From the College Athletics edition of September 11, 2026
Sprint-screening preprint shows low confirmed-sanction precision and contextual gaps
Carnegie Mellon University Africa; Blessed Madukoma and Prasenjit Mitra · Collegiate athletics · International athletics data; authors based in Rwanda
- Publisher
- arXiv:2604.21953v1
- Original publication
- April 23, 2026
- Source retrieved
- 2026-09-12
What happened
Retrospective screening results support caution about treating performance anomalies as evidence of wrongdoing.
Why it matters
Transferable to college track analytics oversight; the paper includes an NCAA race illustration but no NCAA-specific validation.
Evidence and measured results
Table III benchmarks eight methods on 31,604 100-m athletes, 25 with recorded sanctions. Excess Performance flags 226, including 2 sanctioned athletes: precision .009, recall .080, F1 .016. Five methods find no sanctioned athletes. The implemented baseline is static, despite broader trajectory language.
Limitations and uncertainty
Preprint; incomplete sanction labels, zero-imputed missing wind and unadjusted altitude. No prospective NCAA evaluation. The authors' claim that incomplete labels make recall a conservative lower bound is not established; no guilt inference is warranted.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-09-12; this does not change the original publication date. Labels below come from the analysis itself.
Sales
Role takeaway
The relevant problem is whether an analytics service can support careful review without generating disproportionate investigative burden. Engage athletics research, athlete representatives, compliance leadership and institutional privacy staff before discussing a use case. Ask what decision is authorized, which independent evidence is available and whether a screening tool is needed at all. A bounded engagement could audit historical data quality and evaluation methodology without producing individual accusations. Its value hypothesis is better understanding of error and workload, not more sanctions. The retrospective benchmark does not establish savings, NCAA suitability or a market opportunity at any named institution.
Pre-sales engineering
Role takeaway
Start with a reproducible offline benchmark and a restricted review interface. Preserve athlete-identity provenance, event context, label dates and explicit missing-value indicators. Validate separately by event and data completeness, and compare simple baselines under the same protocol. Prevent future information from leaking into a purported prospective test. Measure review workload at a fixed alert budget, report uncertainty from sparse labels and seek independent expert adjudication. Access to public records does not justify unrestricted dissemination of inferences. Infrastructure placement remains an institutional choice; autonomous investigative agents and automatic sanctions have no supported role in this evidence.
Delivery
Role takeaway
A designated research lead should own any feasibility study, with privacy and athlete-welfare review before data linkage. Establish a reproducible dataset snapshot, contextual-quality checks and a written restriction on downstream use. Train reviewers to distinguish anomaly, recorded sanction and unverified status. Proposed acceptance criteria include full provenance for reviewed records, documented missingness, reproducible aggregate metrics and no automated consequential action. Stop if reviewers cannot explain errors or if lawful authority and data rights remain unresolved. Evaluate adoption through reviewer understanding, not alert volume. Incomplete labels and reputational harm are central risks; retrospective success cannot authorize production use.
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?
Preserve source records, missingness and model versions in a reviewable analysis service; benchmark simple methods before complex ones.
Governance
Who approves, reviews and stays accountable for outcomes?
Confine any exploratory use to authorized expert review; never turn an anomaly into an automatic accusation or eligibility decision.
Security and privacy
What data, permissions and controls need testing?
Public results joined to sensitive inferences still require access limits and a correction process.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Make uncertainty understandable and fund qualified reviewers rather than merely exposing a score.
Procurement
What should contracts, pricing and exit terms secure?
Demand prospective validation and contextual-data coverage before considering operational acquisition.
Operating model
Which teams own the service once it runs?
Assign accountable research oversight and maintain a strict separation between analysis and authorized enforcement.
What changed
Absent from all 217 canonical archive records scanned at offsets 0, 100 and 200, including URL and related-title checks. Newly archived historical evidence; no new-since-last-run publication or update to an existing resource is claimed.
Publication history
- 2026-09-11College Athletics · Issue 063 resources
Stable resource ID: madukoma-athletics-anomaly-benchmark-2026