analyse

/analyse — Activity analysis

Analyse a completed training session.

Arguments

$ARGUMENTS Required: intervals.icu activity ID, e.g. i12345678

---

Workflow

Analysis standard (MANDATORY):

• FIT file + sub-laps is the primary data path — intervals.icu streams alone are not enough
• Compliance — one canonical definition: direct compliance = actual vs. planned, computed by the analysing agent itself from activity + plan data. The precomputed compliance property from intervals.icu is unreliable — it is never cited and never used as a gate; coaching feedback never comments on it
• ERG-trainer power is not an athlete signal (MANDATORY). On an ERG-controlled smart trainer the device holds target watts whatever the athlete is doing, so a flat power trace is the trainer working, not the rider. Never report "power held to the last rep", "zero decay", "watts never dropped" as a strength, a finding, or evidence of adaptation, and never gate a progression on it. What stays valid from such a session is repetition count (volume compliance), cadence, HR and RPE — cadence being the variable that actually carries muscular capacity, since it falls when the athlete runs out while power does not. Check config/equipment.md for the trainer and its control mode before analysing any indoor ride; where the mode is unknown, assume ERG for a structured session and state the assumption. Prescribed watt anchors remain valid as dose — only the inference from held watts back to athlete state is invalid.
• GCT on recovery runs: High GCT during slow jogging is biomechanics, not error. Only evaluate GCT as a fatigue indicator when the GCT rise disproportionately exceeds the pace slowdown (pace-normalized). Do not comment negatively on absolute GCT values in recovery phases.
• Cool-down running dynamics are out of scope (MANDATORY). The cool-down is run at a shuffle — far below any pace the athlete trains at — and gait at that speed is a different movement pattern, not a slower version of the same one. Ground-contact time, vertical oscillation, step length, cadence and contact balance measured there describe the shuffle, not the session. Restrict every running-dynamics statement, trend and fresh-vs-fatigued comparison to the warm-up- completed main block; never close a dynamics argument with a cool-down value, and never let a claim about persistence ("the shift had not recovered by the end") rest on cool-down data. State the excluded window rather than silently trimming it. The same applies to the post-interval jog segments inside a session.