Daedalus Fleet
Recon 2026-08-01
9 hosts surveyed
Field Report — Code Flow

Your code flows
forward backwards.

GitHub is supposed to be the original that every machine copies from. Right now it's a photo album of things that already happened somewhere else — and on eight of your nine boxes, it isn't even that.

01

What's happening vs. what should

Same pieces, opposite direction. That single reversal is the root of every problem in this report.

● Today
Edit where it runs
Change made on the boxlive prod
runs immediately
Only copy in existenceno history
sometimes, later
GitHubweeks stale

No undo. No record of what changed or why. Every box holds a slightly different version and none of them is "the" version. Lose the drive, lose the work.

● The model
Edit once, pull everywhere
You change one thingone place
commit
GitHub — the originalfull history
machines pull
VPS 1-5
DSF island
Talos / Nami
Cerberus

One original, nine identical copies. Every change dated, labelled, reversible. A box dies — you rebuild it from GitHub in minutes.

02

The scoreboard

Live survey of all nine machines. Green means the box pulls its code from GitHub. Red means the code lives there and only there.

Machine
Status
Reality on the box
VPS1 · web
Pulls
Checks GitHub every 5 minutes, rebuilds itself. 2,328 successful deploys. This one is already right.
VPS2 · db
No git
JobHinge scraper, parser and database. Hand-placed files. Exists nowhere else on earth.
VPS3 · Luna.DB
Dormant
Old repo, everything switched off. Not pulling.
VPS4 · scout
No git
Files copied in by hand. Ownership fingerprint proves rsync, not a clone.
VPS5 · checker
No git
Same. Runs Helios and n8n from files with no origin.
DSF island
Orphaned
Three real git repos — none connected to GitHub. Islands that link to nothing.
Talos
Drifted
Repos exist but 155 changed files never sent up. Local tree three months stale.
Nanami
No git
Runs lottery forecasts on a timer from a folder that isn't a repo at all.
Cerberus
No git
Fleet-watch every 5 min, galileo 4x/day — all from untracked folders.
Acemagic
419 dirty
Your supposed source of truth, carrying 419 uncommitted files and 11 projects with no git at all.
1
of 9 boxes
pulls from GitHub
419
uncommitted files
on Acemagic
750MB
active work
in no repo at all
0
automated deploys
outside VPS1
03

How a machine knows to pull

This is the part that sounds like magic and isn't. The machine simply asks, on a schedule. Nothing has to reach into your house — the box dials out, which is why this works behind your home router with no tunnel and no open ports.

Every 5 minutes

It asks

The box contacts GitHub: "is there anything newer than what I have?" Costs nothing, takes a second.

If yes

It pulls

Downloads only what changed. If the answer was no, it goes back to sleep and asks again in five minutes.

Then

It restarts itself

Rebuilds and reloads the service so the new code is actually running. No one touches the box.

VPS1 has been doing exactly this, unattended, 2,328 times. The pattern is already proven on your own hardware — it just needs copying onto the other eight.

04

Sandbox, then prod

A branch is a complete parallel copy of everything. You change it freely and production doesn't move an inch. Merging is the moment it goes live — and it's reversible.

1

Sandbox branch

A full parallel copy. Break whatever you want. Prod keeps running the old code, untouched.

2

Staging box

One machine pulls the sandbox branch instead of main. You click around at a test address and see it working for real.

3

The gate

GitHub runs your tests automatically and shows you every changed line. Nothing merges unless it passes.

4

Merge to main

One click. Within five minutes every machine has pulled it. If it's wrong, you roll back to yesterday in seconds.

05

What you actually get

Eight things that turn on the moment this is wired up. Each card shows what today costs you.

01

Nothing is ever lost

Every file on every machine has a second copy off-site, updated continuously.

Today: one dead drive erases 750MB of active work.
02

A time machine

Every version of every file, forever. Go back to how it was last Tuesday in one command.

Today: a bad edit is permanent. There's nothing to go back to.
03

One real version

You stop asking "which box has the current copy?" GitHub is the answer, always.

Today: nine boxes, nine slightly different versions.
04

Safe to experiment

Try the risky idea on a branch. If it's bad, delete the branch. Prod never knew.

Today: experimenting means editing the live thing.
05

Deploys happen without you

Commit once, nine machines update themselves within minutes.

Today: hand-copying files box to box.
06

A record of why

Every change stamped with what, when, and the reason. Six months later it still explains itself.

Today: no way to know why anything is the way it is.
07

Rebuild any box from scratch

A server dies at 2am. Spin up a fresh one, pull, done — instead of reconstructing from memory.

Today: VPS2 dying means rebuilding JobHinge by hand.
08

Work from anywhere

Change things from your phone. Push to GitHub. The fleet updates itself while you're out.

Today: you have to be at the machine.

What's exposed right now

Not theoretical. This is the state of the fleet as of today's survey.

  • VPS2 dies → JobHinge's scraper and parser are gone. They exist on that one box and nowhere else.
  • Acemagic's drive dies → 750MB of active work is gone. Lead-gen, the n8n library, Council, Aesop, the trending watcher you touched today.
  • Kepler has 37 commits that only exist on this machine. Never sent to GitHub.
  • jobhinge-failover runs in production on two machines and has never been in git anywhere, on any box.
  • The DSF island's three repos point at nothing. All that work is one server away from vanishing.