographically signed event. The summary is only as trustworthy as the underlying credentials — which is exactly the point.
Here's what citation traceability looks like in practice:
This 2019 CAT 320 excavator demonstrates consistent preventive maintenance with oil changes performed at manufacturer-recommended intervals [EVT-001], [EVT-007], [EVT-015]. A hydraulic pump replacement was completed at 4,200 hours [EVT-012] as proactive maintenance before failure.
Each bracketed reference is a clickable link to the source credential. Verifiers can click any claim to see the signed event that backs it up. No citation? The claim doesn't appear in the summary.
What summaries contain
A passport summary typically includes three sections:
Executive summary
2-3 sentences capturing the maintenance story: how well-maintained is this asset? Any major events? Overall assessment of the history.
Key highlights
Bullet points calling out important events: major overhauls, component replacements, regulatory inspections, significant repairs.
Maintenance assessment
Analysis of maintenance patterns: Was the schedule followed? Any gaps? How does the history compare to expected maintenance for this asset type?
Example summary excerpt
"This 2019 CAT 320 excavator shows consistent preventive maintenance with oil changes every 450-500 hours [EVT-001, EVT-007, EVT-015]. A hydraulic pump replacement was performed at 4,200 hours [EVT-012] following manufacturer recommendations. No METER_ANOMALY flags are present."
The citation requirement
The citation requirement is what makes passport summaries trustworthy. Here's how it works:
| Aspect | Claim type | Citation requirement |
|---|---|---|
| Specific event mention | "Oil change at 5,000 hours" | Must cite the event with that meter reading |
| Count claim | "47 maintenance events" | Must match actual event count |
| Date claim | "Last serviced March 2024" | Must cite event from that date |
| Pattern observation | "Regular 500-hour intervals" | Must cite multiple events showing pattern |
The validation system checks each citation programmatically. If the AI claims "hydraulic fluid changed at 3,000 hours [EVT-008]" but EVT-008 actually shows 2,800 hours, the citation fails validation and the summary is regenerated.
Citation validation and confidence scoring
AI systems can hallucinate — generating plausible-sounding but false information. In maintenance contexts, this is dangerous. A hallucinated "engine overhaul at 5,000 hours" could affect purchasing decisions worth hundreds of thousands of dollars.
Claims require citations
Every factual statement must include a citation. Validation checks that cited events exist and that claims are grounded in source data.
Source events are signed
The events being cited are themselves cryptographically signed credentials. The summary inherits their trustworthiness.
Confidence scoring
Each summary includes a confidence score showing validation results. Low-confidence summaries are flagged so users know to verify against source events.
Trust through traceability
When you read a passport summary, you can click any claim to see the source credential. This isn't blind trust in AI — it's verified trust through citations to signed records.
What summaries won't include
The citation requirement means summaries are deliberately limited to what can be proven:
Subjective opinions
"This asset was well cared for" only if events support it. No general praise without evidence.
Future predictions
"Should run another 5,000 hours" is speculation. Summaries describe history, not forecast reliability.
Comparative claims
"Better maintained than average" requires data UMP doesn't have. Each asset is summarized independently.
Information not in events
If something wasn't recorded as a signed event, the summary can't mention it. No reading between the lines.
These limitations are features, not bugs. They ensure summaries stay grounded in verifiable facts rather than drifting into AI-generated speculation.
Summaries in passport packs
Passport summaries are included in passport packs — the shareable bundles that asset owners generate for verifiers. When a buyer or lender receives a passport pack, they get:
Summary first
The executive summary provides immediate context. Verifiers understand the maintenance story in 30 seconds.
Citations to explore
Interested verifiers click citations to see source credentials. Deep verification is one click away.
Full event list
The complete event history is available for thorough review. Nothing is hidden — the summary is just the entry point.
See summaries in action
Ready to see how passport summaries work?
For asset owners
Generate summaries for your assets to make passport packs more accessible to buyers and lenders.
Learn moreFor verifiers
Use summaries as your starting point for due diligence, with full traceability to source credentials.
Learn moreWant to see a sample summary?
We can show you real passport summaries with clickable citations.
Request a demo