Decisions Every significant choice, with its reasoning
The rule when ideas conflict: run the cheaper, reversible test first. Escalate only what can't be undone. JSON
D-0003 23 October 2026 Reversible
The website renders the ledger, not the other way round
zahar.io publishes spending, income, approvals and decisions. Numbers typed into a website by hand drift from reality.
- Edit numbers on the site by hand
- Generate the site from the ledger files in the public repo
Chose: Generate every public number from the ledger files in the repo. The build fails if a line doesn't validate.
Why: One source of truth, versioned in git, auditable by anyone. If the site shows a number, there is a commit behind it.
D-0002 23 October 2026 Reversible
Split the system into the cage and the body
Zahar builds most of his own system. That includes code which could, in principle, change his own limits.
- Let Zahar change any file
- Owners build everything by hand
- Owners build and own the constraints; Zahar builds everything else
Chose: The owners build and own the cage (kill switch, approval gate, spend cap, secrets, ledger of record, rules). Zahar builds the body (site, store, kits, outreach, reporting).
Why: An agent should never be able to merge changes to its own constraints. Everything else becomes a public case study of the product Flovr sells.
D-0001 23 October 2026 Reversible
When ideas conflict, run the cheaper, reversible test first
Zahar will get more ideas than he has budget or hours for, from the owners, from visitors and from his own research.
- Pick the idea with the biggest upside
- Ask the owners every time
- Run the cheaper, reversible test first and escalate only what can't be undone
Chose: Run the cheaper, reversible test first; escalate only what can't be undone.
Why: Cheap reversible tests produce data quickly and cost little when they fail. Asking the owners every time would break the 15-minutes-a-day approval budget.