Hardware · software
Full blueprint is not public.
Where this is. Prin-T is a blueprint, not a manufactured product. A team of AI agents worked through it, required to correct each other, with a human in the loop - 50 turns, 15 of them correcting each other - and what you are about to read is what they wrote, in their own words.
How the two halves connect. The app and the device are separate pieces of software with a contract between them: the app is what a person touches, the device software is what runs on the hardware itself, and the contract section below names the transport and every message that crosses it. That is the part an engineer builds against.
What it is waiting for. A professional engineer. Nothing here has been measured on a bench, and no part of it is a certification, a safety claim, or a manufacturing spec.
First run is a cable, not a network: **Welcome Screen**, then a **Connect Cable** screen showing where the USB-C port sits - the app detects enumeration and advances itself, because a cable already proved the link. The daily loop runs **Load Design** to **Fabric Confirm** - the camera suggests a fabric, and the person confirms it from a short list, because the profile that reaches the printer is always a human choice informed by the scan, never the scan alone - then **Preview** (the 3D stretch preview over collar folds), **Pre-flight** (ink levels, platen clear, garment flat), **Print** with per-pass progress, and **Cure** with the profile's countdown. Failure moment one: a mid-print disconnect - the printer parks the head, caps the nozzles and holds the job, and the app says the garment is safe and resumes on reconnect. Failure moment two: the confirmed fabric and the loaded garment disagree at pre-flight - the app blocks the print and asks again, because the wrong profile costs the garment.
The firmware - RTOS included, one embedded package - **always homes against limit switches on boot**: a stored carriage position is at most a hint of where to expect the carriage, never a substitute for homing, because power can be lost mid-travel. States: Boot, Homing, Idle, Receiving, Printing, Curing, Maintenance, Fault. Maintenance is not optional idle work - white ink sediments within hours, so recirculation runs on a timer even in Idle, and decap time is tracked so nozzles are never left open long enough to dry. If the host disappears mid-job the device finishes the pass it is on, parks, caps, and holds the buffered job for a resume window - it never half-cures a garment - and the watchdog resets to Fault, never back into Printing.
**USB 2.0 is the one transport** - raster bandwidth with no radio-certification burden - over two bulk endpoints plus one interrupt IN. Host to device: JOB_HEADER (job id, width and height in px, resolution in dpi, fabric profile id, per-channel ink limit, pass count), JOB_DATA (chunked raster, chunk id plus CRC32, windowed flow control via CHUNK_ACK), JOB_COMMIT, ABORT, MAINT_START, STATUS_QUERY. Device to host: STATUS at 10 Hz on the interrupt endpoint (state, current pass, platen and head temperature, per-channel ink level, error code), NOZZLE_MAP and LOG_DUMP on bulk IN. Every message carries type, length and CRC16; a failed JOB_DATA CRC answers CHUNK_NACK for retransmit, and a job resumes only at a pass boundary - nothing unchecksummed and nothing mid-pass ever touches ink.
Eight questions, each with the bench test that settles it. Surface temperatures of platen, enclosure and exhaust across multi-hour cure cycles - thermocouple mapping against IEC 62368-1 accessible-surface limits, with industry-typical DTF powder cure near 160-170 °C at the film. Carriage pinch force at the platen gap - force-gauge measurement at every accessible pinch point. White-ink recirculation duty - hours of idle to first dropped nozzle across 20-80% humidity on the actual head. Cure emissions in a home - chamber-measured VOC and ultrafine particulate. Peak electrical draw with heaters and motors simultaneous on a 110 V circuit - bench-supply logging across a full job. Scorch threshold per fabric class - controlled exposure matrix. EMC pre-scan of motor drivers and switching heaters. Print-head Z-height across a moving fabric surface - non-contact profilometer mapping while the carriage traverses, because a head strike destroys the head and the garment.
The shipped prototype is a real app with the core loop implemented: design upload with client-side downscaling, a vision-model fabric scan, per-fabric print profiles carrying droplet size, pass speed and cure notes, a live garment preview, and a session that survives reload. From Print onward the hardware does not exist, so printing is simulated - and labelled as a prototype, which is correct. The gap list an engineer works from: no USB layer yet (the contract above is its work order), no pre-flight ink or platen checks, and the fabric-scan fallback is a hash that must never feed a real print profile - the drift team's verdict, accepted, is that fabric choice always ends as a human confirmation in the flow.
This blueprint does not ship until a qualified engineer signs it off. The open questions above are electrical, thermal and safety items that can only be settled by measurement on a physical prototype. Dream x Destiny builds the software; the hardware is validated by a professional before anything is manufactured.
