Adaptive Learning

The Adaptive Learning Layer is where Layerr gets smarter over time. It analyses execution traces, extracts patterns, and adjusts future routing decisions based on what actually worked.

Learning Sources

Source	What Is Learned	Key File
Execution Outcomes	Which provider/model succeeded for which workload	learning/adaptive/engine.ts
User Feedback	Explicit thumbs-up/down on responses	src/features/replay/learning/learningModel.ts
Provider Learning	Provider-specific performance trends	src/features/replay/learning/learningModel.ts
Strategy Learning	Which strategies work best for which intents	learning/adaptive/engine.ts
Failure Patterns	Common failure modes and their predictors	runtime/protection/classifier.ts

The Learning Loop

Execution Trace
    |
    v
[1] Extract Outcome    --> Success? Quality? Cost? Latency?
    |
    v
[2] Correlate Features --> Which workload signals predicted the outcome?
    |
    v
[3] Update Weights     --> Adjust routing and scoring weights
    |
    v
[4] Flag Drift         --> Alert if provider quality degrades
    |
    v
[5] Build Summary      --> Present learning insights to user

Adaptive Classification

The confidence classifier (learning/adaptive/engine.ts) learns to predict routing confidence:

function classifyConfidence(
  workload: WorkloadProfile,
  provider: ProviderMetadata,
  historicalTraces: Trace[]
): ConfidenceLevel {
  // Uses historical success rate, error patterns, and workload similarity
  // Returns: high | medium | low | uncertain
}

Failure Learning

The failure classifier (runtime/protection/classifier.ts) categorises failures and learns from them:

Failure Type	Learning Action
Rate limit	Reduce request rate to this provider for similar workloads
Timeout	Increase timeout or switch to faster model
Content policy	Flag content type, suggest alternative provider
Provider outage	Automatically deprioritise provider for next N hours

Provider-Specific Learning

Provider learning (src/features/replay/learning/learningModel.ts) tracks:

Metric	Description
providerLearning	Historical performance score per provider
strategyLearning	Historical performance score per strategy
Latency trends	Is this provider getting slower?
Error rate trends	Is this provider becoming less reliable?
Cost efficiency	Quality per dollar, trending over time

Learning UI

Component	File	Purpose
ProviderLearningView	src/features/replay/learning/	Shows provider learning curves
LearningReplayInsights	src/features/replay/learning/	Insights from replay analysis
AdaptiveLearningControls	src/features/intelligence/	Toggle learning features on/off

File Reference

File	What It Does
learning/adaptive/engine.ts	Main adaptive learning engine. Updates weights, classifies confidence
runtime/protection/classifier.ts	Failure classification and pattern learning
src/features/replay/learning/learningModel.ts	Frontend data model for learning insights
src/features/intelligence/hooks.ts	React hooks for intelligence and learning features

Integration

1. Replay Layer → provides trace history for pattern extraction
2. Provider Health → receives updated health scores from learning
3. Scoring Layer → receives updated provider quality scores
4. Strategy Engine → receives recalibration recommendations
5. Explainability → includes learning insights in explanations