FitCraft · Coach Intelligence · Brainstorm Results · Internal
How do we make the coach feel genuinely intelligent — adaptive, alive, personal — without cameras, without complicated UX, for beginners? Seven focused brainstorms generated 59 ideas grounded in the real codebase; four adversarial critics (engineering, beginner-UX, retention, content-cost) tested every one. This page is the curated result: the thesis, the tiered roadmap, the mechanics we invented, and the graveyard.
Every validated pattern from the competitive research — JuggernautAI's "you said you were tired, so today is lighter," Fitbod's silent learning made legible by a muscle heat-map, Whoop selling explanation as the product — reduces to one move: take something the user did or said, change something real because of it, and say so in one verifiable sentence.
Deterministic rules read as intelligence when the loop is narrated. The fanciest adaptation reads as a bug — or goes completely unnoticed — when it isn't. Narration is not a feature on this roadmap; it is a shipping discipline attached to every feature on it.
Standing rule: nothing adapts silently, and nothing is spoken that didn't happen. Every line the coach says must be a verifiable fact, kindly framed — one hallucinated claim in the trainer's voice is trust-fatal. (This is the same "receipts, not compliments" law the squad design agreed on.)
Every idea that survived all four critics implements the same five-step loop. It's the spine of the whole roadmap:
Dormant domain data first (difficulty ratings, per-exercise skip statuses, rest deltas, days away) — then near-zero-friction taps: one chip, one button on an alert that already exists.
Rules, not LLM prescriptions. Novices under-report effort, so require converging evidence before adapting anything.
Through proven seams: client-side rest/work scaling, tier switching across the four authored difficulty trees, rule-engine line selection.
One line, one fact. "A few extra breathers," never "three pauses." Framed as smart self-management, never as an audit. Struggle is never counted out loud.
Every adaptation recorded with its input context — debuggable, auditable, and speakable later ("remember when we eased off in March?").
The codebase mapping surfaced the single most actionable fact of the whole exercise: FitCraft's problem is signal consumption, not signal capture.
WorkoutFeedback). No system consumes it. It's the
cheapest intelligence we will ever ship.ExerciseEventRecord knows Skipped /
AttemptedButCouldntDo / CompletedPartialThenSkipped per exercise, plus prescribed-vs-actual rest.WasFirstWorkoutAndQuit sits unused while 13,112
accounts vs 1,090 people-who-ever-worked-out is the biggest leak in the business.All of it is reachable by the rule engine — most of it is one [RuleProperty] attribute away.
Two hygiene fixes ship first: subtract coach-audio time from rest deltas (the rest timer stalls while the
trainer talks, polluting the signal), and reconcile the debug force-complete path that fakes first-workout
completions.
Ranked. Items 1, 2, 4 and 5 ship with their narration lines in the same release — that's the law. Almost everything here is backend + FCAdmin authoring with zero client release.
The existing 1–5 rating finally does something. Phase 1: expose the last rating and a consecutive-too-hard counter to the rule engine; the coach's next preview line references what you said — "you called last one brutal, so we ease in today." Phase 2: two "too hard"s in a row → visibly longer rests next session.
Why it won: the only idea all four critics green-lit. Phase 1 needs no app update (audio is server-selected). And it retrains users that ratings have consequences — which raises signal volume for every other idea on this page.
Watch: ratings are sparse. Bias conservative and cross-check against completion/skip data.
The workout-completed line cites what actually happened. The client already counts pauses, rest skips and +20s presses; those counters ride an existing context dictionary to the completion audio call, and ~12–15 new rules turn them into lines like "taking that breather mid-way was the right call — and you still finished everything."
Why it won: three critics' top pick; the highest wow-per-line-of-code on the list. The coach proves it was watching — kindly.
Watch: the shame trap. Hard content rubric — quantities stay vague ("a few extra breathers"), one behavior per line, always framed as smart self-management, never an audit.
Quit your very first workout → next day, one warm trainer push introducing a purpose-built, finishable 8-minute "Round Two" workout: "most people's first try is a scouting mission."
Why it won: the retention critic's #1 — this is the only idea aimed directly at the 13k→1k activation cliff, and the beginner-UX critic called its framing "the best de-shaming sentence in the list."
Watch: the trigger flag is polluted by debug force-completes — reconcile that data first.
Come back after 2+ weeks away → the server sends a comeback modifier (longer rests, slightly gentler work — heavier the longer the break) that decays over your next 2–3 completed workouts. The coach frames it as muscle memory, and never says how long you were gone.
Why it won: strong across all four critics once three lenses' duplicate comeback ideas were merged into one. ~900 lapsed exercisers are the biggest recoverable pool we have.
The quit-confirmation alert currently plays a guilt clip — a live violation of our own care-over-guilt principle. Replace it with warm lines plus a third door: one tap shrinks what's left to a 2–3 minute finish, ending in a real completion instead of a quit.
Why it won: client-only, 1–2 weeks, sitting on the highest-leverage 30 seconds in a beginner app. Quits become completions.
Watch: decide the product policy first — do short finishes count toward streak/XP? (Recommendation: yes.) Plain quit stays one tap away; never ask more of someone quitting.
Pure rule authoring, zero client changes: dense early milestones (3rd, 5th, 10th workout — where retention is actually decided); "Tuesday morning — that's our time" habit recall; rematch callbacks when you conquer a workout you previously bailed on (fired at completion as redemption, never at preview as anxiety); and Fight Week — training-camp hype in the days before the boss battles the programs already contain.
Why it won: everything here is the rule engine's documented easy extension path. "The fastest genuine ship on the entire list."
The product gets exactly one pre-workout chip surface, launched with a single chip: "Short on time?" One tap trims today's workout to ~10 minutes — full streak and weekly-goal credit, and the coach celebrates that you showed up instead of treating it as lesser. "Low energy" comes later only if tap data proves the surface. "Feeling strong" never ships (it invites beginner overreach).
Why it won: "no time" is the #1 stated skip reason, and this feeds the north-star metric directly. All four critics demanded the five competing chip ideas merge into this.
Watch: structural trims can falsify hand-authored stage audio ("three rounds of this…"). Run the automated audio truth-scan before launch.
The backend watches ratings, completion fractions and skip statuses; at level boundaries it proposes moving up a difficulty tier with one accept tap: "you've been cruising through these." Down-shifts are never proposed by name — a demotion notice is a demotion notice no matter the copy. Struggling users get invisible rest/dose scaling narrated as the coach's plan.
Why it won: the four difficulty tiers are fully authored parallel trees — real adaptation with zero new content. Wrong-tier placement from one-shot onboarding is a silent churn driver.
Watch: define the cross-tier level/XP carry-over policy before building.
There is a literal //TODO - DO CHEERING AND CELEBRATION HERE in the workout
screen. Fill it: 3–5 reaction animations (short cheer, big celebrate, patient idle instead of the current
freeze-frame pause, beckon) triggered at rest, stage boundaries and the finish. Coach Mood Mirror follows
later, reusing the same clips as home-screen idle moods.
Why it won: the 3D coach is the paid differentiator and it never physically reacts. Zero lines, zero scientist bandwidth, zero localization — and the only idea that improves muted sessions. It multiplies every audio line shipped around it.
Mid-workout reactions from buttons users already press: skip rest three times in a row → the coach shifts to a hotter clip pool; repeatedly extend rest → a caring pool plus a silent +10s on later rests. Client-picked clips, no server round-trip, no latency. Bundled: a halfway nod and a final-round intro in three intensities, keyed only to locally verifiable facts.
Why it won: claims the "in-workout affect" lane the research flagged as unowned — in-workout pleasure predicts +38–41 min/week of future activity. This is where dropout decisions form.
Watch: audio stacking during short rests — single-slot priority so whispers replace tips.
Two missed-goal weeks → the coach offers a new deal (one tap lowers the weekly target), framed as the coach's seasonal plan — the failure math is never shown. A 1×/week Maintenance Mode covers busy seasons and suppresses nag pushes.
Why it won: the permanently-unachievable-goal loop is the deadliest quiet churn frame for a population where 74% of active weeks contain exactly one workout — and 1×/week preserves >90% of training gains.
Must scope in: streak logic hardcodes 2 workouts/week and ignores the goal — a renegotiated 1×/week deal with a still-breaking streak is a user-visible contradiction. Reconcile first.
At first-workout completion, the system quietly composes a capsule from the user's own onboarding answers (their goal, their "why") through approved voice templates — and the coach plays it back at workout 10 or first level-up: "Day one, you told me why you started. Listen."
Why it won: plants an open loop at onboarding and cashes it exactly at the milestone cliffs, using data currently rotting unused. Impossible to read as canned — because it isn't.
Watch: v1 uses the goal×trainer matrix with no spoken name (names wait for the Speakable Names infra). Keep the payoff warm, not ceremonial.
Each bet ships in stages with a kill-switch between them. Bet 0 is the gate the others' content appetite depends on.
5–10× the coach's line variety per rule by letting the existing FCAdmin bulk-generation pipeline run at full capacity — with an LLM rubric judge replacing line-by-line human approval. Staged: shadow mode → auto-approve low-stakes segments with a 25% human audit → 10% as the leak rate proves near-zero. Slop in a trusted trainer's voice is worse than repetition, so trust is earned segment by segment. Depth goes to the ~20 highest-traffic rules, not blanket coverage.
Make mid-workout audio rule-driven without mid-session round-trips: one call at workout launch evaluates the rules once and returns a slot-indexed audio plan; static authored audio remains the fallback for every unmatched slot. Start with two slots (stage intro + final-round intro), measure, expand. Corner Coach — a personalized mid-workout receipt line — only builds once the manifest and the render cache (Bet B) are independently proven.
A render cache over the existing template-gated TTS path — never rendered live; a cache miss serves a static line and queues the render. Stage 1: milestone numbers + first names (riding the squad plan's Speakable Names investment, with a pronunciation round-trip check). Stage 2: Nightly Postcards — slot-filled, human-approved sentence frames, text-first, each paired with a trainer push. Stage 3: the end-of-Season recap — generated only when there's a real story to tell, never a padded report card.
The unification: a deterministic per-user program resolver — a pipeline of pure transforms (tier select → dose scale → exercise swap → deload) with an adaptation ledger, and the exercise scientist setting bounds as reviewable config with an offline simulator. No LLM anywhere in the decision path — the critics' verdict on LLM-planned sessions was unambiguous: every useful adaptation was a deterministic function of signals we already have; the LLM added only nondeterminism and review burden. Migrates the already-shipped transforms (comeback, rest-scale, express) into one seam once two show measured lift.
Genuinely novel mechanics that survived critique — none exist in any competitor we studied:
Recurring failure patterns, promoted to design laws:
North star: second-workout-of-week conversion (26% today). Nearly every Tier 1–2 item feeds it directly: rescued quits, express completions, comeback sessions, Fight Week's day two, renegotiated goals.
Complementary tracks that share rails, not surfaces. Tier 1 is deliberately backend and content-ops heavy, so it runs parallel to the Seasons MVP without competing for client engineers — Feedback Loop-Back, Soft Landing and the Memory pack ship with zero client release. Sequencing hooks:
The complete inventory, grouped by brainstorm lens. Each idea shows the four critic verdicts in fixed order — Engineering feasibility · beginner UX · Retention impact · content Cost — and where it landed on the roadmap.
Lens 1 — Adaptive difficulty & progression
The coach quietly tunes rest lengths to how you actually rest. Users who always mash +20s stop having to; users who always skip rest get a snappier workout. The coach mentions it once: "I've given you a bit more breathing room between sets."
→ merged → Rest Whisperer (Tier 2)
Come back after two weeks away and the coach doesn't pretend nothing happened — it hands you a deliberately easier re-entry session and says why: "Muscle memory is real. We ramp back over three workouts and you'll be right where you were."
→ TIER 1 · #4 Soft Landing
On the workout preview, three optional chips: "Short on time", "Low energy", "Feeling strong". One tap reshapes today's session in place, and the trainer's opening line acknowledges the choice.
→ merged → Give Me 10 chip row (Tier 2)
After a level, the coach says: "You've rated the last four workouts easy and finished every one — I think you're ready for Intermediate. Want me to move you up?" One tap and the whole program shifts tier.
→ TIER 2 · #2 — up-only
Fail or skip burpees twice and they're gone — replaced by an easier movement that trains the same thing. The coach: "I traded burpees for step-backs today. We'll build back to them." No shame, no settings screen.
→ deferred → Living Program stage 2 (Bet C)
Push hard three weeks straight and the coach schedules you a lighter week before you burn out: "You've earned a recharge week — same workouts, lighter doses. This is how progress sticks."
→ KILLED — overtraining math for a population that undertrains
Quit Tuesday's workout at 60%? Wednesday the home screen offers a 12-minute session built from exactly what you didn't finish. "Close it out" instead of "do it again."
→ survived as invented mechanic “Close It Out” (backlog)
Whenever the plan changed, the coach says why in one specific, verifiable sentence: "You rated Thursday a 5 and skipped the last round, so today is trimmed. Beat it and we scale back up." The intelligence layer that makes every other feature visible.
→ promoted → the narration law itself, attached to every feature
The bold bet: every user's program becomes a per-user document, re-materialized from the master content each session — tier, swaps, dose scaling, deloads, and comeback ramps applied as one deterministic pipeline, every decision logged and speakable. The scientist authors one program; the system serves thousands of personal variants.
→ BET C
Lens 2 — Coach memory & relationship
The coach knows when you usually train and notices when you break pattern: "Tuesday morning — that's our time." or "An evening session? Look at you, mixing it up."
→ TIER 1 · #6 Memory & Story pack
When you re-enter a workout you previously bailed on or skipped exercises in, the coach calls it: "We've got unfinished business with this one." Finish clean and it pays off: "Last time this workout beat you. Today you beat it."
→ TIER 1 · #6 Memory & Story pack
Your 10th, 25th, 50th, 100th workout and your one-month/one-year 'first workout anniversary' get marked by the coach — optionally speaking your actual name at the big ones.
→ TIER 1 · #6 Memory & Story pack
Coming back after a break, the coach picks up the conversational thread — referencing what you did together last time and reassuring you about muscle memory — without ever mentioning how long you were gone.
→ merged → Soft Landing (Tier 1)
The coach ends a session by setting something up — "Show up Thursday and I'll have something for you" — and actually pays it off next session: "Told you I'd remember." A two-beat memory arc across sessions.
→ invented mechanic — folds into Memory pack once payoff fixes land
At milestones and program completions the coach contrasts present-you with week-one-you using verifiable facts: "Your first session, you tapped out halfway. You haven't done that in weeks."
→ backlog — receipts family, not prioritized this cycle
Bounded personalization inside otherwise-normal coach lines: "{FirstName}, that's workout fifty." or "You skipped {ExerciseName} last time — let's settle the score." Rendered once, cached forever.
→ BET B · stage 1
Once a month your trainer sits you down for a 45-second personal spoken recap of YOUR month — your record week, your comeback day, what changed since last month — in their voice, about facts only you did. Bold multi-month bet.
→ rebuilt → end-of-Season recap (Bet B stage 3)
Lens 3 — In-session dynamism
The workout-completed line proves the coach was actually watching: "Three pauses, two extra breathers — and you still finished every rep. That's grit." No generic "great job" ever again.
→ TIER 1 · #2
Work out at 5:50am when you're normally an evening person and the coach opens with "Before sunrise? That's a new side of you. Let's use it." The coach knows your rhythm and notices deviations.
→ merged → Memory pack time-pattern category (Tier 1)
Skip three rests in a row and the coach says "You're flying today — just don't leave it all in round two." Hammer +20s twice and it says "Take the air you need, I'm not going anywhere" — and quietly gives you slightly longer rests for the remainder.
→ TIER 2 · #4
A live 4-state read of how the session is going — cruising / pushing / grinding / struggling — that every mid-workout voice moment keys off, so tips and round outros match the user's actual state instead of being random.
→ ships logging-only — calibration for Rest Whisperer
At the halfway mark the coach marks the moment with a fact — "Halfway. And you haven't skipped a thing." The final-round intro comes in three heat levels, picked by how the session actually went: earned fire for a crusher, steady resolve for a grinder.
→ merged → Rest Whisperer bundle, facts-only
The 3D trainer physically reacts: claps you through the final rest, flexes when you finish a stage skip-free, and shifts to a patient, arms-crossed waiting stance when you pause for two minutes — instead of freezing mid-burpee at speed zero.
→ TIER 2 · #3
Hit Quit at 80% done and the coach says "You're two minutes from done — two minutes." Hit it early while clearly struggling and it offers a dignified exit: "How about we skip to the finisher so today still counts?" — one tap jumps you to the final stage.
→ merged → Land the Plane v2 (Tier 1)
Every stage intro, round outro, and tip in the entire workout becomes rule-engine-driven and personal — 'Level 3, workout two, and your streak's safe after this one' as a ROUND intro — without a single mid-workout server round-trip.
→ BET A
Mid-rest, in the trainer's actual voice: "Sarah — that's your longest session this month, and it's not over." Real name, real number, generated for this user, dropped at exactly the right breather.
→ BET A · stage 3 — after manifest + render cache prove out
Lens 4 — Zero-friction signal capture
The 1-5 difficulty + 'did you enjoy it' rating users already tap after every workout finally changes something: next session the coach references it out loud and rest times visibly shift. Zero new UI — we close a loop that's already half-built.
→ TIER 1 · #1 — unanimous top pick
On the WorkoutPreview screen, three optional chips — Drained / Ready / Fired up. One tap and the coach opens the workout acknowledging it, and a Drained day gets ~25% longer rests and slightly slower demo pace. The coach says why: 'You told me you're running on fumes — longer breathers today.'
→ merged → chip row (“Low energy”, only if tap data earns it)
The coach starts noticing what you actually did — extra rest on lunges, a skipped plank, breezing through cardio — and mentions it, specifically and kindly, at workout end and before the next one. The user never fills anything in; the log IS the feedback.
→ absorbed → input signals for Tier Autopilot
During rest — dead time the user is already staring at — three small chips ask about the exercise just finished: Too easy / Good / Tough. Flag an exercise too easy three times and it visibly levels up next session; flag it tough and it gets shorter or swapped, with the coach narrating the change.
→ backlog — per-exercise friction, not prioritized
Above the Start button: 'Short on time' / 'Feeling sore' / 'Go easier'. One tap reshapes today's workout in place — fewer rounds, longer rests, or lighter work — and the coach affirms it ('Smart call. Twenty good minutes beats zero perfect ones'). Adapted workouts count fully toward streak and weekly goal.
→ merged → Give Me 10 chip row (Tier 2)
Grant permission once and the coach quietly knows your steps, sleep, and workouts logged in other apps: 'You already got 12k steps in — this'll feel easy,' 'Saturday's run counted toward your week, by the way.' External workouts count toward the weekly goal and streak, so the app stops penalizing people for exercising elsewhere.
→ KILLED — revisit at 10–20× scale
Come back after two-plus weeks away and the coach doesn't pretend nothing happened, and doesn't guilt you either — it quietly serves a gentler version of your workout and explains: 'We're easing back in. Muscle memory is real — you'll be back to full speed fast.'
→ merged → Soft Landing (Tier 1)
Quit a workout at 60% and next time the coach remembers: 'Last time life cut us off partway — zero guilt, today's a fresh run at it.' The coach feels continuous across sessions instead of goldfish-brained.
→ partially absorbed → Personal Rematch (quit-context properties)
Lens 5 — LLM coach brain
The coach stops repeating itself: 5-10x the line inventory per rule by letting the existing FCAdmin bulk-generation pipeline run at full capacity, with an absolute LLM quality gate replacing line-by-line human approval so one exercise scientist is no longer the throughput ceiling.
→ BET 0 — the gate
An offline LLM agent reads the coach's own forensic logs to find the user-moments it currently fumbles with generic fallback lines, then drafts the missing rules for human approval — the rule inventory grows itself.
→ folded → one-time audit query inside Bet 0
The coach speaks to what actually motivates this user — the accountability-seeker hears 'I noticed, I'm watching', the reward-chaser hears 'that unlocked something' — using an onboarding answer the backend already has but never reads.
→ backlog — not prioritized
The coach's completion line references what actually just happened — 'you fought through those mountain climbers even after skipping the rest' — instead of a generic well-done, using per-exercise data the client already holds at the moment it requests the audio.
→ superseded → Finish-Line Receipts (the rule-based version won)
Every night the system writes and voices 1-2 lines just for you — grounded in your real week — so tomorrow's WorkoutPreview greets you with 'third Tuesday in a row, and last week you finished every single exercise' in the trainer's actual voice, with zero serve-time latency.
→ BET B · stage 2
The 20 minutes where users spend the most time with the coach — mid-workout — stops being frozen per-workout audio and becomes rule-selected per-user: stage intros that know it's your streak-saver workout, round outros that know you're about to level up.
→ merged → Session Audio Manifest (Bet A)
The QuickAction buttons grow up: tap 'Ask Coach', pick or type a question ('is it normal to be this sore?', 'what does this exercise work?'), and get an answer in the trainer's persona — text instantly, voice seconds later.
→ KILLED — moderation ops; salvage someday as voiced FAQ chips
The workout itself quietly adapts: you rated Tuesday 5/5 hard and skipped the burpees twice running, so Thursday arrives with longer rests and a burpee swap — and the coach tells you exactly why in her own voice.
→ merged → Bet C — guardrails kept, LLM deleted
Lens 6 — Beginner psychology & plan flexibility
On the workout preview, a single 'Short on time?' chip shrinks today's workout to ~10 minutes. The coach celebrates that you showed up instead of treating the short version as lesser — a shrunken workout still counts toward streak and weekly goal.
→ TIER 2 · #1
Rate a workout 'too hard' in the existing post-workout popup, and the next session opens with 'Last one was rough — I've got you today,' with rests actually a touch longer. The feedback popup finally, visibly does something.
→ merged → Feedback Loop-Back (Tier 1 #1)
Come back after two-plus weeks away and your next workout is quietly re-ramped 10-20% easier, with warm muscle-memory framing ('your body remembers — you'll be back to full speed in two sessions'). Absence length is never mentioned.
→ merged → Soft Landing (Tier 1)
On days you are not training, the coach still shows up: a home card or trainer tap delivers a 15-second rest-day line — 'Two down this week. Today your muscles do the growing. See you Thursday.' Rest becomes prescribed, not absence.
→ delivered through Morning Announcements (squad plan) instead
Miss your weekly goal two weeks running and the coach offers a new deal instead of nagging: 'Want to aim for 2 days a week for a while? We can raise it later.' One tap. A separate 'busy season' toggle drops expectations to 1x/week Maintenance Mode where one workout is a celebrated win.
→ TIER 2 · #5
Hitting Quit mid-workout no longer plays a guilt clip — the coach offers a third door: 'Give me one more round and we'll call it a win.' Tap it and the workout truncates to a 2-3 minute finish that ends in a real completion celebration.
→ TIER 1 · #5
During any rest, a small 'Rough one?' chip. Tap it and the coach immediately eases the remainder — longer rests, final round dropped — and says so out loud within seconds.
→ merged → Rest Whisperer (Tier 2)
Over months, each user's program quietly diverges from the authored template — rests, work times, round counts, occasional exercise swaps — tuned by everything they've ever done, with the coach narrating every change: 'You've stopped skipping rests, so I nudged the pace up.' One authored program becomes thousands of personal variants with zero new authoring.
→ BET C — same bet, second lens
Quit your very first workout and tomorrow your trainer sends one warm push with a purpose-built 8-minute 'Round Two' workout, framed as 'most people's first try is a scouting mission.' No mention of quitting.
→ TIER 1 · #3
Lens 7 — Wildcard
Before the workout the trainer makes a playful, specific prediction ("I'm betting you get through the burpees clean today") — and after the workout they pay it off either way ("Called it." / "Burpees take round one. Rematch Wednesday."). The coach suddenly has skin in the game.
→ KILLED — pass/fail stakes on failure-averse beginners
Every level's day 3 is already a boss battle — but nothing builds toward it. Fight Week turns days 1-2 into training camp: the coach counts down ("Two days out. Friday we take the boss."), a weigh-in-day push lands the morning of, and the boss workout opens with a walkout moment.
→ TIER 1 · #6 Memory & Story pack
As the last stage ends, the coach throws down an optional 45-60 second bonus dare — one exercise, framed as personal ("One more round of mountain climbers. You in?"). Two buttons: Bring it / Done for today. Declining costs nothing; accepting earns a visible flourish.
→ invented mechanic — build after Receipts prove appetite
Each 8-week Season, your trainer lives through their own 8-episode story — Ty preps for a competition, Xayla rehabs an old injury — told one ~25-second chapter at a time, unlocked by YOUR workouts. Miss a week, the story waits for you.
→ KILLED — un-QA-able canon treadmill
The 3D trainer on the home screen physically reflects where you are: bouncing and warming up when your streak is safe and a workout is due, doing quiet stretches on your rest day, and greeting you with an easy, welcoming posture when you come back from a break. You read your status off the coach's body before a single word plays.
→ fast-follow → Cheer Loop (Tier 2 #3)
During onboarding the coach 'records a message for future you' built from your own stated goal. Weeks later — at your 10th workout or first level-up — the coach plays it back: "Remember day one? You told me you wanted to keep up with your kids. Ten workouts in. Look at you."
→ TIER 2 · #6
When you repeat a workout you've already beaten, the coach occasionally hands you a remix instead: same shape, same difficulty, same equipment — but the exercises are reshuffled and a few are swapped. "I rebuilt this one for you last night. Let's see how you handle my version."
→ KILLED — novelty beginners don’t want
Every workout opens with a 12-second locker-room briefing that is never the same twice: a greeting beat, a today's-mission beat, and a stakes beat, snapped together from short clips — "Wednesday. Back again. — Lower body today, we finish on lunges. — Two more workouts and Level 3 is yours."
→ KILLED — value ships as ordinary WorkoutStarting rules