Completing TAG
across Ireland

Section 1

Context: TAG and Northern Ireland

This paper is a companion to Is your parkrun grade the whole picture?, which introduced the Terrain-Adjusted Grade (TAG) — a performance metric that corrects for course difficulty on top of the standard age and gender adjustment. That analysis covered 11.2 million results across 860 UK parkrun courses and showed that TAG reduces course-induced grade variation by approximately 89%.

The UK dataset already includes all 27 Northern Ireland courses. Belfast Victoria parkrun — located in Belfast — emerged as the reference course (factor 1.000), the fastest parkrun in the UK and the venue where both the men's and women's parkrun world records were set. Northern Ireland was always part of the picture.

What was missing was the rest of the island. There are 139 parkruns across the island of Ireland — 27 operated by parkrun.org.uk in Northern Ireland and 112 operated by parkrun.ie. The parkrun.ie results were not available to us at the time of the original UK analysis. This paper adds those 112 courses to the TAG system, completing coverage across the whole island.

The calibration challenge this creates is what makes the methodology interesting. Every existing factor is expressed relative to Belfast Victoria (factor 1.000), a parkrun.org.uk course just over the border in Northern Ireland. To place the 112 parkrun.ie courses on the same scale, we need runners who appear in both datasets. Fortunately, thousands of them do — crossing the border or travelling between Great Britain and Ireland every week.

Section 2

The Problem: Why Not Just Add 112 Courses?

The simplest approach would be to merge the parkrun Ireland and UK data into a single pool of 11.8 million results across 972 courses and rerun the full factor computation pipeline. This would work mathematically — the grade distribution and paired runner methods scale naturally to larger datasets. But it would also change every existing UK factor, however slightly — including the 27 Northern Ireland factors that are already published and working.

The UK factors are published, live in the TAG calculator, and referenced in the companion paper. Any shift — even a 0.003 change at a single course — would invalidate the published numbers and require recalibrating every TAG grade ever computed. For a metric whose value depends on stability and trust, this is unacceptable.

The freeze-and-extend principle

Instead, we freeze the 860 UK factors at their published values and compute only the 112 new Irish factors, calibrated against the frozen UK scale. The shared runners who have completed parkruns in both jurisdictions provide the calibration bridge. Belfast Victoria remains at 1.000. No UK factor moves.

This pattern is designed to be repeatable. When parkrun data from another country becomes available, we would freeze all existing factors and compute the new country's factors against the same scale, using whatever shared runners exist. Each country is an additive layer, never a recalculation.

Section 3

The Shared Runner Bridge

The entire methodology rests on runners who appear in both datasets — identified by their unique parkrun Runner ID, which is consistent across parkrun.org.uk and parkrun.ie. Of the 99,693 unique runners in the parkrun Ireland data, 19,401 (19.5%) also have results in the UK dataset.

These shared runners are not a narrow or unusual population. They visited all 860 UK courses and all 112 parkrun Ireland courses. Their 317,999 UK runs span the full range of course difficulties, from Belfast Victoria (factor 1.000) to the slowest trail courses. This gives us a dense, well-connected calibration network.

The Northern Ireland connection

Geography matters. The strongest calibration links are through the Northern Ireland courses that sit in the UK dataset but share a land border with Ireland. The 27 Northern Ireland courses — from Belfast Victoria (876 shared runners) to Portrush (688) — provide the densest cluster of anchor points. Runners crossing the border for a Saturday morning parkrun are the backbone of the bridge, supplemented by the many thousands who travel between Great Britain and Ireland more broadly.

Crucially, 876 shared runners have completed Belfast Victoria parkrun — the reference course itself. This provides a direct calibration link rather than an indirect chain through intermediate courses. Another 1,244 ran at Ormeau (factor 0.933) and 874 at Stormont (factor 0.946), giving strong additional anchors.

Section 4

Method 1: Grade Distribution

The grade distribution method is identical in principle to the UK analysis. For each parkrun Ireland course, we compute the ALJ 2020 age grade for every experienced runner's personal best (runners with five or more total parkruns, minimum 50 PBs per gender per course). Nine quantiles (P10 through P90) define the course's grade profile.

The difference from the UK analysis is in the reference. Rather than finding the theoretical maximum across all courses (which would now include Irish courses and could shift the reference), we compare each Irish course's quantile profile against Belfast Victoria's frozen profile — extracted from the UK data independently.

This method produced factors for all 112 courses for both genders. It does not require shared runners — only a sufficient local population of experienced runners at each course. It is therefore the method that scales most naturally to countries with no geographic connection to the UK at all.

Section 5

Method 2: Paired Runners

The paired runner method exploits the shared runner bridge directly. For each of the 19,401 runners who appear in both datasets, we know their performance at various UK courses where the factors are already established. We use this to derive their "effective ability" — what pace they would run at Belfast Victoria — and then see how their Irish performances compare.

Step 1: UK-adjusted ability

For each shared runner, we compute the median of (finish_time × uk_course_factor) across all their UK runs. This gives a single number representing their typical pace normalised to the Belfast Victoria scale. We require at least two UK runs for a stable estimate, which 15,010 runners meet.

Step 2: Irish factor derivation

For each parkrun Ireland course, we compute the median of (UK_adjusted_ability / Irish_finish_time) across all shared runners who ran that course. This ratio is the course factor — how fast runners go at this course relative to their Belfast Victoria equivalent pace.

With 19,401 shared runners, this method achieved full coverage — all 112 courses for both genders met the minimum threshold of 10 paired runners.

Section 6

Method Agreement and the Blend

The two methods use different data, different mathematics, and make different assumptions. The grade distribution method measures the shape of the performance curve at each course. The paired runner method tracks specific individuals across courses. Agreement between the two is the primary evidence that the factors measure genuine terrain difficulty.

The correlation between the two methods is 0.705 for men and 0.671 for women. This is lower than the UK figure of 0.886, which is expected — the UK analysis had hundreds or thousands of paired runners per course, while the Irish analysis has tens to low hundreds. The median absolute difference between methods is 0.019 for men and 0.025 for women, meaning the two methods typically agree to within about 2 percentage points of course factor.

The maximum disagreement is 0.058 for men and 0.082 for women — concentrated at small, remote courses where both methods have the least data. The final factor for each course is a simple average of the two methods, which partially cancels the known biases of each: the grade distribution method is sensitive to field composition, while the paired runner method is influenced by tourist bias.

No courses exceeded 1.000

Unlike an earlier iteration (before the Runner ID data cleaning tripled the shared runner population from 5,711 to 19,401), no Irish course produced a blended factor above 1.000. The broader shared runner population diluted the tourist bias that had previously pushed several courses above the Belfast Victoria ceiling.

Section 7

Course Speed Factors — The 112 New parkrun Ireland Courses

The final blended factors for all 112 new parkrun Ireland courses are shown below, complementing the 27 Northern Ireland courses already in the UK dataset. Together, these 139 courses complete TAG coverage across the island of Ireland. The fastest of the new courses is Westport parkrun (combined factor 0.985), run on the old railway line greenway. The slowest is Knockanacree Woods (0.850), a trail course through woodland that BBC Countryfile once listed among the ten toughest parkruns in the world.

The Irish median course factor of 0.943 is slightly higher than the UK median of 0.933, suggesting Irish courses are on average marginally faster. This is consistent with there being fewer extreme hill and beach courses in the parkrun Ireland network compared to the full UK set.

112 courses. Factor referenced to Belfast Victoria (1.000). Blended from grade distribution and paired runner methods.

Section 8

Validation

Applying the TAG formula to all 570,895 parkrun Ireland results using the new factors produces 516,056 valid TAG values (90.4%). The male median TAG is 58.16% and the female median is 53.85% — broadly consistent with the UK figures of 57.99% and 53.76% respectively. The close alignment between jurisdictions is itself a validation: if the Irish factors were miscalibrated, the population-level TAG distribution would diverge from the UK.

The course-level TAG median varies from 50.1% (Inis Meáin parkrun on the Aran Islands — a remote course with a heavily tourist-influenced field) to 59.8% (Macroom Castle Demesne — a small course with a committed local field). The spread of 9.7 percentage points reflects genuine differences in who runs at each course, not factor miscalibration — the same variation exists within the UK when comparing individual courses rather than the broad speed bands reported in the companion paper.

Section 9

Known Limitations

Tourist and visitor effects

Ireland is a popular destination for parkrun tourism, and several courses — particularly scenic coastal and island venues like Inch Beach, Inis Meáin, and Narin Beach — attract a disproportionate number of visiting runners. These visitors tend to be more committed (and therefore faster) than the average participant, which can inflate both the grade distribution and paired runner factors. The effect is moderated by blending two methods but not eliminated.

Multi-lap courses and corner-cutting

A small number of courses — notably Pobalscoil na Tríonóide parkrun (a four-lap school loop rated as the second-fastest course in Ireland) — produce factors that appear higher than their terrain would suggest. We believe this reflects a systematic measurement artefact: on multi-lap courses with tight turns, runners naturally cut corners and run slightly less than 5km. parkrun courses are measured with a surveyor's wheel following a set line; they do not use the IAAF "shortest possible route" protocol used in road race certification. A 1–2% distance shortfall over four laps would produce exactly the kind of factor inflation we observe. The same pattern appears in UK multi-lap courses.

Beach courses

Narin Beach parkrun (factor 0.910) is run entirely on sand, yet its factor is dramatically higher than Great Yarmouth North Beach (0.730) in the UK — also an all-sand course. The difference likely reflects genuine variation in sand conditions (Narin is a wide, flat Atlantic beach that may be run near the waterline on compact wet sand), but visitor composition effects may also contribute. Beach course factors should be interpreted with particular caution.

Method agreement is weaker than the UK

The correlation between the two methods (0.705 male, 0.671 female) is lower than the UK's 0.886. This reflects the smaller parkrun Ireland dataset — 570,895 results versus 11.2 million — and the fact that even with 19,401 shared runners, many courses have only 10–50 paired observations per gender. The grade distribution method carries more analytical weight for Ireland, with the paired method providing validation rather than an equally weighted contribution.

Section 10

Next Steps

The parkrun Ireland course factors complete TAG coverage across the island of Ireland — 139 courses from Belfast Victoria to Knockanacree Woods. Combined with the rest of the UK, the TAG calculator now covers 972 courses. Any parkrunner can look up their TAG at any of these courses and predict their expected time at any other course on the network.

The freeze-and-extend methodology validated here is designed to scale to further countries as data becomes available. The grade distribution method requires no shared runners at all — only a sufficient local population at each course. The paired runner method uses whatever bridge exists between existing and new courses. Even a modest bridge population provides a valuable validation check.