Every state the rider app can be in between opening it and finishing a ride, what each one holds, and where it goes when things fail. Built from the same readiness rule the product runs on: required = max(trip × 1.5, 2 km).
An IA is mostly the consequences of a few early choices. These are the three that produced everything below — and the third one is the reason this document exists at all.
Readiness is trip-dependent, so a bike that clears 3 km does not clear 9 km. That makes the map a result, not a place. If the map were the root, the first screen could only promise a 2 km floor — a weaker guarantee than the product is built to make.
No tab bar once a destination is set. From A7 onward the rider is inside a task holding a payment authorisation and possibly a reserved bike; letting her wander into Account mid-funnel creates states nobody designed. Persistent surfaces hang off A5/A6 only.
The lock's open and close pings define when a ride starts and ends — not the app's buttons. Every entry point is therefore resumable, and a dead phone can't run a fare. This is the single most load-bearing choice in the architecture.
It wasn't a principle I started with. It came out of one edge case — what happens if the rider's phone dies mid-ride — which exposed an assumption buried in my first model: that the app was the source of truth for when a ride ends. It isn't, and it can't be. If the app owns ride state, a dead battery means a fare running until the bike's does.
Moving authority to the lock pings fixed that, and then paid for itself three more times: it is why F12 needs no on-device screen, why L4 can expire a session mid-ride without interrupting anything, and why going offline at A11 is survivable — the rider locks the bike and walks away, and the receipt catches up when the phone reconnects.
The distinction between branch and layer is the one that does the most work here. A branch is somewhere the rider lands; it has a screen and an exit. A layer is a condition — offline, permission denied, session expired — that changes how several steps behave without being a place you can be. Treating connectivity as a branch is how apps end up with a dead-end error screen at every step.
Because ride state is server-side, a cold start can't assume it is starting at the beginning. It resolves identity, then an active ride, then an active hold — in that order — and routes past the funnel entirely if either exists. The rider who force-quits mid-ride and reopens the app lands back on her ride, not on a map.
The spine is linear on purpose. The only genuine fork in the whole product is at A5, where a rider without a destination can fall back to a floor-only browse — and that is labelled as the weaker guarantee it is, not offered as an equal path.
Branches sit to the right of the step that produces them. Every one has a stated return path; none is a dead end, and none requires the rider to start over.
What each node holds, which agent is working behind it, and where it can go. Silent means the agent is acting with no screen; surfaced means it has become visible because confidence dropped low enough that a human should decide.
| ID | Node | Type | What the rider sees | What the system holds | Agent | Exits |
|---|---|---|---|---|---|---|
| Phase 1 — Session · identity only, no trip exists yet | ||||||
| A1 | Phone number | Step | One field, numeric keypad open. Errors resolve on blur, never per keystroke. | Nothing yet. No account is created until A2 succeeds. | absent | → A2 |
| A2 | Verification | Step | Six boxes, one labelled field to assistive tech. Resend reads "Resend in 30s". | Session token on success. | absent | → A3 · wrong/expired code stays here |
| A3 | First name | Step | One field. No surname, email or gender — nothing else is needed to complete a ride. | Display name. | absent | → A4 |
| A4 | Wallet | Step | UPI or card, plus "Skip for now". "No deposit, no cash" stated up front. | Payment method, or a null that A8 will ask for. | absent | → A5 · skip also → A5 |
| Phase 2 — Trip · destination known, so readiness becomes computable | ||||||
| A5 | Destination | Step (funnel root) | Search, recents, saved places. Required range shown per destination before she commits. | Trip distance → required = max(trip × 1.5, 2 km). | readiness · battery | → A6 · browse fallback → A6 on the 2 km floor |
| A6 | Map + list | Step | Only bikes that clear this trip. Range in km on each pin, never a battery percentage. The list mirrors the map exactly. | Cleared set, and the reason every hidden bike was hidden. | readiness | → A7 · zero cleared → F1 · edit → A5 |
| A7 | Preview + hold | Step | "24 km on this bike · your trip needs 9.3 km". Marginal notice if it clears by a thin margin. 5-minute hold. | Held bike ID, countdown, marginal flag. | battery · surfaced when marginal | → A8 · F6 expiry · F7 destination change |
| Phase 3 — Start · money authorised, bike not yet open | ||||||
| A8 | Pay at the bike | Step | Saved method or entry if skipped at A4. Rates stated. "We place a hold now and charge the actual fare when you finish." | Authorisation — a hold, not a capture. | absent | → A9 · declined → F8 |
| A9 | Unlock | Step | Scanner, with manual bike-ID entry at equal weight beside it. "Unlocking…" → "Unlocked. Ride started." | Live readiness re-check. The fare timer does not start until the lock confirms open. | readiness · live re-check | → A10 · F2 no open ping · F10 taken · F9 wrong bike (silent) |
| Phase 4 — Ride · live, and owned by the server | ||||||
| A10 | Live ride | Step | Running fare always visible, live range, ETA. One exit: "Report an issue". | Telemetry recording — GPS trace, lock events, range curve. | battery · billing | → A11 · F3 · F12 |
| A10p | Paused | State of A10 | "₹0.50/min · Yulu #4471 held for you." Lower rate, stated before she commits. | Pause clock. Auto-prompt at 20 min, auto-end at 40. | battery | → A10 resume · F13 at the cap |
| A10r | Report an issue | State of A10 | Four fixed options. She describes nothing. | Full telemetry attached automatically — this is the billing agent's evidence, gathered without an interrogation. | billing | → A10 |
| Phase 5 — Close · bike returned, fare settles | ||||||
| A11 | Park + lock | Step | Position against the zone boundary. "Confirming the lock…" — the app does not take her word for it. | Geofence containment, then the close ping. | readiness · verifying close | → A12 · F11 outside zone · F5 no close ping |
| A12 | Receipt | Step | Itemised fare, pause time, adjustment with its reason, skippable rating. | Capture against the authorisation. Billing agent's decision recorded with its evidence. | billing · surfaced when adjusted | → A5 · ambiguous → F4 |
The screen inventory describes what the rider is looking at. This describes what is true, which is a different thing and the one that money depends on. Two transitions in here cannot be made by the app at all: the bike has to agree.
Twelve branch states. Read the last two columns together: every one of them resolves in the rider's favour by default, which isn't generosity — it's the behavioural rule from the product spec ("prefer a false negative over a false positive; never resolve ambiguity in the company's favour") expressed as structure rather than asserted in a document.
| ID | Trigger | Detected by | What the rider gets | Return path | State effect | Cost to the rider |
|---|---|---|---|---|---|---|
| F1 | Destination set; nothing in range clears it | Readiness, by absence | An empty map that names the check out loud — "There are bikes here, but none have the range for 6.2 km." | Notify me · try a shorter trip · search another area | none | None. This is the product working. |
| F2 | QR scans but no lock-open ping within the timeout | Readiness, live re-check | "We haven't started your ride, and your payment hold has been released." Bike drops off the map for everyone. | Retry ×2, then A6 | authorised → voided | ₹0, stated explicitly — the fear here is being billed for a bike she can't use. |
| F3 | Range exhausted before arrival — the prediction was wrong | Battery Decay, missed | Nearest zone distance and walking directions. Adjustment pre-applied and announced here, not deferred to the receipt. | → A11 on foot | riding → closing, pre-flagged | A walk. Fare stopped at battery death, not at ride end. |
| F4 | Gap in telemetry across the disputed window | Billing Resolution → escalate | "There's a gap in the ride data around 2:14 pm, so we're checking it manually. You haven't been charged for the disputed time while we look." | Human review ≤24h; lands in G3 | settled → review | ₹0 pending. The company absorbs the ambiguity, not her. |
| F5 | No lock-close ping within ~10s at ride end | Readiness | Push the lock and retry — then "End ride anyway", which stops the fare immediately and flags the bike for pickup. | → A12 | closing → settled, flagged | ₹0 extra. This is Ananya's original bad ride, redesigned. |
| F6 | 5-minute hold elapses before she reaches the bike | Hold timer | Toast on return to the map: "Your hold expired. The bike is available again." | → A6 | reserved → none | A wasted walk — the reason hold rate is a tracked metric. |
| F7 | Destination edited while a hold is active; the held bike no longer clears | Readiness, recomputed | The sheet explains the removal rather than letting it read as a bug: "2 bikes dropped off — they can't make the longer trip." | → A6 | reserved → none | One re-pick. |
| F8 | Authorisation declined at the bike | Payment service | "That didn't go through. Try another method." Never blames a specific bank — we don't know. | → A8 | none — nothing held | ₹0. Previously entered values persist; no re-typing. |
| F9 | She scans a different bike than the one she held | Client-side ID compare | Nothing. The scan is accepted and the old hold released silently. | → A10 | hold transfers | None — designed to be a non-event. Don't punish a decision that costs nothing. |
| F10 | Two riders scan the same bike within seconds | Lock state — first confirmed unlock wins | The second rider gets F2: hold released, bike gone from the map. | → F2 → A6 | authorised → voided | Seconds. Reuses an existing failure state rather than inventing one. |
| F11 | End attempted outside every valid zone | Polygon containment (deterministic — never a model) | End blocked, direction to the nearest zone, and any out-of-zone fee stated before she commits. | → A11 | stays riding | A walk, or a fee she agreed to in advance — never discovered afterwards. |
| F12 | Phone dies, or the app is force-quit mid-ride | Server — device heartbeat lost | Nothing on the device, because there is no device. She locks the bike and walks; the close ping ends the ride. E6 hands her the receipt when she's back. | → E6 → A12 | riding → settled, server-side | ₹0 extra. Fare stops at last confirmed movement, not at app reconnection. |
| F13 | Pause exceeds the 40-minute cap | Pause timer | Prompt at 20 minutes, auto-end at 40, with a push explaining it. A held bike is a bike removed from supply. | → A12 | paused → settled | Ride ended for her. A supply decision, communicated — not a punishment. |
Six of the twelve branches cost the rider nothing at all, and the remaining six cost her a walk or a re-pick. None of them cost her money she didn't agree to. That is not twelve separate acts of generosity — it falls out of two structural choices made once: the fare only accrues in riding and paused, and it stops on the failure event rather than when the ride formally closes.
Get those two right and the kind interface writes itself. Get them wrong and no amount of apologetic microcopy saves it.
Conditions, not places. Each one changes how several steps behave without ever becoming a screen the rider is stuck on — which is the trap: treat connectivity as a branch and you end up with a dead-end error page at every step of the funnel.
Denied is a degraded state, not a blocked one — and barely degraded at that.
Barely a layer at all, because the alternative was never a fallback.
The phase you lose signal in decides what happens — and the back half is survivable because the bike holds the truth.
Re-auth returns her to the node she was on — never dumped to home, and never mid-ride.
Everything outside the ride funnel. All five hang off A5/A6 only — never reachable from A7 onward, which is Decision 02 enforced structurally rather than by hiding a tab bar. The single exception is the ride screen's "Report an issue", and that returns to the ride rather than leaving it.
| ID | Surface | Reachable from | What it holds, and why it exists here |
|---|---|---|---|
| G1 | Account | A5 · A6 | Phone, first name, delete account. Deliberately thin — nothing was collected that would need managing. |
| G2 | Payment methods | A5 · A6 | Add, replace, remove. Separate from A8 because A8 is a step in a task and this is maintenance; collapsing them is how riders end up editing cards while a hold is live. |
| G3 | Ride history + receipts | A5 · A6 · push | Every past ride, including anything in review from F4 and anything settled while the phone was dead from F12. This is the landing place for every asynchronous outcome in the system — without it, F4 and F12 have nowhere to resolve. |
| G4 | Saved places | A5 | Home and Work. Text labels, never icon-only. Feeds the destination-first root so the commute case is two taps. |
| G5 | Help | A5 · A6 · G3 | Report a past ride, contact. Notably not reachable mid-ride: during a ride the relevant control is "Report an issue", which attaches telemetry automatically instead of asking her to explain. |
An IA is defined as much by its omissions, and each of these was a live option that got rejected for a reason.
Hiding bikes is a strong move with an obvious failure mode: an empty map can read as a broken app rather than a system exercising care. Absence is paid for in two places — F1 naming the check out loud, and A6 narrating removals when the destination changes — but whether that reads as diligence to a first-time rider is untested, and she is exactly the person with no prior trust to draw on.
Hide rate and F1 frequency are the two numbers that would settle it. They're instrumented for that reason, and they're the dial that would move the 1.5× multiplier if it turned out I'd set it too conservatively.