Plenty · Feature concepts

Thinking. Barcode scanning was considered and dropped: specialty-pharmacy boxes rarely carry per-item consumer barcodes, and half the shipment is supplies like needles, wipes, and mixing caps. The packing list is the one artifact every shipment reliably includes, and it already states quantities. The same flow accepts a photo of the order confirmation email, which covers the partner note about order sheets as input. The skip row matters: parsing everything and tracking only what the schedule uses keeps inventory meaningful.

Thinking. The unit mismatch is the root cause of bad inventory: people count containers, the schedule consumes units, and the two drift apart mid-pen. A “fraction of vial left” slider was rejected because nobody thinks in fractions at 8:30 PM. Instead, the dose log the user already keeps does the work: each logged dose subtracts its IU automatically, so the counts stay right on their own, just from logging doses like always. The pen graphic exists to make the math checkable at a glance, not to be another input. It earns a second job when a dose is due: if the open pen can't cover tonight, we show the split ahead of time (finish the open pen, then the rest from a new one), so two injections is never a surprise at 8:30 PM.

Thinking. The run-out date alone is half the picture: what actually determines your options is the gap between “allowed to refill” and “out.” When that window is one day wide, as in the mock, everything hangs on a single Thursday, and the app's job is to show that days early, not let it be discovered at the pharmacy counter. Showing the date instead of the rule was deliberate: users should never have to compute “last fill plus N days” themselves. The rule stays visible one tap down for verification.

Thinking. Storage advice is most useful exactly once: the minute a cold-packed box lands on the doorstep. So it arrives as a “put these away” card attached to the delivery event, instead of pages someone must know to look for. The what-it-does text rides on the med card because the question “why am I injecting this?” occurs while holding the vial. Both stick to the same rule: describe, never advise, and the clinic's instructions come first.

Thinking. The danger is alarm fatigue: if a skipped gummy looks like a skipped injection, both alarms stop meaning anything, and the strict layer is the one that kills cycles. So supplements are visually second-class on purpose: dimmed, no status color, no repeat reminders, no streaks. They ride the existing morning list because a second list would not get opened. Streaks and adherence scores were rejected outright; this audience does not need another thing to feel behind on.

Thinking. The worst failure mode this product has is silent staleness: a schedule that looks authoritative but stopped updating on Saturday. Compared to that, a loud false alarm is cheap. The “frozen as of Sat Jul 4, 9:12 AM” timestamp is the core of the design because it turns “can I trust any of this?” into “everything through Saturday morning is good, check anything after”. The bar stays until reconnection or an explicit switch to manual entry, and it never blocks the app: a stale schedule you can see beats a modal you must dismiss. The main screen keeps working from the last synced copy, and anything we computed from it, like deadlines and run-out dates, is marked “could be out of date” until the connection is back.

Thinking. Nagging is how tracking apps die, so the audit is risk-based, not scheduled: it only appears when a deadline depends on numbers that have not been touched in days. The framing does the persuading: “your Menopur deadline rests on these numbers” gives a reason, not a chore. Waste reports from the meds list (“lost a vial”) land in this same flow, so counts get corrected one way, everywhere. And the verification stamp closes the loop: every deadline in the app can show when its inputs were last confirmed, which is the same show-your-work idea that protects the reorder math everywhere else.

Thinking. This is really a delivery-risk alert, and weather is just its most dramatic trigger; carrier outages and holidays set it off more often. Two designs were rejected: silently moving the deadline up (the user must see why a date changed, or trust dies) and a weather feed in the app (ambient anxiety with no action attached). The rule we kept: only speak up when the delay threatens your supply, say it once, and put the fix in the same message. The push in the mock names the storm, the med, the slack gained, and the action, in two sentences.

Thinking. Each level gives up a little detail for a little more privacy, and each has a person it is for. Full is the solo household. Abbreviated is most people: the insight is that the reader supplies the context, so lossy text costs them almost nothing while costing an onlooker everything. Discreet is the open-plan office. Coded is the phone a kid or in-law picks up. One binary toggle (names on, names off) was rejected because it forces the office case and the borrowed-phone case into the same setting.

Thinking. One global privacy toggle was rejected: it forces the lock screen down to the setting the calendar needs. The surfaces leak to different audiences at different distances: the lock screen to whoever is nearby, notifications to whatever your screen is mirrored on, the calendar to your employer's Exchange server. So each surface carries its own level, and the defaults get quieter the more people can see a surface: Abbreviated on the lock screen, Discreet on notifications, Coded on calendar export. Partner access sits on the same screen but it's a different thing: a person you chose, not an audience you can't control.

Thinking. Nobody opens a fertility encyclopedia at 7 AM, but everyone wonders “why did my dose change again?” the moment it happens. So knowledge attaches to events (a phase begins, a med starts, a result lands) instead of living in a Learn tab. The library is deliberately tiny: one card per phase, one per med, a handful of nudges, every one clinically reviewed and sourced. The nudges stay physical and non-moralizing: water and activity, never diet culture or “relax, it helps.”

Thinking. The alternative considered was no AI at all: a canned FAQ. It fails at 9 PM, when the question is specific to your protocol and your day (“can I take it an hour late tonight?”). What makes the assistant safe to ship is its limits, and the mock shows a refusal on purpose, because refusing is the point: dosing questions get routed to the clinic with the fastest path to a human. Answers draw only on the user's own schedule, inventory, and clinic messages, so “right” means “consistent with your plan,” never “medically judged.”