The science, honestly

Today's Load is built in three layers. They do not carry the same weight of evidence, and we would rather say which is which than let the whole thing borrow credibility from its strongest part.

Biology provides the principle. Today's Load provides the operationalisation. The decision is the product translation of that principle.

Today's Load is not an algorithm that computes a readiness score for you. It is a system that operationalises a biological principle of regulation, turns it into a usable decision about today's load, and tests the larger proposition that better daily responses can build capacity over time.

Layer one — the measures are textbook

Everything computed per phase is standard heart-rate-variability analysis. RMSSD, pNN50, SDNN, the mean interval and the low- to high-frequency power ratio were defined in the 1996 Task Force standards and have been used unchanged since. Alongside them Today's Load computes the short-term fractal scaling exponent DFA α1, which is not part of that document — it is a later nonlinear measure, established in its own literature.

That RMSSD and pNN50 predominantly reflect vagally mediated short-term cardiac variability, that standing withdraws that vagal influence, and that slow paced breathing amplifies the respiratory component of the rhythm — none of this is our claim. It is ordinary autonomic physiology.

Layer two — the protocol is an established method

Orthostatic testing — lie still, stand, compare — has been used in sports physiology for decades to track autonomic load and fatigue, and one consumer platform (Polar) has shipped a version of it for years. Adding a paced-breathing phase to standardise the respiratory contribution is likewise established practice in heart-rate-variability research.

What is unusual is not the ingredients. It is that almost nobody has put them together into a daily consumer protocol and then shown the user the reasoning.

Layer three — the composition is ours

The specific arrangement is original work by the physiologist who designed the engine: the four-phase lie / stand / lie-paced / stand-paced sequence, reading what happens as the conditions change, not only the phases themselves, deriving four functionally different signals from how the standard measures move across them, and reading those signals together under one fixed interpretation in which a strong signal does not simply cancel a weak one.

Patent pending on aspects of the measurement and decision system.

This composition has no published validation. It has not been tested against a clinical outcome, benchmarked against competing readiness models, or peer reviewed. We do not call it validated, clinically proven, or a lab replacement, and you should treat anyone who uses those words about a consumer readiness score with suspicion.

What it has is a coherent physiological rationale, a frozen implementation, and a growing corpus of real recordings it is regression-tested against. That is an honest position, and it is the one we are in.

The chain

The principle Today's Load is built on is that the quality of regulation is not decided only by how a body answers a disturbance. It is decided just as much by whether the body re-establishes an organised state once the disturbance is over, and whether it holds that organisation when the demand comes again. Today's Load is the attempt to operationalise that principle, and its value depends on how well the whole chain holds in reality:

  1. A biological principle — respond, re-establish, hold.
  2. A measurable physiological phenomenon — the beat-to-beat rhythm and how it moves when you stand, lie back down and breathe on a pacer.
  3. A standardised protocol — twelve minutes, the same every morning, that produces those movements on demand.
  4. An interpretation of the regulatory pattern — the four signals and the fixed logic that reads them together.
  5. A load decision — TRAIN, REDUCE or RECOVER.
  6. A real outcome — whether following that decision, day after day, actually builds the athlete.

The first five links are engineered, frozen and regression-tested on every software change. The sixth is the one nobody has validated, and it is the reason the box above says what it says.

TRAIN, REDUCE and RECOVER are therefore not biological laws in themselves. They are product decisions derived from that principle. The principle is what the measurement actually reads.

What we can prove instead

We cannot claim clinical validation, and we do not publish the exact thresholds inside the engine. What we can show is that the thing is fixed, reproducible, and never arrives without its reason.

It is deterministic

Regression-tested on every change

The same beats always produce the same answer. Nothing in the decision path is random, time-dependent or personalised behind your back. A corpus of 65 real recorded sessions is replayed against the engine on every change, and every one must reproduce its expected decision, mode and score exactly.

That includes independence from your own history. Your morning is scored against fixed limits, not against a personal baseline the app has been building. There is no calibration period, and your first reading is judged by exactly the same constants as your hundredth. The only thing the engine is told about your past is whether your recent mornings were clear, which is what opens or closes the window for a progression day. Everything else history is used for — the trends, the comparisons with your own recent weeks — is shown beside the decision and never enters it.

It is frozen

92 files hashed

The engine is checksummed file by file — 92 of them — and the hashes are verified in the build gate. It cannot be edited quietly. Changing it is a deliberate release with a changelog entry, a version bump, a regenerated manifest, a re-baselined corpus and dual-engine evidence that no previously scored recording moved.

It does not learn

There is no model being retrained on your data, no weights drifting between app updates, no silent personalisation. The constants that decided your morning are the same constants that decided everyone else's, and the same ones that will decide yours next year. They change only in a versioned engine release.

It gives a reason every morning

Every number on screen is the engine's own output — we never compute a friendlier one for display. Alongside the decision you get all four signals and how each came out, what shaped the morning, and a plain statement when the read was of limited quality. The explanation is free and arrives with the decision, not behind a paywall.

What you do not get is the recipe: the engine's exact thresholds are not published. What is published is what makes it trustworthy: the logic is fixed and identical for everyone, and the same beats always give the same decision.

Where the AI sits, and where it does not

Today's Load has a language model in it. It writes the paragraph that explains your morning. It has no part in deciding anything, and the architecture — not a promise — is what keeps it out.

The model never sees the raw engine output. It is given a fact sheet computed deterministically from the result: the decision, the signals, your logged context. It may reorder, synthesise and phrase that material. It may not add to it.

What comes back is checked before you see it. Every number in the text must match a number on the fact sheet, and every directional claim — higher, lower, better than yesterday — must match the sheet's own direction. Text that fails the check is discarded rather than corrected, and you simply get no analysis that morning.

It is free and opt-in: it runs only when you tap Analyse this morning. The decision, the signals and the reasons arrive without it.

Honest limitations

It reads autonomic regulation, and nothing else. It does not measure glycogen, muscle damage, soreness, or sleep itself — those are context you bring to the decision.

A single morning is one data point. The reading is real from day one, but any morning can be moved by a late night, a meal, a rushed start. The direction over weeks carries more than any one day does — that is a limit of the measurement, not a warm-up period.

Published effects of heart-rate-variability-guided training are meaningful but generally modest; they do not justify claims of dramatic transformation.

It is not a medical device. It does not diagnose, treat or prevent anything, it says nothing about illness, blood pressure or heart rhythm disorders, and it is not a reason to delay talking to a doctor.

Known open questions live in the open. Where we have found something we do not like in our own engine, it is documented, reported to the science owner and left untouched rather than quietly patched, because a silent fix would invalidate every reading that came before it.

Sources

The measurement standards and the device-validation evidence Today's Load relies on:

Written for readers who want to check the reasoning. If something here is wrong or overstated, we would rather hear it: support@todaysload.com.