Science — Adaptive Decision Engine (ADE)

5. Adaptive Decision Engine — ADE

The question this layer answers

Given load, physiology, performance, adaptation and the future plan, what should change next?

ADE is the operational decision layer at the end of the Montis Intelligence Stack.

It does not recreate the calculations performed upstream.

It consumes the resolved evidence and produces a base coaching decision, which is then checked against phase and target-event governance.

The core distinction is:

ADE Base Decision = What can the athlete tolerate now?

Phase / Event Governance = What should the athlete do next?

ADE Base Score

The ADE Base Score starts at 100.

The engine then applies explicit support signals and penalties.

The current score considers:

• operational state;
• risk flag;
• forecast fatigue context;
• forecast load trend;
• HRV guardrail;
• target-event / taper context;
• nutrition where confidence is sufficient;
• ESPE adaptation state.

The score is deliberately labelled pre_phase_governance.

This means the number is not the final training instruction.

A high score can still be overridden by strategic phase requirements.

Operational-state effect

Current ADE scoring includes:

• recovery_priority → substantial penalty;
• load_accepting → supporting driver.

This gives the current load/recovery relationship strong influence over the base decision.

Forecast fatigue and productive_fatigue

Future Forecast resolves the projected relationship between planned load and future form.

A state such as productive_fatigue means the plan is expected to create meaningful fatigue while remaining within a productive overload context.

It is not the same as recovery_priority.

• productive_fatigue describes the projected future load/form context.
• recovery_priority describes the current operational state.

An athlete can therefore be load_accepting today while the future calendar deliberately moves toward productive fatigue.

Why Forecast Trend: Increasing does not always lower the score

The current ADE logic is context-sensitive.

If forecast load is increasing while:

• operational state is load_accepting;
• risk is normal;
• forecast fatigue is not amber/red;
• the athlete is not inside an A-race taper window;

the increasing trend can be recorded as a supporting driver:

Forecast load trend is increasing and aligned with load-accepting state.

Increasing load becomes penalised when it conflicts with evidence such as:

• recovery_priority;
• moderate/high risk;
• amber/red forecast fatigue;
• taper requirements.

This is important because progressive training should not be penalised simply for progressing.

Risk Flag

ADE derives an immediate risk flag from the future fatigue classification:

• green / transition → normal;
• amber → moderate;
• red → high.

The risk flag is then scored separately from operational state.

This prevents one metric from carrying the entire decision.

HRV guardrail

When HRV ratio is available, ADE applies an additional autonomic guardrail:

• below 0.90 → suppressed;
• 0.90 to below 1.00 → mildly reduced;
• 1.00 or above → stable/supportive.

HRV modifies the decision but does not replace the other evidence.

Adaptation state

ADE consumes the ESPE adaptation state as a light-touch modifier.

Examples include:

• maladaptation / decline → penalty;
• mixed adaptation → smaller penalty;
• baseline / stable / positive adaptation → support;
• other defined adaptation states → contextual support.

ADE does not recalculate the power curve.

Event and taper governance

Target events introduce additional constraints. Only A races get taper governance.

ADE can consider:

• event priority;
• days to event;
• training bias;
• taper state;
• projected Event TSB;
• target Event TSB range;
• whether projected form is too fatigued, in range or too fresh.

Inside a taper window, increasing planned load can therefore have very different meanings:

• too fatigued + increasing load → strong conflict;
• target form + increasing load → unnecessary fatigue risk;
• too fresh + increasing load → controlled sharpening may be appropriate.

This is why more freshness is not automatically better.

Phase governance

After the base ADE score is calculated, Montis resolves the strategically required phase.

The final outcome can:

• honour the base directive;
• honour it with controlled sharpening;
• override it because phase requirements take precedence;
• suppress live guidance for historical reports.

This gives Montis its "Can versus Should" model.

Example:

Can: current physiology is load accepting.

Should: reduce load because the athlete is in a required recovery phase.

That is not a contradiction. It is the purpose of governance.

Acting on the decision

To turn the decision into today's plan, press WHAT NEXT? on the Overview ADE card or the Micro panel. It restates the final, phase-governed decision in plain words, for example whether to keep, adjust or reduce today's session. See What next?.

Score labels

The current ADE Base Score is labelled:

• 85–100 → excellent;
• 70–84 → strong;
• 50–69 → caution;
• 35–49 → constrained;
• below 35 → blocked.

These labels describe the base operational decision before phase governance.

They should not be presented as a universal readiness percentage.

---

How the five layers connect

The stack should be read from top to bottom:

Training Load
What stress was applied?

↓

Physiology Response
How did the athlete respond?

↓

Performance Intelligence
Is the load being absorbed and expressed effectively?

↓

Adaptation Progression
Is capability changing in the intended direction?

↓

Adaptive Decision Engine
Given the current state, future plan, phase and event context, what should happen next?

The output of one layer becomes governed context for the next.

The purpose of the stack is not to generate more metrics. It is to progressively reduce many pieces of athlete evidence into a traceable coaching decision.

Evidence → Interpretation → Adaptation → Decision