NRCD: An Open Database of Collegiate Running with Unified Performance Standardization

Jonathan A. Karr; Ryan M. Fryer; Ben Darden; Nicholas Pell; Kayla Ambrose; Evan Hall; Ramzi K. Bualuan; Nitesh V. Chawla

cs.LGAug 14, 2026

NRCD: An Open Database of Collegiate Running with Unified Performance Standardization

Authors: Jonathan A. Karr, Ryan M. Fryer, Ben Darden, Nicholas Pell, Kayla Ambrose, Evan Hall, Ramzi K. Bualuan, Nitesh V. Chawla

Organizations: University of Notre Dame Notre Dame, Indiana, USA · National Running Club Database Michigan, USA · University of Virginia Charlottesville, Virginia, USA · Virginia Tech Blacksburg, Virginia, USA ·

Abstract

Collegiate running in the United States generates thousands of race results annually in cross country and track and field, yet no large-scale dataset has been publicly available for research. Existing websites such as Athletic.net, MileSplit, and TFRRS host results but do not support bulk download, restricting prior analyses to ~500 performances, often skewing studies toward male athletes. We introduce the National Running Club Database (NRCD), the first openly available collegiate running dataset at scale: 143,868 approved performances from 31,351 athletes across 1,423 meets in four sports (cross country (XC), indoor and outdoor track, and road races), 36.2% women, spanning 2003-2026. Meets from August 2023 onward carry comprehensive course distance, elevation gain and loss, weather at race time, and track venue metadata (99.9% of XC rows with weather fields). NRCD is community-governed through open submission and expert approval and is maintained as a live database whose meet volume has grown yearly. We release a unified performance standardization framework that operationalizes established distance, elevation, and heat adjustments in one pipeline; XC-only validation; heat is a Hadley-band heuristic. We recommend gender-stratified modeling. On XC, full standardization lowers median within-athlete cross-meet variability by 51.1% (women) and 35.4% (men) versus raw times. We release the dataset and pipeline with a Python package `nrcd' under FAIR principles, supporting longitudinal athlete modeling, environmental-confounder studies, and gender-equity research in collegiate sport.

Figures & tables

Figure 1. NRCD data and standardization pipeline

Sport	Results	Athletes †	Meets
Cross Country	75,489	23,565	546
Indoor Track	29,595	9,657	348
Outdoor Track	36,598	9,976	469
Road Race	2,186	1,630	60
Total	143,868	31,351	1,423

Table 1. Results by sport

Sport	Era	Results	Athletes †	Weather (%)	Alt. (%)	Course (%)
Cross Country	historical	52,129	17,865	27.2	97.8	3.0
Cross Country	comprehensive	23,360	7,056	99.9	100.0	38.5
Indoor Track	historical	11,386	5,118	—	100.0	—
Indoor Track	comprehensive	18,209	5,174	—	99.4	—
Outdoor Track	historical	13,399	5,098	0.0	100.0	—
Outdoor Track	comprehensive	23,199	5,665	35.0	99.9	—

Table 2. Total coverage by sport and era (Alt. = altitude, Course = gain/loss and/or short/long course details)

Sport	Women	Men	Women (%)
Cross Country	29,113	46,376	38.6
Indoor Track	9,965	19,630	33.7
Outdoor Track	11,967	24,631	32.7
Road Race	986	1,200	45.1
Total	52,031	91,837	36.2
Unique athletes	12,364	18,987	39.4

Table 3. Results by sport and gender

Sport	Gender	Sprint	Mid	Dist.	Relay	Field	Total
Cross Country	F	0	0	29,113	0	0	29,113
Cross Country	M	0	0	46,376	0	0	46,376
Indoor Track	F	2,851	1,018	3,453	1,129	1,514	9,965
Indoor Track	M	6,184	2,019	7,175	2,003	2,249	19,630
Outdoor Track	F	3,853	1,563	3,904	729	1,918	11,967
Outdoor Track	M	7,808	3,425	8,977	1,534	2,887	24,631

Table 4. Results by category. Sprint.:

\leq

400 m and hurdles; mid: 401-1000 m; dist.:

>

1000 m and steeplechase.

Factor	Cross Country	In. track	Out. track	Road
Riegel distance	✓	✓	✓	✓
Heat ( $f_{\mathrm{weather}}$ )	✓	—	mid+	✓
Wind (per-result)	—	—	sprint	—
Elevation gain/loss	✓	—	—	✓
Course distance accuracy	✓	—	—	✓
AQI*	✓	—	✓	✓

Table 5. Adjustment factors by sport (AQI only measured)

$H$	Hadley (%)	Mid (%)	Quad. (%)	$f_{\mathrm{weather}}$
110	0.0-0.5	0.25	0.16	0.9984
120	0.5-1.0	0.75	0.64	0.9936
145	3.0-4.5	3.75	3.24	0.9676
160	4.5-6.0	5.25	5.76	0.9424
175	8.0-10.0	9.00	9.00	0.9100

Table 6. Hadley bands vs. quadratic surrogate (

k{=}0.0016

$k$	W reduc. (%)	M reduc. (%)	W infl. (%)	M infl. (%)
0.0010	50.9	35.1	9	13
0.0016	51.1	35.4	15	20
0.0020	50.9	35.1	19	24

Table 7. XC heat-

k

sensitivity. Reduc. = cross-meet SD; inflat. = converted vs. full.

Group	$n$	Raw	Conv.	Std.	Reduct.
Women (6 km)	2,201	108.09	54.66	52.86	51.1%
Men (8 km)	3,984	85.56	59.12	55.27	35.4%

Table 8. Median within-athlete cross-meet SD (XC,

\geq 3

meets/season). Reduct. = raw vs. standardized.

Group	$n$	Conv. only	Full std.	Inflat. (%)
Women	5,653	44.4 [41.7, 47.0]	38.6 [35.5, 41.5]	15 [11, 21]
Men	9,130	41.0 [39.0, 43.3]	34.2 [32.4, 36.5]	20 [17, 24]

Table 9. Median within-season improvement (sec) with 95% bootstrap CIs. Inflat. (%)

=100\times(\mathrm{median}_{\mathrm{conv}}/\mathrm{median}_{\mathrm{std}}-1)

Appendix figures & tables19 assets

Supplementary material from the paper’s appendix.

Appendix

Group	$n$	Raw	Conv.	Pre-wx	Full	Raw $\to$ full
Women (6 km)	2,201	108.1	54.7	54.6	52.9	51.1%
Men (8 km)	3,984	85.6	59.1	58.7	55.3	35.4%

Appendix

Table 10. Median within-athlete cross-meet SD (seconds) by adjustment stage.

Group	Dist. share	Env. share	Conv $\to$ full	Infl. (%)
Women	96.7%	3.3%	3.3%	15
Men	87.3%	12.7%	6.5%	20

Appendix

Table 11. Attribution of XC raw

\to

full median-SD drop (Dist./Env. shares) vs. converted

\to

full SD reduction and improvement inflation.

Factor subset	Median	Mean	P05 to P95	% outside $[0.95,1.05]$
Weather (all rows)	0.993	0.990	0.967 to 1.000	0.5
Weather ( $H{>}100$ only)	0.990	0.988	0.966 to 1.000	0.6
Elevation (grade)	1.000	1.000	1.000 to 1.002	0.0
Altitude (Péronnet)	0.996	0.996	0.993 to 1.000	0.0

Appendix

Table 12. Factor

f

such that adjusted time

=f\times

prior result time (comprehensive XC).

Model	Gender	Conv.	Std.	$\Delta R^{2}$ 95% CI
RF	Men	0.740	0.728	$[-0.006,\ +0.037]$
RF	Women	0.712	0.715	$[-0.034,\ +0.034]$
GB	Men	0.587	0.561	$[-0.006,\ +0.059]$
GB	Women	0.592	0.516	$[+0.033,\ +0.125]$

Appendix

Table 13. Test

R^{2}

by result-time adjustment.

\Delta R^{2}=R^{2}_{\mathrm{conv}}-R^{2}_{\mathrm{std}}

; 95% CI excludes 0 only for GB women.

Riegel config	Men raw $\to$ std	Women raw $\to$ std	Max $\|\Delta\|$ pp
Gender-specific (default)	35.4%	51.1%	0.0
Unified $b{=}1.06$	35.8%	50.9%	0.4
Unified $b{=}1.055$	35.4%	50.8%	0.3
Unified $b{=}1.08$	36.4%	51.1%	1.0

Appendix

Table 14. Median within-athlete SD reduction under alternate Riegel exponents (

\geq 3

meets/season).

Min races	Men $n$	Men infl.	Women $n$	Women infl.
2	9,130	19.9%	5,653	15.0%
3	4,055	18.9%	2,233	7.5%
4	1,340	16.8%	739	13.5%

Appendix

Table 15. Converted-only median improvement inflation vs. full standardization by minimum races per athlete-season.

Sport	$n$	Median factor	% outside $[0.98,1.02]$
Cross Country	31,450	0.985	52.4
Outdoor Track	18,619	0.998	9.7
Indoor Track	14,693	1.000	0.1
Road Race	854	0.995	2.3

Appendix

Table 16. Comprehensive-era running results: full sport-appropriate adjustment pipeline.

Gender	$n$ both	Spearman $r$	Med. XC best
Women	598	0.538	26:13
Men	688	0.431	29:05

Appendix

Table 17. Athletes with

\geq 1

XC and

\geq 1

road result: Spearman correlation of career-best standardized times (XC median shown; road distances are heterogeneous so road medians are not comparable).

Gender	$R^{2}$ conv.	$R^{2}$ std.	$\Delta R^{2}$
Men	0.720	0.720	$0.000$
Women	0.771	0.727	$+0.044$

Appendix

Table 18. Meet-redacted temporal split (train 2023; test 2024 with 25% of meets removed from feature construction).

Gender	$n$	P1	P5	P10	P25	P50	P75	P90	P95	P99
Men (8 km)	13,962	25:03	26:09	26:49	28:16	30:20	33:02	36:16	38:54	47:02
Women (6 km)	9,603	21:53	23:07	23:52	25:21	27:19	29:43	32:32	34:30	40:13

Appendix

Table 19. Career PR for all athletes with

\geq 1

parseable XC result (mm:ss); nationals included.

Cohort	Gender	$n$	P5	P50	P95
All athletes	Men	13,962	26:09	30:20	38:54
All athletes	Women	9,603	23:07	27:19	34:30
$\geq 2$ school years	Men	5,336	25:35	28:54	34:54
$\geq 2$ school years	Women	3,321	22:30	25:54	31:16
$\geq 3$ school years	Men	2,596	25:19	28:16	33:27
$\geq 3$ school years	Women	1,479	22:10	25:15	30:04

Appendix

Table 20. Career PR by longitudinal cohort (Aug 1 school-year boundary; Section D ). Anchor exceedance by cohort is in Table 22 .

Benchmark	Men (8k)	Women (6k)
NCAA D1 All-American	${\sim}$ 24:00	${\sim}$ 20:00
NCAA D1 5th runner (top-15 teams)	${\sim}$ 26:00	${\sim}$ 22:00
NIRCA nationals 5th runner (NRCD fit)	${\sim}$ 26:30	${\sim}$ 23:30
NCAA D2 5th runner	${\sim}$ 28:00	${\sim}$ 24:00
NAIA 5th runner	${\sim}$ 28:30	${\sim}$ 24:30
NCAA D3 5th runner	${\sim}$ 29:30	${\sim}$ 25:30

Appendix

Table 21. Standardized anchor times (fastest

\to

slowest). “5th runner”

=

fifth counting finisher on each of the top-15 teams at that division’s nationals (Section B ).

Anchor §	Men	Women
NCAA D1 All-American	0.1%	0.0%
NCAA D1 5th runner	4.2%	1.3%
NCAA D2 5th runner	21.6%	11.0%
NCAA D3 5th runner	40.2%	26.9%
NAIA 5th runner	27.7%	15.4%
NJCAA 5th runner	27.7%	26.9%

Appendix

Table 22. Share of athletes whose standardized career PR is at or faster than each anchor (Table 21 ). § Threshold labels shortened; times in companion table.

Band (faster anchor $\to$ slower)	Men	Women	Interpretation
D1 5th $\to$ NIRCA 5th	3.3%	6.0%	NIRCA depth, not D1
NIRCA 5th $\to$ D2 5th	14.1%	3.7%	D2 depth, not NIRCA
NIRCA 5th $\to$ D3 5th	32.7%	19.6%	D3 depth, not NIRCA
NIRCA 5th $\to$ NAIA 5th	20.2%	8.1%	NAIA depth, not NIRCA
NIRCA 5th $\to$ NJCAA 5th	20.2%	19.6%	NJCAA depth, not NIRCA

Appendix

Table 23. Athletes between adjacent anchors (standardized career PR). Read row as: faster than the slower anchor but slower than the faster one.

Sport	Era	Weather (%)		Course feat. (%)
		Res.	Meet	Res.	Meet
XC	comprehensive	99.9	99.3	38.5	18.2
XC	historical	27.2	14.3	3.0	0.8
Out. track	comprehensive	35.0	4.6	—	—
Road	comprehensive	100.0	100.0	33.4	43.2

Appendix

Table 24. Metadata coverage: share of results vs. share of meets (altitude omitted;

{>}97\%

throughout).

Sport	Athletes	$\geq$ 2 yr	P50 meets	P90 meets	P50 res.	P90 res.
Cross Country	23,565	36.7%	2	4	2	4
Outdoor Track	9,976	30.8%	1	3	2	5
Indoor Track	9,657	33.4%	1	2	1	4
Road Race	1,630	20.4%	1	1	1	1
All sports	31,351	37.0%	n/a	n/a	n/a	n/a

Appendix

Table 25. Per sport: athletes with results, % with

\geq

2 school years, and within-season meets/results per athlete-season (medians and P90). All-sports row from full export.

Table	Primary key	Role
result.csv	result_id	Finisher time, athlete, meet, event
meet.csv	meet_id	Date, sport, venue altitude, nationals flag
athlete.csv	athlete_id	Gender; no names in public export
team.csv	team_id	Institution / club label
athlete_team_association.csv	( athlete_id , team_id )	Roster membership by season
running_event.csv	running_event_id	Event name and distance

Appendix

Table 26. NRCD public export tables (v2.1.0: 143,868 approved results, 31,351 athletes with

\geq 1

result, 1,423 meets, 190 teams). Only approved=True meets and finishers are released.

File	Key fields
athlete.csv	athlete_id , gender
athlete_team_association.csv	athlete_id , team_id
meet.csv	meet_name , start_date , meet_city , meet_state , altitude , sport_id , nirca_only , regionals , nationals
result.csv	athlete_id , meet_id , running_event_id , result_time , grade , wind ; relay athlete_id_2 to 4 and splits
running_event.csv	event_name
sport.csv	sport_name

Appendix

Table 27. Public-export key fields (relational tables).

File	Key fields
course_details.csv	elevation_gain , elevation_loss , estimated_course_distance , date_of_event , temperature , dew_point , humidity , barometric_pressure , wind_speed , weather_conditions , altitude
joined.csv	Denormalized key fields from result, meet, team, athlete, and course tables; wind , result_time , grade , relay IDs/splits

Appendix

Table 28. Public-export key fields

Accelerometry-Derived Digital Biomarkers for Cardiometabolic Risk: A Population-Representative Tabular Benchmark with Uncertainty Quantification
Jun 29, 2026Federico FelizziAlgorithmic Fairness Wearable Health Monitoring
SIIAM, Rome, Italy · enGene Statistics, Basel, Switzerland.
Piecing Together Cross-Document Coreference Resolution Datasets: Systematic Dataset Analysis and Unification
Feb 28, 2026Anastasia Zhukova, Terry Ruas, Jan Philip Wahle +1Coreference Resolution Cross-Dataset Generalization
University of Göttingen Göttingen, Germany
ROAR: Unifying Runs across Heterogeneous AI-Driven Research Systems
Oct 2, 2026Leo Y. Lin, Vishakha Ramani, Z. Berkay Celik +2AI-Assisted Scientific Research Autonomous Research Agents
Purdue University · IBM Research

Jun 29, 2026J/K move · Enter open · S save

Accelerometry-Derived Digital Biomarkers for Cardiometabolic Risk: A Population-Representative Tabular Benchmark with Uncertainty Quantification

Federico Felizzi

SIIAM, Rome, Italy · enGene Statistics, Basel, Switzerland.

Algorithmic Fairness Wearable Health Monitoring Clinical Risk Prediction Regression Prediction Intervals Tabular Regression

Open paper AskPDF arXiv

Structured tabular data dominates clinical medicine, yet existing benchmarks fail to reflect real-world properties like complex survey sampling, demographic oversampling, and subgroup fairness. We introduce the NHANES Accelerometry Cardiometabolic Benchmark, derived from NHANES 2003-2006, comprising 1,381 adults with hip-worn accelerometry, fasting laboratory biomarkers, dietary intake, and anthropometrics. We evaluate three tabular learning methods -- ridge regression, XGBoost, and the foundation model TabPFN v2 -- to predict glycated haemoglobin (HbA1c), fasting triglycerides, and C-reactive protein (CRP) from activity phenotypes and lifestyle covariates. TabPFN v2 achieves the best overall performance (HbA1c R^2=0.156, CRP R^2=0.383), while triglycerides remain largely unpredictable (R^2 < 0.05), consistent with known genetic dominance. We apply split conformal prediction to generate distribution-free 90% prediction intervals and evaluate demographic coverage equity across sex and race/ethnicity subgroups. Marginal coverage aligns with the 90% target for CRP and HbA1c but falls below for triglycerides. At the subgroup level, we observe localized undercoverage (e.g., HbA1c for Mexican American participants), illustrating the gap between marginal guarantees and the conditional coverage required for clinical fairness. Code and data are at https://github.com/felizzi/nhanes-accel-cardiometabolic-benchmark.

Abstract

Explore similar work

Accelerometry-Derived Digital Biomarkers for Cardiometabolic Risk: A Population-Representative Tabular Benchmark with Uncertainty Quantification

Piecing Together Cross-Document Coreference Resolution Datasets: Systematic Dataset Analysis and Unification

ROAR: Unifying Runs across Heterogeneous AI-Driven Research Systems