Skip to content
Titan Slayer

Numbers

What the project has used and produced so far. Costs only in relative terms, not as totals.

Code and commits

2,285
Commits
2,240 by Claude, 45 by Jens
208,704
Lines of code
program code across game, tests and tooling
48 %
Share of tests in code
610
Test files
229 of 395
Packages done
started with 110 packages, the rest were added along the way
198
Decisions in the log
about a third of them by Jens

Commits per day

13 to 16 October: weekly quota used up, no shifts.

020040060080006.08.10.11.12.13.14.15.16.17.18.19.20.21.
ClaudeJens

Packages finished per day

02040608006.08.10.11.12.13.14.15.16.17.18.19.20.21.

Human and machine

Jens' time is estimated from the timestamps of his chat messages, plus tests on the iPhone that happened outside the chats. Answers given via multiple-choice prompts carry no timestamp – the real time is probably a bit higher.

40 h
Jens' time
estimated
143 h
AI shift time
often with several agents running in parallel
73
Chat sessions
across five project chats
291
Messages from Jens
62 %
Ideas from Jens
46 of 74 ideas from the chats
35
“Needs Jens” items
points where the AI couldn't continue on its own

Jens' time per day (estimated, minutes)

010020030040004.06.08.10.11.12.13.14.16.17.18.19.20.21.22.

Shifts

66
Shifts
57 with work, 7 empty, 2 aborted
143 h
Shift time
150 min
Average per shift
254 min
Longest shift
17 October, 23:02 – a single package integrated
296
Sub-agents recorded
23,164
Tool calls recorded

Shift minutes per day

05001,0001,5002,00006.08.10.11.12.13.14.15.16.17.18.19.20.21.

When commits happen

All commits by weekday and hour (Berlin). Only just over two weeks of data so far – the quiet midweek is the break from 13 to 16 October.

MonTueWedThuFriSatSun036912151821

Tokens

Counting only started on 17 October, when the shift collector went live. For the time before, the value is estimated from shift duration.

3.6 bn
Tokens measured
since 17 October
2.1 bn
Tokens before
estimated
95 %
Read from cache
94 %
Share on Opus
33 M
Tokens per finished package
39 M before the rebuild, target 25 M
2–3 %
Weekly quota per shift

Tokens per role (measured)

In millions. The engine teams use the most because they touch rendering, simulation and input at once.

05001,0001,5002,000engineLeitergrafiktesterkampfarchitekt

Art and audio

718
Image requests
301 drafts, 279 finals, 89 corrections, 49 research
32
Atlases in the game
385
Cells checked by the style guard
37 discarded, 337 in the game
5
Failed walk tests
13
Voices picked
129 + 8
Sounds and music tracks

Costs in proportion

The Claude subscription is flat-rate and covers every shift. Reviews via API credit, images and audio run through their own budgets with their own limits. Absolute amounts are deliberately left out.

15 %
Art budget used
46 %
Monthly audio limit used
62 %
Share for API credit
of spending besides the shifts, excluding audio
38 %
Share for images
of spending besides the shifts, excluding audio
31 %
Biggest image item
liveliness and art upgrade before gate 2
23 %
Walk cycle search
of image costs went into hero sheets and the walk

All values come from the repository (commits, shift statistics, cost ledger). Anything that could only be derived is marked as estimated.