Weekly AU
Also known as: Weekly Load, Weekly Internal Load
The sum of sessionAU across all completed sessions in a calendar training week, optionally split by modality (Strength / Endurance / Mixed). Weekly AU is the primary unit Afitpilot uses to describe "how much load did you absorb this week" — replacing weekly tonnage as the headline weekly number.
Formula
weeklyAU = sum(sessionAU for each completed session in the week)
weeklyAUByModality = { strength, endurance, mixed } [3-bucket coarse split]Example
Strength sessions: 420 + 380 + 460 = 1,260 AU. Endurance sessions: 180 + 540 = 720 AU. Mobility (mixed): 90 AU. Weekly AU = 2,070, split 61% strength / 35% endurance / 4% mixed.
How Afitpilot Uses This
We persist a load_summary doc per week per athlete with weeklyAU, the 3-bucket modality split, weekly tonnage, session count, and a modalitySource flag (explicit when the LLM tagged each session, classified when our deterministic fallback inferred it). The coach drawer renders a stacked-bar visualisation; the athlete Insights view shows the same numbers. Weekly summary LLM prompts now receive weeklyAU as evidence for narrative commentary.
Weekly AU in practice
| Who / Context | Value | Note |
|---|---|---|
| Recreational training (3-4 sessions) | 1,000-2,000 AU/week | Sustainable; modality split varies with goals |
| Hybrid block | 2,500-4,500 AU/week | Strength + endurance both contribute meaningfully; tonnage alone misses 40%+ of the load |
| Endurance athlete (build block) | 3,000-5,000 AU/week | Driven by long-session duration, not high sRPE |
| Taper / deload week | 30-50% drop vs preceding 3-week average | Visible immediately in the modality stack |
Known Limitations
- •Modality is a 3-bucket coarse split — "Mixed" is a catch-all for skill, mobility, hybrid sessions, and anything the classifier couldn't bucket cleanly. Power work and technique sessions all land here.
- •Weekly AU rolls up cleanly only when sessions have sessionAU. Backfilled estimates (planned duration) are included in the rollup but flagged.
- •The classifier (Path B) and the LLM-explicit modality (Path A) can disagree on the same session. Drift target is >=95% agreement on a 50-session validation set; we run the classifier daily in dev as a regression check.
- •Cross-athlete comparison is misleading because AU is sRPE-driven. A 2,000 AU week for one athlete is not the same as 2,000 AU for another — within-athlete trend is what's meaningful.
Science Context
Weekly AU is the foundational input for downstream load-monitoring constructs — acute:chronic ratios, EWMA trends, monotony, and strain. Modality split adds a layer Foster's original method did not have, motivated by Impellizzeri (2020): "the same internal load can be generated by very different external work, and treating that distinction matters for adaptation interpretation." Our 3-bucket coarse split is a pragmatic trade-off — finer splits (e.g. by sport) require more reliable session classification than the LLM currently produces.