Shindo The normal case#
Look at any twelve-week block honestly and the striking thing is how few weeks went as written. Work travel, a sick child, a cancelled pool session, a cold, a late meeting, a morning where you simply could not. Perfect weeks are the exception, and a plan built on the assumption that they are the rule is misdescribing the thing it is planning.
So the interesting question is not what a plan does when you follow it. It is what it does on the Thursday when two sessions are already gone.
The two things it usually does, both wrong#
It carries the debt forward. The missed sessions reappear, squeezed into the days that remain or pushed into next week, and the athlete who missed 25% of a week now faces one that is 25% bigger. This is the exact spike that a ramp cap exists to prevent, and it arrives dressed as diligence. Training load is not a debt. A session you did not do is not owed; it simply did not happen, and the physiological consequence of that is nothing at all.
It paints the week red. Two crosses, five ticks, “71% adherence”. This is a compliance report, not a training assessment, and it teaches the wrong lesson twice over — it implies the plan was the goal (it was not; adaptation was), and it makes the athlete who most needs to keep going feel like someone who is already failing. The reliable predictor of progress over a season is not adherence to any given week. It is still being here in month six.
Five behaviours worth demanding instead#
1. The missed load stays missed#
The week’s target drops by what was not done. Next week is planned from where you actually are — your real accumulated load, your real fatigue, your real form — not from the fictional athlete who completed the sheet. This is the difference between a plan that recalculates and a plan that merely displays.
2. What is left gets re-flowed, not restacked#
There is a difference between shifting everything one day to the right and asking a fresh question: given four days remain and this is the athlete’s current state, what is the best week available from here?
Usually the answer is not “the same sessions, compressed”. Two hard sessions in the remaining four days may be too many. The long session may be worth keeping and the second interval session worth dropping entirely. Something has to give, and choosing what should be a decision, not the accident of a queue.
3. A swap counts as work#
You rode instead of running. You swam instead of riding. You did 40 minutes of a planned 70. None of these are zeros, and a plan that treats them as zeros is simply wrong about what happened to your body.
The honest handling is to match what you did against what was planned and count it for what it was: the same sport but shorter is a partial; a different sport at similar load is a substitution carrying most of the aerobic stimulus and none of the specificity; a session moved by two days is a session that happened. Load is load. Specificity is a separate conversation, and worth having — but not by pretending the work did not occur.
4. Next week still obeys the cap#
The temptation after a broken week is to compensate. It should be refused. Weekly load growth stays capped — roughly 8% — measured from what you actually did, and a light week does not license a heavy one. The constraint that matters most is the one that binds when you least want it to.
The one legitimate adjustment: if the sessions were missed because you were ill, the next week should be smaller, not merely capped, and the return should be gradual regardless of how good you feel on the first clear morning.
5. It says what it did, and why#
A plan that quietly rebuilds itself is not much better than one that quietly guilt-trips you. If the second hard session was dropped, the week should say so and say why — which constraint it hit and what it protected. Then you can disagree with it, which is the only way you ever find out whether it was right.
The checklist#
Whatever tool you use, these are answerable questions:
| Ask | Bad answer | Good answer |
|---|---|---|
| Where does missed load go? | Into next week | Nowhere |
| What happens to the rest of the week? | Everything shifts right | The remaining days are re-planned |
| Is a bike ride instead of a run a zero? | Yes | No — counted as load, flagged as a substitution |
| Can next week be bigger to compensate? | Yes | No, the cap holds |
| Does it explain what it changed? | No | Yes, with the reason |
Who decides — the part worth understanding#
Increasingly, weekly plans are generated by a language model, and that raises a fair question: what stops it inventing a week that looks plausible and is quietly unsafe?
The answer that actually works is not a promise about the model’s judgement. It is a boundary that sits outside the model. A proposed week gets checked, before you ever see it, against limits computed from your own data — the ramp cap, the easy/hard balance, the days you actually have available, the recovery-week cadence, monotony and workload ratio. If the proposal does not pass, it does not get published.
This is worth insisting on because the failure mode of a generated plan is not obvious nonsense — that is easy to spot. It is a week that reads beautifully, is internally coherent, and contains a 30% jump in load because nothing was checking.
The general principle: the numbers that constrain your training should come from deterministic computation, and language should be the last step, not the first. If the same inputs can produce different numbers on different days, there is nothing to check the plan against.
Where the free line usually sits#
Worth knowing before you evaluate anything, because it is the part that tends to be vague in marketing: which of these behaviours require paying.
In Shindo, the weekly plan itself is built by the deterministic engines on the free tier — phases, the 80/20 split, the capped ramp, recovery weeks — and plan-to-actual matching runs there too, so swaps, partials and shifted sessions are counted correctly whether or not you pay. What Shindo Pro adds on top is the AI layer: every session explained in plain words, and the week re-flowed around what is actually left after life interferes. Your most recent 60 days of history are free; Pro removes that 60-day window.
The distinction matters more than the price. A free tier that gives you a static template and a paid tier that gives you the arithmetic is a different product from one where the arithmetic is free and the wording is paid, and it is worth checking which one you are looking at.
Shindo builds your week from deterministic engines on the iPhone, then — if you are on Pro and the AI is turned on — lets a model propose the week instead. That proposal is checked against roughly nineteen named rules derived from your own data, gets exactly one retry if it fails, and is replaced by the engine’s own week if it fails again. Shindo runs on an iPhone 11 or newer and reads your training data from Apple Health. The methodology documents the validator and its rules.
Read next