Why Your parkrun Age Grade Is a Number You Can't Check

22:00 at a parkrun on Saturday morning. 22:00 at a measured 5K three weeks later. Same runner, 48 years old, same legs.

Saturday said 66.0%. The race, graded here, says 65.4%.

Six tenths of a point is nothing. The interesting part is that you cannot find out where it came from, because one of those two numbers was produced by a table nobody outside parkrun has ever seen.

Two percentages, one runner, no way to make them agree

Last time I wrote about why every age-grade calculator gives you a different number, and the argument was that the useful question isn't which calculator is right. It's which one tells you what it used. My ten-miler scores 83.33%, 83.09% or 82.50% depending on the edition, and all three are checkable, because Alan Jones publishes his tables.

parkrun is the largest distributor of age grades in the world. Millions of percentages issued every Saturday morning, free, to people who mostly aren't in a club and don't own a spreadsheet.

It is also the one age grade in the sport you cannot audit.

That isn't an accusation. Read the next section before deciding it is.

What does parkrun say about its own number?

Their support page, article 4.5, last edited on 9 July 2024 and unchanged since. It opens like this.

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).

Then, two paragraphs down:

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.

Read that twice. parkrun tells you the number is rough, tells you not to lean on it, and names two specific reasons why. I've read a lot of age-grade copy this year and almost none of it is that honest. Most calculators hand you four significant figures and say nothing at all.

Then the sentence this whole post hangs on:

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.

Loosely based. Their words, volunteered, on their own help page, not dragged out of anyone.

So parkrun's stated position is that this is an approximate number, from an unpublished table, for casual comparison between people standing in the same field. Everything below is what happens when a reader takes that number somewhere parkrun never said it should go.

Which table is parkrun actually using?

Nobody outside parkrun knows. The help page names WMA as the family and stops there, so every claim about which edition — including the ones below — is inference from outside checks rather than documentation. Here is how close the outside gets.

The most detailed public account is from the volunteer team at Crathes Castle parkrun, who wrote the whole thing up in 2023. Their reading: the grades are "broadly based on the WMA (Word Master Athletics) 2006/2010 tables for track 5k records - to which they have added figures for 4 year olds and over 100s". Separately, the calculator at hometreadmill.co.uk reports checking a real M60 parkrun result against its own 2006/2010 figure and matching it to within about a tenth of a percent.

I can add one more data point, computed here.

That Crathes article lists four parkrunners it says all score exactly 60%: a 9-year-old girl in 29:15, a 23-year-old man in 21:30, a 65-year-old man in 27:57, and a 75-year-old woman in 40:25. Four ages, both sexes, a nineteen-minute spread of finish times.

Run all four through the WMA 2006/2010 factors at 5 km and they come out at 59.99%, 60.00%, 60.01% and 60.00%.

Run the same four through Alan Jones 2025 and they scatter — 58.7%, 59.6%, 59.4%, 55.8%.

That is a very good fit. It is not proof, and there are three reasons to say so out loud.

• The quartet may itself have been generated on the same 2006/2010 calculator I checked it against. If so I've replicated arithmetic, not tested parkrun. I can't rule that out from here.
• Crathes says the basis is the track 5k records. But the fit only appears on the road factor set, whose open standard is 12:54, against 12:37 on the track. So the inference disagrees with itself on the detail that matters most.
• parkrun has added ages at both ends beyond where WMA stops. Those rows can't be checked against anything, because there is nothing to check them against. Fauja Singh's 179.04% at Valentines parkrun, set the day before his 101st birthday, is a number that cannot exist in the WMA tables at all — which is a good moment to remember that age grading isn't a participation trophy, it's just a ratio, and a ratio will do whatever the denominator tells it to.

How far off does it run?

Assume the inference is correct, and grade the same 5K both ways. These are computed from the two factor sets directly — the 2006/2010 arrays out of Howard Grubb's calculator page, the current ones from the Alan Jones 2025 road tables RaceRecords ships. No third-party website was asked for an opinion.

For men that's a rounding error. Six or seven tenths, which is roughly what the running web means when it says parkrun grades run slightly generous.

For women it's four to five times bigger, and I haven't seen anyone say so.

So why does the same table drift cost a woman four times what it costs a man? One number moved. An age-graded percentage is your time measured against the world best for your age, so the world best is the floor of the whole calculation. Here's what that floor did between the two editions.

Five seconds against fifty-four.

So a woman holding a Saturday percentage next to a race percentage is looking at a two-to-three point gap that has nothing to do with her, nothing to do with parkrun, and everything to do with what women's distance running has done since 2006. The men's tables barely noticed the same fifteen years.

What is the course underneath the number?

Now the part that has nothing to do with tables at all. A parkrun course is measured with care and certified by nobody, and that second gap is larger than anything the editions can do to each other.

An age grade assumes the stated distance is real. A certified road course has been measured by an accredited measurer on a calibrated bicycle, along the shortest possible route a runner could take, with a short-course prevention factor added on top. It's paperwork, and the paperwork is the point.

A parkrun is not that and has never said it was. Courses are laid out with real care — Nonsuch parkrun describes plotting the route on mapping software, riding it with a bike computer accurate to ±10 m, then walking it with a surveyor's wheel accurate to ±1 m, and notes that "GPS (either watch or phone) is not used because it is neither accurate nor consistent". That is more diligence than several official 10Ks I've paid to enter. It's still not certification, and parkrun doesn't submit it for any.

Crathes put it in one phrase, explaining why their results carry 100%-plus scores most weeks: "parkrun is not an official event and the distance is 'more or less 5k'".

Then there's the ground itself. The same volunteers estimated their own course's penalty: "our range of age graded percentages at Crathes could be considered 6-10% lower when compared with a fast parkrun". Their average finish time is 30:55. Edinburgh's, flat and fast, is 26:46.

So which of the two problems is bigger, the table or the ground? A table of uncertain vintage, worth around a point. Terrain, worth up to ten. The unpublished table is the small problem, and it's the one everybody argues about.

Two runners with the same parkrun age grade have not necessarily run the same performance. They may not have run the same distance.

What does that mean for a record book?

RaceRecords is a record book for races — measured courses, official timing, one table edition across your whole history. That's the whole product, and the reason is the sentence above. A percentage means something only if the distance underneath it is fixed.

Which is exactly why I'm not going to pretend a parkrun doesn't count.

Log it. The race form has a Certified course distance checkbox, and for a parkrun that hasn't been through official measurement you untick it. The percentage is still computed on Alan Jones 2025, the result still sits in your race history, and it carries an uncertified flag everywhere it appears — race list, detail page, public profile. Your certified-race count doesn't include it. The methodology page says the same thing in fewer words.

That's the honest handling. Not excluded, not quietly promoted.

The part where I concede two things

Two things cut against everything above, and both belong in the post rather than in a footnote. The first one is the bigger of the two, and it's about my own site.

RaceRecords imports races from Strava, and GPS distance is worse than a parkrun course. Not marginally worse. A wheel-measured parkrun laid out to ±1 m beats any watch on any wrist. If I claimed certification as a universal standard here while imported activities walked in the back door, I'd be running the same unverifiable-number problem I've just spent fifteen hundred words describing.

So here's exactly what happens instead. A Strava import never grades on the GPS distance. The activity snaps to a canonical distance only if it lands within ±2% of one, and the grade uses that canonical figure — a watch reading 5.07 km is graded as 5,000 m, not 5,070 m. Imports arrive as drafts you confirm rather than as finished races. And every single import is created with the certified box unticked. Nothing that comes in from Strava is silently promoted to a measured result.

That's a defensible pipeline, not a perfect one. The distance is still an assumption. The difference is that it's a labelled assumption, and you can see the label.

The second concession is shorter and matters more.

parkrun's number is doing its job. A rough, weekly, free comparison between the people standing in the same field on the same morning, on a course parkrun openly tells you it hasn't certified, from a table it openly tells you it hasn't published. It was never trying to be a record-book figure.

So if you brought a parkrun grade to a race history and it wouldn't reconcile, the category error is the reader's, not parkrun's. Mine included. I spent an evening trying to make two percentages agree before I went and read their support page properly.

So what do you do with both numbers?

Keep both. Keep them apart.

The Saturday number is a trend line. Same course, same table, week after week — it is the cleanest free signal you'll get about your own fitness, and the table being unpublished genuinely does not matter for that, because an unknown constant cancels out when you only ever compare yourself to yourself. That's the job it was built for and it's good at it.

The race number is for the record book. Certified distance, chip timing, an edition you can name and a spreadsheet you can open.

They only become comparable when distance, timing and tables are all held constant, and between a parkrun and a road race not one of the three is. Averaging them gives you a number about nothing.

So: 66.0% on Saturday, 65.4% at the race. Both are true statements about 22 minutes. Only one of them comes with an address.

Sources and what they said

Everything below was checked on the date shown. parkrun's help pages move and their volunteers rewrite event news, so treat all of it as perishable.

• parkrun Support, 4.5 Age grading — "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", and the rough-comparison caveat quoted in full above. Article last edited 2024-07-09. Retrieved 2026-08-17. https://support.parkrun.com/hc/en-us/articles/200565263-4-5-Age-grading
• Crathes Castle parkrun, "Age Grading – all you ever wanted to know", 30 August 2023 — the WMA 2006/2010 track-5k reading, the added ages for 4-year-olds and over-100s, the four 60% examples, the 6-10% terrain estimate, the "more or less 5k" line, and the Fauja Singh 179.04%. Retrieved 2026-08-17. https://www.parkrun.org.uk/crathescastle/news/2023/08/30/age-grading-all-you-ever-wanted-to-know/
• Nonsuch parkrun, "Course Measurement", 12 December 2017 — mapping software, bike computer to ±10 m, surveyor's wheel to ±1 m, and why GPS isn't used. Retrieved 2026-08-17. https://www.parkrun.org.uk/nonsuch/news/2017/12/12/course-measurement/
• hometreadmill.co.uk age-grade calculator — reports a checked M60 parkrun result matching its 2006/2010 figure to within about a tenth of a percent, and describes parkrun's grades as running slightly generous. Retrieved 2026-08-17. https://hometreadmill.co.uk/age-grade-calculator/
• Howard Grubb, WMA age-grading calculator (2006/2010 factors) — the factor arrays and open standards used for every 2006/2010 figure in this post. Retrieved 2026-08-17. https://www.howardgrubb.co.uk/athletics/wmalookup06.html
• Alan Jones, Age-Grade Tables — the 2025 road tables, which is what RaceRecords grades on and where every Alan Jones 2025 figure here comes from. Retrieved 2026-08-17. https://github.com/AlanLyttonJones/Age-Grade-Tables

No figure in this post was read off a third-party calculator. Both factor sets were run locally against the same arithmetic, which is the only way to compare two tables without importing a third party's rounding.