From the College Athletics edition of September 21, 2026
NRCD study: collegiate XC ML fails individual improvement forecasts; team race volume still associates with placement
University of Notre Dame / University of Virginia / National Running Club Database · College athletics / cross country scheduling analytics · United States (collegiate club / NIRCA; transfer caution to NCAA varsity)
- Publisher
- arXiv (cs.CY / cs.LG)
- Original publication
- September 15, 2026 (arXiv v6); analysis of 2023–2025 seasons
- Source retrieved
- 2026-09-22
What happened
Using 23,360 comprehensive-era NRCD club cross-country results (7,056 athletes, 2023–2025), leakage-controlled temporal validation shows race-result features do not support out-of-year forecasting of individual improvement (best men’s R²=0.044; women’s −0.018), near a low reliability ceiling (~0.23–0.28). Converted Only times overstate mean first-to-last gains by 15–21 seconds versus weather/elevation Standardized times. Separately, team race frequency associates with nationals placement (pooled RR=2.09; OR=2.56 per SD), but within-team year-to-year schedule changes do not predict placement change and depth signals entangle with club size. Authors advise cautious transfer to NCAA/NAIA/NJCAA varsity contexts.
Why it matters
Newly relevant for fall 2026 XC scheduling and for athletic departments buying “athlete progression AI” pitched on results history alone: a large public collegiate dataset with honest null individual prediction—contrasts vendor accuracy headlines and complements archived injury-AUC papers that warn overall accuracy can hide weak classes.
Evidence and measured results
arXiv abs and HTML v6 (2509.10600): abstract metrics, Table 4 held-out R², weather-inflation 15–21 s, pooled RR/OR team results, discussion on varsity transfer limits, IRB non-human-subjects note.
Limitations and uncertainty
Club not varsity sample; observational team findings; no training logs/wearables; finish times omit injury/tactics; reliability ceiling approximates ≥4-race subgroup.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-09-22; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- XC staff want predictive tools for who will improve and how to schedule meets, but results-only models may look smart while failing true forecasts.
- Stakeholders
- Head XC coach
- Sports performance/science
- Athletics analytics
- Sports medicine
- FAR/compliance
- Discovery questions
- Do current dashboards use weather-adjusted or raw/converted times?
- Have any vendor models been tested with next-season holdout?
- Who approves changes to athlete race frequency?
- Is the population varsity NCAA, club, or mixed?
- Value hypothesis
- If local pipelines adopt Standardized times and leakage-controlled evaluation, departments can avoid false confidence in individual progression scores while still studying team schedule structure carefully.
- Potential engagement
- Three-week methods workshop: replicate weather-inflation checks on two seasons of campus XC results and document whether any athlete-level model beats a null baseline out of year.
- Unsupported claims
- That increasing every athlete to 4+ races causes better nationals placement
- That club NRCD coefficients apply unchanged to NCAA varsity
- Positive individual forecasting skill from finish times alone
Pre-sales engineering
Role takeaway
- Fit
- Evaluation harness and time-standardization layer for collegiate endurance analytics—not a turnkey improvement predictor.
- Architecture and integration
- Ingest results with course/weather fields; compute Standardized times; train/test by season year; block endpoint leakage features; report R², SER, and reliability ceilings.
- Prerequisites
- Multi-season results with high weather/course coverage
- Clear athlete–season unit definition
- Agreement that null results are publishable internally
- Constraints
- Source population is collegiate club XC
- Individual held-out R² near zero
- Team associations are between-program and size-entangled
- Security
- Tokenize or restrict identifiable student-athlete performance exports; no public redistribution of campus PII.
- Proof of value
- Temporal held-out R² with confidence intervals
- Leakage add-back audit table
- Documented Converted vs Standardized inflation on local data
Delivery
Role takeaway
- Work
- Local replication notebook, coach-facing brief on null individual forecasts, and optional prospective schedule pilot design with medical oversight.
- Dependencies
- Results warehouse access
- Sports medicine review for load changes
- Coach willingness to treat nulls as findings
- Ownership
- Sports science owns metrics validation; head coach owns schedule decisions.
- Skills and adoption
- Analysts practice leakage-controlled CV; coaches interpret weather-adjusted times.
- Governance checkpoints
- No athlete-facing improvement score without positive temporal skill
- Medical sign-off before race-frequency experiments
- Written varsity-transfer caveats
- Acceptance criteria
- Replication memo with held-out metrics
- Coach briefing delivered
- Pilot protocol filed if schedule changes proceed
- Risks
- Misreading team RR as causal within-team advice
- Over-racing injury risk
- Vendor models that skip leakage audits
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
If building XC analytics, standardize environmentally adjusted times before modeling; evaluate with temporal holdout and leakage audits; do not ship athlete-level improvement scores from finish history alone.
Governance
Who approves, reviews and stays accountable for outcomes?
Present team-level schedule associations as observational hypotheses for coaches, not causal mandates to maximize race count; require health oversight before changing athlete race loads.
Security and privacy
What data, permissions and controls need testing?
NRCD export is de-identified public data—campus deployments using identifiable student-athlete results need ordinary FERPA/athletics access controls even when methods transfer.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Publish null results internally so less-resourced programs are not shamed into buying weak individual-forecast tools.
Procurement
What should contracts, pricing and exit terms secure?
Require temporal holdout and leakage audits in XC/analytics RFPs; reject results-only improvement guarantees.
Operating model
Which teams own the service once it runs?
Separate research evaluation from coach decision rights; schedule changes remain human-owned with health constraints.
Publication history
- 2026-09-21College Athletics · Issue 154 resources
Stable resource ID: nrcd-xc-schedule-ml-null-2026