Three engines decide your training, and they answer different questions. One
reads your numbers and works out what is actually holding you back. One reads your
night and says how much you have got today. The third works out what to eat to fuel it. Here is
the machinery of each, and an honest account of where it is weak.
Engine one
What is holding you back?
Scores your running and your strength against what your goal demands, and builds the block around whichever is behind.
Engine two
How much have you got today?
Reads your night off your watch and turns it into one call: train hard, steady work, or recover.
Engine three
What should you eat today?
Moves the day’s food up with the day’s training, and holds protein steady underneath it.
The idea · Two questions
What is holding you back
Readiness decides how hard today is. This decides what today is
for. A hybrid race asks you two things, so Ravns asks you the same two.
1Can you run at the level your goal needs?
Scored from a race or time trial you actually ran, against the pace the goal demands — with the weight you would be carrying.
2Are you strong enough for the stations?
Scored from your heaviest lifts against the loads your division puts on the floor. In pounds, because the sled weighs what it weighs.
Whichever answer is further behind is your
limiter, and the block is built to move it. Everything past this point is how those
two scores are worked out — open it if you want it.
Step three · The four answers
The limiter, named
Two scores, two questions, four possible answers. This is the whole diagnosis —
and every corner is named for what to build, never for what is wrong with you.
↑ Strength & stations
Strong, needs the run
The strength is there. The running is what stands between you and the goal.
Limiter: running economy and power-to-weight. Three hard days, run-biased.
Sharpen & optimize
Both levers are close to what the goal demands. Now it is about putting them together.
Limiter: integration and pacing. Four hard days, the most of any corner.
Build the base
Neither lever is where it needs to be yet, so the plan does not pretend otherwise.
Limiter: general base. Two hard days, the longest base phase.
Fast, needs strength
You can run. The strength work and the stations are the ceiling.
Limiter: strength and station capacity. Three hard days, station-biased.
← RunningRunning →
Every corner is named for what to build, never for what is wrong with you. There is
no corner called "weak".
Where this is solid and where it is not
Your position comes from real numbers, not a questionnaire: your running is scored from a
recent race or time trial against the pace your goal actually demands, with body composition
weighted alongside it, and your strength from your logged lifts against your division's
standards. That much is arithmetic.
The choice of axes is the judgement, and it is the honest weak point. Running against
strength-and-stations is the pair this app was built on and the one that has been in front of
athletes longest, but "has held up so far" is not the same as proven, and I am not going to
dress it up as more. If you coach hybrid athletes and you would have picked a different pair,
that is the first argument I want to have.
Step four · The programming
What your corner changes
A diagnosis you cannot act on is a horoscope. Four concrete things move.
Hard daysHow many genuinely hard sessions your week holds: two while you are building a base, three when one lever is the limiter, four when both are close and the work is integration.
Which one leadsWhich hard session gets the freshest day. If strength and stations are your ceiling, the race simulation leads the week; otherwise the quality run does.
Phase lengthsHow your weeks to race day are split. A base-building athlete spends 55% of the remaining weeks on base; a well-rounded one spends 30% and gets a longer peak instead.
What countsWhich sessions your week is graded on. Miss an easy run and the week is fine. Miss the session that moves your limiter and the review says so, even if the total hours looked good.
How the two scores are calculated
Step one · The two levers
What a hybrid race asks of you
Readiness decides how hard today is. This decides what today is for, and
it starts a long way before the diagnosis: with two levers. A lever has to pass two tests
before it is allowed to be one of them.
It has to be a leverSomething you can train this week. Never a state like lean mass, and never an outcome like a finish time. A corner is worthless unless it tells you what to do on Tuesday, and "be lighter" is not a session.
It has to be what the race asksA hybrid race is running, interrupted repeatedly by stations that punish you for being weak. Those two things are what the day costs you, so those two are what a plan is allowed to organise itself around.
Step two · The measurement
Where you sit on each lever
No questionnaire decides this, and neither does the AI. Two scores, each between
nought and one, computed from things you have actually done.
What goes inA recent race or time trial. Your bodyweight, with your height or body fat. Your logged lifts.
Measured, or typed once. Nothing here is a self-rating of how fit you feel.
↓
Two scoresEach lever gets its own number between nought and one, computed its own way. They are below, whole.
↓
The cornerEach axis is asked one question: are you at the level this goal demands? The bar is 90%. Clear both and you are sharpening; clear neither and you are building a base; clear one and that is your answer.
Two questions, four answers. That is the whole quadrant.
↓
The limiterThe binding constraint, named in your words rather than the engine's, and the thing the next block is built to move.
The run axis
outputs 0 → 1
Can you cover eight kilometres at the pace your goal actually demands, carrying
what you are carrying? Two parts, weighted 60% pace and 40% power-to-weight, because on
a hybrid course you take yourself over everything.
Reads
One recent race or time trial, plus your bodyweight with either your height or your body fat.
No treadmill guess and no self-rating. It wants a result you actually produced.
Scored against
The running fitness your goal time demands — not a population average, and not last year's you.7 A deficit of twelve points scores nought; arriving at what the goal needs scores one.
The weight half
A slope, not a threshold. It runs from full credit down to none across a range, rather than at a line you pass or fail, because nobody gets suddenly heavy.
If you did not run it
A row or a bike is converted to a run-equivalent first — and the app says so rather than presenting it as a run you did.
Projections
Longer distances are projected from the one result you gave it.6Riegel fitted his exponent to record-class fields, so a long-distance projection runs optimistic until you have actually run one.
The strength axis
outputs 0 → 1
Are you strong enough to move the implements your division will put in front of
you? Scored in absolute pounds, never as a multiple of your bodyweight — the sled
weighs what it weighs, and what you weigh is not part of that question.
Reads
Your heaviest squat, bench and deadlift, in pounds.
Whatever you have. One lift is enough to place you; three place you better.
Scored against
The absolute load your division demands, in the table below. Meeting it lands you exactly on the strong-enough boundary at 0.50.
Why not a ratio
Because a heavier athlete already pays for those pounds on the run axis, and charging them again here would dock them twice for the same body — while their real advantage, putting more force into a fixed sled, went unmeasured.
If you have no lifts
Max strict push-ups, on the same scale. If neither, how often you lift — the weakest evidence in the app, and treated that way. The chain always prefers a number you have actually hit.
Division standardspounds, not bodyweight
Lift
Men's open
Men's pro
Women's open
Women's pro
Squat
220
264
139
174
Bench
176
220
83
104
Deadlift
264
308
174
209
Ravns asks what you have lifted rather than whether you feel strong, because one
of those is a number and the other is a mood.
The fallback rungif you have not logged a lift
Max strict push-ups
Men
Women
10
0.25
0.38
20
0.40
0.52
35
0.55
0.68
50
0.70
0.80
The same nought-to-one axis, reached a cruder way, with 0.50 still the
line between needing strength and not. Log a single real lift and the lift wins, because a
barbell measures what push-ups only estimate.
The programming, in detail
Adaptation order, by cornerfirst one leads the block
Strong, needs the runAerobic base→Running economy→Threshold→Race integration
Fast, needs strengthMax strength→Station capacity→Race integration→Aerobic base
Build the baseAerobic base→Max strength→Station capacity
The same four adaptations, in four different orders. Your corner does not change
what you are training for — it changes which of these gets the freshest day, and how much
of the week it takes.
Specificity floorhybrid race
Base70%
Build80%
Peak90%
Taper95%
How much of the week has to look like the race itself. In base there is room for
work that only pays off indirectly; by taper almost everything you do should resemble the
day you are training for. The floor rises on a schedule rather than on a feeling, which is
what stops a build quietly turning into whatever you enjoy most.
Step five · The shape of the block
Load rises, then backs off
Without a race date the plan runs in rolling four-week blocks: volume steps up
for three weeks, then the fourth eases so the work lands. With a race date the same logic
stretches into base, build, peak and taper, sized to the weeks you actually have.
Volume across one block4 weeks · rolling
Week 10.90
Week 20.95
Week 31.00
Week 4deload
The three-up-one-back rhythm is conventional coaching practice rather than
something Ravns discovered, and the app says so in its own source. What Ravns adds is that
the cadence is anchored to a fixed calendar, so rebuilding your plan cannot quietly restart
your block and leave you climbing forever.
The mechanism · Why rest is training
You get worse first
A hard session does not make you fitter. It makes you temporarily worse, and the
rebuilding afterwards is what makes you fitter. Whether that rebuild ever happens depends
entirely on when the next session lands.
Capacity over time1 session · 14 days
The green dots are what you brought to each session — your capacity
the instant before it started, not where it left you. That is the number this whole app
exists to tell you.
One assumption is doing the work here, and you should know which. The textbook
fitness–fatigue model gives every session the same adaptation whatever state you
started it in, which is why it cannot tell these two patterns apart at all. This chart adds
the thing that model leaves out and that every coach believes: a session you start flat
gives back less and costs more than the same session started fresh. That is the argument,
drawn. It is not a fitted result, the timescales are illustrative, and the
bottom of this page says where the underlying idea is thin.
This is the whole argument for reading your body every morning.
The same session, on the same athlete, is progress on Thursday and a hole on Tuesday.
A fixed twelve-week plan cannot know which one today is. That is not a failing of the coach
who wrote it, it is a failing of paper.
The inputs · Six signals
What it actually reads
Five come off your watch while you sleep. One is a question you answer with a
single tap. Nothing is compared to a population average: every signal is scored against
your own rolling baseline, so your normal is the only normal that counts.
Heart rate variability
ms · 60-night baseline
The night's reading against the spread of your own last 60 nights. Needs 14 nights before it says anything.
Sleep
hours + efficiency
Duration and how much of the time in bed you were actually asleep. The one signal scored against a fixed ideal rather than you.
Resting heart rate
bpm · 60-day baseline
Drifts up when you are under-recovered, fighting something, or short on sleep.
Training load
recent vs. usual
What you have done this week against what you are conditioned for. The only signal that comes from your training rather than your body.
Respiratory rate
breaths/min · 60-day
The quiet one. Elevated breathing overnight often shows up before you feel anything.
How you feel
one tap · 5 points
Not averaged in with the sensors. Applied after them, because when your body disagrees with your watch you are usually right.
If a signal is missing, it drops out and the rest carry the score. That is
deliberate: plenty of watches never record heart rate variability, and their owners still
deserve an answer.
The score · One call, not a number to decode
How the morning number is built
The five sensor signals are averaged, weighted by how much each one actually
tells us — heart-rate variability counts most, breathing rate least. Then your own tap
moves the result, and a collapsed heart-rate variability can overrule everything.
What comes out is one of three calls: train hard, steady work, or
recover. Not a score out of a hundred you would have to interpret before coffee.
The weights, the bands and the two overrides
The score · Weighted, capped, then adjusted
How the morning number is built
The five sensor signals are a weighted average. Then two things happen that a
plain average would get wrong.
Drop a signal and the others renormalise to fill the gap.
The three callscutoffs at 45 and 65
StrainedRecoverBalancedSteady workPrimedTrain hard
04565100
Three calls, not a hundred shades. A number you have to interpret before coffee
is a number that does not help.
The two overridesA collapsed HRV overrules everything else.
Two standard deviations below your own baseline caps the score at 40 on that signal alone,
however good the rest looks. One and a half deviations down caps it at 55, or at 46 if a
second signal agrees. A great night's sleep cannot argue you out of it.
Feeling bad counts more than feeling good.
Your one tap moves the final score: wrecked takes 18 points off, tired takes 7, good adds 5,
primed adds 10. The asymmetry is on purpose, and a positive tap is ignored entirely when the
score has been capped. You can talk yourself into a worse day. You cannot talk yourself into
a better one.
Step one · The day sets the number
One target a week is the wrong shape
Your training is not the same on Tuesday as it is on Sunday, so your food should
not be either. The number starts from what you weigh and moves with what the day asks.
Energy by training daykcal per lb of bodyweight
Rest13.5
Easy14.5
Quality15.5
Long16.5
The day type comes from the session actually on your plan, not from a weekday.
Move a long run to Wednesday and Wednesday’s food moves with it. And because the target
is computed from the day’s training, it cannot exist before you have a plan —
which is why logging works from the first minute and targets arrive later.
Step two · What the plate is made of
Protein holds still. Carbs do the moving
Protein and fat are set per pound of bodyweight and stay where they are.
Everything the day adds or takes away lands on carbohydrate, because carbohydrate is what
fuels the session.
A 176 lb athletesame ruler, both days
Rest day2,376 kcal
Long day2,904 kcal
Protein 167 g — unchangedFat 56 g — unchangedCarbs 300 → 432 g
Protein is 0.95 g per lb and fat 0.32 g per lb, so both blocks are
the same width on either day. Only carbohydrate grows. Protein is held because it is the
floor your muscle rebuilds from whatever the day was10
— and holding it is what stops a hard training block quietly costing you muscle.
Step three · What it refuses to do
The limits on a cut
Most of the harm in diet advice is in how far it will let you go. These are the
three places Ravns stops, whatever you ask it for.
The deficit is cappedA cut is the smaller of 500 kcal and a quarter of your maintenance. A smaller athlete gets the smaller number, because a fixed 500 is a far deeper cut at 130 lb than at 200 lb and nobody should be handed the same knife regardless of size.
There is a hard floorTargets never go below 1,500 kcal, or 1,200 on the women's standards, no matter what the arithmetic says. When the floor binds, the app tells you it is the floor rather than quietly serving a smaller number.
Recovery outranks the cutThe morning after a race or a near-maximum effort, the deficit is suspended entirely and you eat at maintenance. A cut taken then is paid for in repair and glycogen, not fat — so it is not a cut, it is just a worse week.
Step four · Engines decide, ravens explain
A meal is not a day
The clearest illustration of the whole architecture, and it is a bug story.
Asked to log a quiche, the nutrition raven returned the
quiche’s own macros as the day’s target. Four hundred calories. The athlete’s
goal for the day became one slice of pie.
The fix was not a sterner prompt. The engine now
computes the day’s baseline and the AI is allowed to refine it, never to replace it:
a target landing more than a thousand calories from the engine’s own number is not a
refinement, it is a different kind of thing entirely, and it is rejected. The window is
deliberately generous — wide enough that no honest adjustment is ever blocked, narrow
enough that a single plate cannot pass for a day.
That is the rule the whole app runs on. The engines
decide; the ravens explain, ask, and read your plate. A language model never gets the
final say on a number.
The method · Where the numbers come from
Borrowed, and cited
Ravns did not invent its training theory, and the AI does not improvise your
plan. The decisions are ordinary tested code running published methods. Here is each one,
with its actual source.
Session stressEvery session becomes one number weighted by how hard your heart worked, not just how long you were out.1,2
Your own baselineHeart-rate variability is read against the spread of your own last sixty nights rather than anyone else's average, and needs a fortnight before it will speak.3,4
Your own opinionThe one-tap check moves the score after the sensors have spoken, rather than being averaged into them.5Across 56 studies, self-reported measures tracked training load with greater sensitivity than the objective ones — and the two frequently disagreed.
Your pacesOne recent race or time trial sets every pace you are given, and longer distances are projected from it.6,7Riegel fitted his exponent to record-class fields, so treat a long-distance projection as optimistic until you have run one.
The plan's shapeWeeks are grouped into base, build, peak and taper, sized to the time you actually have. The taper cuts volume by 45% and leaves intensity alone.8,9Block periodization traces to Verkhoshansky and was formalised by Issurin. The taper meta-analysis found two weeks at 41–60% less volume, intensity held, to be the best of the strategies tested — for a gain of a couple of percent, not a transformation.
FuelEnergy moves with the day in front of you instead of sitting at one number all week, and protein is held high enough to defend lean mass in a deficit.10That range is per kilo of fat-free mass in lean, resistance-trained athletes. Ravns targets bodyweight, which is the cruder approximation nearly every app makes.
The weak points · Stated on purpose
Where I'd argue with me
A science page that only lists its strengths is marketing. These are the three
places this system is most likely to be wrong.
The load signalComparing recent training to your longer-term average is the shakiest thing Ravns measures. The method was popularised for injury prevention11,12 and has since been taken apart: the two windows overlap, so the ratio is partly correlated with itself, and one team reproduced the injury association after replacing the long-term half with random numbers.13–15 It stays in Ravns because a sudden jump in training genuinely is riskier than steady hard training — but it is one of five signals, weighted at 19%, and it is never allowed to be the whole answer.
The curve at the topThe rebound-above-baseline diagram is a teaching aid, not a timetable. It is solidly documented for muscle glycogen, where Yakovlev discovered it,16 and it has never been established that strength or endurance follow one tidy dip-and-rebound with a peak you can schedule around. The two-factor model drawn there1 is better, and recent work questions how reliably its parameters can be fitted at all.
The fuel mathsEnergy is bodyweight times a factor for the day, which is a rule of thumb, not a measurement. It ignores how much lean mass you carry, how long the session actually was, and how efficiently you move. A proper answer would measure your resting metabolism and the real work done; this estimates both. It is close enough to aim at and re-check against the scale, which is exactly how it should be used and not a calorie count you should trust to the digit.
The weightsThe numbers in that bar — 28.6% heart-rate variability, 23.8% sleep, and the rest — are reasoned, not fitted. They were never tuned against outcomes, because that needs more athletes than Ravns has. The app's own source calls them placeholders. They are defensible and they are not evidence.
What this is not
These are established methods applied honestly. They are not a claim that Ravns itself has
been through a trial, because it has not. Ravns is a training tool, not a medical device, and
it does not diagnose, treat or prevent anything. If something hurts, see a professional, not
an app.
References
The works cited
Every source the page leans on, in the order it first appears. Each was checked
against the primary literature rather than recalled, which is how three of them came to be
corrected before this page went up.
Bibliography16 entries
Banister EW, Calvert TW, Savage MV, Bach T. A systems model of training for athletic performance. Australian Journal of Sports Medicine1975;7:57–61.
Banister EW. Modeling elite athletic performance. In: Physiological Testing of Elite Athletes. Human Kinetics 1991:403–424.
Plews DJ, Laursen PB, Stanley J, Kilding AE, Buchheit M. Training adaptation and heart rate variability in elite endurance athletes: opening the door to effective monitoring. Sports Medicine2013;43(9):773–781.
Plews DJ, Laursen PB, Kilding AE, Buchheit M. Monitoring training with heart rate-variability: how much compliance is needed for valid assessment? International Journal of Sports Physiology and Performance2014;9(5):783–790.
Saw AE, Main LC, Gastin PB. Monitoring the athlete training response: subjective self-reported measures trump commonly used objective measures — a systematic review. British Journal of Sports Medicine2016;50(5):281–291.
Riegel PS. Athletic records and human endurance. American Scientist1981;69(3):285–290.
Daniels J, Gilbert J.Oxygen Power: Performance Tables for Distance Runners. Tempe, AZ 1979.
Issurin VB. New horizons for the methodology and physiology of training periodization. Sports Medicine2010;40(3):189–206.
Bosquet L, Montpetit J, Arvisais D, Mujika I. Effects of tapering on performance: a meta-analysis. Medicine & Science in Sports & Exercise2007;39(8):1358–1365.
Helms ER, Zinn C, Rowlands DS, Brown SR. A systematic review of dietary protein during caloric restriction in resistance trained lean athletes: a case for higher intakes. International Journal of Sport Nutrition and Exercise Metabolism2014;24(2):127–138.
Hulin BT, Gabbett TJ, Blanch P, Chapman P, Bailey D, Orchard JW. Spikes in acute workload are associated with increased injury risk in elite cricket fast bowlers. British Journal of Sports Medicine2014;48(8):708–712.
Gabbett TJ. The training-injury prevention paradox: should athletes be training smarter and harder? British Journal of Sports Medicine2016;50(5):273–280.
Lolli L, Batterham AM, Hawkins R, et al. Mathematical coupling causes spurious correlation within the conventional acute-to-chronic workload ratio calculations. British Journal of Sports Medicine2019;53(15):921–922.
Wang C, Vargas JT, Stokes T, Steele R, Shrier I. Analyzing activity and injury: lessons learned from the acute:chronic workload ratio. Sports Medicine2020;50(7):1243–1254.
Impellizzeri FM, Woodcock S, Coutts AJ, Fanchini M, McCall A, Vigotsky AD. What role do chronic workloads play in the acute to chronic workload ratio? Time to dismiss ACWR and its underlying theory. Sports Medicine2021;51(3):581–592.
Viru A. Early contributions of Russian stress and exercise physiologists. Journal of Applied Physiology2002;92(4):1378–1382.— the accessible English account of Yakovlev's supercompensation work.
Block periodization is credited above to Verkhoshansky, whose 1970s and 1980s
work on concentrated loading has no single clean English citation; reference 8 is the
formalisation that followed it.
If you coach and something above is wrong, I would rather hear it
than keep shipping it. ryan@lastyeti.com
reaches me, and there is one of me.