Your week, read honestly.

Every Sunday at 6pm (or the moment you ask for it), ZoneTwo generates a structured weekly review for you. It pulls every Strava activity from the last 60 days, computes the metrics that matter, runs them through a rule-based safety engine, and then asks a large language model to write a review that cites your actual numbers.

The six sections you get

1. Summary

Two sentences. What happened this week. What it means for the week ahead. No filler, no “great job keeping it up” — one honest read.

2. Metrics snapshot

CTL (fitness), ATL (fatigue), TSB (form), and an interpretation written in plain English. If you don’t know what TSB = 8 means, the review will say “you’re on the fresh side of fit” and what that implies for the week.

2. Highlights

Up to five things that went well: a PR-pace long ride, a consistent tempo block, a new-peak 20-minute power number. Each highlight cites the specific activity and the metric that proves it.

3. Concerns

Up to five things to watch. Rising monotony, an acute ATL spike, a long-run progression that’s ahead of your CTL curve. Each concern is tied to a rule-based trigger — not a vibe.

4. Next block

One to four weeks of prescribed training, with a progression plan for how the template evolves week-to-week. Each workout has a day, sport, duration, target zone, and an athlete-facing description (watts, HR, pace, RPE, and a feel cue).

5. Red flags

Surfaced by a deterministic rule engine, not the AI: acute spike, prolonged negative TSB, high monotony, zero quality, sudden drop, junk miles, discipline neglect, cross-sport load add. If a red flag fires, the AI has to write a recommendation for it — it can’t skip or invent rules.

6. Disclaimer

A short note on what the review is and isn’t. No medical advice. No injury diagnosis. Consult a sports medicine professional for persistent pain.

Three data pipelines feed the review before the model sees a single token:

• Metrics engine. Computes TSS (Coggan for power, Friel for running pace, sTSS for swim), then 42-day CTL, 7-day ATL, TSB, weekly Foster monotony and strain, and per-zone time distribution.
• Rule engine. Looks forACUTE_SPIKE, HIGH_MONOTONY,NEGATIVE_TSB_PROLONGED, ZERO_QUALITY,SUDDEN_DROP, JUNK_MILES,DISCIPLINE_NEGLECT,CROSS_SPORT_LOAD_ADD, andINSUFFICIENT_DATA.
• Plan continuity. Pulls the previous review, matches prescribed workouts against your actual Strava activities, and computes an execution ratio. The new review knows what you followed and what you skipped.

The model then writes the review inside a strict JSON schema with forced tool use. It cannot write a free-form essay — every claim has to land in a typed field. Red flags are replaced with the rule-engine output before the response reaches you, so the model can’t invent a problem that isn’t there or miss one that is.

The part of the review most coaches phone in. Generic 12-week plans aren’t adaptive. Ours is.

• Block length 1–4 weeks, chosen by the state of your training (volatile → 1 week, classic build → 3 weeks, full cycle → 4 weeks).
• Theme (base, build, peak, taper, recovery) that has to match your current state — an ACUTE_SPIKE forces recovery regardless of what you asked for.
• Weekly template of 3–6 workouts with explicit hard days and 48-hour spacing between intensity sessions.
• Progression notes that describe what changes week 1 → 2 → 3 → 4, and when a deload lands.
• Athlete-facing descriptions: watts, bpm, pace, RPE, and a feel cue. No “Z2 endurance, 60 min” on its own.
• Respects your preferences.Tell us you don’t train Mondays or only have the indoor trainer Tue/Thu and the plan will not violate that.