Our method

A deterministic planning engine

Plans are produced by a pure, rule-based engine (current version 0.3.1). Given the same answers and start date it returns the same plan. It reads no clock, uses no randomness and calls no external service, and there is no machine-learning or language-model output inside the planning logic. It is designed to run the same way in a browser, on a server and, later, in a mobile app.

Every plan records the engine version that produced it. Any change to a rule or to its output bumps that version, and the engine's example outputs are reviewed whenever it changes.

The planning rules

• Week 1 never starts above your current weekly mileage. If your time limits cannot hold it, the plan says so.
• Your experience level is checked against your mileage, run frequency and available days. If they disagree, the plan follows the more cautious of the two.
• Mileage rises gradually: a build week adds at most 8% of the last full week (and never more than 2.5 miles). This is a planning heuristic, not a guarantee against injury.
• Every fourth build week is an easier week at about 80% of the previous full week.
• A week adds at most one new stressor: more mileage, an extra running day, or your first harder session. They are never combined.
• The long run grows by at most 15% a week (and at most 1 mile), and is capped by a share of weekly mileage, a distance limit for your goal, and your time limit.
• Most running is easy. There is at most one harder session a week, kept under about 15% of weekly running time and never beyond 25 minutes of hard effort.
• Harder sessions start only for runners with a base: not before week 5, not in easier or taper weeks, not when too few build weeks remain, and never the day before or after the long run.
• Run frequency rises by at most one run every four weeks, and never beyond the days you said you could run.
• Training tapers before the race. Race week is short, easy and includes rest before race day. If the race falls early in the week, the week before it is also a taper week.
• The last long run is at least 6 days before the race, the last tempo or interval session at least 7 days before, and the day before the race is a rest day.
• Pace ranges appear only when you give a credible race or time trial from the last 8 weeks, and only while that result is no more than 12 weeks old. Otherwise the plan uses effort guidance only.
• A finish-time estimate appears only when the plan fits your timeline well, your result is for a distance close to your goal, and the result is still recent on race day. Otherwise the plan gives none.

Read from the engine at build time.

How paces are estimated

Pace ranges appear only when you give a race or time trial from the last 8 weeks. The engine predicts your times at other distances with the Riegel formula (time grows with distance to the power 1.06), then sets:

• Easy runs at 125% to 140% of your predicted 5K pace
• Tempo runs from your predicted 10K pace to your predicted half marathon pace
• Intervals within 2% of your predicted 5K pace

All are shown as ranges, rounded to the nearest 5 seconds per mile. The training pace calculator uses the same code path, and a test keeps its output identical to the engine's. Without a recent race, plans give effort guidance and assume a deliberately slow easy pace for your experience.

How the engine is checked

• Automated unit tests for each part of the logic, including validation of every input.
• Property-based tests that generate many combinations of inputs and check safety invariants on every plan produced, such as never scheduling a run on a day you did not choose, never exceeding your time limits, and never more than one hard session in a week.
• A fixed set of more than 20 example runner profiles whose complete plans are reviewed whenever the engine changes. A human-readable report of those plans is generated for review.
• Key statements on this site about the planner are backed by tests, so a rule change that breaks one of them fails the tests. Not every sentence is covered.