Check it yourself.
Every claim on this site rests on a base you can check: public code, timestamped test logs, and a build you can replay commit by commit. The bigger numbers elsewhere are maths on top of this, and each says so.
THE PROOF WALL
Every number is a star. Open one to see what it is and where it came from.
Tap a label to filter. A label with a zero has nothing to show yet, and we leave it that way.
MEASURED. Counted or timed on our own system, on the date shown.
Source. Worker fleet job ledger
Date. 2026-10-09
Method. 251 finished
MEASURED. Counted or timed on our own system, on the date shown.
Source. Worker fleet job ledger
Date. 2026-10-09
Method. Read from the source on the date shown.
MEASURED. Counted or timed on our own system, on the date shown.
Source. Worker fleet job ledger
Date. 2026-10-09
Method. Read from the source on the date shown.
MEASURED. Counted or timed on our own system, on the date shown.
Source. Token usage meter and the job ledger
Date. 2026-10-09
Method. orchestrator spend dominates; US$/job is total spend / fleet jobs, an upper bound
MEASURED. Counted or timed on our own system, on the date shown.
Source. Token usage meter and the job ledger
Date. 2026-10-09
Method. orchestrator spend dominates; US$/job is total spend / fleet jobs, an upper bound
MEASURED. Counted or timed on our own system, on the date shown.
Source. Session start audit (tokens read before the first action)
Date. 2026-10-09
Method. Read from the source on the date shown.
MEASURED. Counted or timed on our own system, on the date shown.
Source. Linux memory counters on the build machine
Date. 2026-10-08
Method. Read from the source on the date shown.
MEASURED. Counted or timed on our own system, on the date shown.
Source. Windows process memory counter
Date. 2026-10-08
Method. Read from the source on the date shown.
MEASURED. Counted or timed on our own system, on the date shown.
Source. Windows memory counter
Date. 2026-10-08
Method. Read from the source on the date shown.
MEASURED. Counted or timed on our own system, on the date shown.
Source. Linux load average counter
Date. 2026-10-08
Method. 8 cores
MEASURED. Counted or timed on our own system, on the date shown.
Source. Scheduled job run record (every run is logged)
Date. 2026-10-08
Method. 4686 runs
MEASURED. Counted or timed on our own system, on the date shown.
Source. Scheduled job run record (every run is logged)
Date. 2026-10-08
Method. All scheduled jobs in 24 hours now, against one failing snapshot job before. Different bases, shown separately.
MEASURED. Counted or timed on our own system, on the date shown.
Source. Gate run record
Date. 2026-10-08
Method. Recorded by the gate runner at the time shown in the run record.
MEASURED. Counted or timed on our own system, on the date shown.
Source. Backup log
Date. 2026-10-08
Method. Read from the source on the date shown.
UNKNOWN. We have not measured this yet, so we do not show a number.
Source. No durable run record exists yet
Date. 2026-10-08
Method. That suite does not log its own run time, so we show no figure. A recorder is the next fix.
MEASURED. Counted or timed on our own system, on the date shown.
Source. Kernel compression gate record, 2026-10-08; Sigil ceiling report, 2026-10-08
Date. 2026-10-08
Method. 25 matched jobs: same task text as earlier runs on the full rule book, re-run on the compact core. Cost is API-equivalent token pricing, not a bill.
MODELLED. Computed from measured inputs under stated assumptions. A scenario, not a result.
Source. Sigil ceiling report, 2026-10-08
Date. 2026-10-08
Method. The same measured token counts repriced under five layers of change. The last layer needs a decision not yet taken. A scenario, not a forecast.
MEASURED. Counted or timed on our own system, on the date shown.
Source. Kernel compression gate record, 2026-10-08
Date. 2026-10-08
Method. 50 questions about the rule book, read-only tools, judged. Before is the full rule book, after is the compact core. Bar was 48 or better.
MEASURED. Counted or timed on our own system, on the date shown.
Source. Kernel compression gate record, 2026-10-08
Date. 2026-10-08
Method. Questions written by an independent agent that never saw the compact core.
MEASURED. Counted or timed on our own system, on the date shown.
Source. Kernel compression gate record, 2026-10-08
Date. 2026-10-08
Method. Same 25 tasks. Two full-rule-book results claimed clean lint on files that fail lint. The compact core had none.
MEASURED. Counted or timed on our own system, on the date shown.
Source. PC optimisation report, 2026-10-08
Date. 2026-10-08
Method. Run record average before, a timed run after. The secret check now scans only files that could be committed.
MEASURED. Counted or timed on our own system, on the date shown.
Source. PC optimisation report, 2026-10-08
Date. 2026-10-08
Method. Before: 30 to 45 minutes a run (midpoint shown, 37.5 minutes). After: 14 seconds, incremental with a daily full pass.
MEASURED. Counted or timed on our own system, on the date shown.
Source. PC optimisation report, 2026-10-08
Date. 2026-10-08
Method. One bundle per repository keyed on content, instead of re-bundling about 12 worktrees.
MEASURED. Counted or timed on our own system, on the date shown.
Source. Round-2 brief, 2026-10-08 (raw log attaching)
Date. 2026-10-08
Method. Size of the repository directory before and after history and run-state cleanup.
MEASURED. Counted or timed on our own system, on the date shown.
Source. Round-2 brief, 2026-10-08 (raw log attaching)
Date. 2026-10-08
Method. Wall-clock seconds for the full suite, before and after.
RECORDED
250 of 250
trading app tests passed
RECORDED
83 of 83
SpoofGuard tests passed
RECORDED
57
saved versions replayed, one by one
RECORDED
3.5 hours
idea to a working, tested app
RECORDED
250 of 250 tests, 57 saved versions
the trading app, replayed commit by commit
Like watching a builder's time-lapse where the inspector signs off every single frame, not just the finished house.
Where this number comes from
Method: we checked out each saved version in order and ran its tests, and the command is shown below so you can repeat it. Nothing was touched except our own code, no one is replaced by any of it, and it costs you nothing to look.
MEASURED
6,314 test cases
in the main TITANOS engine, plus automated safety suites on the control layer
A pilot's pre-flight checklist, run before every take-off instead of once.
Where this number comes from
That is the headline. What did the replay find that we did not expect?
1. Public code, every change timestamped
- This website's source, every change dated to the second: github.com/kyle4814/titanos-next
- SpoofGuard, our free open-source email security tool: github.com/kyle4814/spoofguard
2. Full test runs, with timestamps
Raw logs, each line stamped with the time it ran (7 October 2026, Brisbane time). No summary written afterwards, just the output:
3. The 3.5-hour build, replayed commit by commit
Think of a builder's time-lapse. We checked out every one of the 57 saved versions of the trading app in order and ran its tests. Each line of the replay shows when that version was saved, its ID, and how many tests passed. Tests grew from 20 at 05:41 to 250 at 09:37.
Here is the part we did not expect, shown as found: five in-between versions each have one test that fails when replayed today. Those tests depend on the time of day or on live market data (for example, a session-clock label and a live one-tap trade). Every one passes at the final version, 250 of 250.
How the replay was run, so anyone with the repository can repeat it:
for c in $(git rev-list --reverse HEAD); do
git checkout -q "$c"
echo "$(git log -1 --format=%ci) $(node --test tests/*.test.js | grep '^# pass')"
doneWhat this proves, and what it does not
It proves the base: real code, real tests, real timestamps, and the speed of that build. Our larger figures (the compounding, the second account, the engineer-equivalents) are calculations built on this base, and each is labelled as a model. They become facts only as weekly results confirm them, and we publish those results too.
Would it be okay if I showed you what a build like this could look like for your business? A no is welcome.