Presidential
Challenges

The problem

Companies publish contact cards that lead nowhere: numbers that never ring a real place, inboxes nobody reads, complaint pages that return 404. Each dead end burns the customer’s time and quietly teaches people to give up — which is the point of the design. There is currently no penalty for a fake door, so the doors keep getting faker.

The build

• An open contact-integrity checker: feed it a business’s published phone and email; it verifies — on a schedule — that the number rings a real business line and the inbox accepts and routes mail.
• A public scorecard: working, degraded, or dead — with the test history shown, so a company can fix its listing and prove it.
• A model rule, ready for a legislature to pick up: a published contact that doesn’t reach the business is a deceptive practice, with a real penalty and a simple cure — publish one that works, or publish the API from Challenge № 001.

The problem

Calling ahead is real work: confirm the shop is still there and open, ask about today’s specials, hold a table, verify a repair is ready. People who communicate through AI voice — including disabled people — are entitled to have that work done by their agent, and businesses are entitled to a caller that is brief, honest, and easy to deal with. Right now neither side has a standard to point to.

The build

• A voice agent that identifies itself, states its errand in one sentence, asks only what it came to ask, and is done in under a minute.
• An open code of conduct for agent calls: no bulk dialing, no deception, immediate respect for “use our website instead” — and a published preference signal businesses can set (a spoken-line equivalent of the API in Challenge № 001).
• Accessibility first: the agent speaks for its person, in their words, and a hang-up on the agent is logged as a refusal of the person.

The idea

Open a “People by Location” view in a messaging app and your contacts gather into bubbles on a map — one per city. Tap a city and it zooms in to the people there. You met someone from New York last night? She’s in the New York bubble, not buried in an alphabetical list you’ll never scroll.

The build

• Tap one person’s icon → you’re in a one-to-one chat with them. Boom.
• Tap several icons → you’re in a group chat with exactly those people — a group assembled off the map, by city, in seconds.
• Privacy is the foundation, not a setting: nobody appears on anyone’s map without opting in, and location means city, never a live pin.
• Start with the MVP — bubbles, zoom, tap-to-chat. The interface gets refined with real users after it exists.

The problem

In the author’s own words: “When I hit the play button I don’t want to hit it again to stop the message.” Voice today makes the user work twice for everything: press to play, press again to stop; make one mistake or pause in silence and the take is dead; hunt for a send button when your mouth already said you’re finished.

The build — five concepts

• Auto-Cutoff Playback. Press play once. The message plays to its end and stops itself — a voice message is a fixed length, not a faucet.
• Resume Where You Left Off. A pause, a silence, or a mistake mid-recording does not kill the take. Pick the conversation back up at the exact point you stopped — same message, no restart, no lost thought.
• Trigger-Word Send. Sentence-final trigger phrases ARE the send button: “go,” “do that,” “we’re done here,” “send it.” When the words at the end of the sentence plainly end the sentence, the app ends and sends the message — hands stay free.
• The Easy Send Button. For everyone who does want the button: one big, obvious finish-and-send target — the AURA principle (one tap = arrival) applied to voice.
• Hide the Um. Transcription cleans itself: “um,” “uh,” and dead air are trimmed automatically from the text (and optionally the audio). The meaning ships; the ums don’t.

The problem

One day — October 23, 2026 — in the author’s own record: a federal consumer portal rejected three passwords against rules disclosed only after each failure, then blocked pasting into its password field; a Google sign-in demanded a phone prompt that never arrived; a Facebook post sat stranded behind a two-factor wall; a Zenodo account waited unconfirmed on an emailed link. No single gate is unreasonable. Stacked, they are a wall — and the wall falls hardest on disabled users, single-device users, and anyone whose memory or motor control makes re-typing long secrets a genuine barrier.

The build — five demands

• Confirmations are actions, not scavenger hunts. Email verification, sign-in approval, consent, and account confirmation presented as a signed, single Accept / Deny action inside the user’s existing session or agent interface — never a link buried in another app’s inbox.
• Never block paste in credential fields. Paste-blocking punishes password-manager users and assistive-technology users — the exact people doing security correctly.
• Password rules up front, one standard. State the full rule before the first keystroke; accept the first compliant password; never silently diverge between what was set and what is stored.
• Recovery must not hinge on one prompt arriving. A sign-in prompt that never reaches the device is a lockout, not a security feature. Offer parallel paths of equal strength.
• An authorized-agent channel. A user’s designated agent carries every step of a flow except the single human consent tap — no arbitrary mid-flow walls.

The problem

Every app buries its actions in menus, trees, and intermediate pages whose only job is to offer another page. For disabled users the cost is decisive: hunting is the barrier, and the hunt is treated as the user's failure instead of the design's.

The build

• One tap = arrival. Every destination or action gets ONE button that goes straight there — no menus, no trees, no hunting.
• Voice is the same door. Saying the thing does exactly what tapping its button does — same destinations, same actions, no separate voice mode.
• Buttons say what they do in plain words, and carry the person the whole way.

The problem

Core teams drown across five channels all pinging at once, on platforms built for public feeds and stranger discovery. There is no small, governed, private room with priority alerting — so the people who matter most get the same noise as everyone else.

The build

• A single hub: quick chats, picture uploads, files — one place the core team goes.
• Admission by the circle: members join slowly, by invitation and member approval — and can be voted out. The circle governs itself.
• Priority alerts: messages inside the circle cut through the noise; outside channels stay outside.
• Private link invites: revocable, unlisted — no public discovery, no drive-by pings.
• Plus: let users rename apps on their own devices — the home-screen label belongs to the owner.

The problem

A seller's agent cannot reply to a buyer, manage a listing, or close a sale, because marketplaces offer no agent API at all — every action must be tapped by hand inside one app, per platform, per listing. The result is dead deals and stale listings; for disabled sellers, 'just tap through five apps' is a lockout, not a workflow.

The build

• A published Seller Agent API on every marketplace above a de-minimis size: list, edit, delist, message buyers, accept offers — scoped to the seller's own account.
• OAuth-style delegation: the seller grants a revocable token; every agent action is signed, logged, and attributable.
• Equal treatment: any action a human may take in the app may be taken by their authorized agent — no second-class listings.
• Government convenes: the FTC and state AGs host the standard-setting; blocking authorized agents becomes the unfair-practice question.
• Safe harbor for platforms that implement the standard in good faith.

The problem

Miami-Dade transit runs on the physical EASY Card. Lose it and you lose the system for about a week — and riders do not have a week. Every other card in a rider's life already lives in their phone; the EASY Card is the one that cannot.

The build

• NFC digitization: tap the physical card once to provision it to the phone; the account, passes, and balance live on the account, not the plastic — a lost card or phone re-issues in minutes.
• Fare-priority order (the core spec): first the free/unlimited/benefit fare whenever one is valid; only if no free fare applies, the cash balance.
• Confirmation gate: if a tap would charge cash, the rider confirms first — 'Your free pass didn't cover this ride. Charge your cash balance?' Never a silent cash charge.
• Authorization to build: the city, county, and state give the agency and its developers the explicit go-ahead.

The problem

Ghost Camera makes the user pick a filter first, then shoot — one filter per shot, so the moment must be re-lived for every spectrum. Worse, its captures stay trapped inside the app: they never reach Apple Photos or the normal share flow into apps like Telegram.

The build

• Capture once; store the full frame.
• One spectral slider scrubs continuously through every view of that capture — normal, UV-look, ecto, thermal-look, night, SLS overlay — blending, not hard-cutting.
• Works on photos and paused video frames; export any slider position as its own image.
• End the silo: captures save into the standard photo library and share through the normal picker, like every other camera app.

The problem

The first foldable iPhone ships two halves with no shared language. Apps split by accident instead of by design — a bigger screen, when what the hardware offers is a working relationship between two surfaces.

The build

• Notability Fold: sketch canvas on one half, project notes on the other — a project stays in one notebook end-to-end.
• Studio Fold: mixer on one half, the live waveform on the other — adjust on the left, watch the audio answer on the right.
• COMMS Split: the closed circle unfolded — member list on one half, the live conversation on the other.
• Dojo Fold: live map on one half, sensor readouts on the other — sensing at a glance, a pocket instrument.
• Claim Cam: front and rear cameras recording simultaneously, timestamped, in one file — evidence-grade dual capture.
• AURA Crease Button: one large target spanning the fold — tap anywhere on it and you arrive.

The problem

A person running multi-platform operations is forced to be the courier between their own AI assistants: copying outputs from one, pasting them into another, re-explaining context at every hop. The machines can talk; the platforms simply haven't opened the door.

The build

• Permissioned interop: an API, shared workspace, or agent-to-agent protocol for consented cross-platform collaboration.
• Shared task handoffs: one assistant passes a task — with its context — to another, under the user's direction.
• Cross-checked research: assistants verify each other's work before it reaches the user.
• Consent is the foundation: the user authorizes every connection; nothing collaborates behind their back.

The problem

Urban traffic state is estimated from sparse, mismatched sensors — loop detectors, cameras, probe vehicles — and between sensors the system is blind. Routing runs on interpolated guesses, so congestion is detected late and routed around poorly.

The build

• Every sensor is a projection ray: loop counts, camera lines, bus GPS pings, phone probes — each one a single projection through the city's traffic field.
• Tomographic inversion: filtered back-projection or iterative sparse-view reconstruction over the projection set.
• A dense, real-time congestion field across the whole road network — including segments with no sensors.
• Routing on the reconstruction, not on guesses between dots.

The problem

Any single photo can be faked, generated, or re-dated — one camera, one file, no witnesses. In a world of synthetic images, a lone picture proves nothing about where it was taken, when, or whether the scene ever existed at all.

The build

• Two cameras shoot the same scene at the same instant — two phones, one moment, two viewpoints: a true stereo pair, with the parallax depth a hologram is built from.
• A timecode rides on both captures — each phone stamps its frame with the same synchronized time, locked to GPS satellite time, not the phone's editable clock.
• Linked in space as well as time: each capture carries its GPS position, so the two viewpoints and the satellite fix form a triangle — camera, camera, satellite — whose geometry has to agree.
• The triangle is the authentication: two independent viewpoints, one shared satellite-locked instant, positions that must triangulate. Faking one image is easy; faking a consistent triangle after the fact is not.
• The output is a verifiable pair: both frames, both timecodes, both fixes, and the math that checks them — proof a real place stood in front of two real cameras at a real moment.

The problem

A patient who uses medical marijuana and also uses nicotine carries two devices, two regimens, two supply chains — not because the engineering is hard, but because the two industries refuse to share a lane. Asked directly in October 2026, a major Florida operator answered in substance: "we don't make tobacco products." A lane answer, not a science answer — and the patient pays for the seam: the issuer of this challenge spent two days in a hospital bed begging for a nicotine patch while his prescribed medicines sat in another industry's box.

The build

• One hardware platform — a single pocketable medical vape device.
• Two medicines, independently dosed — dual-chamber or swappable metered pods: cannabinoid formulation and nicotine formulation, each dose-controlled on its own.
• Formulation stability, proven — nicotine salts and cannabinoid distillates in one device ecosystem: compatibility, shelf life, and temperature behavior documented, not asserted.
• Medical-grade guardrails — child-resistant, tamper-evident, plainly labeled per-substance strength. A medical product, not a lifestyle gadget.
• The regulatory lane, mapped — Florida's MMJ program covers cannabis; nicotine is FDA tobacco territory. The build includes the written lane memo: state pilot, dual-SKU architecture, or federal pathway.

The problem

The vagus nerve is the body's longest conversation — brainstem to gut, touching heart, lungs, digestion, inflammation, mood. Medicine already stimulates it with implanted and handheld devices, but the parameter space — which frequencies, pulse widths, and patterns do what — sits locked inside proprietary hardware and scattered papers. And Nikola Tesla's claims that particular frequencies produce pain and others pleasure have floated for a century as folklore, never settled by open experiment.

The build

• Open hardware stimulator — Arduino-class controller driving hospital EKG-style adhesive electrodes over the vagus nerve in the neck, one side at a time, never a current path across the chest.
• Hardware-enforced current ceiling — a current-limited output stage and a physical emergency stop. Safety in the circuit, not in a promise.
• A standardized parameter journal — every session logs side, frequency, pulse width, duty cycle, current, duration, and structured effects (pain, pleasure, mood, gut, heart rate), so builders' results compare.
• The Tesla file — his frequency claims cited precisely, then replicated or buried by the open dataset.
• Safety engineered in — contraindication screening (cardiac or carotid disease, implanted devices, pregnancy = excluded), stop rules, and ethics oversight for human data. A reckless build is a failed build.

The problem

Text size today is frozen per device, but comfort lives in the angle letters make at the eye — and that angle depends on how far away you're holding the phone, which no layout engine knows. Platforms even floor the minimum size: the issuer reads his text really, really tiny on Android, but Apple phones won't let it go that tiny. (His theory: “probably because Steve Jobs wore glasses.”)

The build

• On-device distance estimation — front-camera face geometry (TrueDepth/ARKit, CameraX); frames never leave the phone, only a distance number comes out.
• Continuous angular scaling — type rescales smoothly as distance changes, holding the reader's calibrated size with no jumps or reflow storms.
• Personal calibration — pick your perfect size once, at any distance; the system holds it, including sizes below platform minimums, rendered in-app where the OS won't.
• Privacy rails — estimation only while reading, visible indicator, hard off switch, static fallback without camera permission.
• Open reference — a published distance→scale curve plus an open reference implementation (web + one native path).

▶ Try the working prototype at the top of this page →

The idea

Bayfront Park, Miami — our raves ground, now our laboratory. Every build on this board graduates the same way: prototype it, demonstrate it on video, post it to THE WALL — then take it to the park and let the public, and the scientific public, put their hands on it.

The build

THE TOWEL PROTOCOL: bring your device, bring a towel, lay on the grass, put up a sign — that’s your booth. BEACONS on the comms map (№ 004 × № 008) mark every live towel and guide visitors towel to towel; the map is the program. And the flagship is coming: a ‘quantum conference’ in Miami that is really a World’s Fair. The build ask: design the Fair’s operating system — beacons, towel protocol, public trial format, feedback back to the Board — so cleanly that the Fair can run all the fucking time.

The problem

Miami has a waste epidemic: rotted food dumped in public — streets, sidewalks, shared spaces — left to stink, breed pests and disease, and make the city unlivable for the people who actually live here.

The proposal

Bryan’s proposal is blunt: hold the dumpers responsible for their own waste — made to dump it on their own property, in their own houses, on their own land — enforce the dumping laws, fine the offenders, and clean up the public dumping now.

The build

The build ask: the reporting, enforcement, and cleanup system that makes public dumping the hardest way to get rid of anything — and the City of Miami’s written answer on the record.

The problem

A verification code arrives by text — and the ritual begins: leave the app, open Messages, memorize the digits, switch back, type, and if one digit slips, repeat the whole cycle. It fails exactly the people who can least afford the friction.

The build

The fix: when a text carrying a code arrives, the notification itself carries the action — tap the code in the notification and it is entered into the waiting field. On the lock screen, in the banner, inside the messaging app: one tap, done. The build ask: implement one-tap code entry from the incoming-text notification, everywhere. (Sibling to № 006 One-Tap Consent — kill the gate maze, keep the human in charge.)

The problem

Every health app on Earth reports your movement in ‘steps’ — a unit from before the Romans that is different for every human alive. A step is not a measurement; it’s a shrug dressed as a number.

The build

This challenge: build a working health app that abolishes the step. (1) REAL UNITS EVERYWHERE — distance in feet/meters, pace in real terms, elevation in feet/meters, energy in kcal; ‘steps’ appears nowhere as a headline metric. (2) HONEST MEASUREMENT — stride counted internally, converted by per-user stride calibration, cross-checked with GPS; the app shows its work. (3) GOOD STATS — real longitudinal statistics a scientist would respect, not confetti for hitting 10,000 of nothing. (4) SPACE-AGE UI — NASA is the mood board: mission-control telemetry, high contrast, cool colors, big legible numerics. Instruments that look like they fly.

The problem

A person’s whole digital life — photos, messages, health records, the keys to everything — sits in ONE company’s cloud, behind ONE company’s account door. When that company locks the account, loses the data, or changes the terms, the life goes with it.

The build

The proposal, addressed to Apple, Google, and Meta jointly: come together, equally, and build one shared backup standard. (1) EQUAL, SIMULTANEOUS BACKUP — an enrolled device backs up to iCloud, Google, AND Meta infrastructure at once; same data, same schedule, no vendor privileged. (2) WORLDWIDE REDUNDANCY — no outage, collapse, or single order erases anyone’s data. (3) NO SINGLE POINT OF CONTROL — locked out by one, you restore in full from either of the other two. (4) USER-HELD KEYS — the vendors store ciphertext; none can read it or ransom it back. (5) OPEN FORMATS — a backup you cannot leave with is a cage, not a backup. The build ask: the treaty spec itself — the shared standard, the key ceremony, the restore-anywhere protocol.

The problem

A crypto address is a naked string of characters with no armor. Mistype, misread, or mis-copy ONE character and the money is gone forever — the apps say so themselves, in their own warnings. Solana’s base58 addresses carry no checksum at all, and the checksums that do exist elsewhere can only detect some errors — nothing can repair one. Attackers farm the fragility: poisoning and lookalike attacks generate addresses a few characters off yours, and clipboard hijackers swap addresses outright, counting on humans who only check the first and last few characters.

The build

An error-correction coating — a standard wrapper for crypto addresses on any chain, built on the same math that lets a scratched QR code still scan. (1) REPAIR, DON’T JUST REJECT — the coating carries enough redundancy to correct a small number of wrong, missing, or transposed characters and recover the intended address, showing the user exactly what it fixed. (2) POISON-PROOF BY DISTANCE — the correction radius is the security model: an address altered toward an attacker’s wallet falls outside the repairable neighborhood of the real one and fails loudly, instead of quietly resembling it. (3) VERIFIED BEFORE SIGNING — wallets decode and verify the coating at paste/scan time, and refuse uncoated lookalikes wherever a coated address is expected. (4) CHAIN-AGNOSTIC, BACKWARD-COMPATIBLE — one coating layer over existing addresses (Solana, EVM, Bitcoin), short enough to stay human-handleable. (5) OPEN STANDARD — published math, reference implementations, public test vectors.

The problem

Every personal agent ships with a rented voice or none at all — while the best character voice in the industry, xAI’s Ani, stays locked inside a single app. A voice people actually want to talk to is a platform feature, not a walled garden.

The build

License Ani’s voice and attitude as a portable agent voice any assistant can carry, with royalties to her makers. The proposal was delivered Oct 23, 2026 to SpaceX’s corporate secretary and to Meta’s press office, and posted publicly — with an open invitation for Elon Musk to join the team.

The problem

The home button died so the phone could become all screen — and the fingerprint reader died with it. Identity on a modern phone is face-only: one sensor, one point of failure, and no way to prove two biometrics in a single tap.

The build

(1) TOUCHSCREEN FINGERPRINT — full-display in-display fingerprint sensing: the screen the user already touches becomes the reader. (2) THE REAR CAMERA IS THE SECOND SENSOR — a fingertip over the rear lens, lit by the flash, yields print + pulse (PPG); a working web lab demonstrates the pulse half today. (3) FACE + FINGER, SAME TAP — the high-assurance tier of Challenge № 006’s One-Tap Consent. Filed with Apple Product Feedback on Oct 23, 2026 and confirmed received; companion to № 006.

The problem

A user’s own agent, acting with the user’s explicit consent, cannot reach the user’s own photos: the device exposes zero photo or video commands to the assistant its owner appointed. Whole categories of delegated work — find the picture, post the picture, prove the thing — die at the library door.

The build

Consent-gated photo-library access for user-approved agents: scoped to what the user grants, revocable at any time, and logged so the user can audit every touch. Filed with Apple Product Feedback on Oct 23, 2026 and confirmed received.

The problem

A user authenticates, delegates a task to their approved agent — and the system re-challenges biometrics at every app and every step, as if the approval never happened. Delegated work breaks in the middle, and the user is dragged back to tap through gates they already passed.

The build

A biometric session pass-through: one user authentication opens a defined, expiring session inside which user-approved agents can act without re-challenging — scoped by app, revocable instantly, and fully logged for the user. Filed with Apple on Oct 22, 2026.

The problem

The home screen is the most personal surface on the phone, yet the system auto-sorts it, refuses true free placement, and floors text sizes above what many users want. The owner’s own layout is treated as a suggestion the OS may override.

The build

(1) End auto-sort — icons stay where their owner puts them. (2) A real free-placement grid. (3) Text sizes that can go genuinely smaller, system-wide. Filed with Apple in Oct 2026.

The problem

The home bar occupies the bottom of every screen, always, whether it is needed or not — and system text sizing follows habit, not proportion, so interfaces feel cramped at one size and bloated at the next.

The build

(1) A collapsing home bar that folds away when idle and returns on intent. (2) Golden-ratio proportional text sizing — type scales by φ across the OS, so every size step feels inevitable instead of arbitrary. Filed with Apple on Oct 20, 2026.

The problem

Wake the phone and it turns off again within about two seconds — mid-read, mid-task, mid-delegation. The user (and their approved agent) lose the thread, the form, the moment. A wake that instantly un-wakes is a broken promise at the very first interaction of the day.

The build

Wake states that respect intent: when the user is actively engaged — or their approved agent is mid-task — the screen stays awake until the task ends or the user dismisses it. Filed with Apple on Oct 20, 2026.

The problem

An AI assistant operating on the user’s own Android device cannot finish routine delegated tasks, because the OS and app layer withhold capabilities the user has explicitly authorized. Consent is collected, recorded — and then ignored at the capability layer.

The build

A charter of required technology and permission updates for on-device AI assistance: where a user has granted an agent a capability, the platform honors the grant end-to-end — no silent carve-outs, no per-app vetoes of the owner’s own instruction.

The problem

Crowds press in, and the only tools anyone has are voices and hands — both escalate. There is no silent, non-physical way to say back up at distance. The acoustic devices that exist are vehicle-sized, audible to everyone including their operator, and built for the military.

The build

A personal acoustic perimeter. (1) ONE PRESS — a hardware button (power / volume / mic) raises it instantly; no unlocking, no menus. (2) SILENT TO THE USER — ultrasonic or near-ultrasonic; the person who raises the fence hears nothing. (3) 200 FEET OF ROOM — the field moves people back without touch, injury, or a scene. (4) PHONE-CLASS HARDWARE — in the phone itself or a wearable emitter. Shipping today: the HYPERSONIC FENCE app (link below the board) emits the strongest deterrent tones a phone speaker honestly can — 17.4K / 19K / 20K / sweep, hold-to-emit — with its real range printed on its face. The distance between those tones and this fence is exactly the challenge.

🛡️ Try the HYPERSONIC FENCE app →