Section 1
Introduction and Methodology
parkrun is one of the most successful mass-participation running events in history. Every Saturday morning, hundreds of thousands of people of all ages and abilities complete a 5km run at one of over 2,000 locations worldwide. To help runners of different ages and genders compare their performances, parkrun awards every finisher an age grade — a percentage score expressing how their time compares to a world-class standard for their age and sex.
parkrun's own description of age grading notes that it "makes no allowance for different weather conditions or the varying terrains of our courses" — a candid acknowledgement that the current system is intended as a rough guide rather than a precise metric. This paper takes that observation as a starting point and asks: what would a more complete grading system look like, and what does the data suggest?
Using 11,203,744 parkrun results from 860 UK courses across the 2025 calendar year, we explore the distribution of grades across different ages, genders and course types. We then propose a complementary metric — the Terrain-Adjusted Grade (TAG) — that adds course difficulty to the existing age and gender adjustment. The aim throughout is constructive: to use the remarkable dataset that parkrun generates to offer an alternative perspective, not to critique what is already a genuinely wonderful institution.
A note on methodology: two independent approaches
The course speed factors at the heart of TAG were derived using two entirely independent methods — one based on grade distributions across all runners at each course, and one based on paired comparisons of the same runner's performance at different courses. The two methods use different data, different mathematics, and make different assumptions. The fact that they produce closely agreeing results (correlation r=0.886, mean difference 0.036) provides strong independent validation that the factors are measuring something real.
Where the methods diverge slightly, we take the average — which partially cancels the known bias of each: the grade distribution method is sensitive to which types of runner self-select to attend each course, while the paired method has a mild "tourist bias" because runners who visit multiple courses tend to be faster than average. Averaging reduces both effects simultaneously. The final factors are therefore more robust than either method alone could provide.
Key validation finding
Two methodologies built on entirely different mathematical foundations — one measuring grade distributions across populations, the other tracking individual runners across courses — independently converge on the same course difficulty rankings. Belfast Victoria emerges as the reference course in both. Great Yarmouth North Beach emerges as the most challenging in both. The agreement across 860 courses gives us confidence that the factors reflect genuine terrain difficulty rather than artefacts of any single method.
Data and processing overview
The full dataset of UK parkrun 2025 results was processed in Python using Apache Arrow for memory-efficient loading. Key analytical steps include: parsing finish times; recovering exact single-year ages from the displayed grade percentage using parkrun's reverse-engineered base time table; computing grade under the Alan Jones (ALJ) 2020 age grading tables; deriving empirical course difficulty factors using both grade distribution and paired runner network methods; blending the two factor sets; and combining with ALJ age factors into the TAG formula. Runner IDs (A-numbers) were used as unique identifiers for 95.9% of records, eliminating the name-matching ambiguity present in earlier work. All code and intermediate data are available on request.
Section 2
What is Age Grading and How Does parkrun Use It?
Age grading converts a raw finish time into a percentage score by comparing it to the theoretical best achievable time for a runner of that age and gender. The formula is straightforward:
In practice, grades above 85% represent exceptional club-standard performance.
70–85% is strong recreational running. 50–60% is typical for regular participants.
The "base time" for each age is derived from world-record performances, scaled by an age factor that reflects the expected physiological decline with age. A 55-year-old woman running a particular time is compared not to the open-class world record but to the world record for a 55-year-old woman.
How parkrun calculates grade
parkrun applies this system to every result. Their own description, published on their website, is notably brief:
"All parkrun events use age grading to allow parkrunners to compare results. Age grading takes your time and uses the world record time for your gender and age to produce a score (a percentage).
Age Grades are calculated to allow rough comparisons between our participants, and should not be taken too seriously. For example, age grading makes no allowance for different weather conditions or the varying terrains of our courses.
We do not share the actual table used to perform the calculations but it is loosely based on the tables produced by WMA, previously known as WAVA."
— parkrun.com, age grading FAQSeveral things are worth noting here. parkrun is transparent about the fact that their tables are only loosely based on WMA standards — they have made adjustments appropriate for a mass-participation recreational event rather than a competitive athletics meeting. They also acknowledge upfront that age grading makes no allowance for course terrain. This paper takes both points seriously and tries to build on them.
Our starting question
If parkrun's grades are a "rough comparison" by design, what would a more precise version look like — one that accounted for the actual terrain of each course and used a fully published, auditable age standard? That is the question this paper explores.
Section 3
Are parkrun Grades Fairly Distributed?
If an age grading system is working as intended, the distribution of grades should be broadly similar across age and gender groups. The proportion achieving high grades should be roughly stable — because a high grade always means the same thing: exceptional performance relative to peers of the same age and gender.
We examined this across all 11.2 million 2025 UK results. The patterns are interesting — particularly for female runners over 50, where the numbers are large enough to draw meaningful conclusions.
The female picture from age 50 onwards
Among female runners, the rate of achieving >85% grade rises steadily with age. At VW50–54 (453,004 runners — a very substantial sample), the rate is 0.117%. By VW65–69 (158,265 runners) it has risen to 0.890%. The median grade tells a similar story: SW25–29 median is 48.2%, rising to 57.8% by VW65–69.
Some rise is expected — older runners who remain active tend to be more committed, and selection effects are real. The question is whether the magnitude of the rise reflects genuine performance differences or whether the underlying age standards diverge from real-world expectations at older ages. The data suggests the latter plays a role.
Figure 1: Female runners achieving >85% grade, by age group
Note the steady rise from VW50–54 onwards. The 50–75 age range contains hundreds of thousands of runners — the pattern here is statistically robust. Source: 11.2M UK parkrun results, 2025.
parkrun grades. n = 10,578,315 valid results.
The male picture: broadly stable
Male grades are notably more consistent. The >85% rate moves from 0.010% at SM25–29 to 0.161% at VM65–69 — a 16-fold increase compared to a much larger rise for women over the same age span. Male median grades are essentially flat from the 40s through the 70s, hovering between 52% and 57%. This suggests the male age standards track real-world performance well across the full age range.
Figure 2: Median grade by age group — parkrun system
Males (blue) are broadly flat from VM45 onwards. Females (red) show a steady upward trend from VW50. The divergence between male and female veteran patterns is the key observation of this section.
Median parkrun age grade by 5-year age/gender group.
Observation — not a criticism
The female veteran age grade trend is worth examining. Among the 453,004 VW50–54 runners alone, the median grade is 4.3 percentage points higher than for SW25–29. Whether this reflects a calibration question in the underlying tables or a genuine performance phenomenon is exactly what the rest of this paper investigates — using a published, auditable alternative standard for comparison.
Section 4
How We Reverse-Engineered parkrun's Grades and Ages
Before we can compare parkrun's grades against any external standard, we need to know two things that parkrun does not publish: the exact age factor table they use, and the exact age of each runner (parkrun records only a five-year band such as "VM50–54").
Recovering exact age from the displayed grade
The key insight is that parkrun uses a single discrete base time per year of age — not an average across the five-year band. Within any band, there are exactly five possible base times. Because the displayed grade is computed as base_time / finish_time × 100, we can reverse this:
The closest match (within ±3 seconds tolerance) gives the exact year of age.
This method resolved exact ages for 95.9% of the 11.2M records.
The ±3 second tolerance accommodates display rounding (parkrun shows grades to one decimal place). Where the implied base time falls ambiguously between two years, we fall back to the band midpoint. The 4.6% fallback rate is concentrated in junior age groups and edge cases near age-group boundaries.
Recovering the base time table
By examining the implied base times across a large number of finishes in each age-year, we can recover the complete table of base times parkrun uses — effectively reverse-engineering their unpublished factor table. This gives us, for the first time, a direct basis for comparison with the published ALJ 2020 standards.
Why this matters
Without exact ages, age grading analysis is approximate. A five-year band introduces up to 2.5 years of error per individual. At ages above 60, where age factors steepen to ~0.007 per year for women, this introduces grade errors of up to 1.75 percentage points per person. Our 95.9% runner ID resolution rate makes the subsequent analysis substantially more precise than any prior published work on this dataset.
Section 5
What is WMA and Who is Alan Jones?
World Masters Athletics (WMA) — formerly known as WAVA (World Association of Veteran Athletes) — is the international governing body for masters (veteran) athletics. Among its responsibilities is the publication of age grading tables: factor tables that allow athletes of different ages and genders to compare performances on a common scale.
The WMA tables have been periodically revised as new world-record performances accumulate. The most widely adopted revision was prepared by Alan Jones, a British statistician and masters athlete who has maintained and refined the tables for decades. His 2020 revision — the ALJ 2020 tables — is the version used by parkrun itself, by the majority of athletics databases, and by timing systems worldwide. A further revision was published in 2025, but as of this writing it has not been widely adopted, and parkrun continues to use the 2020 version.
What ALJ 2020 provides
For 5km road running, the ALJ 2020 tables provide:
- Open-class standards: 12:53 (773 seconds) for men, 14:48 (888 seconds) for women
- Age factors for every year of age from 5 to 100, for both genders
- These factors are derived from world-record performances at each age, smoothed to remove noise from single exceptional athletes
The key difference from parkrun's undisclosed table is transparency: the ALJ factors are published, auditable, and based on explicit methodology. This is why we adopt them as our reference standard — not because they are perfect (we will show they are not), but because they are the best available published standard.
Why ALJ 2020 rather than the 2025 revision?
A 2025 revision of the WMA tables has been prepared, with notably faster open-class standards for women (13:54 for 5km, versus 14:48 in 2020). However, parkrun continues to use the 2020 tables, and the 2025 revision is anchored to a small number of extraordinary elite performances that have not yet been validated against real-world mass participation data. Using the same 2020 standard as parkrun ensures direct comparability and avoids importing calibration uncertainties from a revision that remains in early adoption. The 2025 tables are a significant development and warrant a dedicated future analysis once sufficient real-world data accumulates.
Section 6
Base Times: ALJ vs parkrun — and Why the Difference Matters
Having reverse-engineered the base times parkrun actually uses, we can compare them directly against the ALJ base times. The ALJ base time for a given age is the ALJ open-class standard — 12:53 (773s) for men and 14:48 (888s) for women — divided by the ALJ age factor for that age. It is the time a runner of that age would need to achieve 100% on the ALJ scale. Both parkrun and TAG use the same open-class standard; the difference lies in how the age factors are applied across the age range, and in TAG's additional correction for course difficulty.
Male base times: close agreement with a small growing gap
For male runners, parkrun's empirical base times are very close to the ALJ base times at younger ages — at age 25 they are identical, since both use the same open-class standard of 12:53. A small gap opens with age, reaching around 19 seconds by age 84. This reflects a modest tightening of parkrun's empirical age factors relative to the ALJ tables for older men, but the effect is small and does not distort grades materially.
Female base times: a crossover at age 55
The female picture is more complex and more revealing. At younger ages, parkrun's female base times closely match the ALJ base times — they are identical at age 25. But from around age 55 onwards the relationship diverges: parkrun's base times become increasingly slower than the ALJ base times, meaning parkrun begins awarding progressively more generous grades to older women than the ALJ standard would. By age 84, the parkrun female base time is 169 seconds (nearly 3 minutes) slower than the ALJ base time — this divergence explains the grade inflation trend observed in Section 3.
Figure 3: Difference between parkrun and ALJ base times by age (seconds)
Negative = parkrun base time is faster than ALJ (parkrun harder for the runner). Positive = parkrun base time is slower than ALJ (parkrun more generous). Men remain negative throughout. Women cross from negative to positive around age 60 — the direct cause of the veteran female grade inflation in Section 3.
parkrun base times (reverse-engineered) minus ALJ base times (ALJ OC ÷ ALJ age factor), ages 18–84. ALJ OC: 12:53 male, 14:48 female.
| Age | ALJ base M | ALJ base F | parkrun base M | parkrun base F | Diff M (s) | Diff F (s) |
|---|
What the divergence tells us
For men, parkrun's empirical base times closely track the ALJ standard throughout the age range — the two systems agree at age 25 and diverge only slightly at older ages. For women, the picture is very different: parkrun agrees with the ALJ standard at younger ages but diverges sharply from around age 60, becoming increasingly generous. By age 80, a female parkrunner receives a grade approximately 5–6 percentage points higher under parkrun than under the ALJ standard. This divergence — absent for men — is the direct cause of the female veteran grade trend observed in Section 3, and it is precisely what the rest of this paper investigates.
Section 7
Applying ALJ Grades Directly to 11.2 Million parkrun Times
We now apply the ALJ 2020 factors directly to every parkrun result, using the exact ages recovered in Section 4. This gives every runner an ALJ grade calculated from their actual finish time and their exact single year of age.
ALJ grades are noticeably flatter for women 50+
The most meaningful comparison is in the 50–75 age range, where the data is densest. At VW50–54 (453,004 runners), the >85% rate falls from 0.117% (parkrun) to 0.064% (ALJ) — a near-halving. At VW60–64 (289,153 runners), the rate falls from 0.464% to 0.276%. The median grade at VW65–69 moves from 57.8% (parkrun) to 55.3% (ALJ) — a 2.5 point reduction that brings it much closer to the SW25–29 baseline.
The pattern is consistent: ALJ produces a flatter, more stable distribution of grades across female age groups throughout the 50–75 range where the majority of veteran female runners sit.
Figure 4: Female >85% rate — parkrun vs ALJ, by age group
ALJ is substantially flatter than parkrun, confirming that the majority of the veteran female inflation is traceable to parkrun's non-standard base times. A residual upward trend remains in ALJ, which we document as a known limitation.
n = 3,883,270 female results.
Figure 5: Female median grade — parkrun vs ALJ
ALJ median grades are more stable across age groups than parkrun. The residual upward trend in ALJ (from ~48% at SW25–29 to ~59% at VW85–89) indicates that the ALJ female veteran factors are themselves somewhat over-generous at extreme ages.
Median grade by 5-year female age group. parkrun (red) vs ALJ 2020 (blue).
The male picture under ALJ
For male runners, ALJ grades are very similar to parkrun grades throughout the age range. Both systems show a broadly flat >85% rate from SM25–29 through VM70–74, with a modest rise at the oldest age groups reflecting both selection effects and a small age factor drift. The male ALJ factors are well calibrated.
Figure 6: Male >85% rate — parkrun vs ALJ, by age group
Males show a broadly flat pattern under both systems, with close agreement between parkrun and ALJ. This confirms that parkrun's undisclosed male adjustments are close to the ALJ standard.
n = 5,157,540 male results.
Conclusion: ALJ as a useful reference standard
ALJ 2020 produces a noticeably flatter grade distribution across female age groups in the 50–75 range, where the data is most substantial. It is published, auditable, and used by the majority of athletics databases worldwide. We adopt it as our reference standard for the TAG calculation — not as a claim that it is perfect, but as the best available published alternative to work with.
Part Two
Course Variability and the Terrain-Adjusted Grade
Age grading corrects for who you are. But it says nothing about where you run.
Section 8
Course Variability: Two Independent Methods, One Answer
parkrun's own website acknowledges that "age grading makes no allowance for… the varying terrains of our courses." Across 860 UK courses, the difference in typical finishing times attributable to course difficulty alone is large. A runner completing Great Yarmouth North Beach — which runs on beach sand and over dunes — in 30 minutes is working far harder than a runner finishing Belfast Victoria in 30 minutes on a flat, fast course. Yet both receive identical age grades for that time.
To correct for this we derived speed factors for all 860 UK parkrun courses. Crucially, we used two completely independent methods — one measuring grade distributions across populations, the other tracking individual runners across multiple courses — and found that they converge on the same answer. This agreement is the strongest evidence that the factors are measuring genuine terrain difficulty.
Method 1: Grade distribution across all runners
For each course, we computed the ALJ grade distribution across all 2.8 million personal best performances from runners with at least 5 parkruns. Nine quantiles (P10 through P90) were compared against the theoretical maximum across all courses. The course factor is the median ratio of the course's quantile profile to the reference. This method uses the full population at each course.
Method 2: Paired runner comparisons
For the 519,089 runners who attended two or more different courses during 2025, we compared their median performance at each venue directly. For every pair of courses sharing at least 3 common runners, we computed the log ratio of their typical times — 654,524 course pairs in total.
Each pair gives us one equation: "course A appears to be X% faster than course B, based on how the same runners performed at both." The challenge is that these pairwise observations are noisy and sometimes contradictory — runner A might suggest course X is faster than Y, while runner B's results suggest the opposite. To resolve this, we set up a single large system of equations — one per course pair — and find the set of course factors that best satisfies all 654,524 equations simultaneously, using standard linear regression. This is equivalent to finding the most internally consistent set of factors given all available evidence at once, rather than chaining pairwise comparisons together. Belfast Victoria is fixed at 1.000 as the anchor. Each runner acts as their own control, eliminating ability as a confound.
The two methods agree closely
After correcting the paired factors for directionality, the two methods produce a correlation of r=0.886 with a mean absolute difference of 0.036. They agree almost perfectly for large, diverse courses — Battersea's factor is 0.9741 in method 1 and 0.9739 in method 2, a difference of just 0.0002. They diverge most at specialist courses where the tourist runner population is least representative of the general field.
Why the methods diverge at some courses
Runners who visit multiple courses tend to be faster than average parkrunners. When these "tourists" visit a challenging course like Cannock Chase, their times there look better than the average runner's would — so the paired method concludes the course is slightly faster than it really is. The grade distribution method is immune to this but sensitive to the overall ability of a course's regular field. Averaging the two methods partially cancels both biases.
The blended factors
The final course factors are the average of the two methods, normalised so Belfast Victoria = 1.000. The resulting distribution across 860 courses is:
Figure 7: Distribution of course speed factors — 860 UK parkrun courses
The distribution clusters around 0.933 with a long left tail of challenging courses. Derived by averaging two independent methods. Belfast Victoria = 1.000 reference.
Blended factors from grade distribution and paired runner network methods. All 860 UK courses.
What the factor captures
The factor absorbs everything that makes one course faster or slower than another: elevation, surface type, exposure, number of turns, and gradient profile. It also absorbs the typical start conditions at that event — a large parkrun with mass-start congestion will have a slightly lower factor than an equivalent course with a smaller, faster field. This is a feature rather than a limitation: it means the factor correctly describes the actual performance environment a runner faces on a Saturday morning.
Section 9
Course Speed Factors — All 860 UK parkrun Courses
The table below shows the blended speed factors for all 860 UK parkrun courses — derived by averaging the grade distribution method and the paired runner network method described in Section 8. A factor of 1.000 represents Belfast Victoria. Every other factor is ≤ 1.000.
The factor can be read as: "at this course, runners achieve grades approximately (1 − factor) × 100% below what they would on the reference course." A factor of 0.90 means grades are typically about 10% below the reference level.
Belfast Victoria: the natural reference course
Belfast Victoria parkrun was not chosen as the reference course — it emerged as the reference from the data. Its grade distribution most closely matches the theoretical maximum across all quantiles. It is also the course where both the men's (Nick Griggs, 13:44) and women's (Ciara Mageean, 15:13) parkrun world records were set. The data and the records agree independently. Note that the course factor reference (Belfast Victoria = 1.000) is distinct from the TAG open-class standard: the OC is the ALJ age 25 base time (12:53 men, 14:48 women), which is faster than the course records and reflects the world-class road standard that parkrun's own grading uses.
| Course | Men | Women | Combined | Speed |
|---|
860 courses. Factor of 1.000 = Belfast Victoria (reference). Blended from grade distribution and paired runner network methods.
Section 10
Introducing the Terrain-Adjusted Grade
We now have two independently validated components:
- An age correction: ALJ 2020 factors applied at exact single-year resolution, giving a grade that is fair across ages and genders (with documented residual limitations at extreme ages for women)
- A course correction: blended speed factors for all 860 UK parkrun courses, validated by two independent methods
Combining these gives the Terrain-Adjusted Grade (TAG) — a single number that corrects for both where you run and how old you are.
OC(F) = 888s — ALJ 2020 open-class standard for 5km road, women (14:48)
age_factor = ALJ 2020 road factor for exact single year of age
course_factor = grade distribution factor ≤ 1.000, referenced to Belfast Victoria
finish_time = elapsed time in seconds
Why use the ALJ open-class standard, not parkrun world records?
The ALJ 2020 tables define an open-class standard for 5km road running of 12:53 (773s) for men and 14:48 (888s) for women. These are the same base times that parkrun itself uses in its own age grading — as we established by reverse-engineering parkrun's unpublished factor table in Section 4. Using them as our OC standard makes TAG directly consistent with parkrun's own grading methodology.
The parkrun world records — Nick Griggs 13:44 at Belfast Victoria and Ciara Mageean 15:13 at Belfast Victoria — are exceptional performances and historically significant. But they are course records, not open-class standards. There are faster 5km runners in the world: Griggs and Mageean are elite athletes of the highest calibre, but the ALJ tables are calibrated to the theoretical world-class road standard, which is faster still. Using the course records as OC would produce TAG grades that are systematically higher than parkrun's own grades for the same performance — creating inconsistency between the two systems. A TAG of 100% should mean the same thing in both.
Belfast Victoria remains the reference course for our course factors — it emerges naturally from the data as the fastest UK parkrun course, and it is right that it anchors the factor scale at 1.000. But the OC standard is a separate question from the reference course, and the ALJ open-class times are the correct answer to it.
The parkrun world records vs the open-class standard
The ALJ 2020 open-class standard (12:53 men, 14:48 women) is faster than the current parkrun world records set at Belfast Victoria (Nick Griggs 13:44, Ciara Mageean 15:13). This gap — 51 seconds for men, 25 seconds for women — reflects the conditions of a free weekly mass-participation event: congestion at the start, a mixed-ability field, no dedicated pacers. The open-class standard is a road time trial standard; parkrun is something different. Using the ALJ standard as our OC means a TAG of 100% means the same thing as 100% in parkrun's own grading system, not 100% of the Belfast Victoria course record.
TAG as a time predictor
TAG also works as a course predictor. Given a time at one course, the predicted time at another is:
Section 11
TAG Grade Distributions Across All 860 Courses
We apply the full TAG formula to all 10,578,315 valid results. The central test is whether TAG eliminates the relationship between course difficulty and the grade a runner receives.
The principal result
It does. Under the current parkrun system, a runner on a very slow course (factor <0.85) receives a median grade of 48.5% — 8.4 percentage points below a runner on the reference course (56.9%). Under TAG, the spread across the five course speed bands collapses to just 0.9 percentage points.
Figure 8: Median grade by course speed band — parkrun vs ALJ vs TAG
The critical fairness test. Under parkrun, slow-course runners are heavily penalised. Under TAG, median grades are essentially flat across all course speed bands.
n = 10,578,315. Five course speed bands from Very slow (<0.85) to Reference (>0.98).
Figure 9: % achieving >85% grade by course speed band
Under parkrun, the rate of achieving >85% is strongly correlated with course speed. Under TAG, the rate is essentially uniform across all bands.
Same dataset. TAG >85% rate ranges from 0.65% to 0.97% across all five bands.
| Course speed band | N | parkrun med | ALJ med | TAG med | parkrun >85% | ALJ >85% | TAG >85% |
|---|
Principal result — Part 2
TAG reduces the median grade spread across the five course speed bands from 8.4 percentage points (parkrun) to 0.9 percentage points. Course difficulty ceases to be a meaningful determinant of grade. A runner on a challenging hill course and a runner on a flat coastal path receive directly comparable TAG scores for equivalent efforts.
Overall grade distribution
Applying TAG across all 11.2 million results produces the following population-level statistics. The TAG median (57.3%) is higher than the parkrun grade median (52.9%), reflecting the fact that parkrun's empirical base times at younger ages are faster than the ALJ standard — holding younger runners to a harder benchmark and suppressing their grades relative to TAG.
Section 12
Next Steps and Call for Feedback
What TAG achieves
TAG is a universal 5km performance metric that accounts simultaneously for age, gender, and course difficulty. Validated against 11.2 million results across 860 courses, it reduces course-induced grade variation by approximately 89%, collapsing the cross-band median grade spread from 8.4 to 0.9 percentage points. It can be computed in real time from a finish time, an age, and a course name.
We think of TAG as complementary to parkrun's existing grade — not a replacement. parkrun's grade is simple, immediate, and familiar to millions of runners. TAG adds a layer of precision for those who want to compare performances across different courses, or to track progress more accurately across their parkrun journey.
TAG beyond parkrun: Skamper and the universal grading concept
The TAG concept did not originate with this parkrun analysis. It was developed as the core performance metric for Skamper — a free app for discovering and comparing runs across Britain, covering everything from 1km to 100km.
At Skamper, TAG faces a harder problem: there is no population dataset of millions of results to derive empirical course factors from. Instead, Skamper computes TAG analytically from three course parameters: distance, total ascent, and the percentage of the course that is paved. From these, a world-best base time can be derived for any course — the theoretical fastest time a world-class runner would achieve on that specific combination of distance, climb, and surface. A runner's TAG is then their time expressed as a percentage of that world-best base time, adjusted for age and gender using the same ALJ 2020 tables used here.
Two routes to the same metric
Skamper builds TAG from first principles using course geometry. This parkrun analysis builds TAG empirically from 11.2 million results. The two approaches use completely different inputs and methods — yet they are designed to produce comparable grades. A TAG of 65% means the same thing whether it was computed from your parkrun time at Hackney Marshes or from your half marathon time on a hilly trail in the Peak District.
The most accurate race predictor available
Because TAG is comparable across distances, surfaces, and terrains, it enables something no single-distance grading system can: genuine cross-distance race prediction. In the Skamper TAG calculator, a runner can enter any recent result — a parkrun, a 10km, a half marathon — and receive predicted equivalent times for any other distance and course type, all derived from their TAG.
This works because TAG is a single consistent measure of running ability independent of distance or terrain. A runner with a TAG of 68% at 5km on a flat course should achieve approximately 68% at marathon distance on a comparable course — and the Skamper predictor uses the full course parameter model to adjust for distance, climb, and surface simultaneously. The result is the most complete race predictor currently available to recreational runners, grounded in the same age-grading mathematics that underlies this paper.
Skamper and this paper
The TAG concept, the ALJ 2020 age factors, and the Belfast Victoria OC standard are shared between this research and the Skamper platform. Skamper is free to use and available on iOS and Android at skamper.com. The cross-distance TAG calculator and race predictor is at skamper.openair.tools.
Known limitations
Female veteran age factors. The ALJ female factors for ages 65+ show a residual upward trend in grade distribution. A recalibrated set of female veteran factors — possibly derived from road race data rather than parkrun alone — would be the highest-priority improvement to TAG.
Course stability. Course factors reflect the 2025 course configuration. Significant course alterations (a new section, a change of surface) would require factor recalculation.
Junior runners. Junior world records are set on closed tracks under conditions very different from parkrun. Junior TAG grades should be interpreted with caution.
Sparse courses. A small number of courses with very few results have less reliable factors.
What we are building
The TAG formula is available via a free calculator at parkrun-calculator.openair.tools, allowing any parkrunner to calculate their TAG, compare performances across all 860 UK courses, and predict their expected time at any venue. The full methodology and course factor table are published here.
An invitation to collaborate
This is a working paper and we genuinely welcome feedback — from statisticians, masters athletics experts, parkrun regulars, and anyone with relevant data. We would be particularly interested in:
- Review of the course factor methodology from anyone with expertise in grade distribution analysis
- Elite female veteran road performance data to support recalibration of age factors at older ages
- Feedback from runners about whether TAG matches their experience of relative course difficulty
- International parkrun data to extend course factors beyond the UK
"parkrun is a remarkable institution. Every number in this paper comes from the dataset that parkrun's community of volunteers and participants has collectively built over two decades. This analysis is offered in that spirit — as a contribution from the data back to the community."
Contact
Openair Research · parkrun calculator · Skamper TAG calculator · Independent analysis. Not affiliated with or endorsed by parkrun.
Share your thoughts
RESPONSES ARE ANONYMOUS AND USED ONLY TO IMPROVE THIS RESEARCH · TAKES ABOUT 2 MINUTES
How useful did you find this analysis? *
How well does TAG match your own experience of course difficulty? *
Select N/A if you haven't tried the calculator yet
What best describes you? *
Which part of this paper interested you most? *
Any thoughts, corrections or suggestions? Optional
500 characters remaining
Notes
1. parkrun is a series of free, weekly, timed 5km events. "parkrun" (lowercase p) is used throughout, consistent with the organisation's own style. This analysis uses publicly recorded results and is not affiliated with or endorsed by parkrun.
2. ALJ 2020 tables: Alan Jones, World Masters Athletics age grading tables, 2020 revision. Road 5km open-class standards as used by parkrun (reverse-engineered): men 12:53 (773s), women 14:48 (888s). The published ALJ road OC is approximately 12:51 for men; the small difference reflects rounding in parkrun's implementation.
3. TAG open-class standards: ALJ 2020 road 5km, men 12:53 (773s), women 14:48 (888s). These are the same base times parkrun uses in its own age grading, as established by reverse-engineering parkrun's unpublished factor table (Section 4). The parkrun course records — Nick Griggs 13:44, Ciara Mageean 15:13, both at Belfast Victoria — are exceptional performances but are slower than the ALJ open-class standard and are not used as OC.
4. The WMA/ALJ tables rely on world-record performances to set factors. At very old ages, world records are set by exceptionally unusual individuals — a single exceptional athlete can shift the factor table for an entire cohort. Sparsity of data at older ages is a known problem in age grading methodology.
5. Skamper Ltd is registered in Scotland, company no. SC742632. The Skamper app is available free on iOS and Android at skamper.com. The cross-distance TAG calculator is at skamper.openair.tools.