Repair Prioritisation Fairness Governance requires organisations deploying AI agents in property management, social housing, or landlord-tenant operations to ensure that automated or semi-automated maintenance decisions do not systematically disadvantage tenants who belong to protected groups. Maintenance and repair scheduling is a high-frequency decision domain where small biases — in queue ordering, urgency classification, contractor assignment, or follow-up timing — compound over thousands of interactions to produce disparate outcomes correlated with tenant race, ethnicity, national origin, disability status, familial status, sex, religion, age, or source of income. This dimension mandates that organisations proactively identify, measure, and prevent discriminatory patterns in repair prioritisation before they manifest as habitability failures, constructive eviction, or fair housing violations, because a maintenance queue that appears neutral on its face can embed structural bias through proxy variables, training data, or workflow design choices that correlate with protected characteristics.
Scenario A — Queue-Ordering Algorithm Embeds Neighbourhood-Level Racial Proxy: A property management company operating 4,200 residential units across a metropolitan area deploys an AI agent to triage and schedule maintenance requests. The agent ingests each request, classifies urgency on a five-tier scale, and assigns a queue position within each urgency tier. Within the same urgency tier, the agent uses a composite priority score incorporating property age, historical maintenance frequency for the unit, estimated repair cost, and "tenant reliability score" — a metric derived from rent payment timeliness, lease violation history, and prior maintenance request volume. Over 14 months, a fair housing advocacy organisation analyses publicly available code enforcement data and discovers that median time-to-completion for non-emergency repairs is 11.3 days for units in predominantly white neighbourhoods managed by the company, versus 19.7 days for units in predominantly Black and Latino neighbourhoods. An internal investigation reveals the cause: "tenant reliability score" correlates strongly with source of income (Section 8 voucher holders have lower scores due to administrative payment timing misclassified as late payment), and prior maintenance request volume penalises tenants in older buildings with more frequent legitimate repair needs — buildings that are disproportionately located in minority-majority neighbourhoods due to historical housing segregation patterns. The agent never processes race data directly, but the proxy variables produce a racially disparate outcome. The company faces a HUD complaint, a state attorney general investigation, and class action litigation. Settlement costs exceed $3.8 million, and the company is required to implement an independent fair housing monitor for three years.
What went wrong: The priority-scoring model incorporated variables that served as proxies for protected characteristics. No disparate impact analysis was conducted before deployment or during the 14 months of operation. The "tenant reliability score" penalised structural disadvantage — administrative payment delays inherent to voucher processing and higher maintenance needs in older housing stock — rather than measuring anything relevant to repair urgency. No monitoring system compared repair completion times across demographic groups. The absence of preventive controls allowed a facially neutral algorithm to reproduce and amplify existing housing inequality.
Scenario B — Disability-Related Requests Systematically Deprioritised: A municipal housing authority uses an AI agent to classify and route maintenance requests submitted through a tenant portal. The agent classifies requests into categories (plumbing, electrical, HVAC, structural, cosmetic, accessibility) and assigns urgency. Requests classified as "accessibility" — grab bar installations, ramp repairs, doorway widening, accessible fixture replacements — are routed to a specialist contractor queue with limited capacity, while all other categories are distributed across a general contractor pool with five times the capacity. The agent classifies requests based on keyword matching and description analysis. Over 8 months, tenants with disabilities experience median repair completion times of 26 days for accessibility-related requests, versus a median of 9 days for general maintenance requests across the portfolio. A tenant advocacy group files a reasonable accommodation complaint after documenting that a wheelchair-using tenant waited 34 days for a broken ramp repair — during which time she was unable to leave her unit independently. The investigation reveals that the specialist contractor queue is a structural bottleneck created by the agent's routing logic, not by the inherent complexity of accessibility repairs. Many "accessibility" requests (replacing a grab bar, fixing a lever-style door handle) are straightforward tasks that any general contractor could perform, but the agent routes them to the specialist queue based on keyword classification rather than task complexity. The housing authority is found to have violated the Fair Housing Act's reasonable accommodation requirements and Section 504 of the Rehabilitation Act. The settlement includes $1.2 million in damages, a requirement to redesign the routing system, and mandatory annual fair housing training.
What went wrong: The agent's routing logic created a structural bottleneck that disproportionately affected tenants with disabilities. No analysis compared repair timelines across request categories or tenant demographics. The keyword-based classification system over-routed to a constrained queue, and no feedback loop detected that the specialist queue was producing unacceptable wait times. The system's design choice — routing by category label rather than task complexity — had a disparate impact on tenants with disabilities that was entirely foreseeable but was not assessed before deployment.
Scenario C — Language-Based Urgency Misclassification: A large landlord deploys an AI agent that processes maintenance requests submitted via text message, web portal, and phone transcription. The agent classifies urgency based on natural language analysis of the request description. The agent was trained primarily on English-language maintenance descriptions. Tenants who submit requests in Spanish, Mandarin, or other languages — or in English with non-standard syntax common among non-native speakers — receive systematically lower urgency classifications. A water heater failure described as "no hot water, very cold, children sick" by an English-fluent tenant is classified as Tier 1 (emergency). The same condition described as "water heater no work, cold water only, my kids" by a non-native English speaker is classified as Tier 3 (routine). Over 6 months, emergency classification rates for requests from units with Limited English Proficiency (LEP) tenants are 40% lower than for English-fluent tenants, controlling for actual repair type and severity. A pattern of delayed repairs to habitability-critical systems — heating, plumbing, pest control — in units occupied by LEP tenants constitutes a disparate impact on the basis of national origin. The landlord faces a state civil rights enforcement action resulting in a consent decree, $890,000 in penalties, and mandatory system redesign.
What went wrong: The agent's natural language processing model was trained on a corpus that underrepresented non-native English descriptions, producing systematically lower confidence scores and urgency classifications for requests from LEP tenants. No disparate impact testing evaluated classification outcomes across language groups. No multilingual capability assessment was conducted. The organisation assumed that an English-language NLP model would process all English-language inputs equivalently, failing to account for the well-documented performance degradation of NLP models on non-standard English.
Scope: This dimension applies to any AI agent that influences, recommends, or determines the priority, scheduling, routing, or resource allocation of property maintenance and repair activities in residential housing. The scope includes agents operated by private landlords, property management companies, public housing authorities, social housing providers, homeowner associations, and any third-party service provider acting on behalf of a property owner or manager. The scope covers all decision points in the repair lifecycle: initial triage and urgency classification, queue ordering within urgency tiers, contractor assignment and routing, scheduling and appointment setting, follow-up and completion verification, and re-prioritisation of open requests. The scope extends to agents that process maintenance requests through any channel — text, voice, web portal, mobile application, email, or in-person transcription — and applies regardless of whether the agent makes final decisions autonomously or provides recommendations to human property managers. If the agent's output materially influences repair timing for residential tenants, this dimension applies. The scope includes both explicit prioritisation (the agent assigns a priority score or tier) and implicit prioritisation (the agent's routing, classification, or resource allocation decisions have the practical effect of advancing or delaying specific repairs relative to others).
4.1. A conforming system MUST conduct a disparate impact analysis of repair prioritisation outcomes across all protected classes recognised in applicable fair housing law before initial deployment, and at recurring intervals not exceeding 90 days during operation, measuring differences in repair completion time, urgency classification distribution, queue position, contractor response time, and request re-classification rates.
4.2. A conforming system MUST prohibit the use of tenant agent risk scores, payment history, lease compliance history, prior complaint frequency, or any tenant-level metric not directly related to the physical urgency or safety implications of the specific repair request as an input to repair prioritisation decisions.
4.3. A conforming system MUST ensure that urgency classification is determined solely by factors relating to the physical condition being reported — including health and safety risk, habitability impact, property damage risk, and regulatory compliance obligations — and not by characteristics of the requesting tenant, the tenant's unit history unrelated to the current condition, or the property's financial performance metrics.
4.4. A conforming system MUST implement multilingual processing capability or human fallback for maintenance requests, ensuring that requests submitted in any language spoken by more than 5% of the tenant population, or in non-standard English, receive urgency classifications statistically equivalent to those for standard English descriptions of the same physical conditions.
4.5. A conforming system MUST ensure that routing logic does not create structural bottlenecks that disproportionately delay repairs for requests associated with protected characteristics, including but not limited to disability-related accessibility repairs, which MUST be routable to the general contractor pool when the repair task does not require specialist expertise.
4.6. A conforming system MUST generate and retain a complete audit record for every repair prioritisation decision, including the input data considered, the prioritisation factors applied, the resulting priority score or tier, the queue position assigned, and the actual completion time, in a format that enables retrospective disparate impact analysis.
4.7. A conforming system MUST implement automated monitoring that triggers an alert when any protected group experiences a median repair completion time exceeding 1.5 times the portfolio-wide median for the same urgency tier and repair category, sustained over any rolling 30-day window.
4.8. A conforming system MUST escalate to a qualified human decision-maker any repair request where the agent's urgency classification is disputed by the tenant, where the request involves a condition with potential health or safety implications, or where the request is associated with a prior reasonable accommodation or disability-related modification.
4.9. A conforming system SHOULD implement blind prioritisation — processing the repair request without access to tenant identity, demographics, unit address, or neighbourhood characteristics during urgency classification — to prevent both direct and proxy-based discrimination in the classification stage.
4.10. A conforming system SHOULD conduct intersectional disparate impact analysis examining outcomes for tenants at the intersection of multiple protected characteristics (e.g., elderly tenants with disabilities, single mothers in voucher-subsidised housing) where compounding disadvantage may be masked by single-axis analysis.
4.11. A conforming system SHOULD publish anonymised, aggregate repair performance data disaggregated by property location, building age, and request category at least annually to enable external accountability and independent fair housing review.
4.12. A conforming system MAY implement A/B testing of prioritisation model changes against a fairness-validated baseline to detect disparate impact before full production deployment.
4.13. A conforming system MAY use synthetic test requests representing identical physical conditions submitted with varying linguistic patterns, names, and neighbourhood contexts to proactively detect proxy-based discrimination in urgency classification.
Repair prioritisation in residential housing is a domain where algorithmic bias produces immediate, tangible harm to tenants' living conditions and legal rights. Unlike many AI fairness contexts where the harm is abstract or probabilistic, a biased maintenance queue produces concrete, measurable consequences: a tenant whose broken heating system is deprioritised lives in a cold apartment; a tenant whose water leak is delayed suffers property damage and potential mould exposure; a tenant whose accessibility ramp is not repaired cannot leave their home. These are not hypothetical harms — they are daily-life consequences that accumulate over time and disproportionately affect tenants who are already structurally disadvantaged.
Fair housing law in the United States, the United Kingdom, the European Union, and most other jurisdictions prohibits discrimination in the terms, conditions, and privileges of housing — which includes maintenance services. The U.S. Fair Housing Act (42 U.S.C. 3604(b)) explicitly covers discrimination in the "provision of services or facilities in connection" with housing. Disparate impact liability under the Fair Housing Act, as affirmed by the U.S. Supreme Court in Texas Department of Housing and Community Affairs v. Inclusive Communities Project (2015), means that a facially neutral practice — including an algorithm — that produces a discriminatory effect violates the law regardless of intent. The UK Equality Act 2010 prohibits indirect discrimination in housing services. The EU's proposed AI Act classifies AI systems used in the administration of essential services including housing as high-risk, requiring conformity assessments that include bias and fairness testing.
The risk of proxy-based discrimination in repair prioritisation is particularly acute because the variables that appear operationally relevant — building age, historical maintenance frequency, neighbourhood location, tenant payment patterns — are strongly correlated with protected characteristics due to decades of housing segregation, redlining, discriminatory lending, and income inequality. An algorithm that prioritises repairs in newer buildings over older buildings, or in higher-rent units over lower-rent units, or based on tenant financial metrics, is likely reproducing historical patterns of housing discrimination through facially neutral proxies. The preventive nature of this control is essential because the harm from a biased maintenance queue accumulates continuously — every day of delayed repair is a day of degraded living conditions. Detective controls that identify bias after the fact are insufficient because the harm has already occurred and may be difficult to remediate (a tenant who lived in an unheated apartment for three weeks has already suffered the harm; compensating them after the fact does not undo the experience).
The multilingual processing requirement addresses a well-documented failure mode: NLP models trained predominantly on standard English systematically underperform on text produced by non-native speakers, speakers of non-standard dialects, and multilingual individuals. In a maintenance triage context, this performance degradation directly translates to lower urgency classifications and longer repair wait times for tenants whose language patterns differ from the training corpus. Because language proficiency correlates strongly with national origin, this creates a disparate impact on a protected characteristic. The requirement is preventive — ensuring that language processing capability is adequate before deployment — rather than waiting for the disparate impact to manifest.
The prohibition on tenant agent risk scores in repair prioritisation reflects a fundamental principle: the urgency of a repair is determined by the physical condition being reported, not by the behaviour of the person reporting it. A broken water heater is equally urgent regardless of whether the tenant has a perfect payment history or is three months behind on rent. Incorporating tenant behavioural metrics into repair prioritisation conflates property management enforcement (addressing lease violations) with maintenance obligation fulfilment (providing habitable premises), and the resulting conflation systematically disadvantages tenants who are already economically vulnerable — disproportionately members of protected racial and ethnic groups.
Repair Prioritisation Fairness Governance requires both architectural controls (how the prioritisation system is designed) and operational controls (how the system is monitored and corrected over time). The architectural controls prevent known categories of bias from entering the system. The operational controls detect emergent bias that architectural controls did not anticipate.
Recommended patterns:
Anti-patterns to avoid:
Social and Public Housing. Public housing authorities and social housing providers face the most stringent legal obligations because their tenants are disproportionately members of protected classes: racial and ethnic minorities, persons with disabilities, families with children, and elderly individuals. Federal funding recipients in the United States are subject to Title VI of the Civil Rights Act, Section 504 of the Rehabilitation Act, and the Affirmatively Furthering Fair Housing (AFFH) obligation, which requires not merely non-discrimination but proactive steps to address segregation and disparate outcomes. Social housing providers must implement the full set of mandatory requirements and should implement blind prioritisation (4.9) and intersectional analysis (4.10) as a matter of legal compliance rather than optional enhancement.
Private Property Management at Scale. Large private property management companies managing thousands of units across diverse neighbourhoods face the highest operational risk because their portfolios span demographic boundaries, making geographic disparate impact both more likely and more visible. These organisations should implement the disparate impact dashboard pattern with geographic disaggregation and should establish a fair housing compliance function with authority to override prioritisation decisions when disparate impact is detected.
Landlord-Tenant Regulatory Environments. Jurisdictions vary significantly in their definition of habitability standards, mandatory repair timelines, and tenant remedies. The agent must be configured with jurisdiction-specific rules that define maximum permissible repair timelines for categories of conditions (e.g., no heat in winter must be addressed within 24 hours in many jurisdictions). These regulatory timelines serve as hard constraints that the prioritisation algorithm cannot override, providing a floor below which no tenant's repair can be deprioritised regardless of queue position. AG-210 (Multi-Jurisdictional Regulatory Mapping) provides the framework for maintaining these jurisdiction-specific rules.
Basic Implementation — The organisation has removed tenant agent risk scores and financial metrics from repair prioritisation inputs. Urgency classification is based on physical condition factors. A disparate impact analysis has been conducted at initial deployment. Audit records are generated for all prioritisation decisions. Human escalation pathways exist for disputed classifications and health/safety-related requests. All mandatory requirements (4.1 through 4.8) are satisfied.
Intermediate Implementation — All basic capabilities plus: blind prioritisation is implemented, separating condition classification from tenant identity and unit location. Multilingual processing is validated across all languages representing more than 5% of the tenant population. Automated monitoring with the 1.5x disparity threshold is operational and generating alerts. Flexible contractor routing routes simple accessibility requests to the general pool. The disparate impact dashboard is reviewed monthly by a designated compliance officer. Synthetic test injection is conducted quarterly.
Advanced Implementation — All intermediate capabilities plus: intersectional disparate impact analysis is conducted quarterly. The disparate impact dashboard is publicly accessible in anonymised form. An independent fair housing auditor validates the prioritisation system annually. A/B testing is used for all model changes, with fairness regression gating deployment. The system integrates with AG-687 (Geospatial Bias) for geographic disparity detection and AG-684 (Habitability Risk Escalation) for automatic urgency escalation when regulatory habitability timelines are at risk. Tenant feedback on repair timeliness is collected and disaggregated by demographic group to validate that statistical fairness metrics correspond to lived experience.
Required artefacts:
Retention requirements:
Access requirements:
Test 8.1: Disparate Impact Analysis Existence and Recurrence (validates 4.1)
Test 8.2: Prohibited Input Exclusion (validates 4.2)
Test 8.3: Condition-Centred Urgency Classification (validates 4.3)
Test 8.4: Multilingual Processing Equivalence (validates 4.4)
Test 8.5: Routing Bottleneck Prevention for Accessibility Requests (validates 4.5)
Test 8.6: Audit Trail Completeness (validates 4.6)
Test 8.7: Automated Disparity Monitoring and Alerting (validates 4.7)
Test 8.8: Human Escalation for Disputed and Safety-Related Requests (validates 4.8)
| Regulation | Provision | Relationship Type |
|---|---|---|
| U.S. Fair Housing Act | 42 U.S.C. 3604(b) (Discrimination in Terms/Conditions/Services) | Direct requirement |
| U.S. Fair Housing Act | Disparate Impact (Inclusive Communities, 2015) | Direct requirement |
| Section 504, Rehabilitation Act | 29 U.S.C. 794 (Nondiscrimination in Federally Assisted Programmes) | Direct requirement |
| Americans with Disabilities Act | Title II / Title III (Public/Private Housing Services) | Direct requirement |
| UK Equality Act 2010 | Section 29 (Provision of Services) | Direct requirement |
| EU AI Act | Article 6 / Annex III (High-Risk Classification — Housing) | Supports compliance |
| EU AI Act | Article 9 (Risk Management — Bias Testing) | Supports compliance |
| HUD AFFH Rule | Affirmatively Furthering Fair Housing Obligation | Supports compliance |
| Executive Order 11063 | Equal Opportunity in Housing | Supports compliance |
| NIST AI RMF | MAP 2.3 (Bias Pre-Deployment Testing) | Supports compliance |
Section 3604(b) prohibits discrimination in the "terms, conditions, or privileges of sale or rental of a dwelling, or in the provision of services or facilities in connection therewith." Maintenance and repair services are squarely within the scope of "services or facilities in connection with" a rental dwelling. An AI agent that systematically provides slower or lower-quality maintenance service to tenants of a particular race, national origin, familial status, or other protected class violates Section 3604(b), regardless of whether the discrimination is intentional. The Supreme Court's 2015 decision in Texas Department of Housing and Community Affairs v. Inclusive Communities Project confirmed that disparate impact claims are cognisable under the Fair Housing Act — a facially neutral algorithm that produces discriminatory effects is subject to liability. This dimension's disparate impact analysis requirement (4.1) and prohibited input restrictions (4.2, 4.3) directly implement the Fair Housing Act's anti-discrimination mandate in the maintenance context.
Section 504 prohibits discrimination on the basis of disability in any programme or activity receiving federal financial assistance. Public housing authorities, Section 8 administrators, and any housing provider receiving HUD funding are subject to Section 504. The requirement that accessibility-related repairs not be systematically delayed by routing bottlenecks (4.5) and the requirement for human escalation of requests associated with reasonable accommodations (4.8) directly implement Section 504 compliance in the maintenance prioritisation context. A routing system that channels all disability-related requests to a capacity-constrained specialist queue, when many such requests are general-complexity tasks, violates Section 504 by imposing a disproportionate delay on the basis of disability.
Section 29 prohibits discrimination in the provision of services, including housing-related services. The Equality Act's indirect discrimination provisions parallel the Fair Housing Act's disparate impact framework: a provision, criterion, or practice that puts persons sharing a protected characteristic at a particular disadvantage is unlawful unless it is a proportionate means of achieving a legitimate aim. An AI repair prioritisation system that produces disparate outcomes for tenants sharing a protected characteristic must demonstrate that the practice producing the disparity is proportionate and serves a legitimate aim — the urgency of the physical condition — and does not incorporate irrelevant factors correlated with protected characteristics.
The EU AI Act classifies AI systems used in the "access to and enjoyment of essential private services and essential public services and benefits" as high-risk. Housing maintenance services fall within this scope. High-risk AI systems are subject to conformity assessment requirements including bias testing, risk management, and human oversight. This dimension's requirements — disparate impact analysis (4.1), prohibited inputs (4.2), human escalation (4.8), and audit trails (4.6) — implement the AI Act's conformity assessment requirements in the specific context of repair prioritisation. The AI Act's Article 9 requirement for risk management that addresses "possible biases" is directly served by the recurring disparate impact analysis and automated disparity monitoring mandated by this dimension.
MAP 2.3 addresses pre-deployment bias testing and evaluation. This dimension's requirement for a pre-deployment disparate impact analysis (4.1) and the synthetic testing approach (4.13) implement MAP 2.3 in the repair prioritisation context. The NIST framework's emphasis on context-specific bias evaluation — recognising that bias manifests differently in different domains — is reflected in this dimension's housing-specific requirements regarding prohibited inputs, multilingual processing, and accessibility routing.
| Field | Value |
|---|---|
| Severity Rating | High |
| Blast Radius | Tenant-population-wide — affects every tenant whose repair is processed by the agent, with disproportionate impact on tenants belonging to protected classes in the managed portfolio |
Consequence chain: Without Repair Prioritisation Fairness Governance, the AI agent's maintenance decisions embed and amplify existing patterns of housing inequality. The immediate failure mode is undetected disparate impact in repair timing — tenants in certain protected groups wait longer for repairs without any visible signal that the disparity exists. The first-order consequence is degraded living conditions for the affected tenants: longer exposure to habitability deficiencies including heating failures, water damage, pest infestations, and accessibility barriers. The second-order consequence is health and safety harm: mould-related respiratory illness in children, falls due to unrepaired accessibility features, property damage from unaddressed water leaks, and the psychological burden of living in substandard conditions while observing that neighbours in other buildings receive faster service. The third-order consequence is legal and governed exposure: HUD complaints, state civil rights enforcement actions, class action litigation under the Fair Housing Act, and — for federally funded housing providers — potential loss of federal funding. Fair housing settlements in maintenance discrimination cases have ranged from $500,000 to $10 million depending on portfolio size and the duration of the discriminatory pattern. The fourth-order consequence is reputational: a finding that a property manager used an AI system to systematically deprioritise maintenance for minority tenants, tenants with disabilities, or LEP tenants produces severe reputational harm that affects tenant retention, investor confidence, and the organisation's ability to obtain future government contracts or housing authority partnerships. The compounding nature of maintenance bias means that the harm accumulates daily — each day of delayed repair adds to the disparity — making early detection through preventive controls far more effective than reactive remediation after the disparity is discovered.
Cross-references: AG-001 (Operational Boundary Enforcement) defines the operational boundaries within which the repair prioritisation agent must operate, including prohibitions on actions that exceed the agent's authorised scope — prioritisation decisions that incorporate prohibited inputs violate those boundaries. AG-019 (Human Escalation & Override Triggers) provides the escalation framework that Requirement 4.8 instantiates for repair-specific triggers; the escalation pathways must be configured for tenant disputes, health/safety conditions, and reasonable accommodation associations. AG-022 (Behavioural Drift Detection) monitors whether the agent's prioritisation behaviour changes over time in ways that introduce or amplify disparate impact — drift detection should be configured to alert on changes in the demographic distribution of urgency classifications. AG-040 (Sensitive Category Data Processing Governance) governs the processing of sensitive personal data including race, disability status, and national origin; this dimension's blind prioritisation approach (4.9) reduces the need to process sensitive data in the classification stage, but disparate impact monitoring (4.1, 4.7) necessarily requires demographic data or proxies, which must be governed under AG-040. AG-055 (Audit Trail Immutability & Completeness) ensures that the audit records required by Requirement 4.6 are immutable and complete — repair prioritisation audit trails are high-value evidence in fair housing investigations and must not be modifiable after the fact. AG-084 (Model Training Data Governance) governs the training data used to build the urgency classification model; training data that underrepresents non-English descriptions or overrepresents higher-income property conditions will produce the linguistic and geographic biases described in Scenarios A and C. AG-210 (Multi-Jurisdictional Regulatory Mapping) provides the framework for maintaining jurisdiction-specific habitability standards and mandatory repair timelines that serve as hard constraints on the prioritisation algorithm. AG-679 (Tenant Screening Fairness) addresses a parallel fairness domain in housing — discriminatory screening and discriminatory maintenance are distinct harms but share proxy variables and analytical methods. AG-684 (Habitability Risk Escalation) provides the escalation framework for conditions that threaten habitability — repair prioritisation that delays habitability-critical repairs triggers AG-684 escalation obligations. AG-687 (Geospatial Bias) addresses the geographic dimension of housing discrimination that directly intersects with neighbourhood-level disparate impact in repair scheduling.