Decisions only you can make
Everything below survived adversarial review and now needs an owner's call. Where the principles already point somewhere, the recommendation is marked — say "agreed" or override. Full evidence for each lives in RESEARCH.md (Parts I–IV, §6 of each).
The three that shape the product
Fixed-clock nap as the product, or onset-anchored?
The original vision was onset-anchored everything. The evidence: sensor noise on onset detection (±13–33 min) is wider than the whole nap decision, while a fixed ~20 minutes lying down works for everyone on day one and automatically shortens your sleep when you take long to drift off — "errors land early" by structure.
Recommendation: fixed-clock is the product; onset detection becomes an invisible refinement that can only shorten a nap, never extend it. This is your works-fully-or-not-offered rule applied honestly.
Do the coaching surfaces ship at all?
Nap-window advisories, evening warnings, bedtime nudges, debt bands — individually defensible, but assembled they are Rise with an alarm attached: the tracker you rejected, arrived at by accretion.
Proposed rule: anything that renders when no alarm is set is out of scope. Keeps invisible calibration and the deadline/buffer disclosure; cuts most of the rest. Yes or no on that rule.
Is the niche worth validation-grade engineering?
The gap is real — nobody combines onset-aware naps + nights + haptic wake + guaranteed deadline + free. But the category's entire visible demand is ~30 app ratings, Apple ships sleep features yearly, and your rule demands measured accuracy before claims. The research tells you exactly what it costs to make the product true; whether that's worth it is strategy, not evidence.
No recommendation — genuinely your call.
Product promises & honesty
- Rename the promise. "Wakes you in light sleep" can't be claimed (misses deep sleep a third of the time or more, zero real-time validation). Offerable: "times your nap and wake from detected sleep onset, does its best to avoid deep sleep." Rec: rename.
- Onset accuracy claims wait for our own measurement. Until on-wrist validation exists, the UI shows detected onset and frames timing as assisted, not guaranteed. Rec: agree.
- The deadline guarantee must be honest about the iPhone. Watch-standalone guarantee covers only the 30-min window; beyond it the phone must be reachable. Onboarding says so. Rec: agree.
- Kill the sleep-debt nap-shortening mechanic. Our signature "smart" idea rests on a number nobody has ever measured, moves less than the sensor noise, and inverts on safety (tired sleepers are hardest to wake). Replace with the inertia buffer: alarm fires at deadline−10 min (−20 after a rough night or before driving). Rec: cut it, ship the buffer.
- Drop "Rise's sleep-need number is unvalidated" from positioning. The literature actually supports their approach. Compete on honest error framing and the deadline guarantee instead. Rec: drop.
Scope conditions & disclosures
- Nap mode is conditional, not committed. If wrist haptics can't reliably wake a sleeper under Sleep Focus (device test), the mode is cut — not shipped weak. Rec: agree.
- Night mode ships windowed or not at all. All-night sensor streaming can kill the battery and the alarm with it. Rec: windowed.
- Naps fill your Activity rings — unavoidable, no workaround. An "alarm, not a tracker" that touches rings every nap: disclose prominently, or consciously accept the contradiction. Rec: disclose.
- Battery gate is a feature, not polish. Session refuses to start, loudly, on low battery or Low Power Mode. Rec: agree.
- Contraindication screen. Ask about insomnia/CBT-I, shift work, jet lag at onboarding; disable coaching (never the alarm) for those users. Rise has this; we had nothing. Rec: required.
- Arrhythmia users: warn at onboarding (heart-rate-based detection unvalidated for them) or fund validation. Silent degradation violates your rule. Rec: warn.
Smaller calls
- Collect age, or label every number "for adults"? Every default is a healthy-young-adult artifact; teens need 8–10 h and a different evening zone. Rec: one onboarding question.
- Allow a one-tap "rough night?" self-report? It's the one input that's actually observed (vs. inferred); it would only widen the inertia buffer. Rec: allow.
- Calibration gate vs. reality: many users won't wear the watch at night; the gate must never block the basic nap, and never nag people into overnight wear. Rec: agree, instrument first.
- Two research chores before defaults ship: re-source the deep-sleep-latency distribution and four broken citations flagged in Part II. Rec: do in build phase.
Phone-only mode (Part III)
- iOS version floor for phone night mode. iOS 26+ gives a clean system-alarm guarantee (survives app death, Silent, Focus); older iOS can only promise "guaranteed while the app stays open." Rec: iOS 26+ only.
- Mattress mode: cut entirely? Worst-validated modality ever tested (κ 0.07), haptics aren't partner-safe on a shared bed, and charging-on-the-mattress contradicts Apple's safety guidance. Rec: cut.
- Restless-moment early wake default. The one bedroom smart-alarm trial found no benefit over a good gentle alarm, and every early fire costs real sleep. Rec: ship narrow, measure grogginess in-product before promoting it.
- Partner mode depth. With a partner or pet in range, breathing features turn off (the sensor measures the bed, not a person) — leaving "excellent alarm + snore events." Accept that, or don't offer night sensing to partnered users. Pretending to track one of two sleepers is banned by the evidence. Rec: accept the honest reduction.
- Accuracy publication policy. Only metric-and-population-matched numbers, each with its conflict-of-interest flag printed — no "independent study" in this field actually is. Rec: adopt.
- Fund our own validation, or two features never ship: sonar-gated nap onset and the "onset errs late, the safe direction" claim both have no borrowable evidence. Rec: fund during build.
- Android timing. Different failure surface (OEM battery killers, sonar whitelists) — needs its own works-fully gate list. Rec: defer until the iOS product earns its accuracy numbers.
- Sonar = a permanent per-handset validation commitment. Every new phone generation needs re-testing, or the in-app self-test must be authoritative enough to admit unknown devices. Rec: build the self-test to be authoritative.
Oura & WHOOP integration (Part IV)
- Build the Oura integration at all? It's calibration-only forever, with real overhead (OAuth, webhook renewals, lapse handling, partner-app approval) — and a week of the app asking "when did you actually fall asleep?" each morning may match its entire value with zero dependency. Decide against that honest baseline, not against zero. Rec: defer until the app's own confirmations prove insufficient.
- Approach WHOOP legally before writing WHOOP code? Their terms include a non-compete "in any manner," a ban on keeping permanent copies, and at-will termination — a feature killable by contract overnight. Rec: legal review + explicit WHOOP approval as a hard gate, or don't start.
- Can a "bonus, when conditions allow" live leg exist under works-fully-or-not-offered? The WHOOP live heart-rate leg can only ship best-effort — but a feature that sometimes silently doesn't run is adjacent to the half-working state your rule bans. Options: truthful nightly preflight status at lights-out, internal pilot only, or drop. Genuinely your call — the research couldn't resolve it.
- Partner-mode consent topology. Her account, her data, on a phone she may not own — consent, revocation, and who sees her sleep-derived schedule are unresolved, and WHOOP's third-party-exposure clause may bear on it. Rec: design this before any partner feature, not after.
- Fund the bench program. Overnight BLE persistence, sonar+Bluetooth coexistence, real webhook latencies, lapsed-account behavior — none of it is measurable from literature; all of it gates claims. Rec: fund during build, with the other validation work.
- Set a vendor re-verification cadence. Both vendors churn APIs fast (WHOOP v1→v2 in 4 months; Oura killed personal tokens; both are moving toward owning the wake moment themselves). Rec: re-check per vendor release; every integration built droppable.