The numbers, warts and all.

Every measured claim we make, in one place — the flattering and the unflattering together. Two rules govern this page: population results are never merged with single-user results, and cold start is never merged with personalized. The short story is on the about page; this is the long one.

The technology, measured

The short form. The unabridged record — every number with every caveat, as we state it to ourselves — is public at /llms.txt, and the raw data is available under NDA.

• The recognizer. A compact quantized convolutional neural network over accelerometer and gyroscope, running on a Nordic nRF54LM20B with its on-chip neural accelerator. 25–28 ms per letter; zero accuracy loss from float to INT8. We don’t publish the architecture or the model size.
• Letters, cold (population). 67.5% per-character top-1 for a first-time user, across a 300+-subject paid program. Attune adds 8–10 points at population level and halves false rejections; on the physical device it has lifted a known hand into the 90s (single user). A practiced single writer through the full pipeline lands 97%+ of words exactly.
• Words, cold (population). 85% of a first-time user’s words come out exactly right — 317 held-out paid writers, leave-one-subject-out, phrases assembled from their real recorded letters. The median writer is at 95.5%; one in ten is under 66%. The spread is the finding: most people need no personalization at all.
• The re-test (August 2026). 28 people wrote twice, on different days. The twelve the model found hard went 82% → 93.5% of words exact after Attune — all twelve improved. The sixteen it already read well moved 95.6% → 96.5% — real but small, and four came out slightly worse. Attune is a fix for the tail, not a general upgrade — and after it, nobody was left above 20% word error. One narrow cohort (US men, 50–80); the full accounting, corrections included, is in the unabridged record.
• Speed. About 15 words a minute in letter mode in regular use — one letter at a time, on the wand itself, where the letter arrives as a keystroke while you write. The 5 standardized words per minute we published until 31 August is the same mode measured on the study rig: a prompt screen between every letter, a network round-trip on each one, and a pause to read the phone. About 70% of that clock is not writing — press, release, wait, read, reposition. Both figures are ours and they measure different things. Treat the 5 as the floor: it is what the conservative path measures, it matches the published Georgia Tech air-writing benchmark of 5.43, and the TV D-pad on our own figure is 5–9. On that same rig the founder’s draw-only bound is ~11 and the paid cohort’s median at real rhythm is 3.9; one early fleet user writes close to 12, letter by letter. The 15 is the founder’s own rate on his own hardware, self-timed — nobody but him has yet been timed on the device, and 15 words a minute is about 12 on the standardized ÷5 convention the other figures here use.
• Staccato, in the field. A whole word drawn in one continuous hold: early real-user sessions pool to ≈8.7 standardized words per minute on the shared model, no personalization (the founder’s median ≈12). Small samples, draw-time estimates — and we judge the mode on correct words per minute, not raw speed. That ≈12 is not comparable with the 15 above: it counts drawing only, and at the drawing tier staccato is in fact slightly slower than one letter at a time. What it buys is writing a whole word in one flow, and at 15 words a minute in that lane our recognizer cannot yet read people reliably — which is why staccato is a claim about flow and letter mode is the one carrying a speed number.
• Hardware status, stated plainly (9 October 2026). Five production-design wandBox boards (Rev A) were assembled in September 2026; the first runs the shipping firmware, pairs and types over Bluetooth, and bench validation of the rest — power, clock, charging, radio, the production inertial sensor — continues through October. The one fault found on all five, a USB-C socket the plug cannot reach, is fixed in the next revision, which also carries a user-replaceable cell. Behind the boards: a schematic complete and electrically clean, a closed 27-line bill of materials, a pre-certified radio module, the factory flow bench-passed end to end — over-the-air update included — and manufacturing quoted, not hypothetical. Prototype wands on development hardware are with outside users, updated over the air. The wandRing’s first board is designed and checked, not built; ten are planned for November 2026. Nothing is on sale — the plan for 2027 is a crowdfunded first run of the device, the wandBox, alongside licensing the recognizer to device makers; the ring follows the handheld.
• The ring, first measurements (October 2026; one wearer, the founder). A 3D-printed test ring around an off-the-shelf motion sensor, 891 letters over three days, every test fitted on one day and read on another. Starting from the letter model 340 paid testers trained with their phones, it read 61% of his ring letters before he had written one on it and 90% after about six minutes of his ring writing; the same network started from nothing needed about nineteen minutes. Cut tighter around each stroke — offline, not yet on the device — the figures are 81% cold, 90% after about three minutes and 95% after about nine, against about twenty-five minutes to 90% from nothing. The ring’s goal is 95 words in 100 once it has learned a wearer’s finger; nobody but the founder has worn it yet. The write-up →
• How we measure. Paid testers only, scored leave-one-subject-out: the model reading a writer has never seen that writer’s handwriting. Nobody coached, nobody hand-picked; the results that went against us are published beside the ones that did not. What we lack is a third party re-running it — the university collaboration is the route, and the standing invitation is the better test: bring your own words.
• Against the published literature. The active research line in IMU handwriting is FAU Erlangen-Nürnberg with STABILO, on the OnHW benchmark, writer-independent: 7.37% character error and 15.12% word error reported, and 12.99% word error in its February 2026 paper with subword tokenization (arXiv 2603.16883). Per character that is much better than ours, and we say so. The difference is the task — OnHW is written with a sensor pen on paper, where the nib marks where every stroke starts and stops and the paper steadies the hand; the wand is written in free space, where nothing segments the strokes and nothing steadies the hand. The other leading number, Meta’s sEMG handwriting at 20.9 words per minute, is also written on a surface: Meta’s own documentation says to trace the letters on a table, a leg or the other hand. At the word level the lines are close but not level — 12.99% word error at best on paper, 14.9% for a first-time user writing in the air with us. (Until 9 October 2026 this page said the two lines meet, against the earlier 15.12%.) The protocols and the corpora are not the same and we are not claiming a head-to-head; what would settle it is a third party running both on the same words.
• The gyroscope, tested properly. Dropping the gyroscope and reading acceleration alone costs the letter model 2.09 percentage points (median 1.39), on a full subject-level leave-one-subject-out — one fold per participant, 354 of them — at p = 4 × 10⁻²⁰. On the staccato model the estimate leans the same way and the study cannot resolve it at n = 41; that is the one place this measurement says we do not know rather than giving a number.

Research & data

• The corpus: 490+ paid, consented participants across 28 studies. The largest set is the Atlas — 349 adults aged 18–78, 52,191 letters with age, sex and handedness. The free games added ~2,000 letters from nine countries; adding them changed accuracy by nothing we could measure, twice.
• Independent evaluation: US university researchers, IRB-approved — a nine-participant pilot on our hardware in spring 2026, a larger study on the latest hardware running fall 2026, peer-reviewed publications in preparation, data open-sourcing planned.
• The Air-Writing Atlas (August 2026): the first public analysis of the corpus. Men and women write the same in the air; older hands don’t shake more, they slow down after 50. Pre-registered, replicated on both phone platforms. The write-up · the release.

Diligence, welcomed

Serious parties can go past this page. Available on request: hands-on device demos — bring your own words, test the runtime offline, put new or left-handed writers on it — and, under NDA, raw anonymized study data, protocols, per-user accuracy distributions and confusion matrices. It is all there to be picked over. Write to john.usher@the-wand.ai.