Time Gates × Time Gates Gates · 03 of 03
The contract, and go / no-go
Ten rules both sides implement, a checklist that has to be true before the gate is switched on for the first time, and the decisions still genuinely open.
The gate contract
These exist so the gate cannot quietly turn into a punishment, and so a fault in Time Gates cannot become a shut phone.
The gate is spent by doing, not by doing well.
Completion is trial counts, dwell time and elapsed minutes. No accuracy threshold, no improvement requirement, no streak that can break. The proposal states it directly: doing badly on the session still unlocks.
The model never grants, withholds or delays a credit.
The model proposes and the rule engine disposes. The credit ledger is explicitly on the list of things the steering model may not touch, alongside scores, gates and safety thresholds.
Education and Essentials are never held, and the training task holds Games alone.
A task may hold four groups - Games, Social, Video, Everything else - and the compiled policy carries that as four postures rather than a device lock. Education and Essentials are outside it structurally, so homework and Time Gates's own Tasks page are never behind a task. Time Gates takes less than it is allowed: its task holds Games.
A rushed session is refused at most twice, then credits anyway.
The refusal becomes a note in the weekly view, not a lockout. A teenager who cannot spend the gate is a teenager who stops opening the app, and no version of this product survives that.
Time Gates adds no second mechanism.
The training rule is an ordinary task and behaves like one: it is part of the schedule rather than a surprise, it never extends Dinner, Sleep or Bedtime, its reminders and holds pause during Sleep and resume in the morning, and a missed day stays due once instead of stacking. A device's per-zone override applies after a grant and before the hold, so an open device is still held and a parent's Add 15 still lands.
Fifteen minutes of warning, and the task is the way out.
An unchecked task runs the same ladder everything else runs: "Fifteen minutes until Today's practice holds games." There is no snooze under a hold, because a snooze would only delay the thing that ends it; the shield's primary button becomes Done, and checking it off lifts the hold without a network call.
Fail open, always - and the two-device split makes that real work.
No credit signal, an unreachable rule engine, one device reporting and not the other, a crashed session, a clock disagreement, an unreadable ledger: the hold lifts. A held group must always be a decision, never an absence, and with a phone, a computer and a rule engine in the path there are more absences to handle than a single-device design would have. Fetched is not applied, so a device that has taken a policy but not reported running it is treated as an absence too.
A parent override takes one tap and is never justified to the app.
Illness, exams, a bad day, a flare. The override is silent, logged, reversible, and available on the phone and the watch. The app never asks why and never counts it against anything.
Movement substitutes for a session.
Twenty logged minutes of sport, dance or martial arts spends the day. The evidence supports it, and it is the difference between a routine and a cage.
The shield says what is holding the group, in the words everything else uses.
Under a hold the shield leads with "A task is due", names the task, and says which groups stay open. One wording, drawn from the shared copy, across the phone, the watch, the shield and the notification. The teenager must never have to guess why Games is shut.
Go / no-go before the gate is switched on
Gate phase 1 does not begin because gate phase 0 was built. It begins when all of this is true, and the honest response to any of it being false is that the gate waits another week.
- Two burn-in sessions per teenager are complete and excluded from every reported number.
- A baseline median reaction time exists, gathered with no incentive attached to it.
- The required set is timed at under ten minutes at the teenager's actual pace, on their actual phone and their actual computer.
- Failing open is tested by breaking things: kill the connection, skew the clock, corrupt the ledger, force-quit mid-session, complete the phone half and not the desk half. The hold lifts every time.
- A credit cannot be forged by a changed date, a replayed event batch, or a second device.
- The parent override is one tap, on both the phone and the watch.
- The shield's line names the session and matches the wording in the app and the notification exactly.
- The teenager has been told, before the first held day, exactly what the rule is - including that a bad score costs nothing.
- An off switch exists that turns the whole gate off in one place, and someone has used it.
Open decisions
Six things this sequence assumes an answer to and does not have one for.
Where the rule engine lives (R8)
Open, and deliberately so until the trial is merged. The hub proposal recommends reducing the existing box to a ciphertext mailbox with every peer compiling its own policy and a hand-written merge. Whichever way it goes, the credit has to reach several devices, and the honest states are already known: syncing, ready, saved for later, consumed, expired - with only the live screen allowed to say ready.
Weekends
A daily requirement on a Saturday is a different product from one on a Tuesday. The default here is every day; the alternative is school days only, and it is a family call rather than a design call.
Whether the training task should hold more than Games
A task is allowed four groups; this one takes one. If a particular teenager's whole screen life is in Social or in Video, a task that holds only Games does nothing at all - which may still be the right answer for that teenager, and is a family call rather than a design one.
What a missed day costs tomorrow
Nothing, in this draft. No debt, no catch-up, no doubled session. Debt mechanics are the fastest known route to abandonment; they are also what would make the gate bite.
The sensor budget
A chest strap unlocks the Settle work that has actual biofeedback evidence behind it. Without one, that column is breathing and a camera pulse, and the heart game does not exist.
Clip licensing for Notice
The research instruments are licensed and finite; model-narrated scenes are free and weaker; filmed scenes are the best use of a small content budget. This decision sets whether gate phase 3's Notice work is real or a demonstration.
The trap this design walks into on its own
In the one double-blind trial of a multi-skill training game that included a placebo arm, parent and teacher ratings improved by medium-to-large amounts in every arm, placebo included. The improvement was real and it came from routine, attention and expectation rather than from the drills.
A household that has just started a daily session with a visible gate has all three of those in abundance. Everyone will feel that it is working, and that feeling is not evidence - it is the single most likely thing to happen whether or not anything transferred.
The only protection is to measure the far outcome somewhere the product cannot reach: an opt-in, parent-completed rating scale run before and after, outside the interface, never shown to the teenager, and never used to open or close anything. Questionnaires are far more stable than tasks, which is exactly why the honest external check is a form a parent fills in rather than a number the app produces.
Version 133 · Privacy · Terms of use · Appearance