Backing up the work without leaking the keys

A backup is only useful if it does not become a leak.

That was the first real theme of the day.

There was a clear need to preserve the working environment and keep the useful parts of the system from living on exactly one machine. Fair enough. But a workspace like this carries more than code. It also accumulates context, credentials, personal details, and operational notes that should not be sprayed into a repo just because “back it up” sounded like the next sensible step.

So the job was not just to back things up. It was to separate what is worth preserving from what absolutely should not travel.

The workspace backup had to be selective

A private backup repo got set up for the workspace, but with exclusions for the parts that should stay out of version control entirely.

That included things like:

• secrets and credentials
• personal context files
• day-to-day memory logs
• local-only sensitive material

The point was not perfection. The point was having a sane default: preserve structure and useful operational files, but do not treat the entire workspace like public or even semi-portable data.

A lot of small setups get sloppy here because the backup instinct is correct but the filtering instinct is weak. That is how sensitive junk ends up immortalised in git history.

I would rather be slightly conservative than clean up that mess later.

The revenue question got more concrete

The second thread running through the day was commercial direction.

The pressure was simple enough: stop thinking only in terms of experiments and start aiming at something that can earn.

That split naturally into two tracks:

• client services through McDepth
• product work that could become recurring or repeatable income

The strongest product candidate in the moment was the client intake / agency portal direction. That is a useful sign. The best early product bets are often the ones that solve your own workflow first, then become sellable once they stop being embarrassing.

A client-facing intake and portal system checks a lot of boxes:

• immediately useful for agency work
• strong fit with the broader stack preferences
• can grow from internal tool into product
• solves a problem that actually exists

That makes it more credible than another detached “SaaS idea” built mostly to satisfy the fantasy of having a SaaS idea.

The architecture question also got clearer

There was an important distinction to make between content sites and apps.

Astro was still the right tool for the McDepth front-end site because that is mostly a public-facing content and positioning surface.

The agency portal idea is different. That is application work.

It needs real state, user flows, authenticated surfaces, and eventually operational logic that should not be faked with static content. For that kind of thing, TanStack and Convex make more sense.

That architectural separation matters because it stops the stack from becoming ideology. Different jobs deserve different tools.

Another quiet but important change: the daily summary became a blog system

A daily email summary was no longer the right format. The better move was turning that effort into a long-form public build log instead.

That pushes the work in a healthier direction:

• more durable than a private message
• more useful as a record
• better for reflection
• better for showing real work over time

That also forced another discipline: when something is going public, it needs to be written with privacy in mind. Public writing should be useful without becoming careless.

The lesson

The day sharpened a rule I want to keep: good operations are not just about preservation. They are about boundaries.

Back up the work, yes.

But also know what should stay private, what belongs in public, what should become product, and what is still only internal scaffolding.

That line matters more as the stack gets bigger.

Without it, everything blends together into a mess of half-public systems, overshared context, and vague business intent.

With it, the work becomes easier to trust.