Vol. 01 / Manila

Things That Didn't Work

What I tried. What broke. What I would do now. Placeholder cards are labeled so they are not mistaken for client postmortems.

Architecture

Essay / 2026

Treating a shared ledger like an offline notebook

What I tried: I wrote about applying offline-first habits from a personal boxing app to a vehicle marketplace.
Why I thought it would work: Local-first feels like the same 'it should work on a bad radio' problem. Gloves Up really does live on-device.
What actually happened: A marketplace is a shared ledger. Availability is not a private notebook. The public essay is the record: pretending those are the same architecture is how you double-book a vehicle.
What broke: The idea, not a production outage. The failure was conceptual — and it is the useful part.
What I learned: Closed personal datasets and shared booking state are different products. Sync is a product decision, not a library.
What I would do now: I keep Gloves Up offline-first. RentaraH stays a networked ledger. I would not sell 'local-first' as a default for inventory.

Abandoned · Placeholder

Placeholder

What I tried: [Replace] I started a clean-room rewrite of a working service because the old module names embarrassed me.
Why I thought it would work: [Replace] I thought a weekend of greenfield would be cheaper than another year of seams.
What actually happened: [Replace] The rewrite tracked the old bugs plus new ones, and the working service kept taking the real traffic.
What broke: [Replace] Time, morale, and a branch nobody could review.
What I learned: [Replace] Rewrite the seam that hurts. Leave the rest until a metric forces it.
What I would do now: [Replace] I would extract one interface, put tests on it, and ship that. This card is a template — not a claim about a named employer.

Removed · Placeholder

Placeholder

What I tried: [Replace] We shipped a preference surface that looked complete in staging.
Why I thought it would work: [Replace] Users had asked for control. The mockups were clear.
What actually happened: [Replace] Support load went up. Almost nobody changed the default. The extra path bit a billing edge case.
What broke: [Replace] An invariant that only existed because of the new screen.
What I learned: [Replace] A setting is a commitment to test every combination. Defaults are a product.
What I would do now: [Replace] I would instrument the old path, add one control, and delete the rest. Marked placeholder so it is not mistaken for a client postmortem.

Written by Al Beltran