Performance · IM Cervia 2026

ClaudeCoach

An AI coaching system built on Claude + Intervals.icu for a 21-week Ironman build. Four automated jobs run daily and weekly — checking readiness, modulating sessions, monitoring load, and summarising the week. Every decision includes a traceable reasoning trail linking a real data point to a rule to an outcome.

Automation — runs without any user action

06:33

Daily

Session prescription

Pulls live readiness signals from Intervals.icu, runs the modulation engine, pushes the adjusted session back to the calendar. Sends a notification only if the session changed.

07:03

Daily

Watchdog check

Evaluates 8 load and recovery triggers. If nothing fires, it produces no output at all — zero noise on good training days.

18:07

Daily

Capture reminder

Checks Intervals.icu for any key session (TSS >40 or duration >45 min) logged today that hasn't been captured yet. Notifies if one is missing.

18:00

Sunday only

Weekly summary

Generates a week card with TSS vs planned, compliance rate, CTL change, flags, and Monday focus. Pushes a single headline notification.

Weekly rhythm

Mon

Check-in

06:33

07:03

Tue

06:33

07:03

18:07

Wed

06:33

07:03

18:07

Thu

06:33

07:03

18:07

Fri

06:33

07:03

18:07

Sat

06:33

07:03

18:07

Sun

06:33

07:03

18:00 summary

Automatic User-triggered Weekly

Session day sequence

06:33 daily

Prescription

Readiness assessed, modulation engine runs, adjusted session pushed to Intervals.icu calendar.

automatic

morning

Train

Garmin records the session. Activity syncs to Intervals.icu via IcuSync after completion.

you do this

after session

Capture

Log RPE, gut comfort, heat tolerance, fuelling adherence. Triggered by you or the 18:07 reminder.

you trigger

after capture

Debrief

Lap drift, aerobic decoupling, power zone distribution, form metrics. One coaching note written to the session log.

automatic

Data flow

Inputs

Live signals

CTL / ATL / TSB
HRV trend (7d)
Sleep last night
Yesterday RPE
Ankle pain score
Temp + dew point

→

Analytics engine

Pure Python primitives

Modulation rules R1–R7
Env pacing (L1)
Banister CTL/ATL model
TSS compliance analysis
Ramp headroom check
Lap debrief metrics

→

Outputs

Actions + reasoning

Adjusted session on calendar
L2 reasoning trail
Push notification
Session debrief note
Weekly summary card
Watchdog alert (if triggered)

Intelligence layers

W2 · Daily prescription

Session modulation

Seven rules applied in priority order at 06:33. Adjusts intensity, interval count, or replaces the session entirely based on fatigue, HRV, sleep, RPE, and heat.

L1 · Environmental

Environmental pacing

Heat and humidity corrections to bike IF and run pace for the Cervia course. Segment-level wind time tax using actual course headings (280° / 100° / 0°).

W4 · Watchdog

Load monitoring

Eight triggers evaluated daily: ATL spike, weekly ramp, HRV trend, sleep deficit, missed sessions, aerobic decoupling, heat dose trajectory, heat recency.

W1 · Re-optimiser

Week re-optimiser

After missed sessions: assesses TSS debt and ramp headroom using the Banister model, then proposes a redistribution that respects quality-session spacing and the ramp cap.

Compliance

TSS compliance

28-day rolling analysis of planned vs actual TSS. Classifies why the gap exists — skipped, duration short, fatigued, or not pushed — then prescribes the right fix for each.

W6 · Post-session

Session debrief

Lap drift (first→last), aerobic decoupling (HR:power ratio across halves), Coggan zone distribution, Stryd form metrics for run. One coaching note per session.

Session modulation rules

Rule	Signal	What happens	Type
R1	Ankle pain ≥ 3/10, or quality movement not yet cleared	Session blocked — no run or brick today	Hard stop
R2	ATL − CTL > 25 (acute fatigue spike)	Quality session swapped to Z2 or easy run; duration preserved	Swap
R3	HRV trend < −7% over 7 days	Intensity −5%, interval count −1	Soft
R4	ATL − CTL between 15 and 25	Intensity capped at 95% FTP	Soft
R5	Yesterday's RPE ≥ 8	Intensity −5%	Soft
R6	Sleep < 6h + HRV depressed (two signals)	Swap to Z2. Single poor night alone: interval count −1 only	Swap / Soft
R7	Ambient temp or dew point above threshold	Intensity reduced by combined heat + humidity correction	Soft

Rules apply in priority order. R1 exits immediately — no other rules run. R2 exits after the swap — soft rules do not stack on top. Intensity floor: 65% FTP regardless of how many soft rules stack.

TSS compliance — why is there a gap?

When planned TSS consistently exceeds actual TSS, the system classifies why before deciding what to do. Scaling targets up is only correct in one of the five cases.

completed

Within 12% of target

No action needed

skipped

No activity logged

Adherence issue — protect calendar time, not a load problem

duration short

Session cut to <80% duration

Scheduling / time pressure — fix the slot, not the target

intensity · fatigued

Full duration, high RPE, low TSS

Load too ambitious — reduce targets or add recovery

intensity · soft

Full duration, low RPE, low TSS

Execution gap — scale targets ×(1/rate), capped at ×1.20

Reasoning trail format — every decision is explained

The L2 format enforces accountability: a real data point, a named rule, a concrete adjustment, and a predicted effect. "Fatigue looks high" is not permitted.

Modulation:
HRV −8.5% over 7d → R3 (HRV below −7% threshold) → drop interval count 4→3, target 95% FTP not 100% → maintain quality stimulus, reduce cumulative strain ~15%

Watchdog:
ATL 148 vs CTL 121 for 4 days → T1 (ATL > CTL +25) → insert recovery day today, swap Thursday threshold to Z2 → TSB recovers ~8 pts by weekend, ramp drops to +3.2/wk

Debrief:
Power fell 14% first→last lap → paced the opening set 8–10W above target → start 5W lower on next threshold session → power holds consistent through the full set

Compliance:
87% compliance over 28d, dominant gap: intensity_short_soft → full duration, low RPE, below target power — execution gap not fatigue → apply ×1.15 correction to quality session TSS targets → actual TSS lands on goal load within 2–3 weeks