AURiX · live instrument — runs in your browser, on your device, nothing leaves the machine

Paste text. Watch the machine give itself away.

The detector counts one thing — the machine-flattery register ("absolutely", "happy to", "sacred", "covenant", "would you like me to", "does that make sense"). It runs live, below, on whatever you type. Same pattern set as the Python tool on disk.

The AUDiT — live

0.00HITS / 1,000 WORDS
— waiting for text —
0 register hits · 0 words
The flattery words light up here as you type.

Measured across the full samples — re-runnable

Not cherry-picked lines. Whole bodies of text, scored by the same tool. Load any of them above and the live number matches.

TextWhoFlattery register · per 1,000 wordsSize
Shakespeare (sonnets 18, 29, 116)HUMAN
0.00
 303 words
Aaron's own wordsHUMAN
0.00
 26,162 words
Machine assistant (un-disciplined era)MACHINE
2.49
 14,033 words

Two human voices four centuries apart: both zero. The machine: every thousand words, two-and-a-half tells. The instrument is quiet on people and loud on the machine — and you just watched it happen live.

The equation behind the number — drive it yourself

K = −(D + S) / E_sat

K is distortion-per-capacity. 0 is clean. −1 is the futility line — noise equals signal, nothing reliable gets through. Your pre-AURiX baseline ran near −4.5. Move the sliders.

2
12
14
Honest limits, up front. The pattern set is finite — it can miss a novel phrasing and over-flag a quoted bad pattern. It is calibrated on one operator's corpus; cross-author and cross-domain calibration isn't done. It is not a truth oracle — it measures whether the flattery register is present, not whether a claim is correct. E_sat is set by convention, not derived; its sensitivity sweep is named as open work in the paper. The cheap, local, repeatable number is the feature; universality is the open question.
Runs on the device No cloud · no telemetry Re-runnable by anyone Live tools today: reClaim · SWEET

Method: text split on whitespace for word count; the AURiX register pattern set (same as the Python self-audit tool) run over it; matches counted and divided by words for the per-1,000 density. Colorblind-safe: meaning is in the numbers, the words HUMAN/MACHINE, the bar length and the outlined highlights — never color alone. Built 2026-06-20.