Systems and records
Building maintenance service credit data: proving response time against the contract
By SourceX Editorial · Reviewed by Noah Loul ·
10 sections
Short answer
A building maintenance service credit turns on whether the provider responded inside the contracted window, so the decision task needs the clause that defines response, the priority and service calendar that applied, and timestamped events from first report to arrival, all in one time standard. Reconcile the contract terms against that timeline, record pauses and disputes, and label the outcome from the credit actually issued.
License terms
Building maintenance service
Key takeaways
- A response-time credit is decided by the contract's clock definitions, priority, calendar and exclusions, not by raw work order times.
- Normalize every event to a UTC offset and record which site calendar the maintenance system applied.
- The source of record named in the agreement governs; other systems corroborate or support a dispute.
- Take the outcome label from credit memos, waivers or billing adjustments, never from ticket status.
- Client confidentiality and personal data in access logs must be cleared before any packet is licensed.
What response-time evidence decides a building maintenance service credit?
Four pieces of evidence decide it: when the clock started under the contract's definition, when it stopped under the same definition, which priority and service calendar applied, and whether any exclusion paused the clock. Everything else in a work order history is supporting detail.
Performance-based government service contracts offer a useful model for reading a commercial maintenance agreement. The Federal Acquisition Regulation requires those contracts to include a performance work statement, measurable performance standards in terms of quality, timeliness or quantity, the method of assessing performance against them, and, where appropriate, incentives that correspond to those standards. A response-time credit is such an incentive, and the method of assessment named in the agreement is the first thing a reconciliation task has to capture.
For an AI developer, these cases are valuable because the answer can be checked. A model reading the contract and the event log should reach the same credit decision the parties reached, and when it does not, the disagreement points to a specific clause or event.
Why a work order timestamp is not yet a response time
Because the fields a maintenance system stamps rarely match the events a contract measures. The created time may reflect when a dispatcher typed the request, not when the tenant called. The technician's status change to on site may be entered from the parking garage, after the fact, or not at all.
Time zones and calendars add a second layer. IBM's Maximo documentation describes service level agreements that calculate target dates from a calendar chosen from the ticket originator, the asset or the location, with a set search order, precisely because a single portfolio spans buildings with different hours and time zones. A credit dataset has to record which calendar the system used, or the reconciled timeline will disagree with the system's own breach flag for reasons that have nothing to do with performance.
Normalize every timestamp to one standard before comparing anything. RFC 3339, the internet profile of ISO 8601, requires every time it expresses to carry a stated offset from UTC; storing events that way, with the building's time zone kept as a separate field, removes the most common source of one-hour and one-day errors around daylight saving changes.
Which contract terms set the response obligation?
Extract the obligation from the agreement and its amendments before looking at any events; the events are judged against the terms, never the other way round. The examples in the last column are illustrative wording, not quotations from a real contract.
| Term | What to extract | Where it usually lives | Example wording (illustrative) |
|---|---|---|---|
| Priority definitions | Each priority level and the conditions that assign it | Service schedule or SLA exhibit | Priority 1: loss of heating, cooling, water or power to an occupied floor |
| Response definition | The event that stops the response clock | SLA exhibit definitions | Response means a qualified technician on site, not a phone call back |
| Clock start | The event that starts the clock | SLA exhibit or help desk procedure | From the time the request is logged in the service desk |
| Service calendar | Hours the clock runs for each priority | Service schedule | Priority 1 measured around the clock; priority 3 in business hours only |
| Exclusions and pauses | Events that stop or pause the clock | SLA exhibit or general conditions | Time waiting for building access provided by the client is excluded |
| Source of record | Which system's timestamps govern | Reporting clause | Measured from the provider's computerized maintenance system |
| Credit mechanism | How a missed target becomes a credit, and any cap or measurement period | Credit schedule | Credit tier set per missed priority 1 response, capped per month |
| Cure and dispute window | Re-performance rights and how long either side may contest | General conditions | Disputes raised within one billing period of the report |
The contract and event timeline reconciliation task
Each task packet joins one service event to the exact terms that governed it and to every timestamped record bearing on the clock, then asks for a credit decision with the evidence that supports it. The fields below are what a team needs to build packets as is.
At a glance
The contract and event timeline reconciliation task
Keep the contract version on each packet. Maintenance agreements are amended often, and a credit decided under last year's priority definitions is a different task from the same event under this year's.
| Field | Content | Source | Validation rule |
|---|---|---|---|
| case_id and site | One service event at one building | Work order and location record | One event per packet; linked follow-up work orders listed, not merged |
| contract_ref | Agreement, amendment and exhibit versions in force on the event date | Contract repository | Effective dates cover the event date |
| priority_assigned and priority_basis | Priority at logging and the facts that justify it | Work order and request text | Basis checked against the priority definitions |
| clock_events | Every candidate start, pause, resume and stop event with source | Phone log, service desk, CMMS, access control, BAS alarms | All times in RFC 3339 with offset; source named for each |
| calendar_applied | Service calendar and time zone used for the clock | SLA configuration and site record | Matches the contract calendar for that priority |
| exclusions_claimed | Pauses asserted and the evidence for each | Technician notes, client emails, access logs | Each pause has a document, not only a note |
| computed_elapsed | Response time under the contract definition | Derived | Recomputed independently from clock_events |
| system_breach_flag | What the maintenance system reported | SLA record | Differences from computed_elapsed explained |
| outcome_label | Credit issued, waived, disputed with breach withdrawn, disputed with credit standing, or not settled | Credit memo or billing adjustment | Taken from billing, never from ticket status |
| rights_screen | Result of the confidentiality and personal data review | Counsel or data owner | Packet excluded unless cleared |
When systems disagree on arrival time, which record wins?
The source of record the contract names decides it, and the other systems serve only as corroboration. When the agreement says response is measured from the provider's maintenance system, an access control entry showing a later arrival does not change the contractual answer, though it is exactly the evidence a client would raise in a dispute.
When the contract names no source, use a corroboration rule fixed before labeling: for example, accept an arrival time when two independent records agree within a stated tolerance, and mark the case disputed when they do not. Record the rule on the packet, because a model's answer can only be graded against the rule that produced the label.
Federal quality assurance practice offers a useful model for the surveillance side. FAR 46.401 says quality assurance surveillance plans should specify all work requiring surveillance and the method of surveillance. A dataset that writes down, per contract, which events are surveilled and how, gives a model the same frame of reference the parties had.
Worked case (illustrative): an after-hours chiller alarm on a two-building campus
A provider maintains a two-building office campus under one agreement. Building A's calendar runs around the clock for priority 1; building B, added by amendment, has a business-hours calendar for all priorities. Every time and duration below is illustrative.
A chiller alarm at building B fires in the evening. The building automation system records the alarm, the tenant calls the help desk a few minutes later, and the dispatcher logs a priority 1 request against building A by mistake, because the two share a chilled water loop.
Under building A's calendar the system flags a breach; under building B's business-hours calendar, the clock does not start until the next morning and no credit is due. The correct label depends on the site correction, which the provider made two days later. A counter-case from the same portfolio shows the opposite pattern: a priority 1 call where the technician waited at a locked loading dock for client security, documented by an email, so the excluded access time pauses the clock and a raw elapsed-time breach is not a contractual one.
| Event (illustrative) | Recorded by | Effect on the clock | Reconciliation note |
|---|---|---|---|
| Chiller alarm at 19:42 | Building automation system | None; the contract starts the clock at logging | Kept as context; shows the delay between fault and report |
| Tenant call at 19:47 | Phone system | None under this contract | Would start the clock under a call-based definition |
| Request logged at 19:55 against building A | Service desk | Starts the clock under building A's around-the-clock calendar | Wrong site; building B's calendar should apply |
| Technician badge at building B door at 21:30 | Access control | Arrival under building B, if corroborated | Supports a later on-site time than the technician's own entry |
| Technician marks on site at 21:10 | Maintenance system | Stop under the source-of-record rule | Earlier than the badge event; flagged for dispute |
Label the outcome from the credit memo, not the ticket status
Take the outcome from billing records, because only they show what the parties actually settled. A ticket closed as breached says what the system computed; a credit memo, or a documented waiver, says what happened.
At a glance
Label the outcome from the credit memo, not the ticket status
Maintenance platforms model this explicitly. IBM's documentation for Maximo for Service Providers describes SLA breach billing, in which penalty fees or credits for breached SLA commitments are billed through sales orders. Where a platform works this way, the outcome label comes from that sales order or adjustment line. On the contract side, the federal inspection clause for fixed-price services shows the usual sequence: the buyer may first require re-performance, and only when defects cannot be corrected that way may it reduce the price to reflect the reduced value of the services. Record which of those paths a case took, because a re-performed visit and a credited one are different outcomes.
- Credit issued: a credit memo or billing adjustment references the event.
- Waived: the client accepted the breach in writing without a credit.
- Disputed, breach withdrawn: the provider contested the breach and the client withdrew it.
- Disputed, credit standing: the provider contested it and the credit stood.
- Not settled: no billing record or written decision exists; keep out of training labels.
What must be screened before facility service records leave the company?
Screen three kinds of content before any packet leaves the provider. Client agreements usually carry confidentiality terms that cover pricing schedules and service levels, so the provider needs a contractual right to use those terms, or the client's consent, separate from any right to its own operational records.
Access control and badge logs, technician names and tenant contact details are personal data, and building security details such as door names and alarm points can be sensitive in their own right. Pseudonymize people consistently across the phone, service desk and access records so the timeline still joins, and generalize site details that a model does not need. Having the data in your systems does not by itself give you the right to license it; ownership, contract rights and privacy permission are separate questions for counsel.
Cases the timeline cannot settle
Leave a case out, or label it not settled, when the evidence cannot support any single answer. The common causes are a missing source of record (the maintenance system was migrated and old SLA records were not carried over), amendments nobody can locate, oral waivers agreed on site, and master agreements that pool performance across a portfolio so no single event decides a credit.
Portfolio-level measurement deserves care. When credits are computed on the share of priority 1 responses met in a month, an individual event never decides the outcome on its own; the task becomes a monthly reconciliation, and the packet should hold the whole month's events for that site rather than one work order.
Licensing reconciled service histories
A facility services provider with several years of agreements, work orders, access and alarm records, and billing adjustments in consistent systems holds decisions with verifiable outcomes, which is uncommon in operational data. Building them into reconciliation packets is what makes them usable for AI training and evaluation.
Facility service providers can have SourceX take such histories through qualification, rights review, contracting, preparation and delivery as a licensed AI dataset. The opening step is a short company assessment that asks for no data, and the provider approves the scope and terms and signs before anything is handed over.
Frequently asked questions
What is building maintenance service credit data?
It is the set of records needed to decide whether a maintenance provider owed a credit for missing a contracted response or resolution target: the agreement and its service levels, the event timeline for each request, and the billing adjustment that settled it.
Which timestamp counts as the response time?
The one the contract defines. Many agreements stop the clock when a qualified technician is on site, others at a call back or remote diagnosis, and the agreement usually names which system's timestamps govern.
How should time zones be handled in a multi-site portfolio?
Store every event with an explicit UTC offset and keep the site's time zone and service calendar as separate fields. Maximo, for example, can take an SLA calendar from the originator, asset or location, so record which one the system used.
Can a ticket marked breached be used as the label?
Not on its own. Use the credit memo, billing adjustment or written waiver, because system breach flags are often corrected, disputed or waived after the ticket closes.
Do service providers need client permission to license these records?
Often yes for the contract terms, which client agreements usually treat as confidential, and personal data in access logs needs its own review. Counsel should confirm the position for each client agreement.
Sources
- FAR 37.601 requires performance-based service contracts to include a performance work statement, measurable performance standards and the assessment method, and incentives corresponding to those standards where appropriate. Source
- FAR 46.401 says quality assurance surveillance plans should specify all work requiring surveillance and the method of surveillance. Source
- FAR 52.246-4 lets the government require re-performance of nonconforming services and, when defects cannot be corrected that way, reduce the price to reflect the reduced value of the services. Source
- IBM Maximo service level agreements can calculate target dates from the calendar of the ticket originator, the asset or the location, in a configured search order. Source
- IBM's Maximo for Service Providers documentation describes SLA breach billing, with penalty fees or credits for breached SLA commitments billed through sales orders. Source
- RFC 3339 profiles ISO 8601 for internet timestamps, and every time it expresses carries a stated offset from UTC. Source
See if your company qualifies
A short company assessment. No data uploads are needed.