AVANTÉ Accuracy Report · Protocol v1.0

What our numbers actually are — measured, not marketed.

Thresholds below were pre-registered before any data was collected. We publish results whether we pass or miss; an honest miss with a correction plan builds more trust than silence.

Headline

Data collection is underway. Headline numbers publish here — with sample size and 95% confidence intervals — once each track reaches its pre-registered cohort. Nothing is claimed before it is measured.

Track 1 — Measurement accuracy

Collection in progress

Your Digital Twin’s measurements vs. a professional tailor’s tape, on your own phone in home conditions. Target cohort 120 bodies · 0 measurements collected · 0 participants.

Within ±2.0 cm of tailor tape

Pre-registered target: ≥85% · Observed: —

Collection in progress

Within ±1.5 cm of tailor tape

Pre-registered target: ≥65% · Observed: —

Collection in progress

High-confidence labels within ±2.0 cm (calibration)

Pre-registered target: ≥92% · Observed: —

Collection in progress

Track 2 — Fit verdict vs. reality

Collection in progress

This is observational beta data, not a randomized controlled trial. Figures describe outcomes for users who followed vs. ignored the recommendation; we never claim a guaranteed reduction in returns.

Fit-driven return rate (followed our size)

Pre-registered target: ≤12% (≤50% of 24% baseline) · Observed: —

Collection in progress

Fit Score discrimination (≥85 vs <70 keep rate)

Pre-registered target: ≥20 pts · Observed: —

Collection in progress

Followed our size

No outcomes recorded yet

Ordered their usual (comparison arm)

No outcomes recorded yet

Track 3 — Render quality panel

Collection in progress

Blind pairwise ratings by independent panelists. Target 200 raters · 0 ratings collected. Parity is not required to win the category — our renders carry fit truth theirs don’t — and we say so.

Win-or-tie vs Google try-on

Pre-registered target: ≥50% · Observed: —

Collection in progress

Win-or-tie vs Doji

Pre-registered target: ≥40% · Observed: —

Collection in progress

Identity preservation (Likert 1–5)

Pre-registered target: ≥4.0 · Observed: —

Collection in progress

What we’re not yet good at

The differentiator no competitor will copy. Every measurement miss is categorized by cause; loose clothing, poor lighting, and phone angle degrade accuracy, and we flag it in-app when confidence drops.

Failure-mode breakdown publishes once Track 1 data is collected.

Methodology & privacy

Full cohort composition, the standardized tailor protocol, statistical methods (mean absolute error with bootstrap 95% confidence intervals; Wilson intervals for proportions), and the pre-registered thresholds shown above are documented in the protocol (AVANTÉ Beta Validation Protocol v1.0).

Written informed consent covers body measurements and photos; data is stored under AVANTÉ's encryption and deletion architecture; the report publishes only aggregates; participants may withdraw and be deleted at any time; no participant images are published without a separate explicit release.

Report generated July 29, 2026. Refreshed as data accrues.