← Red Kit THE RESEARCH, IN REAL-WORLD UNITS · AUG 2026

What does it take to teach a phone first aid?

Everything below is computed from our own logs and ledgers — the same ones our published results come from. Rounded, but never invented.

The bookshelf we trained on

3,781 novels

The training text this lab has built, filtered and frozen across every experiment — about 2.04 billion characters of emergency medicine, survival doctrine and hand-written examples. As paperbacks, that's a bookcase forty shelves long.

2,041,500,000 chars ÷ 6 chars/word ÷ 90,000 words/novel · counted from every frozen dataset on disk

The paper stack

113 meters

Print it all — 1.13 million pages — and the stack stands taller than the Statue of Liberty, torch included. A 35-story building of paper, distilled into a model that fits beside your camera roll.

1.13M pages × 0.1 mm/sheet · Statue of Liberty: 93 m

Books our machines wrote back

230 novels

To find out what the models actually learned, we make them answer — a lot. Their practice answers and teaching drafts total ~21 million words. Nobody reads them all; that's the next number's job.

124M generated chars, counted from every response and teacher file on disk

The strictest grader alive

11,762 graded answers

Every answer is scored by an AI judge against a written checklist for that exact emergency — the judge's grading notes alone run ~50 million words. A human instructor at three minutes per answer would need 3.5 months of full workdays. Ours does it in an afternoon, and the hard calls get a second, stricter judge.

counted from every verdict file on disk · 66.1M judge tokens · human est: 11,762 × 3 min ÷ 8h days

45 answers per emergency

346 scenarios × 45

Our test is 346 frozen emergency scenarios — never trained on, never edited to make a model look good. Across every model and every experiment, each one has now been answered about 45 different ways. We keep the failures on file with the wins.

15,522 evaluated answers ÷ 346 scenarios

An encyclopedia habit

37 years of reading

Our open research line indexed all of English Wikipedia — 7.6 million articles, sliced into 53 million searchable passages. Reading it yourself at a brisk 250 words a minute, around the clock, no sleep: see you in 2063.

≈4.9B words ÷ 250 wpm · corpus figures as published in our lab brief

The arithmetic

428 humanity-years

Our rented GPU time adds up to roughly 1020 calculations. If every person on Earth did one calculation per second — eight billion pencils scratching in unison — it would take about 428 years to match what we bought for the price of a nice dinner.

≈30 rented GPU-hours × ~10¹⁵ ops/s ÷ 8B people · GPU hours from our spend ledgers

The electric bill confession

~2,000 phone charges

All that compute drew roughly 25 kWh — about what your phone uses in five and a half years of nightly charging. Small on purpose: this lab's method is doing more with less, because the product has to run on the least hardware of all — yours.

≈25 kWh ÷ 12.5 Wh per phone charge

And it all fits in your pocket

2.8 GB

The point of the bookcase, the paper tower and the humanity-years: a model smaller than an hour of 4K video that answers in the first 30 seconds — with no signal at all. The mountain goes in, the pocketknife comes out.

shipping model file size, quantized for iPhone · see the measurements
Counted, not conjured: sources are our committed eval files, spend ledgers and the lab brief. Rounding is generous, direction is honest.
Red Kit · Red · Blue · Hardware