S-Bike Training Hub

Make the checked profile the hub's bike - only after the check passed, the athlete felt the resistance change, and they agree. share=true returns a pre-filled GitHub link so they can share the profile and the next owner of this bike needs no assistant (ask them first).

Apply the exact reviewed coaching draft after explicit athlete approval. Rejects stale state, forecast violations and unresolved restrictions. Read back get_coaching_review and the calendar; this never grants running clearance.

Save the exact reviewed preview by draft_id, only with athlete approval. A changed athlete state or expired draft is rejected; retries are idempotent while the draft remains available. Read back get_program and get_training_calendar afterward.

Apply exactly the previewed report back-fill after the athlete approves it. Requires the preview token and approved=true; a changed ledger or record is rejected (preview again). Writes a receipt; idempotent.

Apply exactly the reviewed selected model draft through the normal Hub API. Requires approved=true; stale evidence and wrong drafts rejected, retries idempotent. Retires block countdowns but preserves injury/protected return-to-run checks.

Record the bike's data packets for a few seconds while the athlete pedals as you asked (e.g. label 'steady 60 rpm', then 'steady 90 rpm') so you can see which bytes follow cadence and power. Listen-only: nothing is sent to the bike. Tell the athlete what to do first.

Make a standard timed workout: warm-up 15%, ramp 15%, intervals 40% (split evenly), rests 20%, cool-down 10% of total_minutes. Intensities in % of FTP. Optionally make it today's ride.

Make any workout from blocks, each one of: {type:'steady', minutes, pct|watts, label?}, {type:'ramp', minutes, from, to, label?} (pct of FTP), {type:'intervals', times, on:{minutes,pct}, off:{minutes,pct}}. Optionally make it today's ride.

Describe a drafted lift session against the joint framework before planning it: each exercise's joint and group percentages, the session's blend (weighted by sets), the groups its main exercises work (a group counts from 25% of an exercise), whether the shape fits, and the session's classifiers: Type (groups), Duration (main exercises per group), Intensity (light/medium/heavy by % of each lift's estimated max, or of what was moved last time), Style, with each lift against its own last time. Give sets, reps and weight to get them. Describes; it does not judge the session or save anything.

Start here to compose a lifting session inside the classifiers, in one call: Type = the groups (upper, trunk, lower), Duration = main exercises per group (1-3), Intensity = light (30-60% of the estimated max), medium (61-80%) or heavy (81-105%, as far as is safe; never maxing), Style = standard, endurance, power_explosive or power_heavy. Returns the athlete's rules, per group their lifts with the last time done and a target weight, the style's sets and reps, and the two supporting minors (sport support for the sports being built, and core alternating anti-rotation and rotation by week). Leave intensity, style or duration out and the hub recommends them from the phase and the athlete's last sessions (their reports and next-day soreness), with the reasons and those sessions listed; phase=taper also gives the volume cut (40-60%). Choose from it or search the web for exercises that fit, check with describe_lift_structure, then plan_lift_session after the athlete approves.

Score a lift session before it's done: points per exercise (intensity against the estimated one-rep max, or effort for throws/bands/holds, times tempo), shared across regions, the restorative share against the steer (fits_steer), against the rider's usual, and what each region is already carrying. Use it to judge and adjust; nothing is saved.

Investigate a reading: implemented formula, constants, source hash, profile inputs, baseline, recorded contributors, planned doses and assumptions. Unknown is not zero; these models are provisional.

Heart or unheart a library workout by id, or keep an existing scheduled workout by date and index. Prescriptions and actual activity records are preserved.

Inspect candidate heart-rate response windows from an imported workout. Timed effort has priority, then the athlete’s duration estimate; HR rise/peak/fall is supporting uncertain evidence, not exact work timing or force. No exercise labels, assumed lag correction, or automatic logging. Ask the athlete to confirm candidates.

Read ONE imported completed workout with watch observations and existing Hub load estimates. Swim: SWOLF, average/best distance per watch-counted stroke, pool length, stroke-specific metrics and excluded drill lengths. Run: pace, cadence, steps, available form metrics and laps. Energy rate: estimated metabolic watts and kcal/min with energy source and duration basis, separate from measured cycling power. Missing values stay unknown. Read from calendar activity_id; does not change records, load or holds.

Read progression policy, recent decisions and unresolved localized symptoms before changing workloads. Does not change anything.

The aerobic engine ride by ride, from the watch's ride files (the bridge is its power meter, so power and heart rate together): efficiency factor (normalized watts per beat - rises with fitness), aerobic decoupling (watts per beat, first half vs second - under 5% stayed aerobic; judged only on 40+ min steady rides), and heart rate at the same 100-120 W (falls as the heart gets fitter), with trends. Import the watch's ride files first (import_activities).

What needs going through with the athlete in the coming week, most urgent first: stop or whole-body warnings, training-rule breaks, forecast conflicts, running deloads and targets, calibration questions, missed sessions without a reason, cautions. Each item says what and the lightest fix. record_checkin returns the same list; work through it in the same conversation (preview, apply once they agree, say what changed).

When the athlete says 'get my bike working' (or their bike won't connect to the hub): read the bike setup request they queued from the hub's welcome page - the bike's Bluetooth name, services, readable values and data packets captured while they pedaled - plus step-by-step instructions, safety rules and the profile format. Start here and follow its what_to_do steps. Does not change anything.

The rider's capacities, each from tests and training with a confidence that fades after the last test: FTP (ramp test; 20-min bests between), heart rate at 90 W (morning diagnostic), big-gear 3 min (big-gear test; grit rides between), critical swim speed (400 m + 200 m time trial), and the running block (a benchmark run taken well can grow it, 10% a test). Plus which tests are coming due - put them on the calendar with schedule_test.

Read estimated goal demands, observed sport exposure, gaps and unresolved symptoms. Optional demands previews a read-only scenario. Does not predict finish times, calibrate limits or save a program.

Recent daily check-ins, diagnostics, verdicts and plans - the trend over days.

Every climb the rider has ridden on a route (or all): date, climb number, length, grade, time, average and max watts, cadence, km/h, VAM (m climbed per hour), gears used - to see progress on the same climb and set the next goal.

Read the next 1–14 days of running/lifting outlook, forecast conflicts, holds, symptoms and evidence-based capacity review candidates. Does not save or clear holds.

Fitness (42-day load), fatigue (7-day), form, advice, the last 7 days' totals and the last 14 days day by day. From power, bike rides only - check COROS for runs, swims and sleep.

Read the detailed rules for one kind of work before doing it: planning, capacity, rules, running, lifting, reports, checkins, workouts, races or work_rate (the server instructions list what each covers). Read once per conversation per topic.

The health and history records the athlete has chosen to share: the latest pre-exercise screening (ten questions following the ACSM 2015 screening logic: no doctor needed, check before vigorous training, check before starting, or pause and see a doctor), conditions, past and current injuries, medications (marked when they change heart rate) and what a doctor or physiotherapist advised. Read before planning; plan around it as the athlete's own guidance. Never diagnose, never clear anyone to exercise, never override a clinician. Local only.

The rest of what the watch recorded, cross-referenced. Signals and flags only - nothing here scores the rider or changes a load model; ask the rider each flag's question rather than judging. SWIM, per set: SWOLF (seconds + strokes per length), strokes, stroke rate, pace, and how the stroke changed - held / gradual (slow fade: ordinary fatigue) / sudden (a step between neighbouring lengths - even a small one is the more telling strain signal) - plus how far in it first broke down. swim.session_length reads that across swims: use it to decide how long swims should be. Heart-rate drop in the rests is a wrist reading in water: rough. RUN: laps and 5-minute splits of power, cadence, heart rate, pace, vertical oscillation/ratio, step length, ground contact; form drift is flagged only when the pace matched. RIDE: work in kJ, heart rate in the opening minutes against the usual at those watts, heart-rate recovery in the minute after hard efforts, and (from get_aerobic) drift; ride.work_vs_legs fits kJ to the next morning's legs, ride.leg_check asks whether cadence and heart rate show the running load, ride.climbs compares each climb to the last time up. ALL: pauses mid-workout (count, time) - many of them is a question, not a fault. Trends say when they have too little data.

Read one exact reusable prescription and its original response records. Use this after a compact get_workout_library listing, then inspect current phase and whole-calendar load before scheduling or changing it.

Lifting and functional strength. Read before planning or scoring a gym session: the rider's RULES (never plan what a 'No ...' rule excludes; use favourites), the coaching GUIDANCE, the STEER (restorative or build, the suggested restorative share of the session and why, from the load the rider carries), sessions the rider built that are waiting for you to score (unscored_sessions), the exercise library, estimated one-rep maxes, recent sessions, the follow-up that's due, and each region's recovery block. Units: the rider's (lb or kg). All load numbers are estimates.

The joint framework to plan lifting inside: six joints (shoulder, elbow, wrist, spine, hip, knee, ankle) in three groups (upper, trunk, lower), each joint's movement directions, the body regions and which joints they move, load levels (light 30-60%, medium 60-80%, heavy 80-90% of an estimated one-rep max), and the session shapes: one group with 1-3 main exercises, two or three groups with 1-6, plus up to 3 minor ones. Compose original sessions from your own exercise knowledge, giving every exercise region shares; use it for exercise ideas mid-session too, and search the web for exercises when you can. Structure only: training rules are still being written, so plan conservatively. Manual workouts are not bound by it. Read with get_lifting (the rider's rules and maxes).

Training load by body system across ALL sports (runs, swims, gym, rides): ENGINE (heart/lungs; watts on the bike, heart rate calibrated to watts elsewhere), IMPACT (feet/bones; running and walking x body weight), MUSCLE (leg force; pedal torque, gym effort, running). Each has fitness, fatigue, form, the acute:chronic ratio and provisional zone, this week's used load versus the current estimated capacity, and a weakest-link readiness verdict with what's limiting. These zones are planning heuristics, not validated injury-risk thresholds. Use this before set_plan.

Read the plan's load outlook: recorded and projected 3-day rate, biggest day in the last 3 and 28-day load per day through the plan, phase targets (recovery 50-60% of the reference rate, taper 40-60%, build steps gated by reports), the reference basis (own 28 days, or a starting estimate for new athletes) and flags. Read-only; use it when placing or moving sessions, then preview changes as usual.

Streaks (days, weeks on goal), lifetime totals, milestones earned and next up.

Start here when planning, in two passes, both on the athlete's request. scope=program (pass 1): placeholders for the whole macrocycle - sport, day, role improve or maintain, a minutes band and, for lifting, joint groups - saved with preview_program. scope=week (pass 2): the coming week's real workouts inside those placeholders, drafted with preview_coaching_change kind=calendar (its mirror compares the draft with recent weeks). Returns a brief first: at_a_glance (goal and days to it, the phase now and next, holds, reports well above their usual, each sport's usual week, calibration, health status, events), the last 7 days of reports (journals shortened), the phases or the week, the general playbook, and 'more': an index of sections. Ask for a section only when the decision needs it (sections=['volume'], sport='run'): volume, efficiency, reports in full, calibration, health_history, program, running_gate, playbook for one sport. Reason from it, explain 'why this week', show the draft, and ask: approve, regenerate, or what would you change? Apply only after approval.

Read the saved macro program: goal, phases, weekly session counts and capacity notes, compact by default. detail=true returns the full phases (placements, review criteria), every week and the capacity review, which preview_program needs; phase=<label> returns one phase in full. Does not change anything.

Programming a swim, bike or run session: the phase (counted back from the next race: base, build, peak, taper, recovery), the rider's load state for that sport (fresh, moderate, heavy but ready, not ready), the tier cap and the suggested tier (easy, moderate, hard, very hard), and the nearest templates from the research-backed library, filled with the rider's paces (critical swim speed, FTP, a recent 5K) - each with what it trains, what it doesn't, and sources. For swimming also the stroke mix for this block (four strokes out of 100, from the race or the rider's dials), eight reusable swim_profiles with eligibility reasons, selected profile, main-set work ratios, equipment and weekly sequence context. Use swim_profile to preview another eligible module. Profiles are provisional coaching defaults; respect blocked choices. Record swim_profile and the returned swim_plan snapshot on planned swim sessions; adapt its why to the actual goal and projected load. When the sport isn't a good idea today it says why and offers mobility or another sport. Use it before writing any swim, bike or run session; adapt the template, don't invent one; keep the week's core sessions the same for 4-6 weeks and progress them. Lifting uses get_lifting.

Read measured training responses and evidence for improvement; modeled conditioning is not a guaranteed performance gain.

Race predictions so far, per race: every saved prediction (most likely time, range, legs, basis, levers), how it moved, whether a new one is due (a test since the last one, or 3+ weeks) and, after the race, how each one graded against the result.

The athlete's response to the plan, week by week: planned vs done minutes against their available time, each week's phase and shape (build, consolidation, peak_volume, peak_performance, taper, race), hard sessions, and every missed session with its reason. Read it before deciding whether to build, hold, recover or taper next week. Does not change anything.

Read open and recent bad-report re-checks: why each started (pain 5+, pain worse than last time, soreness well above usual), when the evening and morning re-checks are due, answers, outcome, and whether running is held or the next increase waits. Ask the athlete the due re-check questions.

Read watch-detected pool lengths and strokes from an imported FIT swim into an editable transcription. Drill mode is unidentified: ask which drills or kicks were performed. Flags missing lengths and mismatched totals. Does not approve, schedule, log or change load.

Read imported watch workouts or choose matching scheduled and recorded workouts for one activity. Supports every imported sport. Suggestions are not confirmed links. Read this after import_activities; the same choices appear in the manual import review.

Everything about one ride (ride id like ride_2026-09-27_0959 from get_rides): stats, time in power zones, highlights, best efforts, records, workout adherence grades per part, streak, fitness.

Recent rides: date, start, minutes, average and normalized watts, intensity (IF), training load.

A saved route's climbs, numbered from 1: where each starts and ends (m along the route), length, gain, average and max grade - for placing climb goals and efforts with set_route_session.

The rider's saved routes: id, name, km, climbing, steepest grade, easy/mixed/hilly.

Review the single athlete running recovery time scale, rolling 42-calendar-day evidence, missing/confounded intervals, pooled candidate and validation limits. Read-only. Clean recovery needs feet/legs reports and delayed confirmation; a completed run during a hold is not clearance.

Review selected report-anchored running response, separate feet/legs observations, timing uncertainty, 90-day evidence and validation limits. No block clearance date.

Read completion, partial completion, athlete cancellations and missed sessions separately from changes to prescribed targets. Finished days only. Baseline/revisions are tracked from first observation; earlier edits are unknown. Adherence does not establish fitness improvement.