Short answer
A work order history records maintenance jobs end to end: the reported symptom, the asset, the technician's diagnosis, the parts and labor used, the fix and whether it held. Field service firms, facilities teams and fleet operators hold years of these records in CMMS and field-service systems. For humanoid programs they support task selection, diagnostic reasoning and maintenance language, but they contain no record of how the physical repair was performed.
What buyers use work order histories for
Maintenance is one of the task families humanoid programs target, and work orders are the most complete existing record of what maintenance actually involves. Buyers use them in five ways:
- Choosing what to record. Counting jobs by type, asset and duration shows which repairs are frequent and long enough to justify demonstration capture; see prioritizing maintenance tasks for collection.
- Diagnostic reasoning. Symptom, finding and fix, followed by whether the problem came back, form natural training and test items for models that reason about faults.
- Maintenance language. Asset names, part names and technician shorthand ground the vocabulary an embodied assistant must understand.
- Planning. Parts and tools consumed per job type tell a planner what to fetch before starting.
- Outcome labels for video. When a recorded repair can be joined to its work order, the clip gains a symptom, a diagnosis and an outcome.
Anatomy of a work order
| Field | Typical form | Supports | Common problem |
|---|---|---|---|
| Asset and location | Equipment tag linked to site and system | Grouping jobs by equipment type | Hierarchies differ across sites |
| Request or symptom | Free text from a requester, call center or alarm | Diagnosis input | Vague: "unit not working" |
| Problem, cause, remedy codes | Picklists | Structured labels | Left blank, defaulted or "other" |
| Technician notes | Free text with shorthand | Diagnostic reasoning; language | Abbreviations, misspellings, copied text |
| Parts consumed | Part numbers and quantities | Linking fixes to parts; job planning | Parts drawn but unused; van stock unrecorded |
| Labor | Time on job, crew size | Duration estimates | Travel and waiting mixed in |
| Status timestamps | Opened, dispatched, on site, closed | Workflow timing | Closed in batches at day end |
| Follow-up | Linked repeat visit | Outcome label | Repeat visits often not linked |
| Attachments | Photos, checklists, meter readings | Visual evidence | Sparse; premises and people in frame |
The structure that makes these fields interpretable, the asset hierarchy, is covered in CMMS asset hierarchies. Outcome fields are covered in first-time fix and callback labels.
Maintenance text needs its own preparation
Technician notes are not ordinary prose. MaintNet assembled maintenance logbook data from aviation, automotive and facilities domains and released abbreviation lists, morphosyntactic lists, termbanks and preprocessing tools alongside it (Akhbardeh et al., 2020), a sign of how much preparation this text needs before standard language tools handle it well. Expect to normalize abbreviations, part names and misspellings before training; maintenance logbook text covers the free-text side in detail.
Job types also need a shared vocabulary when records come from several suppliers. O*NET publishes task statements and Detailed Work Activities with a file that maps one to the other (O*NET Resource Center), which can serve as a neutral layer for mapping each supplier's job codes.
Supplier types
| Supplier | Typical systems and records | Strength | Watch for |
|---|---|---|---|
| Field service organizations | Field-service management and dispatch systems | Many similar jobs across many sites | Equipment and premises belong to their clients |
| Facilities management firms | CMMS work orders across client buildings | Broad building-maintenance task mix | Client contracts limit reuse |
| Fleet maintenance operations | Vehicle repair orders | Standardized assets; consistent codes | Driver and telematics data |
| In-house plant maintenance | CMMS with preventive and corrective orders | Rich asset hierarchy | Confidential process details |
| MRO providers | Task cards and sign-offs | Disciplined documentation | Regulatory and contractual constraints |
What work orders cannot support
A work order says "replaced run capacitor"; it does not show how the panel was opened, which hand held the meter, or how much force the connector needed. There is no motion, force or view of the repair, and the diagnosis is usually written after the job, shaped by what worked. Work orders tell a humanoid program which repairs matter and how they turn out; they cannot teach a robot to perform them.
For the physical side, pair the records with recordings: repair video demonstrations covers existing and new footage, and existing workplace footage vs commissioned recording covers the buy-or-record decision.
Illustrative record schema
{
"_note": "Illustrative only — synthetic record, field names are examples",
"wo_id": "WO-000-118734",
"site_id": "site_041",
"asset": {"asset_id": "AST-RTU-0007", "class": "rooftop_hvac_unit", "hierarchy": "site_041/roof/hvac/rtu_07"},
"opened_at": "2024-07-02T09:14:00Z",
"request_text": "Unit 7 blowing warm air, east offices",
"codes": {"problem": "NO_COOL", "cause": "CAP_FAIL", "remedy": "REPLACE"},
"tech_notes": "chkd cont, comp not starting, run cap out of spec, rplcd, unit cooling",
"technician_id": "tech_B19",
"parts": [{"part_no": "PN-CAP-0045", "qty": 1}],
"labor_min": {"on_site": 55, "travel": 30},
"closed_at": "2024-07-02T11:02:00Z",
"follow_up": {"repeat_visit_within_30d": false},
"attachments": ["photo_000913.jpg"]
}
A delivery would add a data dictionary, code lists with definitions, the asset hierarchy, a crosswalk for any system migration, and fill-rate statistics per field and year.
Quality and acceptance criteria
- Fill rates are reported per field, per site and per year, not as a single average.
- Code quality. The share of blank, default and "other" codes is stated; a sample of codes is checked against the notes.
- Linkage. Asset IDs resolve to the hierarchy at a stated rate, and repeat visits are linked or linkable.
- Coverage. Job counts by type and asset class match the summary the supplier provided.
- System history. Migrations and code-list changes are documented with dates.
- De-duplication. Auto-generated preventive orders closed without work are flagged.
- Evaluation split. Held-out records are separated by site or time, depending on the claim being tested.
Limitations and preparation
Expect to normalize text, map codes across sites, resolve asset hierarchies, link repeat visits, and replace customer and technician identifiers through pseudonymization. Records also carry bias: jobs done informally are never logged, preventive orders can inflate counts, and notes favor the explanation that matched the fix.
Questions to ask suppliers
- Which systems hold the records, for which years, and were there migrations?
- How often are problem, cause and remedy codes used, and how often are they blank or "other"?
- Do technicians write free-text notes, and in what style and language?
- Are repeat visits linked to the original order?
- Are parts recorded per order, with part numbers?
- Is there an asset hierarchy, and is it consistent across sites?
- Are photos, meter readings or checklists attached?
- Under your service contracts, who controls the records: you or the owner of the building or equipment?
Permissions and provenance
Work orders often name the client, the address, the people who reported the problem and the technicians who fixed it. Pseudonymization reduces exposure but does not by itself settle whether the records may be licensed; service contracts may restrict reuse of client information. Ask early. This is not legal advice.
Discuss your requirements
If you need maintenance records for task selection or diagnostic reasoning, describe the asset types, fields and years you need to SourceX. SourceX looks for US businesses that hold suitable records, checks the data and the supplier's licensing permissions, and coordinates pricing, allowed uses and delivery in a license. It reviews the request and follows up about potential sources; a request does not guarantee a match.
The catalog entry for field service and maintenance work orders and the page on licensing maintenance logs describe the record type. The maintenance, repair and tool-use hub and the humanoid training data guide cover the rest of the section.
Explore sourcing and licensing options with SourceX
Sources
- MaintNet: A Collaborative Open-Source Library for Predictive Maintenance Language Resources — maintenance logbook data and the normalization resources it needs.
- ONET Database (ONET 29.1 per search snapshot): Task Statements and Detailed Work Activities — task statements and work activities as a shared task vocabulary.