How EnduranceIQ works

Overview & narrative summaries

The weekly surface merges deterministic rule-engine findings with optional Claude Haiku prose generated only on the server from numeric aggregates: HR summaries by workout, intensity percentages, load indices, and rules-output snippets—never free-form athlete notes or Strava titles. Outputs pass validator gates before persistence so blocked prose falls back to template paragraphs grounded in the same findings JSON.

Intensity distribution

Each timed running interval contributes classified HR buckets grouped into easy (zones 1–2), moderate (zone 3), and hard (zones 4–5) using athlete-relative thresholds anchored on observed max HR or configured ceilings from onboarding data. Benchmark heuristic aligns loosely with polarised prescriptions—a directional compass rather than laboratory-derived physiology labels.

• Seiler (2010)
• Stöggl & Sperlich (2014)

Training load (acute vs chronic)

Acute stress aggregates roughly trailing-seven-day totals versus chronic rolling averages derived from synced endurance workloads with measurable strain proxies (runs contributing HR-derived strain estimates). Elevated ratios flag spikes needing pacing restraint relative to historical norms—not Garmin readiness scores.

• Gabbett (2016)
• Windt et al. (2017)

Heart rate zones

Zones approximate Daniels-style proportional anchors scaled against empirical maximum HR captured across uploads plus validated onboarding overrides where athletes capture lactate-threshold estimates elsewhere. Field zones tolerate noisy telemetry—prefer drift-aware pacing cues paired with perceived exertion rather than rigid BPM policing alone.

Canonical physiological thresholds demand metabolic lab testing; EnduranceIQ flags deviations versus heuristic envelopes rather than diagnosing physiology.

Concurrent training & interference windows

Heavy neuromuscular sessions overlapping endurance stimuli inside acute physiological recovery arcs elevate cumulative fatigue risk; EnduranceIQ highlights narrowly spaced resistance-plus-interval stacking scenarios surfaced via deterministic timestamps rather than subjective readiness guesses.

• Fyfe et al. (2014)
• Wilson et al. (2012)

Strength for runners

Programmable lifting prescriptions referencing plyometrics, heavy compounds, and injury-prevention circuits arrive alongside roadmap integrations tying biomechanical weaknesses to prescription tweaks—planned downstream phases populate richer workout widgets referencing citations similar to running insights below.

• Blagrove et al. (2018)
• Beattie et al. (2017)

Session classification

Strava-derived session taxonomy mixes importer-normalised labels (`easy_run`, `long_run`, `interval`, …) with sport modality inference plus HR-relative adherence badges summarising drift versus endurance envelopes described earlier—classification informs downstream UX colouring rather than autonomous scheduling prescriptions yet.