How we make our plans and pages

Runsy is a running app for iPhone, founded by Claudio Fuentes, who is responsible for what this site publishes. We do not employ a certified running coach, and no plan or page on this site has been reviewed by one. We would rather tell you that than imply otherwise.

So instead of a name, this page gives you the thing a name is supposed to stand for: the rules our plans follow, the research those rules rest on and where it runs thin, how every number on a factual page is traced and checked, and how to get a mistake fixed.

Every training plan on this site comes from one small engine: plain code, no AI, the same rules for every distance, level and length. It covers the 5K, 10K, half marathon and marathon, at three levels (three, four or five runs a week), in lengths from 8 to 20 weeks. Because it is code, every plan can be regenerated and tested.

One of those rules exists because we got it wrong. An early version scaled race week off the race itself and prescribed a 19 km run five days before a marathon. A test now fails if race week is ever bigger than it should be.

Two honest limits. The 20% cap applies to the long run, which sets the week, so the week after a cutback is a real step up from the cutback week by design. And these are templates: they cannot know you missed week nine.

The app's plan is built for you rather than picked from a table. An AI model writes it from your goal, your race date, your recent runs (read from Apple Health with your permission) and anything you tell it. Code that uses no AI then does the parts that should never be a judgement call:

• It sets your pace zones as fixed ratios of your measured easy pace. Tempo is 6% to 10% faster than easy, threshold 10% to 14%, intervals 14% to 20%. That is a rule of thumb, not a lab test, and we say so.
• It builds a first week instantly, from the same kind of rules as above, while the full plan is written.
• When a tempo or interval session lands the day before or after another hard day, it swaps it with an easy run in the same week, so you keep the same running days. If no swap fits, the session stays and the check below flags it.
• It pins race day to your actual race date.

Every generated plan is also checked for weekly jumps of more than 10% and for back-to-back hard days, and anything that trips the check is logged so we can find it and fix the cause. The 10% line is a deliberately cautious tripwire for review, not a claim that 10% is the safe limit: as the next section says, the evidence for that rule is weak.

Before any of your data reaches an AI model, the app names the providers and asks for your consent. Which data goes where is in our privacy policy.

Mostly easy. Successful endurance athletes do roughly 80% of their sessions at low intensity (Seiler, 2010). That is why extra days in our plans are easy days.

Taper. A meta-analysis of 27 studies found the best results from about two weeks with volume cut 41% to 60% and intensity kept (Bosquet and colleagues, 2007).

How fast to build. This is where the evidence is weakest. A trial of 532 novices found a programme built on the 10% rule had the same injury rate as a faster one (Buist and colleagues, 2008), while a cohort of 874 novices linked jumps of more than 30% to some overuse injuries (Nielsen and colleagues, 2014). Our 20% cap on the long run sits between the two, and it is a judgement, not a finding.

Cutback weeks and long-run ceilings are coaching convention. Most published plans use them and we think they are sound, but no trial we know of has tested the exact numbers.

• Averages and standards. Every figure is traced to the study that produced it, with its sample size and years. Where sources disagree we print the disagreement, and where we combine them we say how. We do not use parkrun data, because its terms prohibit reuse. The average marathon time page shows the method in full.
• Race facts. Dates, qualifying times and course details come from the race organiser's own pages, and course certification from the USATF certificate where one exists. Race cards show the date their facts were last checked and the site they came from.
• Comparisons. Prices are the price each company publishes, with the date we checked it, and a price table more than a year old fails our build. Every comparison says when to choose the other app, and our rule is that any page comparing Runsy says plainly that we make Runsy.
• Training advice. Physiological claims link the study behind them, inline or in the page's source list.

A page that says Checked with a date means that on that day we re-opened the sources behind its figures and confirmed they still match. Pages without that line list their sources but have not been through a dated re-check.

It does help draft the words on this site, and in the app it writes your plan and the coach's replies.

It does not supply facts. A figure on a factual page has to trace to a source we list; when one did not, we corrected it and the page says so. It does not set the safety rules above, which are code. And it cannot publish: a draft becomes a page only through a separate publishing step, which refuses a page whose links point nowhere or a ranking page with no recorded search check, and our build fails on any comparison whose prices are more than a year old.

If something on this site is wrong, email support@runsy.ai. We fix a confirmed mistake within a day and note the correction, with its date, on the page itself.

Recent corrections, all on Oct 10, 2026, found in our own re-check:

• Average marathon time: the averages for men and women, after we found one report's 2001 figures being read as 2018 figures.
• Average half marathon time: the women's average, by two seconds, and the gap between sources.
• 5K, 10K and half marathon pages: several “top X%” claims, recomputed from the published percentile tables.
• Running cadence: what a higher step rate does to Achilles and calf load, which the research shows we had backwards.

No certified running coach has reviewed these rules. If that changes, the reviewer's name, credential and the date of the review will be on this page. Until then, judge the rules on what they are: everything above is stated precisely enough to disagree with.